Comparative Studies in Germanic Syntax (Linguistik Aktuell 97)

Comparative Studies in Germanic Syntax

<DOCINFO AUTHOR ""TITLE "Comparative Studies in Germanic Syntax: From A(frikaans) to Z(urich German)"SUBJECT "Linguistik Aktuell/Linguistics Today, Volume 97"KEYWORDS ""SIZE HEIGHT "240"WIDTH "160"VOFFSET "4">

Linguistik Aktuell/Linguistics Today

Linguistik Aktuell/Linguistics Today (LA) provides a platform for original monograph studiesinto synchronic and diachronic linguistics. Studies in LA confront empirical and theoreticalproblems as these are currently discussed in syntax, semantics, morphology, phonology, andsystematic pragmatics with the aim to establish robust empirical generalizations within auniversalistic perspective.

Series Editors

Werner AbrahamUniversity of Vienna

Elly van GelderenArizona State University

Advisory Editorial Board

Cedric BoeckxHarvard University

Guglielmo CinqueUniversity of Venice

Günther GrewendorfJ.W. Goethe-University, Frankfurt

Liliane HaegemanUniversity of Lille, France

Hubert HaiderUniversity of Salzburg

Christer PlatzackUniversity of Lund

Ian RobertsCambridge University

Ken SafirRutgers University, New Brunswick NJ

Lisa deMena TravisMcGill University

Sten ViknerUniversity of Aarhus

C. Jan-Wouter ZwartUniversity of Groningen

Volume 97

Comparative Studies in Germanic Syntax: From Afrikaansto Zurich GermanEdited by Jutta M. Hartmann and László Molnárfi

Comparative Studiesin Germanic SyntaxFrom Afrikaans to Zurich German

Edited by

Jutta M. Hartmann

László MolnárfiTilburg University

John Benjamins Publishing Company

Amsterdam�/�Philadelphia

The paper used in this publication meets the minimum requirements8 TM

of American National Standard for Information Sciences – Permanenceof Paper for Printed Library Materials, ansi z39.48-1984.

Library of Congress Cataloging-in-Publication Data

Workshop on Comparative Germanic Syntax (20th ; Tilburg, Netherlands)Comparative studies in Germanic syntax : from Afrikaans to Zurich German /edited by Jutta M. Hartmann, László Molnárfi.

p. cm. (Linguistik Aktuell/Linguistics Today, issn 0166–0829 ; v. 97)Selected papers presented at the 20th Comparative Germanic Syntax

Workshop held in June, 2005, in Tilburg.1. Germanic languages--Syntax--Congresses. I. Hartmann, Jutta. II.

Molnárfi, László, 1971-

PD361 .W67 2005435--dc22 2006050841isbn 90 272 3361 6 (Hb; alk. paper)

© 2006 – John Benjamins B.V.No part of this book may be reproduced in any form, by print, photoprint, microfilm, orany other means, without written permission from the publisher.

John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The NetherlandsJohn Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa

JB[v.20020404] Prn:25/09/2006; 10:39 F: LA97CO.tex / p.1 (47-96)

Table of contents

From Afrikaans to Zurich German: Comparative studiesin Germanic syntax 1

Jutta M. Hartmann & László Molnárfi

I. Studies on predication

The Nom/Acc alternation in Germanic 13Halldór Ármann Sigurðsson

Shape conservation, Holmberg’s generalization and predication 51Olaf Koeneman

Quirky verb-second in Afrikaans: Complex predicates and head movement 89Mark de Vos

Nominal arguments and nominal predicates 115Marit Julien

II. Studies on the (pro)nominal system

Pronominal noun phrases, number specifications, and null nouns 143Dorian Roehrs

Toward a syntactic theory of number neutralisation:The Dutch pronouns je ‘you’ and ze ‘them’ 181

Gertjan Postma

Long relativization in Zurich German as resumptive prolepsis 201Martin Salzmann

III. Historical studies

Auxiliary selection and counterfactuality in the history of Englishand Germanic 237

Thomas McFadden and Artemis Alexiadou

JB[v.20020404] Prn:25/09/2006; 10:39 F: LA97CO.tex / p.2 (96-105)

Table of contents

The loss of residual “head-final” orders and remnant fronting in LateMiddle English: Causes and consequences 263

Mary Theresa Biberauer and Ian Gareth Roberts

Syntactic sources of word-formation processes: Evidence fromOld English and Old High German 299

Carola Trips

Index 329

JB[v.20020404] Prn:22/09/2006; 8:33 F: LA97IN.tex / p.1 (46-118)

From Afrikaans to Zurich German

Comparative studies in Germanic syntax

Jutta M. Hartmann & László MolnárfiTilburg University

The present volume contains a selection of papers presented at the 20th Com-parative Germanic Syntax Workshop in Tilburg, June 2005. While – following atradition of earlier CGSW-proceedings – the contributions cover a wide range ofGermanic languages as well as a wide range of current topics in modern syntactictheory, the selection also shows a strong comparative commitment. Such commit-ment might seem evident. Indeed, the relevance of the comparative methodologyfor modern syntax, and more specifically for a theory of UG, can hardly be dis-puted. To some extent, syntactic theorizing is meaningless without observing,describing, comparing (and, in the ideal case, explaining) varieties or variationsof a specific language phenomenon occurring cross-linguistically or in differenthistorical stages of a given language. The aim is to find and to control the cross-linguistically relevant contrasts that do not go back to external factors, but can beexplained as reflexes of the same difference in the grammar of the given languages,contributing to our better understanding of the architecture of UG.

Yet, the editors of the present volume feel (and, as we believe, this sentimentis shared by many) that the comparative aspect of CGSW has somewhat weakenedin recent years and needs to be addressed in a proper way. Only if comparativestudies offer more than just non-systematic side-glances to other languages canimportant generalizations be captured and real explanatory power achieved (cf.especially Haider 1993 or Abraham 1995 in this regard). For that reason we wantedto take the notion “comparative” and “Germanic” in the title of the Workshopseriously, and asked for contributions that address at least two Germanic languages(or different diachronic stages of the same Germanic language) in depth, or discussa specific grammatical phenomenon of a given language in the overall Germanicperspective.

Heeding this truly comparative perspective, the essays in this volume celebratevariety both with respect to the languages investigated and the topics addressed.



The editors particularly welcome that besides the “usual suspects” (i.e. Englishand German) and recurring guests (i.e. the Scandinavia) of the CGSW-series,we could include here studies on lesser-investigated languages such as modernAfrikaans and Zurich German. The volume has also benefited from a strong his-torical component, including studies on linguistic aspects of various diachronicstages of English and German. Here, the emphasis often lies on intralinguistic,rather than cross-linguistic variety, the methodology of comparison facing partic-ular challenges in terms of adequate collection and evaluation of historical data(see particularly McFadden & Alexiadou’s contribution in this respect).

While covering a wide range of current issues in linguistics, the present col-lection of essays can be subsumed under three major thematic headings. The firstpart of the volume contains comparative studies on predication in Germanic, ad-dressing issues such as case dependency in the domain of predication (Sigurðsson),constraints of movement preserving or distorting thematic relations (Koeneman)or a “quirky” case of complex predicate formation in Afrikaans (de Vos). Thesecond part of the volume contains papers on the (pro)nominal domain in Ger-manic, including studies on the licensing conditions of pronominal noun phrases(Roehrs), number neutralization in the Dutch pronominal system (Postma) andresumptive pronouns in Zurich German (Salzmann). The last part of the vol-ume looks at Germanic syntax from a diachronic perspective, taking up on issuessuch as auxiliary selection in the history of English and, more generally, in Ger-manic (McFadden & Alexiadou), remnant fronting in Middle English (Biberauer& Roberts) and syntactic sources of word-formation processes in Old English andOld High German (Trips). Thus, the volume presents a wide range of studies thatenrich both theoretical understanding and empirical foundation of comparativeresearch on the Germanic languages.

“The Nom/Acc alternation in Germanic” by Halldór Ármann Sigurðsson de-scribes the distribution of accusative case and discusses the nature of the nomi-native/accusative distinction in the standard Germanic languages. In addition, itillustrates and discusses the well-known fact that inherent accusatives and certainother types of accusatives do not behave in accordance with Bruzio’s Generaliza-tion. In spite of these Non-Burzionian accusatives, there is a general dependencyrelation between the so-called ‘structural’ cases, Nom and Acc, here referred toas the relational cases, such that relational Acc is licensed only in the presence ofNom (as has been argued by many). This relation is here referred to as the SiblingCorrelation, SC. Contrary to common belief, however, SC is not a structural cor-relation, but a simple morphological one, such that Nom is the first, independentcase, CASE1 (‘an only child’ or an ‘older sibling’, as it were), whereas Acc is thesecond, dependent case, CASE 2, serving the sole purpose of being distinct fromNom – the Nom-Acc distinction, in turn, being a morphological interpretationor translation of syntactic structure. It has been an unresolved (and largely a ne-



glected) problem that the Germanic languages split with respect to case-markingof predicative DPs: nominative versus accusative (It is I/me, etc.). However, themorphological approach to the relational cases argued for in this paper offers asolution to this riddle: The predicative Acc languages have extended the domainof the Sibling Correlation, such that it applies not only to arguments but also toadjacent DPs in general. That is, the English type of predicative Acc is not ‘default’,nor is it caused by ‘grammatical viruses’, but a well-behaved subtype of relationalAcc. The central conclusion of the paper is that one needs to abandon the struc-tural approach to the relational cases in favor of a more traditional morphologicalunderstanding. However, this is not a conservative but a radical move. It requiresthat we understand morphology (and PF in general) not as a direct reflection ofsyntax but as a translation of syntax into an understandable but foreign code or‘language’, the language of morphology. Nom and Acc are not syntactic featuresbut morphological translations of syntactic correlations. It is thus no wonder thatthey are uninterpretable to the semantic interface.

Olaf Koeneman’s “Shape conservation, Holmberg’s generalization and predi-cation” builds on all the previous approaches to Shape Conservation and tries tosolve some problems that arise with them (in particular, related to A-movementin passive constructions). It is argued that Holmberg’s Generalization is a syntac-tic and not a phonological phenomenon. This view allows the author to generalizeover a larger set of facts in the following way: Within the thematic domain, it is im-possible to invert the relationships of thematic categories, i.e. categories assigningor receiving a θ-role. The reason is that the grammar wants the interface interpre-tations at LF and PF to be uniform. It is shown that notorious counterexamplesto thematic isomorphism, such as passivization and short verb movement, can bedealt with in a unified way by making reference to predication theory.

Mark de Vos’ “Quirky verb-second in Afrikaans: complex predicates and headmovement“ discusses a special case of complex predicate formation in modernAfrikaans. The central aim of the paper is to give a novel account for ‘Quirky VerbSecond’, a peculiar construction in Afrikaans which optionally pied-pipes a coordi-nated verbal cluster to verb-second position. Afrikaans is a verb-second languagethat also allows the formation of a coordinated verb cluster: [POSTURE VERB][AND] [LEXICAL VERB]. The construction is putatively pseudo-coordinative incharacter and typically occurs with aspectual verbs of posture. Either the postureverb may undergo verb-second individually or, alternatively, the entire coordi-nated verbal complex may undergo verb second. This construction is puzzlingon a number of grounds. If verb second includes head movement from at V toat least T (Den Besten 1989), then the optional pied-piping of a phrase-like el-ement is puzzling. However, if verb-second involves phrasal remnant movement(Müller 2004), then the optional ability of the posture verb to be extracted from acoordinate structure (in violation of the Coordinate Structure Constraint (Ross



1967)) is equally puzzling. This dilemma places this construction in a uniqueposition of being able to distinguish between these two opposing views of verb-second. The paper proceeds by outlining the properties of the pied-piped vs. thenon-pied-piped construction. It is demonstrated that the pied-piped coordinatedconstituent is indeed a verbal head. It is also shown that the base, non-pied-pipedstructure is phrasal. A variety of tests are used to provide converging evidence forthese claims. Crucial evidence from separable particle placement is used to demon-strate that a remnant-movement analysis would be untenable. The analysis iscouched in terms of ‘true’ coordination – in other words, the pseudo-coordinativecharacter of the construction is derived from the properties of the phrase struc-ture itself rather than the properties of the coordinator. Coordination is argued toscope over individual aspectual features within the verbal cluster itself. This meansthat under certain special conditions, individual phonological features are notwithin the scope of coordination, allowing them to undergo verb-second withoutviolating the Coordinate Structure Constraint. Thus, it is argued that verbal head-movement may indeed be phonological feature movement (Boeckx & Stjepanovic2001; Chomsky 2000; Zwart 1997), but with the added caveat that it can also betrue syntactic movement in certain instances. The proposal has implications fortheories of head movement, excorporation and coordination.

In “Nominal arguments and nominal predicates”, Marit Julien argues that theclaim that predicative nominal phrases are structurally smaller than argumentalnominal phrases is not corroborated by Scandinavian. For one thing, singularnominals without determiners, which are structurally smaller than DPs, can bepredicates or arguments. Even more strikingly, it appears that full DPs, and evenlarger phrases, can also be predicates as well as arguments in Scandinavian. Toshow this, Julien first sets out to identify a number of predicate tests, and thenapplies these tests to Scandinavian nominal phrases of various sizes. The conclu-sion is that DPs can clearly be predicates, and so can phrases where a universalquantifier has a DP as its complement. Hence, the difference between nominal ar-guments and nominal predicates cannot be linked to the presence or absence ofa D-projection. Nominal phrases containing demonstratives are however not ac-ceptable as predicates. The reason might well be a purely semantic one, having todo with the deictic content of the demonstrative. The conclusion will be that thecontrast between nominal arguments and nominal predicates is not structural butsemantic. If the lexical content of a nominal phrase is such that the phrase can geta purely intentional interpretation, the phrase can be a predicate, but if its lexicalcontent requires an extensional reading, the phrase is necessarily referential andcannot be used to predicate.

Dorian Roehrs’ “Pronominal noun phrases, number specifications, and nullnouns” deals with the licensing conditions of pronominal nouns phrases in Ger-manic. According to standard assumptions, the determiner and the head noun



in the DP exhibit morphological agreement. Adopting the Postalian view, Roehrsstarts with the observation that pronominal determiners require semantic, ratherthan morphological, agreement. Concentrating on number, he demonstrates indetail that these standard assumptions are not only too weak, allowing ungram-matical cases such as *du verdammtes Pack ‘you(SG) damn gang’, but also toostrong, disallowing grammatical examples such as ihr verdammtes Pack ‘you(PL)damn gang’. In order to provide a uniform and homogenous account of the de-terminer system, he proposes that pronominal determiners must agree with theirhead noun not only semantically but also morphologically. Morphologically dis-agreeing nouns are argued to be in a Specifier position and the head of theextended noun phrase hosting that Specifier is a null noun. Specifically, Roehrsproposes that both regular and pronominal determiners are the same with regardto morphological agreement; however, they differ in their semantic denotationsand syntactic selectional features: while regular determiners are defined as general“totality extractors” and may select AgrP and NP, pronominal ones “pick out” sin-gle or multi-member sets and select not only AgrP and NP but also the phrase withthe dis-agreeing Specifier. He concludes that regular determiners are less specifiedpronominal determiners. More generally, arguing that semantic number is partof the semantics, he proposes that morphological and semantic numbers are tobe dissociated from one another. Another consequence of the discussion is thatthe inventory of null nouns is extended from null countable and mass nouns (cf.Panagiotidis 2002) to collective nouns, pluralia tantum, and proper names.

Martin Salzmann’s “Long relativization in Zurich German as resumptive pro-lepsis“ addresses the issue that, standing out among Germanic languages, ZurichGerman (ZG) employs resumptive pronouns in relativization. There is an in-triguing asymmetry in the distribution of resumptives: while resumptives arelimited to oblique positions in local relativization, they appear across the boardin long-distance relativization. This suggests that there is a fundamental differ-ence between the two constructions. The paper reanalyzes a previous approachby van Riemsdijk where long-distance relativization in ZG is re-interpreted as lo-cal aboutness relativization plus binding. The construction can be shown to haveparadoxical properties: On the one hand, there is reconstruction into the positionoccupied by the resumptive pronoun, on the other hand, the complement clauseturns out to be an island for extraction. This paradox is resolved by assuming atough-movement style analysis: Operator movement in the complement clause de-rives a predicate and licenses an extra argument in the matrix clause, the prolepticobject. This in turn is A’-moved and deleted under identity with the external head.This predication analysis makes an alternative reconstruction strategy available asin tough-movement and accounts for the opacity of the complement. The link be-tween the proleptic object and the operator in the complement clause is an ellipsisoperation. Together with concomitant Vehicle Change effects this nicely explains



the intricate Condition C pattern in both the proleptic construction and in tough-movement. The presence of a resumptive pronoun follows from a constraint thatrequires specific chains to be phonetically realized in ZG. The entire structure rep-resents what Salzmann calls “resumptive prolepsis”. On a more theoretical level,this approach suggests a straightforward way of handling exceptional and hithertoill-understood cases of reconstruction within a theory that makes crucial use of fullcopies of the antecedent. It unifies resumptive prolepsis with tough-movement incrucial respects and thereby provides a fresh look at the latter.

In Gertjan Postma’s “Toward a syntactic theory of feature neutralization”Kayne’s (2000) syntactic theory of feature neutralisation is adopted and adapted toaccount for two cases of number neutralisation in Dutch, as well as a correlationacross Germanic between the presence of number neutralisation in the nomina-tive paradigm and the type of V2 attested in those languages. The weak Dutchobject pronoun, oblique pronoun, and possessive pronoun je ‘you’ is both singu-lar and plural. In traditional terms: je exhibits number neutralization. However,this property of je is dependent on the syntactic context: only if je is bound, itcan be both singular and plural. If not, only the singular reading is retained. Toget a plural reading the use of the complex plural form jullie ‘you.PL’ is the onlyoption. One way to handle this theoretically is to assume two distinct forms jewith the same phonological matrix, an anaphoric pronoun je which has numberneutralisation, and a pronominal pronoun je which is singular. It is shown thatthis option leads to problems with the binding theory and needs various unattrac-tive ad hoc stipulations. Postma follows Kayne (2000), who shows that Italian séis part of the singular paradigm. Nevertheless, it can be used as a plural: it ac-quires plural readings by an abstract distributor, DIST, which occupies a syntacticslot and has syntactic properties, such as the requirement that it must be boundby a plural antecedent. Kayne’s theory can be considered as a syntactisation ofmorphological neutralisation. This theory is straightforwardly applicable to theDutch data listed above. It predicts a deep link between anaphoric behaviour andnumber neutralisation. The main objective of the paper is to apply Kayne’s theoryto a diachronic problem. The Middle-Dutch 3rd person pronoun hem ‘him’ dis-played number neutralisation (it could mean both ‘him’ and ‘them’), and couldbe used anaphorically (‘himselves’/’themselves’). Modern Dutch lost this prop-erty. Recent data (Postma 2004) show that the loss of number neutralisation inhem goes hand in hand with the loss of its anaphoric use. To fill the gap left byanaphoric hem, Dutch borrowed the reflexive zich from German border dialects.This newly acquired form, once again, has number neutralisation. This confirmsthe link between number neutralisation and anaphoricity, as suggested by Kayne.

In the paper “Auxiliary selection and counterfactuality in the history of En-glish and Germanic” by Thomas McFadden and Artemis Alexiadou, the retreatof be as perfect auxiliary is examined in a diachronic perspective. Corpus data



are presented showing that the initial advance of have was most closely connectedto a restriction against be in past counterfactuals. Other factors which have beenreported to favor the spread of have, are either dependent on the counterfactual ef-fect, or significantly weaker in comparison. It is argued that the effect can be tracedto the semantics of the be perfect, which denoted resultativity rather than ante-riority proper. Related data from other older Germanic and Romance languagesare presented, and finally implications for existing theories of auxiliary selectionstemming from the findings presented are discussed.

Theresa Biberauer’s and Ian Roberts’ “The loss of residual ‘head-final’ ordersand remnant fronting in Late Middle English: causes and consequences” is a fur-ther contribution to the ongoing discussion of the possible triggers of word orderchange in Middle English (ME). The primary empirical focus of the paper is theresidual “head-final” orders found in ME. The usual chronology for the generalchange from OV to VO in English situates it in Early ME (Canale 1978; van Ke-menade 1987; Lightfoot 1991; Roberts 1997; Kroch & Taylor 1994; Fischer et al.2000), but as various authors have pointed out, orders which are indicative ofsome kind of persisting OV grammar are found, albeit at rather low frequencyand somewhat disguised by other factors, until the 15th century (see Fischer et al.2000:177 for a summary and references). Here the authors will propose an analysisof these orders which supports the novel proposal in Biberauer & Roberts (2005;henceforth: B&R) that the loss of residual “head-final” orders is related to the in-troduction of obligatory clause-internal expletives. The reason for this is that bothdevelopments result from the loss of vP-movement to SpecTP and its replacementby DP-movement to that position. The orders that are investigated include the so-called Stylistic Fronting (Styl-F), SVAux sequences and what has been analysed asVerb Projection Raising (VPR), i.e. AuxOV sequences. Following and developingthe proposals in B&R, new analyses of these orders are proposed. B&R also inte-grate the observations and analysis of van der Wurff (1997, 1999) regarding thelast attested OV orders with non-pronominal DPs. Furthermore, it is shown howthe changes that B&R propose for Late ME created some of the preconditions forthe well-known development of a syntactically distinct class of modal auxiliariesin the 16th century (Lightfoot 1979; Roberts 1985, 1993; Warner 1997; Roberts &Roussou 2003; Biberauer & Roberts 2006a, b).

Carola Trips’ “Syntactic Sources of word-formation processes” surveys word-formation from a diachronic perspective and the question of whether word-formations are built by the same principles that govern syntax. It is assumed thatword-formations like compounds and derivations historically start out as syntac-tic phrases and in the process of becoming morphological phrases lose structuralsyntactic properties like maximal projections and functional categories as well assemantic properties like e.g. referentiality. This is shown with diachronic datafrom German and English focusing on the phenomena of the development of



suffixes like Modern English -hood, Modern German -heit, and the rise of geni-tive compounds. Based on these findings it will be claimed that an analysis likeLieber (1992) or Ackema (1999) assuming that morphological operations are gov-erned by syntactic principles is not borne out and that word-formation operationshave to be attributed to an independent module of word-formation subject to itsown governing principles. Nevertheless, the rise of genitive compounds shows thatnew syntactic structures can occur once old syntactic structures have developedinto morphological structures implying that there is interaction between syntaxand morphology. Thus, looking at word-formation from a diachronic perspectiveprovides new insights into the nature and place of morphology.

References

Abraham, W. (1995). Deutsche Syntax im Sprachenvergleich: Grundlegung einer typologischenSyntax des Deutschen. Tübingen: Gunter Narr.

Ackema, P. (1999). Issues in Morphosyntax [Linguistik Aktuell/ Linguistics Today]. Amsterdam:John Benjamins.

Biberauer, T. & Roberts, I. (2005). Changing EPP parameters in the history of English:accounting for variation and change. English Language and Linguistics, 9, 5–46.

Biberauer, T. & Roberts, I. (2006a). Cascading parameter changes: Internally driven changein Middle and Early Modern English. Forthcoming in T. Eythórsson & J. T. Faarlund(Eds.), Grammatical Change and Linguistic Theory: the Rosendal Papers. Amsterdam andPhiladelphia: John Benjamins.

Biberauer, T. & Roberts, I. (2006b). Subjects, tense and verb movement in Germanic inRomance. Forthcoming in Proceedings of GLOW V in Asia.

Boeckx, C. & Stjepanovic, S. (2001). Head-Ing Toward PF. Linguistic Inquiry, 32, 345–355.Canale, M. (1978). Word Order Change in Old English: Base reanalysis in generative grammar.

PhD dissertation, McGill University.Chomsky, N. (2000). Minimalist inquiries: The framework. In R. Martin, D. Michaels, & J.

Uriagereka (Eds.), Step by Step (pp. 89–155). Cambridge MA: The MIT Press. Reprinted.Den Besten, H. (1989). Studies in West Germanic Syntax. Ph.D. thesis, Katholieke Universiteit

Brabant, Tilburg, Amsterdam.Fischer, O., van Kemenade, A., Koopman, W., & van der Wurff, W. (Eds.). (2000). The Syntax of

Early English. Cambridge: CUP.Haider, H. (1993). Deutsche Syntax Generativ. Tübingen: Gunter Narr.Kayne, R. S. (2000). Person morphemes and reflexives in Italian, French and related languages.

In R. S. Kayne, Parameters and Universals [Oxford Studies in Comparative Syntax] (pp.131–162). Oxford: OUP.

van Kemenade, A. (1987). Syntactic Case and Morphological Case in the History of English.Dordrecht: Foris.

Kroch, A. & Taylor, A. (1994). Remarks on XV/VX alternation in Early Middle English.Unpublished ms., University of Pennsylvania.

Lieber, R. (1992). Deconstructing Morphology. Chicago IL: The University of Chicago Press.Lightfoot, D. (1979). Principles of Diachronic Syntax. Cambridge: CUP.



Lightfoot, D. (1991). How to Set Parameters: Arguments from language change. Cambridge MA:The MIT Press.

Müller, G. (2004). Verb-Second as vP First. Journal of Comparative Germanic Linguistics, 7, 179–234.

Panagiotidis, P. (2002). Pronouns, Clitics and Empty Nouns. ‘Pronominality’ and licensing insyntax. Amsterdam: John Benjamins.

Postma, G. (2004). Structurele tendensen in de opkomst van het reflexief pronomen ‘zich’ in het15de-eeuwse Drenthe en de Theorie van Reflexiviteit. Nederlandse Taalkunde, 9, 144–168.

Roberts, I. (1985). Agreement parameters and the development of the English modal auxiliaries.Natural Language and Linguistic Theory, 3, 21–58.

Roberts, I. (1993). Verbs and Diachronic Syntax. Dordrecht: Kluwer.Roberts, I. (1997). Directionality and word order change in the history of English. In A. van

Kemenade & N. Vincent (Eds.), Parameters of Morphosyntactic Change (pp. 397–426).Cambridge: CUP.

Roberts, I. & Roussou, A. (2003). Syntactic Change. A minimalist approach to grammaticalization.Cambridge: CUP.

Ross, J. R. (1967). Constraints on Variables in Syntax. PhD dissertation, Massachusetts Instituteof Technology.

Warner, A. (1997). The structure of parametric change and V-movement in the history ofEnglish. In A. van Kemenade & N. Vincent (Eds.), Parameters of Morphosyntactic Change(pp. 380–393). Cambridge: CUP.

van der Wurff, W. (1997). Deriving object-verb order in late Middle English. Journal ofLinguistics, 33, 485–509.

van der Wurff, W. (1999). Objects and verbs in Modern Icelandic and fifteenth-century English:A word order parallel and its causes. Lingua, 109, 237–265.

Zwart, J. W. (1997). Morphosyntax of Verb Movement: A minimalist approach to the syntax ofDutch [Studies in Natural Language and Linguistic Theory]. Dordrecht: Kluwer.

JB[v.20020404] Prn:23/08/2006; 15:15 F: LA97P1.tex / p.1 (47-74)

Studies on predication

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.1 (47-153)

The Nom/Acc alternation in Germanic

Halldór Ármann SigurðssonLund University, Sweden

This paper describes the distribution of accusative case and discusses the natureof the nominative/accusative distinction in the standard Germanic languages. Inaddition, it illustrates and discusses the well-known fact that inherent accusativesand certain other types of accusatives do not behave in accordance with Bruzio’sGeneralization. In spite of these Non-Burzionian accusatives, there is a generaldependency relation between the so-called ‘structural’ cases, Nom and Acc, herereferred to as the relational cases, such that relational Acc is licensed only in theprecense of Nom (as has been argued by many). This relation is here referred toas the Sibling Correlation, SC. Contrary to common belief, however, SC is not astructural correlation, but a simple morphological one, such that Nom is thefirst, independent case, case 1 (‘an only child’ or an ‘older sibling’, as it were),whereas Acc is the second, dependent case, case 2, serving the sole purpose ofbeing distinct from Nom – the Nom-Acc distinction, in turn, being a morphol-ogical inerpretation or translation of syntactic structure. It has been an unre-solved (and largely a neglected) problem that the Germanic languages split withrespect to case-marking of predicative DPs: nominative versus accusative (It isI/me, etc.). However, the morphological approach to the relational cases arguedfor in this paper offers a solution to this riddle: The predicative Acc languageshave extended the domain of the Sibling Correlation, such that it does not applyto only arguments but to adjacent DPs in general. That is, the English type ofpredicative Acc is not ‘default’, nor is it caused by ‘grammatical viruses’, but awell-behaved subtype of relational Acc. The central conclusion of the paper isthat we need to abandon the structural approach to the relational cases in favorof a more traditional morphological understanding. However, this is not aconservative but a radical move. It requires that we understand morphology (andPF in general) not as a direct reflection of syntax but as a translation of syntaxinto an understandable but foreign code or ‘language’, the language of morphol-ogy. Nom and Acc are not syntactic features but morphological translations ofsyntactic correlations. It is thus no wonder that they are uninterpretable to thesemantic interface.

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.2 (153-203)

Halldór Ármann Sigurðsson

. Introduction*

In this paper I discuss the distribution of accusative case and the nature of thenominative/accusative distinction in the Germanic languages. In generative ap-proaches (Chomsky 1981; Burzio 1986, etc.), three different kinds of accusativeshave been generally assumed: Structural (object) accusatives, default accusatives,and other non-structural accusatives, as described with English examples in (1):

(1) a. She saw me. structural Accb. It is me. default Accc. I arrived the second day. other non-structural Acc

The class ‘other non-structural Acc’ includes not only adverbial accusatives butalso inherent accusatives.

I will here adopt a different view, arguing that there are basically only two ac-cusative types: Relational Acc, and Non-relational Acc, where the notion ‘relational’means dependent on the presence of a nominative DP. On this view, so-called de-fault, predicative accusatives are a well-behaved subtype of Relational Acc. Many ofthe Germanic languages, however, apply nominative case-marking of predicativeDPs. This predicative Nom/Acc variation is a central topic of this work.

In Section 2, I discuss Burzio’s Generalization (BG) and describe accusativecase-marking in the Germanic languages, concentrating on accusatives that areapparent or real exceptions to BG, in particular accusative subjects and the abovementioned predicative accusatives. Section 3 argues for a morphological, non-syntactic understanding of the relational (‘structural’) cases, where Nom is seen

* I would like to thank two anonymous reviewers for insightful comments. For helpful discus-sions, comments and corrections, many thanks also to Anders Holmberg, Andrew McIntyre,Cecilia Falk, Christer Platzack, Heidi Quinn, Janne Bonde Johannessen, Joan Maling, JóhannaBarðdal, Kjartan Ottosson, Lois Lopez, Marit Julien, and Verner Egerland. The ideas pursuedhere have, to a varying extent, been presented at several occasions: The 19th Grammar in Focus(GIF) in Lund, February 2005, CGSW 20 in Tilburg, June 2005, and the Linguistics Departmentin Konstanz in July 2005. Many thanks to the organizers of these events and to the audiences.In particular, I wish to express my gratitude to Ellen Brandner and her co-workers in Konstanz.For generous help with data, many thanks to: Heidi Quinn, Andrew McIntyre, Joan Maling, Di-anne Jonas (English), Marcel den Dikken, Sjef Barbiers, Hilda Koopman, Jan-Wouter Zwart(Dutch), Jarich Hoekstra (North and West Frisian), Theresa Biberauer (Afrikaans), BeatriceSantorini, Sten Vikner (Yiddish), Valentina Bianchi (Italian), Ellen Brandner, Gisbet Fanselow,Josef Bayer, Markus Benzinger, Philipp Conzett, René Schiering (German and German vari-eties), Marit Julien (Norwegian), and Ulf Teleman and other friends and colleagues in Lund:Camilla Thurén, Cecilia Falk, Christer Platzack, David Håkansson, Henrik Rosenkvist, MartinRingmar, and Verner Egerland (Swedish and Swedish varieties). A preliminary version of thiswork was published in Working Papers in Scandinavian Syntax 76:93–133 (“Accusative and theNom/Acc alternation in Germanic”).

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.3 (203-256)


as simply the first, independent case, case1, and Acc as the second case, case2,dependent on Nom being present in the structure. Section 4 argues that this mor-phological understanding enables us to analyze the English type of predicativeAcc as involving an extension of the general Nom/Acc distinction between argu-ments to DPs. In the concluding Section 5, I suggest, on the basis of the presentedfacts and analysis, that we need to abandon the view that morpho(phono)logy is astraightforward reflection of syntax. Rather, we must see morphology and syntaxas distinct ‘languages’ or codes, mutally understandable but foreign to each other.That is, morphology does not mirror or ‘show’ syntax, it translates it into its own‘language’, which is radically different from the ‘language’ of Narrow Syntax (inthe sense of Chomsky 2000 and subsequent works).

. The distribution of Nom/Acc across the Germanic languages

In this section, I will describe the distribution of accusative case and how it in-teracts with nominative case in the Germanic languages, mainly the standardones. Three major domains will be considered. In 2.1, I discuss the relational or‘structural’ cases in the sense of Burzio (1986) and the scope of his famous gen-eralization. In 2.2, I discuss argumental and adverbial accusatives that do not fallunder BG, above all certain Icelandic accusative subjects that have sometimes beenconsidered to be mysterious and a major challenge to BG. Finally, in Section 2.3, Idescribe the Germanic predicative Nom/Acc variation.

Sections 2.1 and 2.3 lay the foundations for the discussion in later sections,whereas Section 2.2 is more of an intermezzo, a long detour I have been forced tomake in order to be able to later proceed on the main road, so to speak. Many ofthe accusatives discussed in 2.2 are problematic and interesting, but those readerswho are only interested in the predicative Nom/Acc variation might opt for takinga bypass more or less directly from Section 2.1 to Section 2.3.

The Germanic languages divide into (relatively) case-rich and case-poor lan-guages, the former having (at least some) case-marking of full NPs, whereas thelatter have Nom/Acc marking of only pronouns. In addition, the case-rich lan-guages have morphological dative and genitive case (to a varying extent).

Case-rich: Icelandic, Faroese, German, YiddishCase-poor: Norwegian, Swedish, Danish, North Frisian, English, West

Frisian, Dutch, Afrikaans

Many of the Germanic languages show considerable dialectal variation with re-spect to the distribution of nominative and accusative case. Thus, some Swedishand Norwegian varieties have partly neutralized the Nom/Acc distinction (see Ek-lund 1982; Holmberg 1986), while other Swedish and Norwegian varieties have

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.4 (256-315)


retained even dative case (Reinhammer 1973), some German varieties have someinstances of accusative instead of the general German type of nominative predica-tive DPs, and so on. Also, many varieties that are often referred to as dialects aremore properly regarded as separate languages, from a linguistic point of view, in-cluding for instance the ‘Swedish’ Älvdalsmålet (see Levander 1909) and the ‘Ger-man’ Cimbrian in northernmost Italy (see Tyroller 2003). I will however largelylimit the present study to the 12 above listed ‘standard languages’, only mentioningother varieties occasionally.

. Germanic relational case-marking

All the standard Germanic languages show the core properties of accusative sys-tems, assigning nominative to (non-quirky) subjects and accusative to most ob-jects. This is illustrated below for three of the languages:

(2) a. She(/*Her) had seen me(/*I). English nom . . . accb. Hun(/*Hende) havde set mig(/*jeg). Danish nom . . . accc. Hún(/*Hana) hafði séð mig(/*ég). Icelandic nom . . . acc

A basic (and a generally known) fact about the standard Germanic languages isthat they all adhere to Burzio’s Generalization. The nontechnical version of BG isas follows (Burzio 1986:178; for exceptions, see below and, e.g., Burzio 2000):

(3) All and only the verbs that can assign a θ-role to the subject can assignaccusative case to an object

An alternative simple formulation of this correlation is given in (4):

(4) Relational Acc is possible only if its predicate takes an additional, externalargument

As I have argued in earlier work, however, the true generalization is evidentlynot about the relation between the external role and the internal case, but be-tween the cases themselves, nominative versus accusative. I have referred to thisas The Sibling Correlation (in e.g. Sigurðsson 2003:249, 258), formulating itas follows:

(5) (acc → nom) & ∼ (nom → acc)

In other words, a relational (‘structural’) accusative is possible only in the presenceof a nominative, whereas the opposite does not hold true, i.e. the nominative isthe first or the independent case (an ‘only child’ or an ‘older sibling’, as it were). Asimilar or a related understanding has been argued for by others, most successfullyby Yip et al. (1987), but also by, e.g., Haider (1984, 2000), Zaenen et al. (1985), andMaling (1993). Importantly, however, the Sibling Correlation only makes sense if

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.5 (315-382)


it applies generally, in non-finite as well as in finite clauses (see Sigurðsson 1989,1991). I will discuss the nature of the Sibling Correlation in Section 3.

In accordance with BG or SC, unaccusative (or ergative) verbs like arrive,unergative verbs like run and raising verbs like seem take nominative rather thanaccusative subjects in nominative-accusative languages like English. This is illus-trated by the following examples.

(6) a. She arrived late. / *Her arrived late.b. She ran home. / *Her ran home.c. She seemed to be shocked. / *Her seemed to be shocked.

More tellingly, an accusative object argument of a transitive verb turns up asnominative subject in passive and unaccusative constructions:

(7) a. They fired her.b. She was fired. / *Her was fired.

(8) a. They drowned her.b. She drowned. / *Her drowned.

These facts are well-known and have been widely studied and discussed (for a re-cent detailed study of case-marking in English, see Quinn 2005a). As one wouldexpect, much the same facts are found in the other Germanic languages. This is il-lustrated below for only transitive/passive pairs in German, Swedish and Icelandic,respectively:

(9) a. Siethey

habenhave

ihnhim.acc

gewählt.chosen

‘They chose him.’b. Er

he.nomwurdewas

gewählt.chosen

//

*Ihn wurde gewählt.*acc

(10) a. Dethey

valdechose

honom.him.acc

b. Hanhe.nom

valde-s.chose-pass

//

*Honom valde-s.*acc

‘He was chosen.’

(11) a. Þeirthey

völduchose

hana.her.acc

b. Húnshe.nom

varwas

valin.chosen

//

*Hana var valin.*acc

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.6 (382-441)


. Non-Burzionian accusatives

A priori, it is not clear why BG or SC should hold, that is, it is not obvious why thesubjects in the examples above cannot be accusative. It is appropriate to furtherhighlight this seemingly unexpected fact:

(12) a. *Her arrived late.b. *Her ran home.c. *Her seemed to be shocked.d. *Her was fired.e. *Her drowned.

Why is this the case in not only the other Germanic languages but in accusativelanguages (and accusative subsystems) in general? We shall return to this ques-tion in Section 3. Irrespective of the answer, these facts illustrate a truly strikinggeneralization, and it is indeed proper that it has a name of its own.

As acknowledged by Burzio (1986, 2000), however, it is not the case that allaccusatives fall under his generalization. Adverbial accusatives in languages likeGerman and Icelandic are perhaps the most obvious case of Non-Burzionianaccusatives:

(13) a. Dannthen

regneterained

esit

denthe.acc

ganzenwhole.acc

Tagday

//

*der*nom

ganze Tag.

‘Then, it rained all day.’b. Þá

thenrigndirained

allanall.acc

daginnday.the.acc

//

*allur*nom

dagurinn.

‘Then, it rained all day.’

Accusative adverbial NPs most commonly have a temporal reading, as in theseexamples, but local (path) readings also occur, as illustrated below for Icelandic:

(14) a. Húnshe

syntiswam

heilanwhole.acc

kílómetrakilometre.acc

//

*heill*nom

kílómetri.

b. Hannhe

gengurwalks

alltafalways

sömusame.acc

leiðroute.acc

//

*sama*nom

leið.

As discussed by (Zaenen et al. 1985:474–475), path adverbials of this sort oftenshow up in the nominative in passives, thus behaving similarly as Burzionian ac-cusatives.1 In contrast to argumental accusatives, however, path accusatives mayalso be retained in impersonal passives, that is, the Acc passive (?)Það er/var gengiðþessa sömu leið til baka daginn eftir ‘it is/was walked this same route.acc back the

. In Finnish, this even applies to temporal adverbials (Maling 1993).

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.7 (441-496)


day after’ is fairly acceptable, whereas, e.g., *Það er/var teiknað þessa sömu leið ‘itis/was drawn this same route.acc’ is impossible.2

Another type of Non-Burzioninan accusatives is accusative complements ofprepositions. As illustrated below for English, German, Swedish and Icelandic, inthat order, accusative prepositional complements are well-formed in the absenceof an external argument:

(15) a. There is much talking about him here.b. Hier

herewirdis

(*es)(*it)

vielmuch

überabout

ihnhim.acc

gesprochen.talked

c. Härhere

talasis-talked

(det)(it)

mycketmuch

omabout

henne.her.acc

d. Hérhere

eris

(*það)(*it)

talaðtalked

mikiðmuch

umabout

hana.3

her.acc

These types are not problematic for BG, as it is formulated specifically for argu-ments of verbs, but they illustrate that morphological accusatives can be used forNon-Burzonian purposes, even in basically accusative systems.

On the other hand, quirky accusatives are unexpected under BG and SC.Consider the Icelandic examples below:

(16) a. Migme.acc

vantarlacks

peninga.money.acc

‘I lack/need money.’b. Mig

me.acclangarlongs

heim.home

‘I want to go home.’c. Mig

me.accfurðarsurprises

áin

þessu.this

‘I’m surprised by this.’

As seen, the accusatives in these exampels are well-formed irrespective of whethertheir predicate takes an additional argument. That is, BG and SC would seemto make a wrong prediction for these predicates (but see below for a differentinterpretation).

. This applies to my own grammar, which, as far as I can tell, is the standard variety in this re-spect. In the so-called ‘new passive’ variety, on the other hand, Það er/var teiknað þessa sömuleið ‘it is/was drawn this same route.acc’ would be grammatical (see, e.g., Maling & Sigur-jónsdóttir 2002).

. The d-example illustrates the well-known fact that the Icelandic expletive can only occurclause-initially (Thráinsson 1979; see also Sigurðsson 2004a for a feature based approach to thisClause Initial Constraint, CLIC).

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.8 (496-559)


Jónsson (1998:35f.) lists almost 60 predicates that take an accusative sub-ject in (standard) Icelandic. As demonstrated below, Faroese (Thráinsson etal. 2004:253f.) and German also have examples of this sort, albeit much lessfrequently:

(17) a. Megme.acc

grunaðisuspected

hetta.this.acc

Faroese

‘I suspected this.’b. Mich

me.acchungert.hungers

German

‘I am hungry.’

The German construction is peripherical (see, e.g., Wunderlich 2003), and it seemsto be rapidly disappearing from Faroese as well (Eythórsson & Jónsson 2003). It isalso losing some ground in colloquial Icelandic, through so-called ‘dative sickness’,whereby accusative experiencer subjects in examples like (16a–c) are replaced withdatives (see Smith 1996 and the references there).

Icelandic has a second type of quirky accusative subjects, where the subject isnot an experiencer but a theme or a patient, as illustrated below (Zaenen & Maling1984 and many since):

(18) a. Okkurus.acc

rakdrove

aðto

landi.land

‘We drifted ashore.’(drove = ‘got-driven’)

b. Bátinnboat.the.acc

fylltifilled

áin

augabragði.flash

‘The boat swamped immediately.’(filled = ‘got-filled’)

c. Migme.acc

tóktook

út.out

‘I was swept overboard.’(took = ‘got-taken’)

d. Menninamen.the.acc

barcarried

aðtowards

íin

þessu.that

‘The men arrived then.’(carried = ‘got-carried’)

As we shall see shortly, this second, theme/patient construction has an uncon-trolled process or fate reading. For convenience, we may thus refer to the ac-cusatives in (16)–(17) versus (18) as Psych Accusatives and Fate Accusatives,respectively.4 While Psych Accusatives tend to get replaced by datives, Fate Ac-cusatives often give way to the nominative in (mainly) colloquial Icelandic (see

. As pointed out to me by Kjartan Ottosson, the notion ‘fate’ may not be entirely satisfactoryhere. The most common type of these predicates typically involves the natural forces as thesource or the ‘hidden agent’ of the event (as discussed in Ottosson 1988). However, this does

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.9 (559-618)


Eythórsson 2000), in which case they behave like ordinary unaccusatives in thelanguage (see below).

As discussed by Haider (2001) and Kainhofer (2002), German also has FateAccusatives of a similar sort, as illustrated in (19) (ex. (7a) in Haider 2001:6):

(19) Oftoften

treibtdrives

esit

ihnhim.acc

insinto-the

Gasthaus.bar

‘He often ‘drifts’ into the bar.’

However, the German construction has an expletive, which perhaps or even plau-sibly may be analyzed as carrying nominative case.5 An expletive is excluded in theIcelandic construction:

(20) a. Mannone.acc

hrekurdrives

stundumsometimes

afoff

leið.track

‘Sometimes one loses one’s track/gets carried away.’b. *Það

ithrekurdrives

mannone.acc

stundumsometimes

afoff

leið.track

Thus, the Icelandic construction differs from the German one. However, Icelandichas another construction that is to an extent similar to the German construc-tion. This is the Impersonal Modal Construction, IMC, discussed in Sigurðsson(1989:163ff.), with an arbitrary external role and an optional expletive (the expli-tive is generally only optional in Icelandic, see Thráinsson 1979). IMC is exem-plified in (21); as suggested by the postverbal position of the accusatives, they areregular objects and not quirky subjects (in contrast to the quirky accusatives in(16), (18) and (20)):

(21) a. Þaðit

áhas

aðto

byggjabuild

húsiðhouse.the.acc

hér.here

‘They are going to build the house here.’b. Það

itþarfneeds

aðto

aðstoðaassist

hana.her.acc

‘One needs to assist her.’c. Hér

heremámay

ekkinot

reykjasmoke

vindla.cigars.acc

‘One may not smoke cigars here.’

not extend to all examples of this sort, for instance (18d) and (20) below. I therefore take theliberty of using the notion ‘fate’ as a cover term for forces that are not in human power.

. This might extend to the ‘new passive’ in Icelandic (type ‘It was hit me.acc’). I will not dis-cuss this here, but see, e.g., Sigurðsson (1989:355f.), Sigurjónsdóttir & Maling (2001), Maling &Sigurjónsdóttir (2002).

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.10 (618-769)


Possibly, however, both IMC and the German construction throw a light on theorigin of the Icelandic Fate Accusative, that is, it may have grown out of a similartransitive construction, with an unexpressed ‘fate subject’, as it were.

As discussed by Zaenen & Maling (1984) and by Sigurðsson (1989), ordinaryunaccusatives have similar properties in Icelandic as in related languages, show-ing the familiar Acc-to-Nom conversion when compared to homophonous orrelated transitives, much like passives. Consider the following transitive-passive-unaccusative triple:

(22) a. Hún stækkaði garðinn.she enlarged garden.the.acc

Transitive: Nom-Acc

b. Garðurinn var stækkaður.garden.the.nom was enlarged

Passive: Nom

c. Garðurinn stækkaði.garden.the.nom enlarged

Unaccusative: Nom

In contrast, Fate Accusative predicates, like the ones in (18), show an unexpectedand (what seems to be) a cross-linguistically very rare behavior, in taking anunaccusative accusative, as it were:

(23) a. Hún fyllti bátinn.she filled boat.the.acc

Transitive: Nom-Acc

b. Báturinn var fylltur.boat.the.nom was filled

Passive: Nom

c. Bátinn fyllti.boat.the.acc filled

Unaccusative: Acc!

In contrast, datives and genitives are regularly retained in both passives and unac-cusatives:6

(24) a. Hún seinkaði ferðinni.she delayed journey.the.dat

Transitive: Nom-Dat

b. Ferðinni var seinkað.journey.the.dat was delayed

Passive: Dat

c. Ferðinni seinkaði.journey.the.dat delayed

Unaccusative: Dat

On a lexical approach to quirky and inherent case-marking, we would seem to beforced to analyze the accusative in (23c) (and the ones in (18)) as lexical, that is,as selected by an quirky case requiring feature or property of the predicate (see thediscussion in Sigurðsson 1989:280ff.). As seen in (23b), however, this accusative

. As opposed to nominalizations and the so-called ‘middle’ -st construction (see Zaenen &Maling 1984 and many since, e.g., Svenonius 2005).

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.11 (769-839)


is not retained in the passive, instead undergoing the Nom-to-Acc conversion reg-ularly seen for ordinary relational, non-inherent accusatives, as in (22b). That is,what would seem to be ‘one and the same’ accusative shows paradoxical behavior.We may refer to this state of affairs as the Fate Accusative Puzzle. As we shallsee soon, however, the puzzle is in a sense not real, as the unaccusative accusativeis arguably not the ‘same’ accusative as the transitive one.

As recently discussed by McFadden (2004), McIntyre (2005) and Svenonius(2005), there are reasons to believe that the inherent cases are in fact structurallymatched against syntactic heads or features rather than lexically licensed.7 In thisvein, Svenonius (2005) argues for a structural solution to the Fate Accusative Puz-zle, suggesting that the predicates in question “have a cause component but onlyan optional voice, in the sense of Kratzer (1996) and Pylkkänen (1999)” – wherevoice is the head that licenses agent. In addition, Svenonius (2005) suggests thatcause is “implicated in the licensing of accusative case, and is absent from normalunaccusatives”. That is, predicates are varyingly complex, transitives having bothvoice and cause, Fate Accusative predicates or ‘accusative unaccusatives’ havingonly cause, and regular unaccusatives having neither:8

(25) a. [VoicP DPnom voic [CausP caus [VP V DPacc]]] Transitive Nom-Accb. [CausP caus [VP V DPacc]] Acc unaccusativesc. [VP V DPnom] Nom unaccusatives

Dative taking unaccusatives, like seinka ‘delay’ in (24c), also have the cause com-ponent plus a special dative or dat feature, “necessary for the assignment of da-

. Thus, as has been observed in the literature every now and then, there is generally no fixedlinking between lexical roots and specific cases, as illustrated by numerous minimal pairs likethe following one (involving various types of predicates):

(i) a. Veðriðweather.the.nom

eris

kalt.cold

b. Mérme.dat

eris

kalt.cold

‘I’m freezing.’(ii) a. Húsið

house.the.nomvarwas

lokað.closed

‘The house was (in the state of being) closed.’b. Húsinu

house.the.datvarwas

lokað.closed

‘The house was (in the process of being) closed (by someone).’

. Svenonius assumes a slightly more complex analysis (where active versus passive or act andpass play a crucial role), but the presentation in (25) is sufficiently detailed for our purposes.

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.12 (839-902)


tive case” (ibid). The transitive and the dative unaccusative in (24a, c) thus haveroughly the following structures:

(26) a. [VoicP DPnom voic [CausP caus [DatP dat [VP V DPdat]]] Transitive Nom-Datb. [CausP caus [DatP dat [VP V DPdat]]] Dat unaccusatives

Icelandic has many kinds of datives (Barðdal 2001; Maling 2002a, 2002b; Jóns-son 2003; Sigurðsson 2003:230ff.), so we must understand dat as a shorthand foran array of syntactic features (‘heads’) or feature combinations, each such featureor feature combination leading to dative case-marking in Icelandic morphology.9

With that modification, it seems to me that Svenonius has developed an interestingapproach to many of the numerous facts known from the voluminous literatureon Icelandic case. However, while a structural approach to the inherent cases ispromising, such an approach to the relational, so-called ‘structural’ cases (Burzio-nian Nom/Acc) is fundamentally mistaken, I believe, contradictory as that mayseem (see also Sigurðsson 2003, 2006). I will return to the issue in Section 3.

As mentioned above, the peculiar ‘accusative unaccusative’ construction inIcelandic has a special uncontrolled process semantics, a get-passive fate readingof a sort, hence the term Fate Accusative. Consider (18) = (27):

(27) a. Okkurus.acc

rakdrove

aðto

landi.land

‘We drifted ashore.’(drove = ‘got-driven’)

b. Bátinnboat.the.acc

fylltifilled

áin

augabragði.flash

‘The boat swamped immediately.’(filled = ‘got-filled’)

c. Migme.acc

tóktook

út.out

‘I was swept overboard.’(took = ‘got-taken’)

d. Menninamen.the.acc

barcarried

aðtowards

íin

þessu.that

‘The men arrived then.’(carried = ‘got-carried’)

Importantly, this fate reading is not shared by the transitive or passive counterpartsto these (or other Fate Accusative) predicates (as already pointed out by Ottosson1988:148). Thus, Icelandic “we filled the boat” and “the boat was filled” has muchthe same expected readings as English We filled the boat and The boat was filled,that is, it means that the boat was deliberately filled in some usual, expected man-ner, with fish or some cargo. Icelandic “the boat filled”, in contrast, has only onevery specific meaning, namely that the boat unexpectedly and dangerously got

. The same features are arguably present in the syntax of languages, such as English, that ‘keepquiet’ about them in their morphology (cf. Sigurðsson 2003, 2004d).

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.13 (902-960)


filled with water, i.e. that it ‘swamped’. Similarly, Icelandic Mig tók út in (27c), lit-erally ‘me took out’, cannot possibly mean that ‘somebody took me out’. It has onlyone, very specific meaning, the fate reading that I ‘accidentally swept aboard’. In allcases of this sort, the transitive and passive versions have much the same general,broad semantics as in English and other related languages, whereas ‘accusative un-accusatives’ always have a narrow, semi-idiomatic fate meaning, absent from thetransitive and the passive.

This important fact has not been generally noticed or highlighted, so one com-monly sees pairs like the following in the literature (here taken from Sigurðsson1989:216, but see also similar examples in e.g. Zaenen & Maling 1984; Jónsson1998; Svenonius 2005):

(28) a. Bátinnboat.the.acc

rakdrove

áon

land.land

‘The boat drifted ashore.’b. Stormurinn

storm.the.nomrakdrove

bátinnboat.the.acc

áon

land.land

This description is however misleading. As pointed out already by Ottoson(1988:147f.), transitive clauses like (28a) are semantically anomalous, since tran-sitive verbs like reka ‘drive’ usually require an animate agent. The same holds forother apparent pairs or sets of transtives/passives and ‘accusative unacusatives’, asillustrated below for fylla ‘fill’:

(29) a. Bátinnboat.the.acc

fylltifilled

(af(with

sjó).sea)

‘The boat swamped.’b. ?Sjórinn

sea.the.nomfylltifilled

bátinn.boat.the.acc

(30) a. Viðwe.nom

fylltumfilled

bátinn.boat.the.acc

‘We filled the boat (with cargo, fish, etc.).’b. Báturinn

boat.the.nomvarwas

fylltur.filled

‘The boat was filled (with cargo, fish, etc.).’

Thus, the ‘accusative unaccusatives’ require a special fate or uncontrolled processfeature to be present or active in their clausal structure. Call this feature simplyfate. There is no doubt, as we have seen, that this feature is precluded in relatedtransitives and passives, and the natural interpretation of that fact is that fate is avoice feature of a sort, blocking or ‘turning off ’ the usual voice feature that other-

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.14 (960-1004)


wise introduces agent in both transitives and passives.10 As Ottosson (1988:148)puts it, Fate Accusatives occur “in a construction that stands outside the regu-lar voice system”. That is, the nature of Fate Accusatives is quite different fromthat of normal relational (not semantically linked) accusatives, and hence the FateAccusative Puzzle is not real.

The fate feature is largely (but not entirely) specific for the ‘accusative un-accusatives’, that is, it is not generally active in structures with either regular un-accusatives or dative taking unaccusatives. This is illustrated for the dative takingljúka ‘finish, end’ (discussed in, e.g., Sigurðsson 1989 and Svenonius 2005):

(31) a. Húnshe.nom

laukfinished

sögunni.story.the.dat

‘She finished the story.’b. Sögunni

story.the.datlauk.finished

‘The story ended/came to an end.’

The transitive means ‘end/finish something’, and the unaccusative also has the gen-eral core meaning ‘end’, without any special reading being added. While Fate Ac-cusative predicates yield information about the power (fate/natural forces) causingthe event, such semantically narrowing or specifying information is absent frommany or most other unaccusative predicates. It is of course logically possible thatan event expressed by predictes like ljúka ‘finish, end’, seinka ‘delay’ in (24) andstækka ‘enlarge’ in (22) can be due to fate or natural forces, but this reading is notforced for these predicates, in contrast with Fate Accusative predicates.

In sum, there is no doubt that Fate Accusatives relate to semantics of a ratherspecial sort. However, this does not alter the fact that these peculiar accusatives arelike Psych Accusatives in that they do not comply with BG or the Sibling Correla-tion. Restaurant Talk Accusatives, recently discussed by Wiese and Maling (2005),on the other hand, can be anlyzed as involving deletion, as sketched below forGerman and Icelandic, respectively:11

. An issue of general theoretical interest is whether inactive features are syntactically absent,or present but default or not activated. I assume the latter (following Cinque 1999:127ff.; for ageneral discussion, see also Sigurðsson 2004d).

. Alternatively, one can assume silent functional categories in examples of this sort, includingthe subject number and person, a modal head commonly expressed by verbs meaning ‘want’ anda silent main predicate commonly expressed by verbs meaning ‘get’. Under such an approach(tallying with the approach to morphosyntactic silence argued for in Sigurðsson 2004d), theproblem of recoverability does not arise.

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.15 (1004-1089)


(32) a. IchI.nom

möchtewould-like

Einena

Kaffeecoffee.acc

bekommen,get,

bitte.please

‘One coffee, pleaseb. Ég

I.nomvilwant

fáget

Tvotwo

stóralarge

bjóra,beers.acc,

takk.please

Similarly, accusatives in PRO infinitives are unproblematic if nominative is activein such infinitives (as argued in Sigurðsson 1989, 1991):

(33) a. Mirme.dat

grautdreads

[PROPRO

denthe

Briefletter.acc

zuto

schreiben].write

German

‘I find it dreadful to write the letter.’b. Mér

me.datleiðistbores

[aðto

PROPRO

lesaread

bókina].book.the.acc

Icelandic

‘I find it boring to read the book.’

Icelandic has many predicates that take a dative subject and a nominative object(Thráinsson 1979; Zaenen et al. 1985; Sigurðsson 1989, 1996, and many others),and German has some similar Dat-Nom predicates (usually taken to have differentproperties as regards subjecthood versus objecthood, but see Eythórsson & Barð-dal 2005 and Barðdal 2006 for a different view). In contrast, some related and/orsimilar predicates in Faroese are Dat-Acc predicates, as illustrated in the examplesbelow (from Thráinsson et al. 2004:255ff.):

(34) a. Mærme.dat

líkarlikes

hanaher.acc

væl.well

‘I like her.’b. Henni

her.datvantarlacks

góðagood

orðabók.dictionary.acc

‘She needs a good dictionary.’

The Icelandic equivalent of (34b) can in fact also be heard in substandard Ice-landic. Conversely, Faroese also has some Dat-Nom predicates.

In Sigurðsson (2003), I argued that accusatives in examples of this sort arerelational, the structures in question involving an invisible nominative, triggeringor licensing the accusative (in accordance with the Sibling Correlation; for relatedideas see Haider 2001, but for a different approach, see, e.g., Woolford 2003). Thiswould seem to get support from the historical development in English in general(Allen 1996) and partly in Faroese, where numerous predicates have altered theircase frames in the following manner:

(35) Dati-Nomj > Dati-Accj (or Obliquei-Obliquej) > Nomi – Accj

Alternatively, one might want to suggest that the accusative in Faroese Dat-Accconstructions is licensed by the external dative or that it is some sort of a default

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.16 (1089-1148)


case, but that would seem to raise even more difficult problems and questionsthan the invisible nominative case approach, most simply the question of why theFaroese subject dative should license or allow object accusative any more than ex-ternal datives in e.g. standard Icelandic, German and Old English. Also, invokingthe notion of ‘default case’ amounts to giving up any hope of an insightful ac-count. If Faroese resorts to default case in its Dat-Acc constructions, the questionarises why it does not in e.g. predicative constructions (see the next subsection).Moreover, the change Dati-Accj > Nomi-Accj would involve two changes on thedefault case approach (inherent-default > relational-relational), whereas it involvesonly one change on the relational case approach (inherent+relational-relational >relational-relational). In addition, the accusative in the Faroese Dat-Acc patternseems to be like regular relational accusatives in not being semantically linked,unlike both Psych Accusatives and Fate Accusatives.

Regardless of how we account for the exceptional Dat-Acc pattern in Faroese,it is clearly unexpected under any straightforward morphological understand-ing of Burzio’s Generalization (BG) and the Sibling Correlation (SC), like PsychAccusatives and Fate Accusatives. Yet another type of unexpected argumental ac-cusatives is found in a peculiar (and lexically a very limited) raising constructionin Icelandic (discussed in Sigurðsson 1989, e.g., 218f.), where accusative is retainedor ‘fossilized’, in contrast to, e.g., both German and English:

(36) a. Ólaf/*ÓlafurOlaf.acc/*nom

varwas

hverginowhere

aðto

finnafind

__.

b. Er/*Ihnhe/*him

warwas

nirgendsnohwere

zuto

findenfind

__.

c. He/*Him was nowhere to be found __.

As mentioned above, adverbial and prepositional accusatives are not really prob-lematic for BG or SC. On the other hand, all the argumental accusatives that arewell-formed in the absense of an external nominative argument are unexpectedunder BG/SC:

– Psych Accusatives in Icelandic and to an extent in German and Faroese– Fate Accusatives in Icelandic (and possibly in German varieties, depending on

whether or not the expletive carries nominative case)– The ‘fossilized’ accusative in Icelandic examples like (36a) (perhaps only a

subtype of the Fate Accusative)– The accusative in Faroese Dat-Acc constructions

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.17 (1148-1193)


Moreover, English allows (subject and) object accusatives in gerunds like the fol-lowing ones (see e.g. Quinn 2005a:Section 8.6), where there is no visible well-formed nominative:12

(37) a. I was embarrassed [by him seeing me there].b. [His accusing me] surprised me greatly.c. [Him seeing me there] was unfortunate.d. *[He seeing me there] embarrassed me.

There seems no doubt that the object accusative in examples of this sort is a regularaccusative, much as in subjectless gerunds and PRO infinitives:

(38) a. Seeing me there suprised him.b. To see me there surprised him.

On the relational view of the so-called ‘structural’ cases, the object accusativein all these cases is licensed by an active nominative case feature, even thoughthe nominative is morphologically invisible. The same applies to the matrix ac-cusative in examples like (38), where the gerund seeing me there and the infinitiveto see me there receive invisible nominative case, thereby licensing the matrix ac-cusative him (see Sigurðsson 2003:248). An alternative view would be that theseobject accusatives are exceptional in one way or another. That does not seem tobe the case.

One way around the problem raised by Non-Burzionian accusatives in at leastIcelandic (see e.g. Burzio 2000; Sigurðsson 1989) is to say that these accusativesare inherent, like datives and genitives, and to formulate BG such that it applies torelational (‘stuctural’) accusatives only, as I did in (4) above. This might seem to becircular and vacuous. First, we adopt a broad generalization over accusative case,and then, when we learn about accusatives that do not behave as the generalizationwould lead us to expect, we reformulate it such that it does not apply to these‘exceptional’ accusatives. By also excluding other types of ‘different’ accusatives(prepositional, adverbial) we end up with a notion of a relational accusative thatsimply means ‘an accusative that behaves in accordance with BG/SC.’

However, in spite of the seemingly ad hoc flavor of this approach, it is em-pirically quite correct, as far as can be seen. Outside the domain of exceptionalor Non-Burzionian accusatives, Burzio’s Generalization really does hold true for

. Joan Maling and Andrew McIntyre, p.c. The fact, illustrated in the d-example, that the sub-ject of the gerund cannot be nominative seems to suggest that the accusative of the gerundsubject in the c-example is a prepositional accusative of a sort (like the subject accusative in thea-example), assigned by a deleted or a silent for-type preposition (‘for him seeing me there . . .’, asit were). I refrain from taking a stand on the issue, though (according to Huddleston & Pullumet al. 2002:460, at least some gerunds can have a nominative subject in formal style).

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.18 (1193-1250)


all Germanic varieties and, quite presistently, across accusative systems in gen-eral.13 Thus, even in Icelandic, Non-Burzionian accusatives are generally excluded,as illustrated below for the most central predicate types:

(39) a. Hún/*Hanashe/*her

komarrived

seint.late

b. Hún/*Hanashe/*her

hljópran

heim.home

c. Hún/*Hanashe/*her

virtistseemed

verðabecome

skelkuð.shocked

d. Hún/*Hanashe/*her

varwas

rekin.fired

e. Hún/*Hanashe/*her

drukknaði.drowned

Thus, in spite of the fact that Icelandic and some other languages have some typesof Non-Burzioninan accusatives, Burzio’s Generalization or the Sibling Correla-tion expresses a striking thruth about accusative systems and must therefore betaken seriously as a ‘revelation’ of the nature of the relational cases.

Having established that most of the Non-Burzioninan accusatives discussedin this section can be specially accounted for and are thus not counterevidenceagainst the Sibling Correlation, we can take the ‘main road’ again, turning to theGermanic predicative Nom/Acc variation.

. The Germanic predicative Nom/Acc variation

The case-rich Germanic languages all have nominative predicative DPs (in non-ECM):14

(40) a. IkhI

binam

aa

gutergood

yidJew.nom

//

*a gutn yid.*acc

Yiddish nom

b. Dasthat/this

sindare

wirwe

//

*uns.*us

German nom

‘That/This is us.’

. Outside the Germanic languages, exceptional or Non-Burzionian accusatives are found in,e.g., Tamil (Dat-Acc, see Lehmann 1993:184ff.) and Modern Greek (Gen-Acc, see e.g. Anagnos-topoulo 2003).

. See Maling & Sprouse (1995) on predicative case in some of the Germanic languages. I havenot been able to track down any other studies of predicative case from a comparative/generativeGermanic perspective.

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.19 (1250-1414)


c. Taðit

eruare

vitwe

//

*okkum.*us

Faroese nom

‘It is us.’d. Það

iterumare

viðwe

//

*okkur.*us

Icelandic nom

‘It is us.’

The case-poor Germanic languages, on the other hand, show a remarkable varia-tion with respect to predicative case, that has, to my knowledge, never been care-fully studied. English, Danish, most varieties of Norwegian (Norw1) and NorthFrisian have accusative marking of predicative NPs:

(41) a. It is us. English accb. Det er os.

it is usDanish acc

c. Det er oss.it is us

Norw1 acc

d. Dåt as üs.it is us

N. Frisian acc

In English, predicate nominative examples like It is I do occur, above all in thewritten language (e.g., Quinn 2005a:233ff.).15 Corresponding Danish examples,*Det er jeg, etc., are unacceptable in all contexts (Allan et al. 1995:143).

The other case-poor Germanic languages normally have nominative marking:

(42) a. Det är vi/*oss.is is we/*us

Swedish nom

b. Det er vi.16

is is weNorw2 nom

c. Dat zijn wij/*ons.is are we/*us

Dutch nom

d. Dat bin ik/*my.it is I/*me

W. Frisian nom

e. Dit is ek(ke)/*my.it is I/*me

Afrikaans: nom

The English type accusative marking in (41) has not generally been taken seriouslyas a linguistic fact, at least not within generative case theory. As Quinn puts it, in

. Examples of this sort have variable status, that is, fixed expressions like This is she (as aformula used to answer the telephone) and It is I are better or more familiar than for instance??It is we (Joan Maling, p.c., Heidi Quinn p.c., see also Huddleston & Pullum, et al. 2002:459).However, the expressions in question seem to be kept alive mainly or exclusively by prescriptivistinfluences (Joan Maling, p.c., Andrew McIntyre, p.c.).

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.20 (1414-1458)


her detailed study of English case-marking (2005a:1), it has often been assumed orsuggested “that pronoun case selection in English is largely unsystematic, and besttreated as the product of local rules, grammatical viruses, and hypercorrection”(see also the historical opinion survey in Visser 1963:241ff.). The fact that Englishshares many of its case properties with some other Germanic varieties immediatelysuggests that this view must be mistaken (as also pointed out by Visser 1963:244).Nonetheless, the Germanic predicative Nom/Acc variation has remained largelyneglected within generative case theory, for the embarrassing reason, I believe, thatthere is really nothing interesting to be said about it under a structural approach tothe relational cases – under such an approach the predicative case variation makesno sense, it is just unexpected and mysterious. In contrast, I will here argue thatthe English type of case facts are indeed serious and ‘intelligent’ data that can besystematically accounted for to a much greater extent than often assumed.

For convenience, I will refer to the Germanic languages that (normally orcentrally) apply nominative versus accusative marking of predicative DPs as pred-icate nominative languages versus predicate accusative languages.

Most of the predicate nominative case-poor languages are actually partly pred-icate accusative. Thus, in the third person, both Dutch and West Frisian allowaccusative marking, provided that the pronoun is a weak one (in which case fo-cal stress moves from the pronoun onto the verb), as illustrated in the c-examplesbelow:17

(43) a. Datit

isis

hij.he

Dutch

b. *Datit

isis

hem.him

c. Dat IS ‘m.

. My informants cannot use the weak het ‘it’ here (but for a discussion of a Dutch gram-mar where this is possible, see van Gelderen 1997:152ff.). With a weak it-pronoun, many ormost speakers of both Dutch and German accept only the West-Germanic Inverted Predica-tive Construction (IPC) ‘we are it’, etc. (German: Wir sind es; Dutch: Wij zijn het). This typewas also prevailing in Old English, with the non-verb-second order ‘we it are’ (see the exam-ple in (55a) below), but it is not generally acceptable in the Scandinavian languages and ModernEnglish. I don’t have anything interesting to say about the correlation between IPC and the regu-lar, non-inverted predicative construction. Let me just emphasize that verb agreement is clearlynot decided by ‘subjecthood’ in the Germanic languages. Thus, predicative NPs control verbagreement in Icelandic examples like the following (Sigurðsson 1996, 2004b, 2004c):

(i) Þaðit

hafiðhave.2pl

þáthen

sennilegaprobably

baraonly

veriðbeen

þið.you.nom.pl

‘It has then probably only been you.’

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.21 (1458-1538)


(44) a. Datit

isis

hy.he

W. Frisian

b. *Datit

isis

him.him

c. Dat IS ‘m.

In the first and the second person, on the other hand, accusative is always ungram-matical (irrespective of focal stress and verb agreement), as illustrated for only thefirst person in only Dutch below:

(45) a. Datit

benam

ik.I

//

*Datit

ben/isam/is

mij.me

b. Datit

zijnare

wij.we

//

*Datit

zijn/isare/is

ons.us

This is reminiscent of 1st/2nd versus 3rd person contrasts in other languages, forinstance the fact that agreement with nominative objects in Icelandic is possibleonly in the 3rd person (Sigurðsson 1996; Boeckx 2000, etc.), and, in particu-lar, the fact that accusative clitics can only be in the 3rd person in the presenceof a dative clitic in the double object construction in numerous languages, forinstance Romance and Slavic languages (the Person-Case Constraint, see Perlmut-ter 1971; Bonet 1991, and, e.g., Anagnostopoulou 2003). It is not clear though,whether there is a correlation between these widely discussed phenomena andthe Dutch/West Frisian facts above. I have not been able to find any obviousconnection.

In spite of generally being predicate nominative languages, both Dutch andWest Frisian apply accusative in modal contexts, as opposed to Afrikaans, as il-lustrated below:

(46) a. IkI

wilwant

jouyou.acc

zijn,be,

jijyou

kancan

mijme.acc

zijn.be

Dutch

b. IkI

wolwant

dyyou.acc

wêze,be,

doyou

kinstcan

myme.acc

wêze.be

W. Frisian

c. EkI

wilwant

jyyou.nom

wees,be,

jyyou

kancan

ekI.nom

wees.be

Afrikaans

Similarly, Dutch and West Frisian opt for the accusative in conditionals, again asopposed to Afrikaans:

(47) a. Alsif

ikI

jouyou.acc

waswere

. . . Dutch

b. Atif

ikI

dyyou.acc

wiewere

. . . W. Frisian

c. Asif

ekI

jyyou.nom

waswere

. . . Afrikaans

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.22 (1538-1603)


It seems, then, that Dutch and West Frisian are or have been developing in the‘English direction’, so to speak, along several different paths. A similar process canbe observed for Swedish (as discussed in part by Teleman 2001), but, interestingly,the paths of the change are not always the same as in Dutch and West Frisian. Thus,most speakers (who regularly distinguish between nominatives and accusatives)accept only the nominative in conditionals (although the accusative is clearly lessmarked here than in, e.g., (49) below):

(48) Omif

jagI

vorewere

du/?dig . . .you.nom/?acc

In general, the nominative is the only option, it seems, when the predicative con-struction expresses plain identity, as in the following examples (the first one ismodelled on an example in Teleman 2001):

(49) a. Denthis

härhere

lillalittle

pojkenboy.the

påon

bildenphoto.the

äris

jag/*mig.I/*me

b. Han/Dethe/it

är/ÄRis/IS

intenot

han/*honom.he/*him

If, on the other hand, the predicative construction gets a reading where the pred-icative DP takes on the role of the subject DP, rather than its identity, then thepredicate DP can be accusative for some speakers. This is illustrated below, wherethe % sign indicates that the acceptability is subject to some speaker variation(again, the a-example is based on Teleman 2001):

(50) a. %JagI.nom

låtsaspretend

intenot

varabe

dig.you.acc

‘I’m not pretending to be you.’b. %Jag

I.nomvillwant

varabe

dig,you.acc,

duyou.nom

kancan

varabe

mig.me.acc

Also, as mentioned by Teleman (2001), accusative is the only option in mostreflexive predicates, for most or all speakers:18

(51) a. JagI

äram

intenot

längrelonger

migme.acc

själv(my)self

//

*jag*I.nom

själv.(my)self

‘I’m not myself any more.’

. Compatible facts are found for at least some Alemannic varieties. Philipp Conzett providesthe follwowing example from his Alemannic (spoken in Graubünden in easternmost Switzer-land):

(i) II

binam

nümmanot-more

mimy.acc

selber.self

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.23 (1603-1679)


b. Hanhe

äris

intenot

längrelonger

sigrefl.acc

själv(him)self

//

*han*he.nom

själv.(him)self

‘He is not himself any more.’

In addition, there is considerable case variation (both individual and for-mal/informal) in comparative predicative phrases introduced by än ‘than’ and som‘as, like’, accusative in general being more common than nominative in the spokenlanguage, whereas nominative is more typical of the written language (Teleman etal. 1999, Vol. 3:672ff.):

(52) a. Honshe

äris

störretaller

änthan

jag/mig.I/me

‘She is taller than me.’b. Karlar

chapssomas

vi/osswe/us

gördo

intenot

så.so

‘Chaps like us don’t behave like that.’c. Hon

shekancan

intenot

älskalove

ena

karlchap

somlike

jag/mig.I/me

‘She can’t love a chap like me.’

Among the case-poor Germanic languages, Afrikaans is the most persistent pred-icate nominative language. As for the case-rich Germanic languages, predicativeDPs in finite clauses in Icelandic, Faroese and standard German are, to my knowl-edge, exclusively nominative.19 Interestingly, however, some German varieties dis-play at least some ‘accusative tendencies’. Thus, many speakers of the Ruhr-dialectshow the following pattern, with nominative pronominal predicative DPs but(preferably) accusative full predicative DPs:20

(53) a. Datthat

binam

ich.I

//

*Dat*that

isis

mich.me

b. Datthat

isis

er.he

//

*Dat*that

isis

ihn.him

c. Datthat

isis

‘na

feinenfine

Kerl.chap.acc

. I have not been able to get reliable information on Yiddish in this respect.

. René Schiering, p.c. For full DPs, as in the c-example, the nominative is also grammatical,but it is clearly marked as standard, non-colloquial, at least to fluent speakers of the dialect. Thiskind of Nom/Acc variation is probably found in other parts of the Low German area, but I haveno accurate information on its geographical distribution. Much descriptive work on the syntaxof German dialects or languages remains to be done.

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.24 (1679-1717)


In contrast, speakers of Alemannic varieties accept some accusative pronominalpredicates, in many of the same environments as discussed above for Dutch andWest Frisian:21

(54) a. Desthat

ischis

(e)nhim.acc

doch.though

‘That is him, for sure.’b. Wenn

ifichI

dichyou.acc

wär, . . .were

c. IchI

willwant

dichyou.acc

si,be,

duyou

kaschcan

michme.acc

si.be

The historical development of the English predicative construction is usually de-scribed in categorically distinct steps, as demonstrated below for the first personsingular (based on the overview in Quinn 2005a:243):

(55) a. Old English and Early Middle English: ic hit eom (‘I it am’)b. Late Middle English (Chaucer): it am Ic. Early Modern English: it is Id. Modern English: it is me

What this describes is actually several changes. It is a well-known fact that eachchange of this sort is typically a gradual process, spreading to different construc-tions or grammatical domains at different times (see, e.g., the general discussionin Fischer et al. 2000:17ff.). The Germanic predicate Nom/Acc variation, so brieflyoutlined here, suggests that this also applied to the English changes sketched above.The fact that we know only very few details of this history (see Visser 1963:236ff.,244ff.) is revealing about the limitations and challenges of diachronic linguis-tics. Conversely, however, there is also an important positive lesson to be learnedabout language change from the present study: Comparative studies of contempo-rary language variation are likely to yield important, complementary informationabout the paths and the nature of language change that is not readily accessiblethrough historical studies.

More importantly for our present purposes, the Germanic predicate Nom/Accvariation suggests that the structural approach to the Nom/Acc alternation gener-ally adopted in generative syntax must be abandoned. I discuss this issue in more

. Markus Benzinger, p.c. (his Alemannic variety is spoken in the southernmost Baden regionin Germany). As indicated, the pronoun is weak and the verb has focal stress in the a-example,much as in corresponding examples in Dutch and West-Frisian (Ellen Brandner, p.c.). Also asin Dutch and West Frisian, first and second person predicative DPs must be nominative in plainindicative, non-modal examples of this sort. – Philipp Conzett tells me that only strong nomi-native forms are possible in examples like (54a) in his Alemannic variety (Graubünden), wheresaccusative is the only option for him in conditionals and modal constructions, like (54b, c).

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.25 (1717-1772)


detail in Section 4, after having discussed the nature of the relational (‘structural’)cases in Section 3.

. Relational case

On the relational view of the so-called ‘structural’ cases, formulated in the SiblingCorrelation in (5) above, relational accusative is dependent on nominative beingpresent or active in the structure. That is, the true correlation Burzio (1986) triedto capture with his famous generalization is not between the external role and theinternal case of a predicate but between its cases, nominative versus accusative orcase1 versus case2. Hence, it is not surprising that nominative is independent ofthe other cases and also the simple case, normally used for DPs in lists and otherisolated DPs, as well as for DPs in simple structures (unergative, unaccusative, and,in many languages, predicative), whereas other cases are typically added in morecomplex structures (transitive, ditransitive).

Relational case assignment is seemingly a problem for the minimalist bottom-to-top approach to the derivation. The standard minimalist assumption (Chom-sky 1995, 2000, 2002) is that accusative is assigned lower in the structure thannominative and hence also before nominative is assigned. However, as we haveseen, accusative is dependent on nominative being present or active in the struc-ture, that is, the accusative would have to ‘know’ that nominative is going tobe assigned later on in the derivation. In other words, this forces us to assumelook-ahead, a fatal problem in minimalist approaches.22

As I have argued elsewhere, however, the problem vanishes under theLow Nominative Hypothesis (Sigurðsson 2004b, 2006), under which it isactually the nominative argument that is merged lower and earlier than theaccusative argument:

(56) Nom is the first case, Acc is the second case:a. V [θ1] (and θ1→ Nom in morphology)b. θ2 [V θ1] (and θ2→ Acc in morphology)

. Since the computation is entirely ‘dumb’ or ‘blind’, that is, does not ‘plan’ things (eventhough we can plan our sentences, which is a different but not an irrelevant issue). ‘TheNom/Acc problem’ might seem to be avoidable in a top-to-bottom approach (e.g. Phillips 2003;Bianchi and Chesi 2005), but, in fact, such an approach fares no better than a bottom-to-topapproach. Thus, in a German clause beginning with a DP like Peter, the derivation would haveto look ahead ‘downwards’, as it were, in order to know if the DP is an experiencer dative, avocative (Peter!), a psych-accusative or a nominative.

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.26 (1772-1845)


As indicated, I assume that the cases temselves are not realized or assigned until inmorphology.

Later on in the derivation the low argument, θ1(→ Nom), is raised across thehigh argument, θ2(→ Acc), to mach subject number and subject person, yielding thenormal surfcase order Nom-Acc:

(57) θ2(→ Acc) [V θ1(→ Nom)] → nom . . .. acc . . .. NOM

The reason why the low argument, θ1(→ Nom), can raise across the higher argument,θ2(→ Acc), is that the features of the latter get fully matched or interpreted locally,thus becoming syntactically inactive and invisible to external probing (“frozenin place”, as Chomsky puts it (2001:6)). Quirky systems differ from accusativesystems in this respect, arguably because quirky arguments do not get fully inter-preted locally, but must match some relatively high feature in the clausal structure.Hence, the quirky argument raises, leaving the nominative behind:

(58) θ2(→ Quirky) [V θ1(→ Nom)] → quirky . . .. QUIRKY . . . nom

Thus, the nominative argument is the first argument merged in both accusativeand quirky systems, but its order with respect to the second argument is reversed.

The evidence I have presented in favor of this approach is strong, but it is alsocomplicated, so I will not go through it here (but see Sigurðsson 2004b, 2006).What matters for our purposes is only the following two points:

– The cases themselves are not assigned until in morphology (an issue I’ll returnto)

– It is the argument that is merged first that gets assigned nominative in mor-phology

On the present relational approach, regular nominatives and accusatives functionas to make a morphologically visible distinction between distinct arguments ofa predicate. Hence, they do not depend on or match functional categories in theclausal structure and are thus truly non-structural, although they morphologicallyinterpret or translate structural correlations. ‘Nominative’ and ‘accusative’ are sim-ply traditional labels for the notions case1 versus case2. It follows, that there is noinherent connection between nominative and finite Tense, contra the most popu-lar generative view on case (Chomsky 1981, etc.). That is, the alleged Nom-FiniteT Connection is illusory (Sigurðsson 1989 and subsequent work; but see, e.g.,Nomura 2005 for a somewhat different interpretation).

On this account, one would expect to find nominatives in infinitival construc-tions. However, they cannot be expected to show up in the subject position of PROinfinitives in, e.g., the Germanic languages, since that position may never be lexi-calized, irrespective of case (Sigurðsson 1991). Thus, we have to look for differentoccurrences of infinitival nominatives. Three types come into question:

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.27 (1845-1923)


A. Nominative subjects in infinitives that do allow subject lexicalizationB. Nominative objects (quirky)C. Nominative predicative DPs

Icelandic is renowned for having certain exceptional nominative case-marking in-finitives, (Sigurðsson 1989, 1996 and many others). Such infinitives allow nomi-native subjects (raised to the infinitival subject position or not), as well as regularnominative predicative DPs and quirky nominative objects:

(59) a. Mérme.dat

hefurhas

alltafalways

virstseemed

[ÓlafurOlaf.nom

verabe

snjallclever

maður].man.nom

‘To me, Olaf has always seemed to be a clever man.’b. Henni

her.datvirtistseemed

[hafahave

leiðstfound-boring

strákarnir].boys.the.nom

‘She seemed to have been bored by the boys.’c. Það

therevirtustseemed

[hafahave

veriðbeen

veiddircaught

fjórirfour

laxar].salmon.nom

‘There seemed to have been four salmon caught.’d. Það

therevoruwere

taldirbelieved

[hafahave

veriðbeen

veiddircaught

fjórirfour

laxar].salmon.nom

‘People believed there to have been four salmon caught.’

As we would expect, Icelandic also has nominative objects and predicative DPs inPRO infinitives:

(60) a. Húnshe

vonaðisthoped

tilfor

[aðto

leiðastfind-boring

ekkinot

allirall.nom

stákarnir].boys.the.nom

‘She hoped not to find all the boys boring.’b. Hún

shereynditried

[aðto

verðabecome

prestur].priest.nom

‘She tried to become a priest.’c. [Að

toviljawant

verðabecome

kennari]!teacher.nom

‘To want to become a teacher!’

Moreover, Icelandic has nominative floating quantifiers in infinitives, apparentlyagreeing in case with nominative PRO (Sigurðsson 1991):

(61) a. Strákarnirboys.the.nom

vonuðusthoped

tilfor

[aðto

verðabe

ekkinot

allirall.nom

reknir].fired

‘The boys hoped not to all get fired.’b. Strákana

boys.the.acclangaðilonged

ekkinot

tilfor

[aðto

verðabe

allirall.nom

reknir].fired

‘The boys did not want to all get fired.’

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.28 (1923-1994)


As seen in (61b), with an accusative matrix subject, the matrix subject is not thesource of the nominative of the quantifier. Rather, the source must be internalto the infinitive. – In all the examples in (59)–(61), accusative, dative or genitiveforms instead of the bold-faced nominatives would be ungrammatical.23

Most of this evidence is unavailable in the other Germanic languages, asthey lack the constructions bearing on the issue. One construction they all have,however, is the predicative DP construction. As we would expect, the predicatenominative languages generally apply nominative marking of predicative DPs in(non-ECM) infinitives as well as in finite clauses. This is illustrated below forstandard German, Swedish and Afrikaans, in that order:

(62) a. Esit

reichtsuffices

mirme

ich/*michI/*me

zuto

sein.be

‘It is good enough for me to be myself.’b. Det

iträckersuffices

förfor

migme

attto

varabe

jag/*mig.I/*me

c. Ditit

isis

virfor

myme

genoegenough

omcomp

ek/*myI/*me

teto

wees.be

Interestingly, though, Dutch and Alemannic shift from nominative to accusative in(at least many) infinitives and so does e.g. Italian, as illustrated below, in that order:

(63) a. Datit

benis

ik.I

b. Hetit

isis

nietnot

makkelijkeasy

omcomp

?mij/*ik?me/*I

teto

zijn.be

‘It is not easy to be me.’

(64) a. Desit

binis

ich.I

b. Esit

langtsuffices

mer,me.dat,

michme.acc

selberself

zumto

si.24

be

(65) a. Sonoam

io/*me.I/*me

b. Mime.cl

bastasuffices

esserebe

meme.acc

stessa.myself.fem.sg.

. Whereas the floating quantifier shows up in an oblique form if the infinitive hasoblique PRO.

. Baden Alemannic, Markus Benzinger, p.c. However, Benzinger also tells me that the nomi-native is possible or even preferred for at least some infinitival predicative DPs, perhaps becauseof a pressure from standard German. – Graubünden Alemannic also shifts from nominative toaccusative in infinitives like (65d) (Philipp Conzett, p.c.).

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.29 (1994-2049)


All this shows two things. First, there is no inherent or general Nom-Finite T Con-nection, that is, the common assumption that nominative case is dependent onfinite tense is plainly wrong, based on the absence of evidence in case-poor predi-cate accusative languages like English. Second, however, languages that are devel-oping from being predicate nominative into being predicate accusative languagesof the English type, may opt for accusatives instead of nominatives in infinitivalconstructions, much as they may opt for accusatives specifically for weak thirdperson pronouns (Dutch, West Frisian, Baden Alemannic), for reflexive predica-tive DPs (Swedish, Graubünden Alemannic), or even for full DPs as opposed topronominal DPs (Ruhr-German). Thus, again, we see interesting evidence that ahistorical change may develop along different paths in different languages (and indifferent varieties of ‘a’ language).

The inherent cases are ‘semantically linked’ (Chomsky 2002:113). As men-tioned above, however, the semantics in question is not lexical semantics butsyntactic semantics, as it were. That is, the inherent cases generally seem not tomatch or link to features of lexical roots, but rather to syntactic-semantic fea-tures, for instance aspectual features and voice features. Plausibly, such featuresare syntactic, being interpreted at both the interfaces, that is, semantically at theconceptual interface but by case in the morphology (‘deep PF’) of case languages.If so, the inherent cases are ‘structural’, in a sense. In contrast, the relational or so-called ‘structural cases’ are not semantically linked, and they are not structurallylinked either, that is, they are not related to or licensed by functional categorieslike Tense, Aspect, etc. Their sole function is to be morphologically distinctive(Sigurðsson 2003), that is, they make an overt, morphological distinction betweendistinct event participants (‘arguments’), participant1 versus participant2, thefirst being assigned case1 and the latter being assigned case2 – in morphology. Inother words, the ‘virtue’ of the relational cases is that of making an overt distinc-tion, for communicative and processing purposes, and not that of expressing anysemantic-syntactic functions.

. The nature of the predicative Nom/Acc variation

Predicative constructions involve only one event participant (viewed from two an-gles, though). Thus, it is not surprising that many languages do not make any casedistinction between DPs and their predicates. Instead, case-rich languages com-monly apply case agreement in predicative constructions. This is illustrated belowfor the case-rich Germanic languages; the underlined subjects are all nominative:

(66) a. Erhe

istis

eina

gutergood

Schülerstudent.nom

//

*einen*acc

guten Schüler. German

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.30 (2049-2104)


b. Erhe

izis

aa

gutergood

shilerstudent.nom

//

*a*acc

gutn shiler. Yiddish

c. Hannhe

eris

góðurgood

nemandistudent.nom

//

*góðan*acc

nemanda. Icelandic

d. Hannhe

eris

eina

góðurgood

stúdenturstudent.nom

//

*góðan*acc

stúdent. Faroese

Ancient Greek and Latin also applied this case copying or case agreement strategyfor predicates, whereas some other case-rich languages, including Finnish, Polishand Russian, have a mixed system of case agreement versus a special predicativecase (partitive in Finnish, instrumental in Polish and Russian), depending on fac-tors that I cannot go into here (but see, e.g., Karlsson 1985:98ff.; Comrie 1997;Bailyn 2001).

Predicative case agreement is not confined to finite clauses or to nominativecase. This is partly illustrated for Icelandic below:

(67) a. Húnshe.nom

eris

góðurgood

prestur.priest.nom

‘She is a good priest.’b. Við

weteljumconsider

hanaher.acc

verabe

góðangood

prest.priest.acc

‘We consider her to be a good priest.’c. Við

wekynntumstgot-to-know

henniher.dat

semas

góðumgood

presti.priest.dat

‘We got to know her as a good priest.’d. Við

weleituðumseeked

tilto

hennarher.gen

semas

góðsgood

prests.priest.gen

‘We confided in her as a good priest.’

English type accusative marking of predicatative DPs seems to be highly excep-tional in case-rich languages (i.e., languages that have some case-marking of fullDPs). That is, case-rich languages quite generally seem to apply either case agree-ment in predicative constructions or a special predicative case. As we saw above,Ruhr-German and Allemannic varieties are exceptional in this respect, and Stan-dard Arabic is another language with case-marking of full DPs and (some) Accpredicates (see Benmamoun 2000:43), but these are the only exceptions I am awareof. Thus it seems that we can formulate the following tentative ‘Greenbergian type’universal (for finite constructions):

(68) With much greater than chance frequency, case-rich languages do not assignaccusative case to predicative nominals

On the present approach to the relational cases, this tendency is a rather naturalone. Assigning object case to predicative DPs is incompatible with the basic func-

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.31 (2104-2163)


tion of the relational cases, that of making a morphological distinction betweendistinct arguments or event participants. In many case-rich languages, not onlypredicative DPs but also adjectival and participial predicates agree in case, as illus-trated below for Icelandic (the same was true of, e.g., Latin and Ancient Greek):

(69) a. Húnshe.nom

varwas

snjöllclever.nom.f.sg

//

kosin.elected.nom.f.sg

b. Viðwe

töldumblieved

hanaher.acc

hafahave

veriðbeen

snjallaclever.acc.f.sg

//

kosna.elected.acc.f.sg

In a system like this, going from case agreement in predicates to general ac-cusative marking would be a major change.25 In case-poor languages, on the otherhand, accusative marking of predicates is a peripheral phenomenon, since it af-fects only a fraction of all predicates, namely those rare predicates that consist ofpronominal DPs/NPs.

On the popular assumption that the relational cases are syntactic elements orfeatures (Chomsky 2000, etc.), predicative accusatives in languages like Englishand Danish are left unaccounted for. The literature on the English predicativeconstruction is of course enormously voluminous (see Quinn 2005a), but it istruly amazing how little interest the Germanic predicative Nom/Acc variation hasraised within generative approaches to case (with Maling & Sprouse 1995 as anexception). As mentioned in Section 2.3, the reason is presumably that syntacticapproaches to the relational cases have little or nothing to say about this variation:it is unexpected and mysterious under such approaches. Assuming that predica-tive accusatives in English, Danish, Norwegian and North Frisian, is ‘default’ (seesome of the references in Maling & Sprouse 1995:167 and the discussion in Schütze2001) is not insightful either. If accusative is ‘default’ in the English type of lan-guages, then nominative must in some sense be ‘non-default’ or marked, whichwould imply that these languages should be unique among the languages of theworld in assigning a marked case to subjects (in finite clauses) across the board.

Under the present, morphological approach to the relational cases, on theother hand, the Germanic predicative case variation results from a simple and a‘reasonable’ historical change. Languages like English and Danish have extendedthe distinctive function of the relational cases from arguments to DPs in general,that is, case-marking in these languages is not mysterius or due to grammaticaldiseases but well-behaved, at least by and large. We can describe this informally interms of the following rules or directives of the two different grammars, stated in

. In West-Germanic varieties, in contrast, only predicate DPs get case marked. Thus, the abovementioned accusative marking of certain predicates in Ruhr-Germanic and Alemannic varietiesis not nearly as pervasive a change as accusative marking of predicates would be in a languagelike Icelandic.

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.32 (2163-2209)


(70). I refer to the grammars as ‘Grammar q’ and ‘Grammar w’, abstaining fromcalling them ‘Grammar 1’ and ‘Grammar 2’, because, as we have seen, evidencefrom related languages suggests that these stages have been separated by severalintermediate grammars, whith the change from q to w only partly completed:

(70) Grammar q: Apply Nom/Acc to morphologically distinguish between thefirst and the second argument of a predicate.

Grammar w: Apply Nom/Acc to morphologically distinguish between thefirst and the second DP of a predicate.

However, Grammar w is evidently not the end-point of the development, as sug-gested by the fact that many speakers of both English and Danish tend to make acase distinction between adjacent DPs in general, even conjoined DPs, as in himand I, ham og jeg, etc. (e.g. in the subject function as in Him and I went home, Hamog jeg gik hjem).26 Even though this is generally considered to be substandard inEnglish (see Huddleston & Pullum, et al. 2002:462ff.), it is evidently frequent inthe spoken language (Quinn 2005a, 2005b).

Many factors interact with the basic function of the relational cases in theEnglish/Danish type of languages (see Henry 1995 and Quinn 2005b on BelfastEnglish versus New Zealand English).27 Thus, as studied by Quinn, the relevantfactors in English include, e.g., the distinction between weak and strong pronounsand the phonological ‘heaviness’ or ‘robustness’ of strong pronouns. Similarly,many factors, above all the inherent cases, interfere with the basic function ofthe relational cases in the Icelandic type of languages. The fundamental differ-ence between these language types with respect to the distribution of Nom/Acc,however, is that the English/Danish type has extended the distinctive function ofthe relational cases from arguments to DPs as such.

. See Johannessen (1998) for a discussion of this phenomenon in some other languages,including Norwegian.

. Conjoined subjects like Him and I have agreement correlates in Belfast English as describedby Henry (1995), in contrast to New Zealand English, where conjoined subjects always seemto trigger plain plural agreement, regardless of case and other factors (Him and I have . . . , Meand him have . . . , etc.; Heidi Quinn, p.c.). This fact about New Zealand English suggests thatthe nominative-verb agreement connection observed in many languages is a more superficialphonomenon than often assumed (see Sigurðsson 2003 and subsequent work on this connectionin Icelandic morphology).

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.33 (2209-2270)


. Conclusion

The ‘language’ of Narrow Syntax is understandable but foreign to both the in-terfaces (or, if one likes, to the articulatory and conceptual organs). That is, theinterfaces do not ‘speak Narrow Syntax’, they interpret it and translate it into their‘own languages’.

This is perhaps not so clear if one only considers morphologically poor lan-guages like English. However, a simple comparison of English with a language likeIcelandic immediately reveals that some languages have an extremely complexmorphophonological component that is largely absent from analytic languageslike English. Consider the following rather simple examples (N, A = nominative,accusative, m.pl = masculine plural):

(71) a. AllirallN.m.pl

þessirtheseN.m.pl

dómararjudgesN.m.pl

voruwereind.3pl

taldirbelievedN.m.pl

verðabeN.m.pl

útnefndir.nominated

‘All these judges were believed to be nominated.’b. Við

weN

töldumbelievedind.1pl

allaallA.m.pl

þessatheseA.m.pl

dómarajudgesA.m.pl

verðabeA.m.pl

útnefnda.nominated

‘We believed all these judges to be nominated.’

Icelandic and English are relatively closely related languages, but the morpholog-ical differences between them are tremendous. Some of these differences can beseen as resulting from Icelandic being explicit about categories that English is ‘cool’or reluctant about expressing. Thus, English arguably has a syntactic-semanticdistinction between ‘subjunctive’ and ‘indicative’, although it mostly keeps quietabout it in its morphology. In contrast, it is not clear that the abundant case andagreement morphology of Icelandic is telling us something about the syntactic-semantic structure of both languages that English is being silent about. Icelandicexplicitly distinguishes between full DP arguments by relational case-marking,as in (71), but this distinction is evidently an unnecessary extravagance.28 Theagreement in case, number and gender is even less meaningful, that is, it ar-guably arises through feature copying processes in (‘deep PF’) morphology (asargued in Sigurðsson 2004b). From a linguistic point of view, this copying is mor-

. Or else we should find such marking in all languages. It has sometimes been suggestedthat Icelandic case-marking is ‘decorative’, as opposed to case-marking in related languages.This is rather obviously wrong. Case-marking in all languages is ‘useful’ or functional from aparsing/processing point of view – but unnecessary or ‘decorative’ from a strictly linguistic orgrammatical point of view (i.e. it is not indispensable for a message to get through). Icelandiccase-marking is clearly no different from e.g. Latin, German or Russian case-marking in thisrespect.

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.34 (2270-2327)


phological noise. It probably has the socially important function of signaling thegroup identity of speakers, but it has arguably no linguistic function. If it had, wewould expect English to be seriously inferior to Icelandic. That does not seem tobe the case.

The present study highlights the fact that morphological case interprets syntaxin its own terms or its own ‘language’ rather than directly expressing or mirroringit. Thus, to mention only one of the many case contrasts we have seen, there isarguably no Narrow Syntax difference between, e.g., Danish Det er os (‘it is us’)and its Swedish translation Det är vi (‘it is we’). Rather, exactly the same syntacticstructure gets different interpretations in morphology.

We need to return to the traditional view that case is a morphological (PF)phenomenon.29 Both relational and inherent cases are morphological translationsor interpertations of syntactic structures, but they are not present or active in syn-tax themselves, that is, there are no Nom or Acc features in syntax (or Dat orGen, for that matter). All the alleged syntactic effects of relational or ‘structural’case stem from matching of interpretable features, typically person and number inaccusative systems (Sigurðsson 2003, 2004a, 2004b, etc.) but often aspect, tenseor focus in different systems (cf. Miyagawa to appear). Nom and Acc as such arenonexistent in Narrow Syntax. It is thus no wonder that they are invisible to thesemantic interface (as discussed by Chomsky 2000, 2001, 2002, etc.).

This entails that the overt morphophonological forms of languages are onlyindirect and incomplete reflections of the language faculty (reminding of Plato’scave allegory). The inevitable conclusion, unwelcome and distressing as it may be,is that there is no such thing as ‘plain, overt evidence’ in syntax. The study of syntaxis a much more difficult task than most of us would like to believe, demandingthat we compare a range of languages and ‘listen’ to the semantic intepretation ofsyntax no less than to its widely differing morphological translations. If a categorydoes not get a semantic interpretation in any of the languages we study, then itis probably just a language specific morphological (PF) category, with the basicfunction of making an overt distinction, a noble and useful enough function assuch. Like a phoneme.

References

Allan, R., Holmes, P., & Lundskær-Nielsen, T. (1995). Danish: A comprehensive grammar.London: Routledge.

. See also Sigurðsson 2004b for a parallel conclusion regarding gender and number, argu-ing that we must distinguish syntactic-semantic gender and number from their morphologicaltranslations, that is, from formal gender and number.

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.35 (2327-2444)


Allen, C. L. (1996). A change in structural case-marking in Early Middle English. In H.Thráinsson, S. D. Epstein, & S. Peter (Eds.), Studies in Comparative Germanic Syntax II(pp. 3–20). Dordrecht: Kluwer.

Anagnostopoulou, E. (2003). The Syntax of Ditransitives: Evidence from clitics. Berlin: Mouton.Bailyn, J. (2001). The syntax of Slavic predicate case. ZAS Papers in Linguistics, 22, 1–23.Barðdal, J. (2001). Case in Icelandic – A Synchronic, Diachronic and Comparative Approach. Lund:

Department of Scandinavian Languages.Barðdal, J. (2006). Construction-specific properties of syntactic subjects in Icelandic and

German. Cognitive Linguistics, 17, 39–106.Benmamoun, E. (2000). The Structure of Functional Categories: A comparative study of Arabic

dialects. Oxford: OUP.Bianchi, V. & Chesi, C. (2005). Phases, strong islands, and computational nesting. Ms.,

University of Siena (presented at GLOW 28, 2005, Geneva).Boeckx, C. (2000). Quirky agreement. Studia Linguistica, 54, 354–380.Bonet, E. (1991). Morphology after Syntax: Pronominal clitics in Romance languages. PhD

dissertation, MIT.Burzio, L. (1986). Italian Syntax. Dordrecht: Reidel.Burzio, L. (2000). Anatomy of a generalization. In E. Reuland (Ed.), Argument and Case:

Explaining Burzio’s generalization (pp. 195–240). Amsterdam: John Benjamins.Chomsky, N. (1981). Lectures on Government and Binding. Dordrecht: Foris.Chomsky, N. (1995). The Minimalist Program. Cambridge MA: The MIT Press.Chomsky, N. (2000). Minimalist inquiries: The framework. In R. Martin, D. Michaels, & J.

Uriagareka (Eds.), Step by Step: Essays on minimalist syntax in honor of Howard Lasnik (pp.89–155). Cambridge MA: The MIT Press.

Chomsky, N. (2001). Derivation by phase. In M. Kenstowicz (Ed.), Ken Hale: A life in language(pp. 1–52). Cambridge MA: The MIT Press.

Chomsky, N. (2002). On Nature and Language. (Ed. by A. Belletti & L. Rizzi). Cambridge: CUP.Cinque, G. (1999). Adverbs and Functional Heads: A cross-linguistic perspective. New York NY:

OUP.Comrie, B. (1997). The typology of predicate case marking. In J. Bybee, J. Haiman, &

S. A. Thompson (Eds.), Essays on Language Function and Language Type (pp. 39–50).Amsterdam: John Benjamins.

Eklund, B. (1982). Jag såg han: Om objektsformer av personliga pronomen i nordsvenskan [Isaw he: On object forms of personal pronouns in Northern Swedish]. In C. Elert & S. Fries(Eds.), Nordsvenska: språkdrag i övre Norrlands tätorter (pp. 161–173). Stockholm: Almquist& Wiksell.

Eythórsson, Th. (2000). Dative versus nominative: Changes in quirky subjects in Icelandic. LeedsWorking Papers in Linguistics, 8, 27–44.

Eythórsson, Th. & Barðdal, J. (2005). Oblique subjects: A common Germanic inheritance.Language, 81, 824–881.

Eythórsson, Th. & Jónsson, J. (2003). The case of subject in Faroese. Working Papers inScandinavian Syntax, 72, 207–231.

Faarlund, J., Lie, S., & Vannebo, K. (1997). Norsk referensegrammatikk (Norwegian ReferenceGrammar). Oslo: Universitetsforlaget.

Fischer, O., van Kemenade, A., Koopman, W., & van der Wurf, W. (2000). The Syntax of EarlyEnglish. Cambridge: CUP.

Fossum, S. (1995). Kasusvakling ved personlige pronomen (Case Variation of PersonalPronouns). MA thesis, University of Oslo.

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.36 (2444-2557)


van Gelderen, E. (1997). Verbal Agreement and the Grammar behind its Breakdown: Minimalistfeature checking. Tübingen: Max Niemeyer.

Haider, H. (1984). The case of German. In J. Toman (Ed.), Studies in German Grammar (pp.65–101). Dordrecht: Foris.

Haider, H. (2000). The license to license. In E. Reuland (Ed.), Argument and Case: ExplainingBurzio’s generalization (pp. 31–54). Amsterdam: John Benjamins.

Haider, H. (2001). How to stay accusative in insular Germanic. Working Papers in ScandinavianSyntax, 68, 1–14.

Henry, A. (1995). Belfast English and Standard English: Dialect variation and parameter setting.New York NY: OUP.

Holmberg, A. (1986). The distribution of case-neutral pronouns in a Swedish dialect. In Ö.Dahl & A. Holmberg (Eds.), Scandinavian Syntax (pp. 88–100). Institute of Linguistics,University of Stockholm.

Huddleston, R., & Pullum, G. et al. (2002). The Cambridge Grammar of the English Language.Cambridge: CUP.

Johannessen, J. (1998). Coordination. New York NY: OUP.Jónsson, J. (1998). Sagnir með aukafallsfrumlagi (Verbs with a quirky subject). Íslenskt mál og

almenn málfræði, 19–20, 11–43.Jónsson, J. (2003). Not so quirky: On subject case in Icelandic. In E. Brandner & H. Zinzmeister

(Eds.), New Perspectives on Case Theory (pp. 127–163). Stanford CA: CSLI Publications.Kainhofer, J. (2002). Monadische Akkusativ-Subjekt-Konstruktionen im isländischem

(Monadic Accusative-subject Constructionen in Icelandic). Magister thesis, University ofSalzburg.

Karlsson, F. (1985). Finsk Grammatik (Finnish Grammar). (6th edition). Helsinki: SuomalaisenKirjallisuuden Seura.

Kratzer, A. (1996). Serving the external argument from the verb. In J. Rooryck & L. Zaring (Eds.),Phrase Structure and the Lexicon (pp. 109–137). Dordrecht: Kluwer.

Lehman, T. (1993). A Grammar of Modern Tamil. Pondicherry: Pondicherry Institute ofLinguistics and Culture.

Levander, L. (1909). Älvdalsmålet i Dalarna: Ordböjning ock syntax. Stockholm.Maling, J. (1993). Of nominative and accusative: The hierarchical assignment of grammatical

case in Finnish. In A. Holmberg & U. Nikanne (Eds.), Case and other Functional Categoriesin Finnish Syntax (pp. 49–74). Berlin: Mouton de Gruyter.

Maling, J. (2002a). Icelandic verbs with dative objects. Working Papers in Scandinavian Syntax,70, 1–59.

Maling, J. (2002b). Það rignir þágufalli á Íslandi. Verbs with dative objects in Icelandic. Íslensktmál og almenn málfræði, 24, 31–105.

Maling, J. & Sprouse, R. (1995). Structural case, specifier-head relations, and the case ofpredicate NPs. In H. Haider, S. Olsen, & S. Vikner (Eds.), Studies in Comparative GermanicSyntax (pp. 167–186). Dordrecht: Kluwer.

Maling, J. & Sigurjónsdóttir, S. (2002). The new impersonal construction in Icelandic. TheJournal of Comparative Germanic Lnguistics, 5, 97–142.

McFadden, T. (2004). The Position of Morphological Case in the Derivation: A study on thesyntax-morphology interface. PhD dissertation, Univeristy of Pennsylvania.

McIntyre, A. (2005). The interpretation of German datives and English have. To appear in D.Hole, A. Meinunger, & W. Abraham (Eds.), Datives and Other Cases: Between argumentstructure and event structure. Amsterdam: John Benjamins.

Miyagawa, S. (To appear). On the EPP. MITWPL: Proceedings of the EPP/Phase Workshop.

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.37 (2557-2681)


Nomura, M. (2005). Nominative Case and AGREE(ment). PhD dissertation, University ofConnecticut.

Ottosson, K. (1988). A feature based approach to thematic roles. In V. Rosén (Ed.), Papers fromthe Tenth Scandinavian Conference of Linguistics, Vol. 2 (pp. 136–150). Bergen.

Perlmutter, D. (1971). Deep and Surface Structure Constraints in Syntax. New York NY: Holt,Rinehart and Winston.

Phillips, C. (2003). Linear order and constituency. Linguistic Inquiry, 34, 37–90.Pylkkänen, L. (1999). Causation and external arguments. Papers from the UPenn/MIT Round-

table on the Lexicon, MITWPL, 35, 161–183.Quinn, H. (2005a). The Distribution of Pronoun Case Forms in English. Amsterdam: John

Benjamins.Quinn, H. (2005b). Licensing, blocking, and English pronoun case. Paper presented at the

International Conference of Historical Linguistics in Madison, August 2005.Reinhammar, M. (1973). Om dativ i svenska och norska dialekter. 1. Dativ vid verb (On dative in

Swedish and Norwegian dialects. 1. Dative with verbs). Uppsala.Schütze, C. (2001). On the nature of default case. Syntax, 4, 205–238.Sigurðsson, H. (1989). Verbal Syntax and Case in Icelandic. Lund University. (Republished 1992

in Reykjavik: Institute of Linguistics).Sigurðsson, H. (1991). Icelandic case-marked PRO and the licensing of lexical arguments.

Natural Language and Linguistic Theory, 9, 327–363.Sigurðsson, H. (1996). Icelandic finite verb agreement. Working Papers in Scandinavian Syntax,

57, 1–46.Sigurðsson, H. (2003). Case: Abstract vs morphological. In E. Brandner & H. Zinzmeister (Eds.),

New Perspectives on Case Theory (pp. 223–268). Stanford CA: CSLI Publications.Sigurðsson, H. (2004a). The syntax of person, tense, and speech features. Italian Journal of

Linguistics / Rivista di Linguistica, 16, 219–251 (Eds. V. Bianchi & K. Safir).Sigurðsson, H. (2004b). Agree in syntax, agreement in signs. Working Papers in Scandinavian

Syntax, 74, 1–42 (A revised version to appear 2006 in C. Boeckx (Ed.), Agreement Systems(pp. 201–237). Amsterdam: John Benjamins).

Sigurðsson, H. (2004c). Agree and agreement: Evidence from Germanic. In W. Abraham (Ed.),Focus on Germanic Typology (pp. 61–103). Berlin: Akademie Verlag.

Sigurðsson, H. (2004d). Meaningful silence, meaningless sounds. Linguistic Variation Yearbook,4, 235–259.

Sigurðsson, H. (2006). The nominative puzzle and the low nominative hypothesis. LinguisticInquiry, 37, 289–308.

Sigurjónsdóttir, S. & Maling, J. (2001). Það var hrint mér á leiðinni í skólann: Þolmynd eða ekkiþolmynd? (It was pushed me on the way to school: Passive or not?). Íslenskt mál og almennmálfræði, 23, 123–180.

Smith, H. (1996). Restrictiveness in Case Theory. Cambridge: CUP.Svenonius, P. (2005). The nanosyntax of the Icelandic passive. Paper presented at the Lund

Grammar Colloquium, May 26, 2005.Telemann, U. (2001). Mig vara du (Me be you). Språkvård, 1–2001, 26–27.Telemann, U., Hellberg, S., & Andersson, E. (1999). Svenska Akademiens grammatik, 1–4.

Stockholm: Svenska Akademien.Thráinsson, H. (1979). On Complementation in Icelandic. New York NY: Garland.Thráinsson, H., Petersen, H., Jacobsen, J., & Hansen, Z. (2004). Faroese: An overview and

reference grammar. Tórshavn: Foroya Frodskaparfelag.

JB[v.20020404] Prn:25/09/2006; 12:01 F: LA9701.tex / p.38 (2681-2725)


Tyroller, H. (2003). Grammatische Beschreibung des Zimbrischen von Lusern. Suttgart: FranzSteiner Verlag.

Visser, F. (1963). An Historical Syntax of the English Language. Part 1: Syntactic Units with OneVerb. Leiden: E. J. Brill.

Wiese, H. & Maling, J. (2005). Beers, kaffi, and Schnaps: Different grammatical options forrestaurant talk coercions in three Germanic languages. Journal of Germanic Linguistics, 17,1–38.

Woolford, E. (2003). Burzio’s Generalization, markedness, and locality constraints on nomina-tive objects. In E. Brandner & H. Zinzmeister (Eds.), New Perspectives on Case Theory (pp.301–329). Stanford CA: CSLI Publications.

Wunderlich, D. (2003). Optimal case patterns: German and Icelandic compared. In E. Brandner& H. Zinzmeister (Eds.), New Perspectives on Case Theory (pp. 331–367). Stanford CA: CSLIPublications.

Yip, M., Maling, J., & Jackendoff, R. (1987). Case in Tiers. Language, 63, 217–250.Zaenen, A. & Maling, J. (1984). Unaccusative, Passive and Quirky Case. Proceedings of the Third

West Coast Conference on Formal Linguistics, 317–329. Stanford University.Zaenen, A., Maling, J., & Thráinsson, H. (1985). Case and grammatical functions: The Icelandic

passive. Natural Language and Linguistic Theory, 3, 441–483.

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.1 (48-132)

Shape conservation, Holmberg’s generalizationand predication*

Olaf KoenemanMeertens Instituut, Amsterdam

It is argued that Holmberg’s Generalization is a syntactic rather than phonolog-ical phenomenon. This view allows us to generalize over a larger set of facts in thefollowing way: Within the thematic domain, it is impossible to invert the rela-tionships of thematic categories, i.e. categories assigning or receiving a θ-role.The reason is that the grammar wants the interface interpretations at LF and PFto be uniform. It is shown that notorious counterexamples to thematic isomor-phism, such as passivization and short verb movement, can be dealt with in aunified way by making reference to predication theory.

. Introduction

A central issue in Scandinavian linguistics concerns observable restrictions onmovement of object DPs. It is well known that object shift is contingent on verbmovement, as shown in (1). In (1a), the verb has left its base position due to verbsecond and we observe that the object can precede negation, which by hypothesismarks the left edge of VP. In (1b), however, the verb second position is occupiedby an auxiliary and the lexical verb remains in its base position. Therefore, objectshift is impossible. In Swedish, verb second only regularly occurs in main clauses.This means that in embedded clauses the lexical verb remains in situ. As expected,object shift again leads to ungrammaticality (cf. (1c)).

* I would like to thank Ad Neeleman, Kriszta Szendroi, Anders Holmberg, Øystein Nilsen, themembers of the Groningen syntax seminar (especially Mark de Vries), the audience at the 20thComparative Germanic Syntax Workshop in Tilburg, Andrew McIntyre and an anonymous re-viewer for valuable criticism, suggestions and/or providing data. I take full responsibility for theway I made use of them.

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.2 (132-187)

Olaf Koeneman

(1) a. JagI

kysstej

kissedhennei

herintenot

tj ti (Swedish)

b. *JagI

harhave

hennei

herintenot

kysstkissed

ti

c. *DetIt

äris

troligtprobable

attthat

dethey

deni

itlästeread-past

ti

The debate about object shift largely concerns two issues: (i) what is the propergeneralization and (ii) what explains it? Obviously, one’s view on (i) has impli-cations for one’s view on (ii), as the generalization may suggest the nature ofthe account. In recent theorizing, however, consensus has not been reached as towhether the phenomenon is for instance phonological or syntactic. Two views aredominant.

The most pervasive one is probably the view holding that the phonologicalcomponent cannot handle (1b) and (1c). Holmberg (1999) argues that what is atstake is some form of ‘adjacency’ between the object and its trace, which is dis-rupted in the ungrammatical cases. Bobaljik (2002) argues that the lexical verbmust be ‘adjacent’ to a higher functional head, INFL or PART, and that the objectin (1b) and (1c) is the element disrupting this adjacency. Fox and Pesetsky (2005)adopt a cyclic spell-out model and argue that object shift in (1b) and (1c) leads toPF-ordering paradoxes in a way that will be discussed later.

The second view centres around the notion ‘shape conservation’. The mainidea here is that object shift without verb movement leads to a reversal of the mu-tual order between the verb and object. What rules this out, however, is not somuch that some spell-out procedure is disrupted but that the movement of theobject across the verb itself distorts isomorphism. Although one could in princi-ple imagine a theory in which detection of this distortion takes place at PF, it isnot for reasons of PF itself that (1b) and (1c) are ungrammatical. Rather, objectshift has to obey particular syntactic rules guiding the derivational process or, in amore representational view on this, there should be isomorphism between a repre-sentation before and after the relevant movements. Although some formulation ofshape conservation has been entertained as the relevant notion for the data in (1)by Müller (2000) and Kathol (2000) and for Continental Germanic scramblingby Williams (2003), the idea has not been fully worked out for the larger set ofdata that normally enters the discussion of object shift. The purpose of this pa-per is to do just that and present shape conservation as a viable alternative to thePF-approaches. I will argue that object shift and data pertaining to particle con-structions, double object constructions and object-to-subject raising require anadditional constraint, probably syntactic, if the accounts by Holmberg, Bobaljikand Fox and Pesetsky are adopted as an explanation for Holmberg’s generalization.In the alternative offered here, shape conservation, formulated as a syntactic map-

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.3 (187-240)

Shape conservation, Holmberg’s generalization and predication

ping rule operating between PF and LF, does the work. A central role is played bypredication theory (Williams 1980 and subsequent work). Note that in a VO lan-guage, analytic passive is not shape conserving, as the object clearly ends up at theother side of the verb. Although this might seem enough to rule out shape conser-vation as an explanation for the data, I will show that adopting predication theoryallows us to let shape conservation make the right decisions as to which move-ments Scandinavian syntax allows and which not. In addition, the combination ofshape conservation and predication theory has a few favourable consequences, asI will show.

This paper is organized as follows. In Section 2, I will look in detail at the PF-oriented proposals mentioned above. This discussion will lead to a clearer under-standing of the details involved and introduce the relevant data. Section 3 exploresthe syntactic alternative for object shift in terms of shape conservation. Sections 4and 5 extend the analysis to particle constructions and object-to-subject raising.Section 6 addresses two problems for the analysis. Finally, Section 7 provides asummary and outlook on further research.

. Restrictions on object shift as a phonological phenomenon

In this section, I will discuss in more detail the three phonological approaches toobject shift mentioned in the introduction. The purpose is to highlight the datathat have (or have not) been taken as part of the generalization accounting forobject shift, as well as to point out the limitations that phonological approachesencounter.

There are two interpretations of what is now known as Holmberg’s gener-alization. The narrow interpretation is that verb movement feeds object shift,illustrated by the data in (1). In fact, Holmberg (1986, 1999) originally proposeda broader definition which also pertains to particle and double object construc-tions. For him, (1b) and (1c) are ungrammatical for the same reason that (2a)and (2b) are.

(2) a. *DomThey

kastadethrew

mejime

intenot

utout

ti (Swedish)

b. *JagI

gavgave

deni

itintenot

ElsaElsa

ti

What unites them is that object shift across a VP-adjoined adverb or negation leadsto ungrammaticality if some VP-internal element, x, disrupts adjacency betweenthe moved object and its trace, as schematized in (3):

(3) object [VP adverb [VP x tobject ]

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.4 (240-320)

Olaf Koeneman

In (1b, c) x is the unmoved verb, in (2a) it is a particle and in (2b) it is the in-direct object. Evidence for the blocking nature of these elements comes from theobservation that moving them to a position to the left of the shifted object leadsto a grammatical result. In (1a), we already saw that V-to-C movement of themain verb feeds object shift. So does topicalizing the verb (phrase), the particleor indirect object to spec-CP, as shown in (4):

(4) a. Kysstj

Kissedharhave

jagI

hennei

herintenot

tj (Swedish)

b. UTj

Outkastadethrew

domthey

mejime

intenot

tj ti

c. Hennej

Hervisarshow

jagI

deni

ithelsrather

notnot

tj ti

Further evidence comes from particle constructions in Norwegian, Danish andIcelandic. In contrast to Swedish, these languages have a Part-DP/DP-Part alter-nation (sometimes referred to with the non-neutral label ‘particle shift’), i.e. theparticle optionally (Norwegian, Icelandic) or obligatorily (Danish) appears afterthe object. This is true even if the object is a full DP, rather than a pronoun. Asin Danish and Norwegian only pronouns but not full DPs can undergo objectshift, the facts in (5b–d) show that the order object-particle can be independentlygenerated in these languages and that this order is not a result of object shift.

(5) a. Viwe

släptelet

<ut>out

hundenthe.dog

<*ut>out

(Swedish)

b. JegI

skrevwrote

<*op>up

nummeret/detnumber.the/it

<op>up

(Danish)

c. JegI

skrevwrote

<opp>up

nummeret/detnumber.the/it

<opp>up

(Norwegian)

d. ViðWe

hentumlet

<út>out

hundinumdog.the

<út>out

(Icelandic)

As expected, object shift in these contexts is only allowed in Danish, Norwe-gian and Icelandic, as the object does not have to cross the VP-internal particle.Compare the fact in (2a) with the data in (6):

(6) a. JegI

skrevwrote

deti

itmåskemaybe

ikkenot

ti opup

(Danish)

b. DeThey

kastethrew

megi

meikkenot

ti utout

(Norwegian)

c. ViðWe

hentumlet

hundinumi

the.dogekkinot

ti útout

(Icelandic)

Thus, Holmberg’s original generalization accounts for more facts than those in(1). It conjures up two issues. First, note that in order to observe object shift in

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.5 (320-386)


contexts that do not contain a particle, indirect object or main verb in its base po-sition (that is, in legitimate contexts), a VP-adjoined adverb is necessary. In thatevent, we see the object appear to the left of this adverb, showing that it has shifted.However, this implies that, for some poorly understood reason, adverbs are al-lowed to appear between the object and its trace: they are PF-invisible. This posesa problem for an analysis based on PF-adjacency (cf. Bobaljik 2002 for a possiblesolution). A second issue is what to make of object-to-subject raising (as in pas-sive and unaccusative constructions), a process that Holmberg does not discuss.Since the object will have to cross the main verb on its way to the subject position,the verb will subsequently intervene between the subject and its trace in objectposition in Holmberg’s analysis.

(7)

Like Germanic languages in general, Swedish readily allows passivization, however:

(8) HanHe

blevwas

överkördhit

avby

ena

bil icar

gåryesterday

(Swedish)

Obviously, there is no a priori guarantee that object shift (OS, from now on) andpassivization should be similar processes for the present purposes but there is,however, an interesting fact pointing towards their similarity. As shown in Bobaljik(2002), just as object shift is ungrammatical in a Swedish particle construction, sois passivization:

(9) *SkräpetScrap.the

måstemust

blibe

kastatthrown

utout

(Swedish)

In contrast, passivization is grammatical in Icelandic, Norwegian and Danish,the languages that also allow object shift in particle constructions. Examples ofNorwegian and Danish are given in (10):

(10) a. Hundendog.the

blevwas

smedetthrown

udout

(Danish)

b. Hundendog.the

blewas

sluppetlet

utout

(Norwegian)

On the basis of this parallel, Bobaljik argues that the particle and passivizationdata should not be included in Holmberg’s generalization, which is thus narroweddown to the data in (1). He proposes to treat these core data on a par with hisearlier (1995) analysis of English do-support. Just as negation distorts adjacencybetween inflection in I and the verb in (11a), so does the object between participialinflection and V in (11b (=1b)) and between finite inflection and V in (11c (=1c)):

(11) a.

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.6 (386-451)

Olaf Koeneman

b. (Swedish)

c. (Swedish)

In passive and unaccusative constructions, the object lands in a position higherthan I, so that it will not interfere with I/Part-V adjacency. The particle and pas-sivization data, however, should now be covered by yet to be determined syntacticrestrictions on A-movement (cf. especially Anagnostopoulou 2002, 2003 for an ex-plicit account). In addition, Bobaljik assumes that illegitimate object shift acrossan indirect object in Swedish (cf. (2b)) is ruled out by similar syntactic restrictions.Just as with particle constructions, he argues, there is a parallel between OS andpassivization. He shows that the indirect object not only blocks OS but also blockspassivization. This can be observed in Danish and Icelandic.

(12) a. *Bogenbook.the

blevwas

givetgiven

JensJens

(Danish)

b. *PeningunumMoney.the

varwas

skilaðreturned

JóniJon

(Icelandic)

In this case, however, it is less clear what the correlation is. The presence of an in-direct object can indeed block both OS and passivization but does not necessarilydo so within the same language. In fact, Swedish allows passivization of a directobject across an indirect object.

(13) MedaljenMedal.the

blevwas

föräradpresented

JohanJohan

(Swedish)

Hence, there is no correlation between (un)grammaticality of OS across an indi-rect object and (un)grammaticality of passivization of a direct object across anindirect object. As passivization possibilities in double object constructions donot correlate with other facts here discussed, I take them to be dependent onlanguage-specific properties unrelated to the generalization that we are trying toestablish and explain (cf. Anagnostopoulou 2003 who argues that the ungram-maticality of DO-to-subject raising in double object constructions has differentsources cross-linguistically).

Holmberg’s and Bobaljik’s proposal have two features in common. Both haveto assume that adverbs are invisible. After all, adverbs can appear between the headof IP or PartP and V in Bobaljik’s account without any problem. Second, both needtwo rules, one phonological and one syntactic, to capture the data discussed so far.The main difference is that they draw the line differently.

Fox and Pesetsky basically draw the line where Holmberg draws it but offer adistinct explanation. They argue that moving an object across an unmoved verbcreates an ordering conflict. They adopt a multiple spell-out approach, in which

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.7 (451-531)


pieces of structure are spelled out before the complete derivation of the sentence isfinished. Intermediate spell-out points are determined by phases, which they taketo be VP and CP. At the level of VP, the order V>object will be added to a list ofordering statements in Swedish (cf. (14a)). If the object subsequently moves outof the VP, spell-out of a higher phase, CP, will add the order object>V to the list ofordering statement (cf. (14b)) and an ordering conflict ensues.

(14) a. [VP V object] Ordering statement: V > objectb. [CP . . . objecti . . . [VP V ti ]] Ordering statement: object > V

To stay out of such a paradoxical situation, the derivation must somehow preservethe order within VP at the level of CP. Hence, object shift is only allowed if theverb moves to C and ends up in a position again preceding the object.

What in addition to the data in (1) follows from this approach without furtherassumptions is the fact that OS across a particle or indirect object is just as bad,leading to ordering conflicts between object and particle or object and indirectobject. Hence, Fox and Pesetsky capture Holmberg’s original generalization. Whatalso follows is that OS is possible across a particle, indirect object or a main verb,if these elements are subsequently topicalized to spec-CP after OS (cf. (4)). Spell-out of VP will give IO/particle/V>object, and object shift plus topicalization ofIO/particle/V(P) to spec-CP will repeat this same ordering at the CP-level. Hence,no conflict arises. In other words, topicalization here restores order preservationthat is distorted at an intermediate state by object shift.

What does not follow immediately is the fact that wh-movement of the ob-ject to spec-CP is allowed. After all, this operation reverses the order between verband the object as established within VP. Fox & Pesetsky therefore assume that wh-movement first targets the left edge of the VP, so that spell-out of VP gives the orderobject-V. Movement to spec-CP then does not create a conflict, as the final ordermimics the order spelled out at the VP-level. It has been noted (cf. Bobaljik 2005;Anagnostopoulou 2005; Nilsen 2005) that the unavailability of the left edge land-ing site for OS (and the availability of it for wh-movement) now boils down to anecessary stipulation for the proposal to work, which makes the explanation ratherad hoc. This becomes a real drawback once subject-to-object raising is brought in.Like Holmberg, Fox and Pesetsky do not discuss passivization. This process ob-viously reverses the VP-level ordering statement, V>object, at the next phase. Inorder to allow it, F&P are forced to assume that passivization involves movementthat first makes the object land at the VP-edge, just like objects undergoing wh-movement. This raises two problems. Conceptually, the account would again boildown to stipulating for a particular movement operation whether it can land atthe VP-edge or not. Second, if passivization were movement to the VP-edge, justlike wh-movement, object-to-subject movement should be able to cross not onlythe main verb but also a particle. As we have seen (cf. (9)), Swedish shows that this

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.8 (531-579)

Olaf Koeneman

prediction is wrong.1 By necessity, then, F&P cannot offer a uniform explanationfor why particles in Swedish block both OS and passivization. Hence, it is unclearwhether F&P will assume passivization to be movement to the VP-edge or not buteither position is problematic.

Having discussed three PF-oriented approaches, we can now draw up thefollowing overview.

(15)

Pattern Example Holmberg Bobaljik Fox & Pesetsky this article

OK V OB (1a) PF PF PF SC* OBi V ti (1b, c) PF PF PF SC* OBi Prt. ti (2a) PF syntax PF SC* OBi IO ti (2b) PF syntax PF SCOK Vi Vfin. OBj ti tj (4a) PF PF PF SCOK Prtj Vi OBk ti tj tk (4b) PF PF PF SCOK IOj Vi OBk ti tj tk (4c) PF PF PF SCOK DOi ti Prt (6) PF syntax PF SC* SUi AUX V Prt ti (9) syntax syntax syntax SCOK SUi AUX V ti Prt (10) syntax syntax syntax SCOK SUi AUX V IO ti (13) syntax PF syntax SC

Given the discussion above, I conclude that there are basically two generalizationsto be accounted for: (i) verbs and indirect object DPs block OS but allow object-to-subject raising (ii) pre-object particles block both OS and object-to-subjectraising. In the next section, I will start developing an approach that explains bothgeneralizations in a unified way. The goal is to show that, in combination with

. Richards (2004) points out an important conceptual flaw in F&P’s analysis. The rationale be-hind a phase is that it is a computational unit. Once finished, the computational system sends itto the interfaces and can forget about its content. In F&P’s approach, however, the content of afinished phase must remain accessible for ordering paradoxes to arise. This raises the questionof why one should want to have phases in the first place. Richards offers an alternative phasal ac-count of Holmberg’s generalization and argues for word order parameters at PF. Within the firstphase (which for him is vP), the word order between V and object is set to be either consistentlyVO or OV. Hence, movement of an object across a verb will clash with the parameter-setting ofa VO-language under the assumption that OS is phase-internal movement. Wh-movement ofthe object and passivization differ from OS in that they move the object first to the phase-edgeand then to a position outside of the phase. All movements that leave the phase thus make anintermediate landing, a possibility that had to be stipulated for some movements in F&P’s ap-proach. Under the assumption that intermediate traces are irrelevant, these movements do notclash with the parameter-setting of a VO language. Although this approach solves the problemof passivization, it is less clear how the particle and double object facts can be incorporated. LikeF&P, Richards predicts passivization across a particle in Swedish to be possible. Hence, Richards’approach is superior to F&P’s but similar in not aiming at unification of all the facts discussed.

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.9 (579-642)


predication theory, shape conservation pertaining to thematic categories makesthe right predictions.

. Shape conservation as an explanation for object shift

The main idea of this section is that, within a particular domain, the mutualrelationships between thematic elements, whether they are θ-role assigning or re-ceiving, should remain constant in some sense. The grammar likes to conserve theshape of thematic elements relative to one another. It is for this reason that ad-verbs are ‘invisible’: they do not assign or receive a thematic role. From the wayshape conservation is defined, we can conclude what kind of rule it should be. Asthe constraint explicitly refers to thematic elements, it cannot hold at PF, whichis not concerned with the semantic content of the elements it has to align. This iswhy ‘shape conservation’ should be taken as a concept clearly distinct from ‘orderpreservation’ as used in Fox and Pesetsky. It is not clear either that LF should careabout shape conservation. DPs and verbs that move leave traces and the presenceof these will be enough to ensure the appropriate interpretation. That is, it shouldnot matter at LF if the syntactic relation between a DP and a verb is inverted byOS. Neither of the interfaces, then, has any principled problems with shape distor-tion, which makes it unlikely that shape conservation should hold at any of thesetwo levels.

The relevant principle must therefore be syntactic. The main question is whyshape conservation should hold in syntax if neither of the interfaces it feeds trulyrequires it. The answer lies in the fact that syntax is not an interface itself but amodule of grammar that mediates between levels of grammatical representation.In essence, then, shape conservation is not a rule that is beneficial to the syntaxitself but a mapping rule, as argued for in Williams (2003).2 Let us focus in on theinterfaces and establish how to formalize shape conservation.

Syntactic representations must be mapped into a phonological and into a se-mantic representation. It is generally understood that elements relevant at PF arenot necessarily elements that play a role at LF and vice versa. Even more to thepoint, if PF only deals with order and LF only with hierarchy, there is nothing

. The shape conservation rule agued for applies at the point where the syntactic derivationhas been created and is mapped onto a phonological and semantic representation. In termsof the T-model (Chomsky & Lasnik 1977), shape conservation would count as an S-structureconstraint. This is at odds with the general tendency to state constraints at the interfaces asmuch as possible. It may be that shape conservation as a general property of grammars is morenaturally couched in a less ‘syntactico-centric’ model (cf. Jackendoff 2002) in which levels ofrepresentation – syntactic, phonological and semantic – are more directly linked to one another.

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.10 (642-693)

Olaf Koeneman

to conserve, it seems. For shape conservation to exist, we will have to allow forsome minimal similarity between the interfaces. Let us therefore assume, perhapscontrary to some popular beliefs, that order is already established in the syntaxand that the job of the interfaces is to interpret these structures (cf. Bobaljik 2002;Haider 2004). Just as syntax already determines hierarchical relations, needed byLF, it already gives instructions for ordering. This means that a syntactic objectthat is minimally complex consists of a left and a right branch. Structure can becharacterized as a set of “>”-relations that the interfaces will have to interpret.Let us assume that in the syntax, α > β means (i) that α was merged with β or iscontained in the category merged with β, and (ii) that β was merged with α oris contained in the category merged with α. For a simple structure in (16), thesyntactic “>”-relations are: α > Z, α > β, α > γ and β > γ.

(16) Y

Z�

� �

Both interfaces will subsequently interpret these syntactically encoded relations.Suppose that in (16) α, β and γ are thematic categories (receiving or assigninga θ-role). What the interfaces now have in common is that both LF and PF willestablish “α > β” and “β > γ”, although the consequences are different. At PF, itentails that α is spelled out to the left of β, etc. LF-interpretation of structural rela-tions is a bit more complicated and also depends on the content of the nodes andthe nature of the relationship, which in turn depends largely on your theoreticalassumptions. Let us make a few explicit. Assuming that c-command is relevant fordetermining scope, LF will conclude from (16), for instance, that α scopes overβ. With respect to thematic categories, these scopal relations will be important ina theory that assumes a thematic hierarchy to be relevant for the assignment ofθ-roles (cf. Grimshaw 1990; Jackendoff 1972; Bayer & Kornfilt 1994). Moreover,since Williams (1989) it has become standard to assume that predicates can onlybe related to an element that c-commands it. I will furthermore assume that θ-role assignment is phrase-bound and that a predicate in turn must m-commandthe argument. Hence, α > γ can be interpreted as a thematic relation, but onlyif Z is not a maximal category categorically distinct from γ. See Neeleman andWeerman (1999) for a more recent motivation and more radical exploitation ofthematic mapping in these terms.

Although LF-interpretation is thus less straightforward than PF-interpretation,there is sufficient reason to conclude that for (16) both interfaces have the inter-pretations “α > β” and “β > γ”. Where LF and PF principally differ is in their

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.11 (693-742)


interpretation of movement chains. LF interprets the foot of a chain, as it needs toestablish thematic relations. PF, on the other hand, interprets the head of a chain.One may wonder why PF should do that. Several answers can be given. One fun-damental answer would be that if both LF and PF interpret the foot of the chain,the movement would not have any effect on the interfaces, hence it would be un-motivated. However, as PF cannot ‘know’ what LF interprets, this reasoning maynot be fully satisfactory (but see Note 2). For the purposes of this article, I will as-sume that movement leaves a trace and not a copy. Hence, there is nothing that PFcan choose: it has to spell out the head of the chain (cf. Bobaljik 2002 for furtherdiscussion and a different view on this issue). Example (17) shows a structure witha movement chain, as well as the interface interpretations of the relation betweenα and β. As can be established, PF and LF interpret the relation differently, and ina non-uniform way.

(17) Y

Z

�

�

t�

PF: >� �

LF: >� �

�

�� non-uniform: *

Suppose now that we extend this structure by moving up α. As PF interpretsthe moved constituents and LF their traces, the interface interpretations are nowuniform again.

(18) X

Y�

�

t�

PF: >� �

LF: >� �

�

�� non-uniform: OK

Z

t�

The structure in (17) is ruled out, those in (16) and (18) are ruled in. Hence, shapeconservation is about uniform interface interpretations. For our purposes, we canformalize it as follows:

(19) Thematic Shape Conservation (TSC):Within a thematic domain containing α and β (thematic categories), tβ cannotbe (contained in) the right-hand daughter of node N, where N is the minimalnode dominating both α and tβ.

To state (19) more intuitively: A structure in which α and β have changed placeswithin the thematic domain (to be defined) leads to ungrammaticality.

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.12 (742-817)

Olaf Koeneman

(20) *[ . . . [θ-domain . . . β. . . α. . . tβ . . .]]

The intuition behind (19) is thus that shape conservation is a syntactic mappingrule that minimizes interpretive differences at PF and LF. This desire for isomor-phism makes the prediction that syntactic movement may not invert thematicelements. It is now easy to see how the analysis should work for OS. Examples(21a) and (21b) are simply concrete examples of (16) and (17). (I will remain ag-nostic about the landing site of OS. For our purposes, it could be adjunction to VPor movement to a functional specifier.)

(21) a. VP

V DP

b.

DPi VP

V ti

* ?

In (21a), interface interpretation relative to the pair V and DP is uniform and willbe V > DP. In (21b), the minimal node N dominating the two thematic elementsis VP and the right branch of VP contains the trace of one of them. PF interpretsthe structure as DP > V, whereas LF interprets it as V > DP. Hence, TSC as statedin (19) is violated.

It is important to realize that (19) does not require of α and β that they arein a thematic relation with respect to one another. They can be, as is the case in(21), but inversion of two thematic categories without one assigning a θ-role tothe other is generally just as bad. One example is movement of the direct objectacross an indirect object (cf. (3), here repeated as (22)):

(22) *JagI

gavgave

deni

itintenot

ElsaElsa

ti

Here, both α and β are θ-role assignees and (19) correctly predicts that Elsa andden cannot be inverted. A second example is an ECM construction.

(23) *JagI

harhave

hennei

herintenot

hört ti

heardhållagive

föredragtalk

Take the thematic domain to roughly be the area in which thematic relations areestablished (I will define it more clearly in Section 6.) Then we have two thematicdomains here, one directly embedded in the other. Apparently, the fact that onethematic domain is an argument within the embedding domain has as a con-sequence that we have what counts as one complex thematic structure here andthematic categories of the embedded domain are by necessity part of it. Althoughthe matrix verb does not assign a θ-role to the embedded subject, inverting theirrelation is still ungrammatical. Hence, shape conservation does not so much look

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.13 (817-884)


at individual thematic dependencies but rather requires that the overall thematicstructure receives a uniform interpretation at the interfaces.

One property of (19) is already noteworthy. Shape conservation refers to the-matic elements. Adverbs are not included in this category. Hence, OS across anadverb is straightforwardly allowed. Remember that both Holmberg and Bobaljikneed an extra stipulation added to their analyses of OS to the effect that adverbsare PF-invisible. Of course, the constraint in (19) does not in itself explain whyadverbs do not enter the calculus: An adverb entertains a semantic relationshipwith other elements in the clause as well. Although that relation will be modifica-tional rather than thematic, it is not clear at this point why isomorphism shouldnot pertain to such a relationship. What is essential is that the constituent enteringinto a semantic relation with the adverb is not likely to be a thematic category,such as an object. After all, an adverb predicates of an event (Davidson 1966).Hence, we at most expect an effect of shape conservation between an adverb andfor instance a VP, not between an adverb and an object. Therefore, the fact that anobject can cross an adverb but not a verb is expected. Although the consequencesof shape conservation for elements other than thematic ones obviously needs tobe investigated in more detail, I believe this view holds promise.

We have now seen how shape conservation accounts for the observation thatverb movement feeds OS. In Sections 4 and 5, I will turn to particle constructionsand passivization and explain how shape conservation deals with the relevant facts.

. Particle constructions

The leading assumption here will be that the particles under discussion are el-ements with θ-role assigning qualities. Before entering into the analysis of thefacts related to particle constructions, we must make explicit some workable as-sumptions about the analysis of their structure and of what allows the optionalityof placing the particle before or after the DP-object. As the literature on particleconstructions is simply overwhelming, an extensive discussion of well-formulatedproposals in this area well exceeds the purposes of this paper and one has to be abit pragmatic. I will therefore opt for an approach to particle constructions thatseems viable to me and will at the same time allow a straightforward account ofthe central facts.

Leaving a lot of details aside, we can postulate the complex predicate analysisand some version of the small clause analysis as two viable and distinct options.These are, perhaps simplified, given in (24a) and (24b) respectively (cf. Dehé, Jack-endoff, McIntyre, & Urban 2002; and Haiden 2006 for a more detailed overview).

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.14 (884-934)

Olaf Koeneman

(24) a. VP

V DP

V Part(P)

b. VP

V SC

DP Part(P)

Assuming that either (24a) or (24b) is right, the alternation between part-DP andDP-part orders must be a consequence of some movement (with or without someflexibility of argument placement). If we take the view that movement is invariablyto the left, a complex predicate analysis could analyze the DP-Part order as a resultof base-generating DP as a specifier rather than as a complement. Verb move-ment to a ‘Larsonian’ shell or to v, then gives the surface order V-DP-Part order.Adopting a small clause analysis, we can derive the Part-DP order by assumingthat the particle incorporates into the verb or into some higher functional headc-commanding PartP. Alternatively, one could assume that DP starts out moreembedded within the PartP, for instance as the complement of the particle head,and optionally moves up. In that case, DP-Part(P) is actually a derived order. Un-der both approaches, movement is a necessary ingredient. Further theorizing willthen have to reveal why this optionality, which goes without a truth-conditionaldifference in meaning (cf. Svenonius 1996), exists or why the optionality is onlyapparent (cf. Neeleman 1994 and Svenonius 1996 for some completely differentaccounts and for further references).

Given the existence of two competing theories, there are two possibilities. Thefirst is that one of the theories is right and the other one wrong. On the basis of theextensive literature, however, it is extremely hard to make a choice. Both theoriesprovide evidence and offer solutions to deal with the counterevidence. As Haiden(2006:370–372) recently concludes: “The term ‘verb particle construction’ cov-ers too many distinct elements and configurations to allow for a single, consistentanalysis. [. . . ] The strategy which has been followed, explicitly in some and im-plicitly in most accounts, is to single out consistently behaving subclasses of data.”As there are no clear-cut arguments either way, there is legitimacy in adopting thesecond possibility, namely that both are correct to some extent. We conclude that(24a) is the most natural analysis for a particle-object order and (25b) the mostnatural analysis for an object-particle order. In other words, the choice betweenbase-generating (24a) or (24b) is precisely what constitutes the alternation be-tween Part-DP and DP-Part within a language. The evidence for this assumptionis then the sum of the evidence brought forward in the literature for either ap-proach. No movement has to be considered. The bottom line is that the grammarmust create structures in which the particle is able to assign thematic informationto the object. This can be achieved in two ways. The first is through complex pred-icate formation, in which two predicates (the verb and the particle) are combined

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.15 (934-1011)


within the morphological component and the particle adds thematic informa-tion to the verb. Alternatively, the particle can be projected into a full predicateseparately from the verb and assign its thematic information to the DP-object bypredication. This is then similar to the way in which a VP predicates over a subject(cf. Williams 1980).

(25) a. VP

DP VP

b. PartP

DP PartP

In both structures, the subject is predicate-external, the argument c-commandsthe predicate and the predicate m-commands the argument. In the choice between(24a) and (24b), the first option leads to morphological complexity, the second tosyntactic complexity. The fact that complexity is unavoidable but can be carried bytwo different components is then what constitutes the choice leading to the part-DP/DP-part alternation. Note that it is not obvious that an extra cost is involved inallowing both (24a) and (24b) as two base-generated structures, as it obviates theneed to postulate movement (and a concomitant trigger). In Section 5, I will be abit more explicit about my assumptions regarding the morphological component.

With this in mind, let us now turn to the particle facts. There are two facts thatneed to be explained. First of all, we saw that having the DP-Part order, which wenow define as the possibility of base-generating (24b), feeds OS. If a language onlyallows the order Part-DP, which is the case in Swedish, OS is blocked. The secondfact to be explained is that an obligatory Part-DP order also blocks passivizationof the object. If a language allows the order DP-Part, as in Danish, Icelandic andNorwegian, object-to-subject raising is unproblematic. These facts follow, as bothOS and passivization cause the object to cross the particle in Swedish only. There-fore, interface interpretation will not be uniform. According to the definition in(19), the structure of a Swedish particle construction will after movement of theobject contain a node N, the minimal node dominating both the object trace andthe particle, of which the trace is on a right branch. This node is circled in (26a).In Danish, Norwegian and Icelandic, on the other hand, the object trace is a leftbranch of the minimal node dominating the particle and the object trace. Hence,only (26a) is ruled out by the definition of TSC, which is the desired result.

(26) a. VP

V tDP

V Part(P)

* b. VP

V Part P

tDP Part(P)

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.16 (1011-1070)

Olaf Koeneman

Whether DP undergoes OS or passivization, they both contain the violating sub-structure in (26a) and are therefore ruled out for the same reason. Obviously, aquestion that arises is why regular passivization in a non-particle constructiondoes not lead to ungrammaticality. After all, subject-to-object raising crosses theverb. I will postpone answering this question until the next section.

It is important for the analysis that particles are indeed thematic elements,assigning a θ-role to a DP. This idea is rather uncontroversial in both the com-plex predicate and small clause approach and many analyses treat particles on apar with resultatives. Evidence for their predicative status comes from examplesshowing that resultatives and particles can license an additional DP:

(27) a. *I let the dogb. I let the dog out

(28) a. *I walk my shoesb. I walk my shoes to pieces

Admittedly, it is not always clear what kind of thematic information the particleassigns to the object (take to call versus to call up), but there is evidence that evenparticles that do not license an additional argument are still in a licensing relation-ship with the internal argument (cf. Zeller 2001 and McIntyre 2004 for detailedargumentation). At the same time, there is evidence against the stronger claimthat all particles are θ-role assigners. Resultatives are probably the clearest exam-ples of predicates adding thematic information to the structure. This informationis without exception assigned to the object, which therefore must be present. Ifparticles and resultatives should be treated on a par, we now expect particles toonly occur in the presence of an object. This turns out to be incorrect:

(29) a. You must hurry up (English)b. dat

thatJanJan

enen

PietPiet

samenwerkentogether-work

(Dutch)

‘that Jan and Piet co-operate’

Suppose we conclude from this that the particles in (29) are not argument li-censers, hence no θ-role assigners. This turns out to be harmless for the analysis.The only prediction that we now make is that a particle without θ-role assigningcapabilities should feed OS as well as passivization. However, since it is the ab-sence of the object that indicates the lack of these capabilities, there is nothing totest. Therefore, no wrong prediction is made.

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.17 (1070-1127)


. Predication theory

In the previous section, it was argued that in Swedish passivization of the object ina particle construction leads to ungrammaticality, as the object is forced to crossthe particle. Therefore, LF and PF interpretations will be non-uniform and shapeconservation is violated. An immediate question that this analysis conjures up iswhy regular object-to-subject raising does not lead to ungrammaticality. After all,these constructions have the following (reduced) structure:

(30) DP [VP V tDP]

The mutual relation between V and object is reversed. Therefore, one would expectLF to interpret this structure as V>DP and PF as DP>V. Whereas object-to-subjectraising forms a direct problem for analyses capitalizing on order preservation, foran analysis of shape conservation the problem is only as serious as the necessity toadopt the syntactic assumptions that create it. I propose that they disappear onceobject-to-subject raising is couched in predication theory. This approach will haveseveral favourable consequences.

Chomsky (1970) has suggested that verbal categories are inherently predica-tive. This suggestion has been taken up in Williams’ (1980 and further) predicationtheory, which holds that VP, consisting of the verb and internal argument(s), al-ways functions as a predicate, assigning a θ-role to the subject. It is this propertythat has non-trivial consequences for the way LF interprets thematic relations. Re-call what both interfaces, among other things, have to do: to interpret syntactic“>”-relations pertaining to thematic categories. As a subgroup, D-categories haveto be related to V-categories. A simple example is that of a verb and an object. Asthe object is contained within the VP, the verb and the DP-object are the categoriesentering the TSC calculus. A more complicated example is one involving a subject.If a subject occupies a VP-external position, as in predication theory, then morenodes potentially count as thematic categories. These are boxed in (31). Which arerelevant for the interfaces?

(31)

VPDP

V …

The proposal here is that for the purposes of shape conservation, category α onlyhas to be related to category β once. If β is structurally complex (i.e. consists ofmore categories), the node relevant for PF will ultimately be the terminal, as PFbasically spells out terminals. For LF, however, the relevant node is the one that isthematically related to α, if there is one. This entails that for the structure in (32),

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.18 (1127-1182)

Olaf Koeneman

PF interprets DP>V and LF interprets DP>VP. As there is no conflict in theseinterpretations, shape conservation is respected.

(32)

DP VP

V …

VP PF: DP>VLF: DP>VP

If different nodes of a particular category are relevant for each interface (here, Vand VP), some freedom arises. If due to thematicity LF selects VP as the node towhich the VP-external DP must be related and the relation between this DP andV is consequently ignored, we predict that DP and V can be inverted within thethematic domain. Two possibilities arise. First of all, a DP is allowed to cross V ifit lands in a position that is thematically related to VP. This will be our analysis ofpassivization. Second, the reverse is also possible. If in a predication relation DP isthematically related to VP rather than V, then V is allowed to cross DP. This will beour analysis of ‘short’ verb movement in double object constructions. Let us startwith passivization.

According to the standard analysis of object-to-subject raising, an object re-ceives its θ-role from the verb and then moves to the subject position for myste-rious reasons, usually captured under the label ‘epp’. There is, however, a viablealternative, holding that there is something that John walks, an intransitive con-struction, and John falls, an unaccusative construction, have in common, namelythat both contain a subject that is predicated over and that predication is essentiallyan external relation between maximal projections (Williams 1980 and subsequentwork; Neeleman & Weerman 1999 and others). This is schematised in (33):

(33) [ DP . . . [VP V ]θDP

The main intuition expressed here is that the subject entertains a thematic rela-tion with VP, in unaccustive constructions as well as in transitive and unergativeconstructions.

Now, the evidence for the different origins of the θ-roles assigned to John inthe walk- and fall-case is robust. That is, the idea that John starts out as an objectof fall but not of walk is well-motivated, certainly if one takes into considerationcross-linguistic findings such as ne-cliticization in Italian (cf. Burzio 1986) andauxiliary selection in a language like Dutch (where unaccusatives have a complexpast tense with a form of zijn ‘to be’ and transitives and unergatives with a form ofhebben ‘to have’). How can these findings be united with the idea that neverthelessit is the VP headed by fall which entertains a thematic relation with John in subjectposition? This can be achieved once it is recognized that θ-roles are epiphenom-

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.19 (1182-1246)


enal. Ultimately, thematic relations will have to be interpreted at LF, so that theyhave to be expressed in terms that LF recognizes. One way of doing this is to thinkof a thematic relation as a logical formula consisting of a lambda operator and avariable. Saturation of an internal θ-role can then be seen as the application of theformula in (34) to an argument:

(34) λx [V (x) ]

If predication is an external relation, it must be the case that at the level of VPa lambda operator is introduced in the mapping to semantic interpretation, asproposed by Partee (1973) and Williams (1977). This is schematized in (35).

(35) λx[VP . . .(x). . .]

So an external θ-role, in traditional terms, differs from an internal θ-role withrespect to which node introduces the lambda-operator: The lexical entry of theverb provides the operator of an internal role (cf. (34)), but VP introduces theoperator for an external role (cf. (35)). For a transitive predicate, the question nowis as follows. If VP introduces a lambda operator, what introduces the variable xin (35)? Note that not every verb assigns an external θ-role in traditional terms, sothat the lexical entry of a verb that does must be distinct anyway from the entryof a verb that does not. We can thus hypothesize that lexically such an externalθ-role is represented as a variable. A lexical verb selecting an agent and a themethen looks as in (36a). An unergative verb only electing an agent would have theentry in (36b):

(36) a. λy [V (xy) ]b. [V (x) ]

At the level of VP, a lambda operator is introduced that binds the variable x in bothrepresentations. This enables the VP to predicate over a VP-external subject.

Note that the lambda-calculus now distinguishes between an unaccusative(34), unergative (36b) and transitive predicate (36a). Note also that in this anal-ysis we can formulate a contentful version of the epp. If VP inherently acts as apredicate, there will always be a lambda operator introduced at the level of VP thatenables the VP to predicate over a subject. The rule that introduces the lambdaoperator and the rule stating that every clause should have a subject can now becollapsed:3

. An anonymous reviewer wonders how an unergative and unaccusative VP can be distin-guished in such a way that a subject counts as VP-external in the unergative case, therebypointing out a problem of syntactic theory in general. After all, a verb is merged with a DPin both cases and it is unclear how in a bare phrase structure grammar a complement can bedistinguished from a specifier. We could either assume that unergatives have an implicit ob-

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.20 (1246-1304)

Olaf Koeneman

(37) Extended Projection Principle:Map VP onto λx[|VP|]

There is, however, one remaining question, which pertains to an unaccusative(or passive) predicate. Applying the logic entertained so far, we take the verbs inclauses like John falls or The clothes are washed to be represented as in (34). Merg-ing these verbs with an object, John and the clothes, respectively, will entail thematicsaturation. At the same time, merger of these verbs and their objects, creates a VPand this triggers (38), the introduction of an lambda-operator. We now encountera problem: there is no variable left for the lambda operator to bind and VP can-not act as a predicate, as required. This is exactly where object-to-subject raisingbecomes important. After DP-movement, the structure of VP looks as in (38):

(38) [VP V tDP]

Kitagawa (1989) and Chierchia (1995) propose that traces of DP-movement areinterpreted as variables, so that the semantic representation corresponding to thesyntactic representation in (38) looks as in (39).

(39) a. λy [V (y)] x

(39) states that VP-internal saturation is achieved by applying the lambda-variablecombination to a variable. It will be clear that this does not provide the correct in-terpretation of John falls, or The clothes are washed, yet. The representation in (39)reduces to (40a) after lambda conversion. At the level of VP, a lambda operator isintroduced and the representation obtained looks as in (40b).

(40) a. V(x)b. λx[VP . . .(x). . .]

As can now be observed, the semantic representation of a VP with an unassignedexternal θ-role (cf. (35)) and that of a VP containing a DP-trace (cf. (40b)) areidentical. This means that a VP containing a DP-trace can act as a predicate andselect a VP-external subject. It is the moved DP that will fulfil this function. AsVP must be able to enter into a thematic relation with the moved DP, the twonodes must be local to one another. I will follow standard assumptions withinpredication theory and assume that the moved DP must be either adjoined to the

ject (cf. Hale & Keyser 1993), so that an unergative subject is merged with a projection of V,or that this subject is external as it occupies the specifier position of a functional head select-ing VP, for instance v. In the latter case, we could even assume that it is v which introducesthe lambda-operator, thereby bringing the proposal closer to current thinking. See Neelemanand Van de Koot (in preparation) for a solution in terms of θ-theory. I will not explore thesepossibilities further.

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.21 (1304-1364)


VP or occupy the specifier of the first functional projection dominating VP (cf.Williams 1994 and Koeneman & Neeleman 2001 for discussion).4

To state this analysis of object-to-subject raising in more intuitive terms: whatunergative, transitive and unaccusative predicates have in common is that VP con-tains an information gap. This gap is required for the VP to act as a predicate.With unergatives and transitives this gap coincides with the variable present inthe lexical entry of the verb, whereas with unaccusatives, this gap has to be syn-tactically created through object-to-subject raising.5 The conceptual advantage isthat we can then generalize over base-generated and derived subjects and state animportant rule, the revised epp, to capture this in a contentful way. In addition,it provides a handle on some empirical data. Note for instance that an active andpassive predicate can be easily conjoined.

(41) John [VP walked into the manager’s office] and [VP was fired t]

The structure in (41) blatantly violates the Coordinate Structure Constraint, whichdisallows extraction from one of the conjuncts. If, however, this constraint es-sentially requires that the conjoined phrases have similar selectional demands, itsgrammaticality is expected: Both VPs in (41) need to predicate over the same DP.More empirical advantages become available after some fine-tuning of the theory(cf. Neeleman & Weerman 1999; Neeleman & Van de Koot 2002/in preparation),

. In clauses with modal or aspectual verbs, the subject will overtly appear in a position that isnot m-commanded by the main VP-predicate. I will assume that modal and aspectual verbs donot assign a θ-role to a VP-predicate but modify it. That is, may read has the same thematic prop-erties as reads or probably reads (cf. Bobaljik 1995). Modifying verbs simply extend the predicatein the same way as adverbs do. The subject is then either merged after the last modifying verb oris generated lower and allowed to cross it because it is not a thematic category.

. Note that it is hard to translate this analysis in terms of traditional θ-roles and one could eas-ily be under the impression that the same role is assigned twice. Although such a translation mayperhaps be harmless (after all, one θ-role would assigned twice but to the same DP, so that noobvious violation of the θ-criterion ensues), it is more correct to think of the lambda-calculusas doing the following: VP-internal saturation by a variable has as a consequence that the the-matic information originating low in the structure is promoted to the level of VP. Employingthe lambda operator is only one way of ensuring that VP assigns the internal θ-role to an exter-nal argument. See Neeleman and Weerman (1999) for an approach using empty operators andWilliams (1994), where the notion of ‘vertical binding’ is used. The requirement that VP shouldenter into a predication relation with a thematic subject ultimately holds at LF, not in the syntax.Hence, there must be some LF computation involved in structures where an expletive occupiesthe subject position. See Williams (1994) and Koeneman and Neeleman (2001) for two possibleapproaches.

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.22 (1364-1421)

Olaf Koeneman

but this will get us too far removed from the issues at hand.6 Let us therefore turnto the relevance of this for shape conservation.

Object-to-subject raising in terms of predication theory explains why this op-eration does not violate shape conservation. Unlike OS, passivization is movementof a DP to a position interpreted by LF as thematic. Thematic Shape Conservationcan be ensured, as DP and VP are the relevant nodes for LF, just like in the activecounterpart. The trace of DP can consequently be ignored, at least for the purposesof TSC, and there is no ‘crossing’ registered. Interface interpretation is uniform asDP > VP (LF) and DP > V (PF) do not clash.

Obviously, this solution for passivization is only as strong as the (predication)theory it uses, but this criticism is in turn only as strong as the more standardtheory that this criticism would adopt. I do not consider an explanation for object-to-subject movement in terms of a strong epp-feature very insightful, for instance,and believe it is actually in the spirit of minimalist goals to try and define triggersin terms that are transparently relevant for one of the interfaces (here LF). In theabsence of consensus about the nature of epp-effects, therefore, such criticism isbearable. Moreover, I believe that the assumptions of predication theory can benaturally extended, with favourable consequences.

First of all, predication theory provides us with an analysis of short verb move-ment in double object constructions, an operation generally ignored in the orderpreservation literature. Note that in a VP-shell analysis of these constructions, theverb crosses the indirect object, as in (42b), and thereby seemingly violates shapeconservation. The solution is to extend predication theory to these constructionsand take the relation between IO and its sister-VP to be a predication relation (cf.(42a)). As can be seen in (42b), the consequence is that the interfaces again pickdifferent verbal categories to relate to the indirect object and no conflict arises.

. If the epp is about VP being inherently predicative rather than about there being a structuralsubject, we expect cases in which a verbal category functions as a predicate without having astructural subject. Higginbotham (1985) argues that gerunds and participles are a case in point.The prenominal V-projections in (i) enter into a thematic relationship with NP, but they lack astructural subject.

(i) a. A [VP quietly whistling] postmanb. A [VP carefully composed] letter

In addition, (i) shows that both an active and passive VP are capable of doing this.

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.23 (1421-1481)


(42) a.

IO VP

V DO

VP b.

V VP

IO VP

V’

tv DO

PF: V > IOLF: IO > VP

That IOs behave like subjects (rather than internal arguments) has been observedby Larson (1988), Bowers (1993) and Basilico (1998). Evidence for this idea comesfrom all directions. IOs trigger agreement in for instance South-Asian languages(Hermon 1985; Verma & Mohanan 1991; Mishra 1991) and obligatory clitic dou-bling in Bantu, Spanish and (for some speakers of) Greek (cf. Marantz 1993;Ngonyani 1996; Anagnostopoulou 2003). In Icelandic, IOs but not DOs are ca-pable of binding anaphors that are bound by ‘subjects’ only (cf. Maling 1990) andof disturbing the standard agreement pattern between the verb and a nominativeDP. Brandt (2003) argues at length for the view that many asymmetries betweenIOs and DOs (such as idiom formation, binding possibilities and the fact that IOsact as semantic definites) are best understood if the IO is not licensed by the verbbut external to the constituent consisting of the verb and DO, a view shared byPylkkänen (2000) and Anagnostopoulou (2003). Whereas these authors argue foran approach in which IO is licensed by a VP-external functional head (cf. also Note5), I will for the moment stick to my earlier assumptions and take the observationsabove to indicate that IO is predicated over by VP.

A second prediction is made. As the trace of DP can be ignored by TSC if DPmoves to a position that LF interprets as thematic, we expect that passivizationof a direct object across an indirect object should in principle be allowed by thegrammar. This prediction is correct, as this option is realized in at least Swedishand Norwegian.

(43) a. MedaljenMedal.the

blevwas

föräradpresented

JohanJohan

(Swedish)

b. EnA

bokbook

blewas

gittgiven

JonJon

(Norwegian)

This possibility is, however, not used by for instance Danish and Icelandic. What-ever the cause for that, shape conservation cannot be responsible for it. Cf. Holm-berg and Platzack (1995) and Anagnostopoulou (2003), who argues that failure ofthis passivization can have different causes cross-linguistically.

A third instance where predication theory does the right job involves a casewhere the grammar has found an interesting way of circumventing a violationof shape conservation in a passivization construction. Recall from the previous

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.24 (1481-1537)

Olaf Koeneman

section that passivization of an object was ungrammatical in a Swedish particleconstruction. As Swedish lacks the possibility of syntactically projecting the par-ticle into a predicate, complex predicate formation in the morphology is the onlyoption. In that event, there are two predicative categories that must be related tothe DP, namely V and PartP. Now, we have seen that with respect to the thematicinformation on the verb LF interprets the predication relation holding betweenDP in subject position and VP. For the second predicative category, PartP, a prob-lem arises. The reason is that PartP and DP in subject position cannot enter into athematic relation directly, because they are not in a local relation.

(44)

DP VP

V tDP

VP

V Part(P)

The first maximal projection dominating Part(P) is the lower VP-node and thiscategory excludes the subject position. Hence, Part(P) does not m-command thesubject position. The only way in which LF can therefore relate PartP and DPis via DP’s trace. Hence, the LF interpretation will have to be PartP>DP, whichconflicts with the PF interpretation DP>PartP. Predication theory thus explainsthat object-to-subject raising across a verb or indirect object is allowed but thatthe same movement across a predicative particle is ruled out. More generally put,predication theory brings in the possibility of distorting shape but in a very lim-ited way. Familiar locality conditions on predication explain what goes wrong in aSwedish particle construction.

There is, however, an alternative way of generating both the verb and theparticle that does not lead to the problem mentioned, namely through lexicalcompounding. Swedish can prefix the particle to the verb and in that event pas-sivization is allowed (cf. (45)):

(45) Skräpeti

the scrapmåstemust

blibe

utkastat ti

out-thrown

(Swedish)

‘The scrap must be thrown out’

The crucial difference between a complex predicate and a compound can be char-acterized as follows. According to Ackema and Neeleman (2004), a complex wordis created in the morphological component but this component has access tosyntactic as well as purely morphological rules (cf. the reference for argumentsagainst deriving complex words in the syntactic component proper). For complexverbs, which consist of a verbal and a nonverbal part (a particle or resultative), the

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.25 (1537-1608)


morphological component also has the choice between creating these objects bysyntactic or morphological rules and the results can be distinguished. If in Swedishthe morphology puts together a particle and verb by syntactic rules, the order isverb-particle, in accordance with the VO-setting of the language. If it is put to-gether by morphological rules, the order is particle-verb, in accordance with theRight-hand Head Rule. Hence, we get the minimal pair bryter av ‘break off ’ andavbryter ‘interrupt’. These entities behave differently in the syntax. As bryter av isput together by syntactic means, it is syntactically transparent. Hence, the particleis stranded when the verb undergoes verb second:

(46) Jag bryteri inte ti av kvisten (Swedish)I break not off the.branch

If the complex verb is put together by morphological rules, the result is an atomand syntactic rules cannot see the parts it is made of. Hence, the particle is pied-piped when the verb undergoes verb second:

(47) Jag avbryteri inte ti samtalet (Swedish)I offbreak the.conversation‘I interrupt the conversation’

If a particle-verb order reveals that the complex is put together by morphologi-cal rules, the complex verb in (45) must be a compound and hence syntacticallyopaque. As no syntactic rule can refer to a prefixed particle in a compound, thiselement cannot be input to TSC either. Hence, the syntax does not register a cross-ing effect taking place when the object is raised to subject position, which explainsthe grammaticality of (45). The thematic information carried by the particle, Iwill assume, is percolated to the top node of the compound in the morphologicalcomponent and hence available in the syntax.

Svenonius (1996) reports that in other Scandinavian languages the possibil-ity of generating these [Part.-V] structures varies greatly. It seems to be stronglypreferred in some Norwegian dialects, impossible in Danish and Icelandic andsubject to substantial speaker variation in Faroese. As the possibility of the Part-DP/DP-Part alternation cuts across the languages that allow [Part.-V] structures,Svenonius concludes that there is no correlation between having structures like(45) and the surface position of the particle in active sentences (as either preced-ing or following the object). It seems to me, however, that we can at least drawa distinction between Swedish on the one hand and the other Scandinavian lan-guages on the other in that Swedish is the only language in which both Part.-DPorders in active and [Part.-V] compounds in passive constructions are obligatory.We can thus draw the overview in Table 1.

Hence, with the exception of Swedish, both [Part.-V] structures and [Part.-DP] structures are non-obligatory in Scandinavian, meaning either possible or

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.26 (1608-1653)

Olaf Koeneman

Table 1. [Part.-V] and [Part.-DP] structures in Scandinavian

Language [Part.-V] in passive [Part.-DP] in active

Swedish obligatory obligatoryDanish non-obligatory non-obligatoryNorwegian non-obligatory non-obligatoryFaroese non-obligatory non-obligatoryIcelandic non-obligatory non-obligatory

impossible. This pattern is not unexpected, as Swedish is the only language inwhich a [Part.-V] compound has a clear purpose: the obligatory [Part.-DP] orderin the active sentence necessitates compounding for the passive to be possible.

To conclude this section, we have seen three cases where shape conservationseems to be violated: (i) DP movement across the verb in subject-to-object raising,(ii) short verb movement across an indirect object in a VP-shell structure and (iii)passivization of a direct object over an indirect object. All three potential viola-tions are resolved once predication theory is brought into the picture. At the sametime, predication theory explains why OS and passivization across a particle areruled out.

. Shape distortion

If shape conservation were an inviolable principle of UG, we would not know itexisted. Holmberg’s generalization is noteworthy in the face of data that stronglysuggest that initial orders can be destroyed, at least in principle. Hence, shape con-servation holds when everything else is equal and the job is to find out when thisis not the case.

There are two types of shape distortion that I would like to distinguish, namelyshape distortion within and outside of the thematic domain. In the first case,thematic category β crosses thematic category α and lands within the thematic do-main (cf. (46a)). This is what is ruled out by our constraint, the TSC. Alternatively,thematic category β can cross α but land in a position outside of the thematic do-main (cf. (46b)). In that event, there is no visible inversion of thematic categorieswithin the thematic domain and TSC does not apply.

(46) a. *[ . . . [θ-domain . . . β. . . α. . . tβ . . .]]b. [ . . . β. . . [θ-domain . . . α. . . tβ . . .]]

The thematic domain is most naturally defined as the area in which all thematicrelations are established. We can give a more precise definition. Recall that in pred-ication theory VP assigns a θ-role to a VP-external subject, which is either adjoinedto VP or occupies the specifier position of the first projection dominating VP (IP

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.27 (1653-1717)


or TP). The thematic domain therefore comprises everything dominated by thenode that immediately dominates the subject position. Hence, crossing the subjectentails leaving the thematic domain.

Conceptually, it makes perfect sense to restrict shape conservation of thematiccategories to that domain in which thematic relations are established. After all, it isnot obvious why it would hold outside this domain. Moreover, we independentlyknow that the left periphery of the clause has a highlighting function. The CP-domain is used for topics and focus constituents, elements that are set apart fromthe rest of the clause. For a domain to have this function, it must be allowed to dis-obey shape conservation, as its main function is to destroy it. These conceptualarguments for restricting TSC to the thematic domain tie in with the empiri-cal observation that movements to CP can be shape distorting without giving anungrammatical result. An example is given in (47).

(47) Vadi/Bokeni

what/the.bookköptebought

UlfUlf

intenot

ti (Swedish)

Here, both the verb and the object are thematic categories that have moved acrossthe subject to the C-projection. As their landing sites lie outside of the thematicdomain, TSC does not apply. Legitimate shape distortion does not necessarily in-volve movements to the C-projection. As long as the thematic category crossesthe subject, the result should be grammatical. This is supported by the followingdata. Example (48a) shows an object pronoun that has crossed the subject, a phe-nomenon known as ‘long object shift’. Example (48b) from Icelandic involves aDP-object that has moved to a position between the subject and the fronted verb.

(48) a. %Därförtherefore

gavgave

mejme

MaritMarit

intenot

någonany

presentpresent

(Swedish)

b. Þaðthere

stingerput

smjorinuthe butter

einhversomeone

Iin

vasannthe pocket

(Icelandic)

What also follows is that moving categories to CP that in their base position wouldgive rise to a violation of shape distortion, such as a particle, gives a grammaticalresult (cf. (4) and (49)):

(49) UTj

Outkastadethrew

domthey

mejime

intenot

tj ti (Swedish)

Examples (47) to (49) are instantiations of (46b) and therefore instances of shapedistortion that do not affect our analysis. More problematic would be (apparent)cases of shape distortion within the thematic domain. These certainly seem to ex-ist. In Section 6.1, I will look at examples of a direct object that lands between asubject and an indirect object and ask why the grammar does not register these asungrammatical. Section 6.2 looks at negative constituents in Icelandic that seem

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.28 (1717-1790)

Olaf Koeneman

to be obliged to violate Holmberg’s generalization, a phenomenon known as the‘inverse Holmberg effect’. Each section will offer a suggestion of how to tackle theproblem in a way compatible with the general approach.

. DO-IO reorderings

Recall that movement of a direct object across an indirect object is ruled outby TSC, as one thematic category crosses another. This correctly rules out (22),repeated here as (50):

(50) *JagI

gavgave

deni

itintenot

ElsaElsa

ti (Swedish)

It turns out, however, that at least for a significant number of Swedish speakers(50) becomes grammatical if the indirect object DP (here, Elsa) is replaced by apronoun (cf. Hellan & Platzack 1999 for a good discussion of the data). In thatcase, OS is legitimate (cf. (51a)). The direct object must, however, move acrossnegation. VP-internal crossing of the indirect object pronoun is out (cf. (51b)).If the indirect pronoun independently undergoes OS across negation, the directobject can precede it without any problem (51c).

(51) a. %Hanhe

gavgave

denit

intenot

henneher

(Swedish)

b. *JagI

gavgave

intenot

denit

honomhim

c. JagI

gavgave

denit

honomhim

intenot

Let’s start with (51c). Recall that what TSC basically rules out is (52):

(52) *[θ-domain β [ α tβ ]]

This leaves one important option open: β can legitimately cross α if β does notleave a trace. This happens when β moves as part of a larger, and crucially non-thematic, category γ:

(53)

Formally, (53) obeys TSC: tγ is on a right branch but does not violate TSC as longas γ is not a thematic category. PF will order the terminals as β>α. LF, however,cannot interpret the relation of α and β if we assume that β is contained in γ andhence does not scope over α (taking c-command to be the syntactic configurationdetermining scope). This means that TSC is unable to detect non-uniformity. Evenif we alternatively were to assume that everything contained in γ scopes over α

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.29 (1790-1848)


(e.g. by broadening the syntactic configuration determining scope), TSC is notendangered. In that event, LF will interpret the relation as β>α, which is similarto the one obtained by PF. Obviously, establishing α>β can and will eventuallybe achieved by LF but only after reconstruction, i.e. after “late” LF computation.This computation, however, is something that TSC cannot “know”. TSC applies atthe point in the derivation where the structure is sent to the interfaces and simplyevaluates the structure at that point. Object-to-subject raising, which is triggeredin overt syntax, can therefore be input to TSC, but reconstruction cannot.

What is the relevance of this? Let us start with (51c). Assume that pronounsare weak elements that for prosodic reasons like to leave their canonical base po-sition. Richards (2004) argues that the main characteristic of OS is that it involvesphonologically deficient (i.e. defocused) elements. The most natural trigger forOS, therefore, is one that refers to PF rather than to LF. Let us therefore assumethat pronouns must be move out of their base position as it is the position to whichneutral stress is assigned. Then, after short verb movement and merger of an ad-verb/negation, there are two categories that can be moved to obtain this result,namely the direct object itself or VP containing the direct object as its only overtcategory. These two categories are circled in (54):

(54)

ADV VP

V VP

VP

IO VP

DOtv

Both of these movements, however, violate TSC. DO cannot cross IO, as it wouldentail that one thematic category crosses another. IO receives a θ-role from VP, sothat IO is related to VP. V is therefore allowed to cross IO, but VP is not. Hence,there is no way to move DO from its base position. However, we can derive (51c)by first moving the IO pronoun out of the VP (cf. (55a)). This creates a largerVP-remnant, circled in (55b). This node differs from the circled VP in (54) in notbeing a thematic category. It does not, for instance, assign a θ-role to its sister,the moved V.

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.30 (1848-1890)

Olaf Koeneman

(55) a.

ADV VP

V VP

VP

IO VP

DOtv

b.

IO VP

ADV VP

VP

V VP

VPtIO

tv DO

Movement of this remnant VP crossing the IO then gives the surface order DO-IO-ADV-V.7 Subsequently, a subject will be merged and the finite verb will undergoverb second. The end result is a structure in which DO has legitimately moved

. An anonymous reviewer wonders what happens with selected adverbs in ‘behave badly’-typeof constructions. Here, Scandinavian uniformly puts the adverb after the object, it seems.

(i) Hanhe

harhas

<*dårlig>badly

oppføbehaved

rtrefl.

segbadly

<dårlig> (Norwegian)

(ii) DeThey

uppfördebehaved

<*illa>badly

sejrefl.

<illa>badly

(Swedish)

Although in Swedish the syntax of adverbs is apparently different form the syntax of particles,it is not clear what to conclude from this. Either adverbs in Swedish, in contrast to particles,can head a sort of small clause or Scandinavian simply allows adverbs to right-adjoin to VP.Under either hypothesis, it is hard to test if selected adverbs are input to TSC, as no obviouscrossing will occur. If the remnant movement approach suggested in this section is on the righttrack, there is a reason to prefer the second option. In double object constructions that includea clause-final adverb/PP (for example: ‘They sent her the flowers to her office’), a right-adjoinedadverb/PP can have a remnant VP as a sister. This constituent can then undergo object shift‘stranding’ the adverb/PP. A ban on right-adjunction, therefore, will make it more complicatedto uphold the present analysis or would at least invoke more movement operations.

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.31 (1890-1938)


across IO. The thematic domain does not contain structure that violates TSC, asDO does not leave a trace. Only the remnant VP leaves a trace but it is a trace of anon-thematic category, hence irrelevant for TSC.

This brings us to (51a). The question is why the direct object can undergoobject shift across a VP-internal IO-pronoun but not across a VP-internal IO-DP.I believe the crucial difference is that the pronoun is a head and a DP is not. Instructure (55), there is only one position below the adverb accessible as a landingsite for a head but not for a maximal projection, and that is V. Suppose that theIO-pronoun adjoins to the verb.

(56)

IO

VPADV

VP

VP

V

VP

DOtV

V tIO

As a result, the sister of [V-IO] is now a remnant VP that only contains the directobject. This VP, which is non-thematic, can again be moved so as to get the directobject out of its canonical position. Obviously, the verb must be able to undergoverb second later, thereby stranding the IO-pronoun (cf. (57)).

(57)

VPVP

VPADV

VP

tVP

IOV

…DO…

V

Hence, we must conclude that for these Swedish speakers these pronouns are cliticsthat can survive adjoined to a phonologically unfilled head (cf. Roberts 1991; Josef-sson 1992). This actually accounts for a more general fact of Swedish. In contrastto Norwegian, a weak object pronoun can remain within the VP for some speakerseven if it can legitimately undergo OS. In Norwegian, this shift is obligatory (cf.Hellan & Platzack 1999).

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.32 (1938-2003)

Olaf Koeneman

(58) a. %S V ADV Opronoun (Swedish)b. *S V ADV Opronoun (Norwegian)

The Swedish fact is puzzling if pronouns must in general leave their base position.We can now analyze it as just another instance of an adjoined pronoun that issubsequently stranded by the verb.

(59) VP

V tDO

DOV

Hence, it does leave its base position after all. Nothing of course forbids pied-piping. The verb and the adjoined direct object are allowed to cross the indirectobject, as LF will relate IO to its sister, VP, rather than to V. If the verb moves to C,piedpiping the object pronoun, both will eventually precede the subject.

(60) VP

IO VP

V tDO

V DO

Example (48a) showed that such Long Object Shift was indeed possible in Swedish(cf. Josefsson 1992 for conditions).

To sum up, we have suggested to solve the problem of IO-DO reorderingwithin the thematic domain by making use of a mechanism which TSC allows;movement of larger constituents containing nothing but the direct object. As longas the remnant category is non-thematic, the trace it leaves is irrelevant for theTSC calculus.

. The ‘inverse Holmberg effect’

A strange phenomenon in Icelandic poses a serious problem for many analyses ofHolmberg’s generalization, including the one presented here. Negative (and op-tionally quantified) phrases appear in positions unavailable to regular DPs andvice versa. (cf. Rögnvaldsson 1987; Svenonius 2000; F&P 2005). More precisely, itseems that they undergo movement in contexts where the unmoved verb normallyblocks OS (cf. (61a)) and that they cannot move when the verb has undergone verbsecond (cf. (61b)), in contrast to regular OS:

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.33 (2003-2063)


(61) a. JónJon

hefurhas

ekkertnothing

sagtsaid

SveiniSveini

b. *JónJon

sagðisaid

ekkertnothing

SveiniSveini

For Holmberg’s and Bobaljik’s approaches, the question is why these negativephrases do not disturb the adjacency that blocks OS with an unmoved verb in(61a) and why shift is impossible in (61b), where adjacency should not be dis-turbed. For the shape conservation approach, the question is why shape distortionhere gives a grammatical outcome and shape conservation an ungrammatical one.

For Fox and Pesetsky, (61a) follows under the stipulation that negative phrases,unlike regular DPs, can target the left edge of VP before moving to their landingsite. Hence, spell-out of VP gives object>verb, and this statement is repeated at thelevel of CP. Example (61b) is then correctly ruled out as follows. If spell-out of VPprovides the ordering statement object>verb and the verb subsequently undergoesverb second, spell-out of CP gives V>object and a paradox ensues. Hence, move-ment of a negative phrase and V-to-C movement cannot co-occur. As it turns out,the F&P analysis cannot be maintained. The ungrammaticality of (61b) seems tobe caused by the presence of two internal arguments rather than by verb move-ment to C. If the Negative Phrase is the only internal argument, V-to-C movementis unproblematic. Consider the following examples from Svenonius (2000):

(62) a. Viðwe

buðuminvited

engumno

lögregluÞjónumpolice.officers

b. *Viðwe

getumcan

boðiðinvite

engumno


c. Viðwe

getumcan

engumno


boðiðinvite

The contrast between (62b) and (62c) shows that the negative phrase mustmove. It must therefore have moved string-vacuously in (62a).8 Nevertheless,the verb has undergone verb second in (62a), which should not be possibleaccording to F&P.

In short, the ‘inverse Holmberg effect’ poses a problem for all the theories con-sidered. Although I do not have a worked out account for these data, I will sketchwhat seems to me a viable direction to move into. According to Svenonius (2000),movement of a negative phrase is an instance of quantifier raising. An obvious wayto distinguish QR from OS is to argue that QR adds to LF-interpretation, whereasOS does not. Recall that I assumed earlier, following Richards (2004), that OS takes

. Movement of negative phrases targets a lower position than object shift does. As a conse-quence, adverbs are a useless diagnostic.

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.34 (2063-2115)

Olaf Koeneman

place for prosodic reasons, and hence has no bearing on LF. What LF interprets isthe base position of a chain created by OS, as this base position is associated withthematic information. QR is like wh-movement in creating an operator-variablechain. Put differently, wh-movement and QR create interpretable landing sites.LF must interpret both the head and the tail of the chain. This allows the systemto maintain uniform interface interpretation, as LF now minimally interprets theposition that PF interprets as well. This explains why (61a) is grammatical.

This still leaves (61b) unaccounted for. As said, the problem for its ungram-maticality seems to lie in the presence of an indirect object. Svenonius (2000)argues that the negative phrase and the indirect object compete for the same struc-tural position. This problem does not arise in a complex tense, where two verbalheads are generated and an additional specifier position is consequently available.For lack of an alternative stated directly in terms of the present proposal, I willassume that something along these lines is correct.

. Conclusion

In this paper, I argue that Holmberg’s Generalization should not be limited tojust the observation that verb movement feeds object shift. Like Holmberg, I be-lieve that the generalization is broader, including double object and particle data.Moreover, I believe that object-to-subject raising and short verb movement, factsusually ignored, should not be excluded from the discussion. The broader general-ization proposed (‘thematic elements should obey shape conservation within thethematic domain’) is taken to be a mapping rule between PF and LF and is forthis reason stated in syntactic terms. This enables us to account for a number offacts in one location of the grammar instead of several. The two generalizationsformulated at the end of Section 2 are accounted for. Object shift across a verbor indirect object is illegitimate because it violates TSC. Object shift across a pre-object particle is out for the same reason. Object-to-subject raising across a verbor indirect object is grammatical, because the object moves to a position that isinterpreted as thematic. Object-to-subject movement across a pre-object particleis ungrammatical, because the landing site of the object is by necessity outside ofthe thematic scope of the particle. Hence, LF will need to interpret the object traceand a violation of shape conservation results. So although predication theory doesnot categorically forbid what looks like shape distortion, it places the expectedrestrictions on the extent to which this is possible.

A fair question to ask is whether we need an independent rule of shape con-servation. How fundamentally part of UG is it, for instance? First of all, it willbe clear that the present proposal has not addressed various kinds of shape dis-tortion. What do we do with languages showing free argument placement, for

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.35 (2115-2177)


instance? There are several possibilities. We could take shape conservation to bea meta-parametric property, not part of a non-configurational language. Alter-natively, other properties, such as morphological case, may give rise to a moreflexible syntax, where different word orders can be base-generated (an examplecould be German). It is also possible that extensive word order variation withina language signals the availability of designated focus and topic positions, per-haps even within the thematic domain (an example could be Hungarian). If LFinterprets these positions, TSC is not violated.

Second, one could hold against TSC that it is a syntactic rule that cannot be de-rived from either of the interfaces. In contrast, one could look at a PF-approach asan attempt to account for a set of data by referring to properties of the PF-interface,so that the facts here discussed are ultimately a by-product of constraints that areconceptually necessary. An attempt, however, is not the same as an achievement.Just as one can try to deduce properties of the PF-interface from Holmberg’s Gen-eralization, one can try to deduce properties of the mapping system (i.e. syntax)from them. Whether either of the attempts is superior, in the sense of only hav-ing to make use of independently required – or even conceptually necessary –mechanisms, is something that cannot be concluded at the moment. It is, for in-stance, just an hypothesis that grammar makes use of phases. It is not a conceptualnecessity. Hence, any claim about PF that hinges on this assumption cannot beconceptually necessary either. The debate about where in the grammar to accountfor Holmberg’s Generalization therefore continues and will benefit from investi-gating whether isomorphism in general is a matter of ordering (such as in thePF-approaches) or a matter of mapping (as in the shape conservation approach,cf. Williams 2003).

References

Ackema, P. & Neeleman, A. (2004). Beyond Morphology: Interface conditions on word formation.Oxford: OUP.

Anagnostopoulou, E. (2002). Movement across phonologically visible categories. Handout of atalk at the MIT ling-lunch, December 5th 2002.

Anagnostopoulou, E. (2003). The syntax of Ditransitives: Evidence from clitics. Berlin: Moutonde Gruyter.

Anagnostopoulou, E. (2005). Holmberg’s generalization and cyclic linearization: Remarks onFox and Pesetsky. Theoretical Linguistics 31 (Special issue on object shift, K. É. Kiss (Ed.)).

Basilico, D. (1998). Object position and predication forms. Natural Language and LinguisticTheory, 16, 541–595.

Bayer, J. & Kornfilt, J. (1994). Against scrambling as an instance of Move-alpha. In N. Corver &H. van Riemsdijk (Eds.), Studies on Scrambling: Movement and non-movement approachesto free word-order phenomena (pp. 17–60). Berlin: Mouton de Gruyter.

Bobaljik, J. (1995). Morphosyntax: The syntax of verbal inflection. PhD dissertation, MIT.

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.36 (2177-2303)

Olaf Koeneman

Bobaljik, J. (2002). A-chains at the PF-interface: Copies and ‘covert’ movement. NaturalLanguage and Linguistic Theory, 20, 197–267.

Bobaljik, J. (2005). Re: CycLin and the role of PF in object shift. Theoretical Linguistics, 31, 111–125. (Special issue on object shift, K. É. Kiss (Ed.)).

Bowers, J. (1993). The syntax of predication. Linguistic Inquiry, 24, 591–656.Brandt, P. (2003). Cipient predication: Unifying double object, dative experiencer and existen-

tial/presentational constructions. PhD dissertation, Utrecht University.Burzio, L. (1986). Italian Syntax: A government-binding approach. Dordrecht: Kluwer.Chierchia, G. (1995). Individual-level predicates as inherent generics. In G. Carlson & F. Pelleties

(Eds.), The generics book (pp. 176–233). Chicago IL: Chicago University Press.Chomsky, N. (1970). Remarks on nominalization. In R. Jacobs & P. Rosenbaum (Eds.), Readings

in English Transformational Grammar (pp. 184–221). Waltham MA: Ginn.Chomsky, N. (1995). The Minimalist Program. Cambridge MA: The MIT Press.Chomsky, N. & Lasnik, H. (1977). Filters and control. Linguistic Inquiry, 8, 425–504.Davidson, D. (1966). The logical form of action sentences. In D. Davidson (1980), Essays on

Actions and Events (pp. 105–122). Oxford: Clarendon Press.Dehé, N., Jackendoff, R., McIntyre, A., & Urban, S. (Eds.). (2002). Verb-particle Explorations.

Berlin: Mouton de Gruyter.Fox, D. & Pesetsky, D. (2005). Cyclic linearization of syntactic structure. Theoretical Linguistics,

31, 1–45. (Special issue on object shift, K. É. Kiss (Ed.)).Grimshaw, J. (1990). Argument Structure. Cambridge MA: The MIT Press.Haiden, M. (2006). Verb particle constructions. In M. Everaert & H. van Riemsdijk (Eds.), The

Blackwell Companion to Syntax, Vol. 5 (pp. 344–375). Malden MA: Blackwell.Haider, H. (2004). How to turn German into Icelandic – and derive the OV-VO contrasts.

Journal of Comparative Germanic Linguistics, 8, 1–53.Hale, K. & Keyser, S. J. (1993). On argument structure and the lexical expression of syntactic

relations. In K. Hale & S. J. Keyser (Eds.), The View from Building 20 (pp. 53–109).Cambridge MA: The MIT Press.

Hellan, L. & Platzack, C. (1999). Pronouns in Scandinavian languages: An overview. In H.van Riemsdijk (Ed.), Clitics in the Languages of Europe (pp. 123–142). Berlin: Mouton DeGruyter.

Hermon, G. (1985). Syntactic Modularity [Studies in generative grammar 20]. Dordrecht: Foris.Higginbotham, J. (1985). On semantics. Linguistic Inquiry, 16, 547–593.Holmberg, A. (1986). Word Order and Syntactic Features in the Scandinavian Languages and

English. PhD dissertation, University of Stockholm.Holmberg, A. (1999). Remarks on Holmberg’s generalization. Studia Linguistica, 53, 1–39.Holmberg, A. & Platzack, C. (1995). The Role of Inflection in Scandinavian Syntax. Oxford: OUP.Jackendoff, R. (1972). Semantic Interpretation in Generative Grammar. Cambridge MA: The MIT

Press.Jackendoff, R. (2002). Foundations of Language: Brain, meaning, grammar, evolution. Oxford:

OUP.Josefsson, G. (1992). Object shift and weak pronominals in Swedish. Working papers in

Scandinavian syntax, 49, 59–92.Kathol, A. (2000). Linear Syntax. Oxford: OUP.Kitagawa, Y. (1989). Deriving and copying predication. Proceedings of the North East Linguistics

Society, 19, 270–300.Koeneman, O. & Neeleman, A. (2001). Predication, verb movement and the distribution of

expletives. Lingua, 111, 189–233.

JB[v.20020404] Prn:24/10/2006; 15:40 F: LA9702.tex / p.37 (2303-2422)


Larson, R. (1988). On the double object construction. Linguistic Inquiry, 19, 335–392.Maling, J. (1990). Clause-bounded Reflexives in Modern Icelandic [Syntax and Semantics 24],

277–287. San Diego CA: Academic Press.Marantz, A. (1993). Implications of asymmetries in double object constructions. In S. A.

Mchombo (Ed.), Theoretical Aspects of Bantu Grammar (pp. 113–150). Stanford CA: CSLI.McIntyre, A. (2004). Event paths, conflation, argument structure and VP shells. Linguistics, 42,

523–571.Mishra, M. K. (1991). Dative/experiencer subjects in Maithili. In M. K. Verma & K. Mohanan

(Eds.), Experiencer Subjects in South Asian Languages (pp. 105–118). CSLI. (Distributed byUniversity of Chicago Press).

Müller, G. (2000). Shape conservation and remnant movement. In M. Hirotani, A. Coetzee, N.Hall, & J.-Y. Kim (Eds.), Proceedings of the North East Linguistics Society, 30 (pp. 209–241).Amherst MA: GLSA.

Neeleman, A. (1994). Complex Predicates. PhD dissertation, Utrecht University.Neeleman, A. & Koot, H. Van de (2002). The configurational matrix. Linguistic Inquiry, 33, 529–

574.Neeleman, A. & Koot, H. Van de (In prep). Theta Theory. Ms. University College London.Neeleman, A. & Weerman, F. (1999). Flexible Syntax: A theory of case and arguments. Dordrecht:

Kluwer.Ngonyani, D. (1996). The Morphosyntax of Applicatives. PhD dissertation, UCLA.Nilsen, Ø. (2005). Some notes on cyclic linearization. Theoretical Linguistics, 31, 173–185.

(Special issue on object shift, K. É. Kiss (Ed.)).Partee, B. (1973). Some transformational extensions of Montague grammar. Journal of Philo-

sophical Society, 2, 509–534.Pylkkänen, L. (2000). What applicative heads apply to. In M. Williams & E. Kaiser (Eds.),

Proceedings of the 24th Annual Penn Linguistics Colloquium [Upenn Working papers inlinguistics 70] (pp. 197–210). Philadelphia PA: University of Pennsylvania.

Richards, M. D. (2004). Object Shift and Scrambling in North and West Germanic: A case studyin symmetrical syntax. PhD dissertation, Cambridge.

Roberts, I. (1991). Excorporation and minimality. Linguistic Inquiry, 22, 209–218.Rögnvaldsson, E. (1987). On word order in Icelandic. Proceedings of the seventh Biennial

Conference of Teachers of Scandinavian Studies in Great Britain and Northern Ireland, 33–49.London: University College London.

Svenonius, P. (1996). The verb-particle alternation in the Scandinavian languages. Draft,University of Tromsø, Norway.

Svenonius, P. (2000). Quantifier movement in Icelandic. In P. Svenonius (Ed.), The Derivationof VO and OV (pp. 255–292). Amsterdam: John Benjamins.

Verma, M. & Mohanan, K. (1991). Experiencer Subjects in South Asian Languages. CSLI.(Distributed by University of Chicago Press).

Williams, E. (1977). Discourse and logical form. Linguistic Inquiry, 8, 101–139.Williams, E. (1980). Predication. Linguistic Inquiry, 11, 203–238.Williams, E. (1989). The anaphoric nature of θ-roles. Linguistic Inquiry, 20, 425–456.Williams, E. (1994). Thematic Structure in Syntax. Cambridge MA: The MIT Press.Williams, E. (2003). Representation Theory. Cambridge MA: The MIT Press.Zeller, J. (2001). Particle Verbs and Local Domains. Amsterdam: John Benjamins.

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.1 (55-354)

Quirky verb-second in Afrikaans

Complex predicates and head movement

Mark de VosRhodes University, South Africa

The central aim of this paper is to account for ‘Quirky Verb Second’, a peculiarconstruction in Afrikaans which optionally pied-pipes a coordinated verbal clus-ter to verb-second position. This is unique among the Germanic verb-secondlanguages. It is demonstrated that narrow syntax can operate, not only over fea-ture bundles, but over features within feature bundles. It is argued that verbalhead-movement may indeed be phonological feature movement (Boeckx &Stjepanovic 2001; Chomsky 2000; Zwart 1997), but with the added caveat that itcan also be true syntactic movement in certain instances. The proposal has impli-cations for theories of head movement, excorporation and coordination.

. Introduction

Cross-linguistically, posture verbs tend to become grammaticalized and to encodeaspectual information such as durativity (Kuteva 1999). The fact that this occursin a wide-range of unrelated languages makes posture verbs particularly interest-ing objects of study. Afrikaans uses a small set of auxiliary-like posture verbs asmarkers of durativity in a particular type of restructuring configuration.

(1) Waaromwhy

salwill

JanJan

diethe

olifanteelephants

sitsit

enand

wegjaag?away.prt-chase

‘Why will Jan chase the elephants away’

This example illustrates a verb-second sentence with a coordinated verbal string insentence-final position (bold). Like Dutch, Afrikaans is an OV language Barbiers(2000) with verb-second in matrix clauses – and in embedded clauses in someregisters and varieties (Biberauer 2002, 2003). Following established wisdom, Itake verb movement to involve at least head movement from V to T along the

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.2 (354-420)

Mark de Vos

lines suggested by Zwart (1997).1 Whether T-to-C movement specifically involveshead movement (Den Besten 1989) or not (Biberauer 2003; Nilsen 2003; Zwart1997) is a question that will not be directly relevant to the issue at hand. Whatevermechanism is ultimately responsible for T-to-C movement is dependent on initialhead-movement to T.

Returning to example (1), the clause-final verb string consists of a lexical verb,wegjaag ‘chase away’ coordinated with a verb of posture, sit ‘sit’. The posture verbdoes not necessarily imply that Jan was seated but rather denotes durativity. Thus,the coordinated predicates together refer to a single event of chasing the elephantsaway; not to a discrete event of sitting and to another of chasing.

The posture verb is known in the Afrikaans literature as an Indirect LinkingVerb (ILV) or Indirekte Skakelwerkwoord and is part of a closed class of such verbsincluding loop ‘walk’, sit ‘sit’, lê ‘lie down’ and staan ‘stand’. The fact that the objectoccurs to the left of the posture verb (in Spec vP or alternatively Spec AgrOP)indicates that this is an instance of restructuring. What is remarkable about thisAfrikaans construction is that the coordinated verbal string can be pied-piped aspart of verb-second, stranding the separable particle. Henceforth, the pied-piped,complex, coordinated predicate will be called a Complex Initial. The non-pied-piped complex, coordinated predicate will be referred to as ‘the verbal string’. Thephenomenon itself will be called Quirky verb-second.

(2) a. Waaromwhy

sitsit

enand

jaagchase

JanJan

diethe

olifanteelephants

sit en wegaway.prt

jaag?

‘Why does Jan chase away the elephants?’b. Waarom

whysitsit

JanJan

diethe

olifanteelephants

sit enand


‘Why does Jan chase away the elephants?’

Example (2a) illustrates how the Complex Initial (i.e. the fronted, coordinatedverbal string) can undergo ‘inversion’, occurring in second position and to the leftof the subject.2 That this appears to be optional is demonstrated by (2b) which hasthe same denotation.3

There are several reasons why this phenomenon is fascinating. There is theobvious fact that a complex, coordinated predicate is pied-piped as part of verb-

. But also see (Nilsen 2003; Biberauer 2003).

. The phenomenon is not limited to coordinated predicates (i.e. ILVs) but can also occur witha subset of restructuring, raising verbs of an aspectual nature (i.e. Direct Linking Verbs/DLVs).This will be discussed in Section 6.

. That both sentence types have the same meaning is attested to by (Donaldson 1993; Ponelis1993; Robbers 1997).

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.3 (420-470)


second. Such pied-piping is optional and does not lead to a semantic difference.This pied-piping is unique among the Germanic verb-second languages and raisesimportant questions about the nature of head movement. To this end, it is nec-essary to determine whether the pied-piped constituent is indeed a head. Thereis a variety of evidence showing that pied-piped, coordinated predicates do in-deed act as a single verbal head.4 First, there is the fact that the Complex Initialdisplays the same distribution as an ordinary verbal head: it displays a matrix-embedded asymmetry with respect to verb-second. Examples (3a, b) illustrate thematrix-embedded asymmetry characteristic of simplex predicates.

(3) a. JanJan

jaagchase

diethe

olifanteelephants

wegaway.prt

‘Jan chases away the elephants’b. . . . dat

. . . thatJanJan

diethe

olifanteelephants

wegjaagaway.prt-chase

‘. . . that Jan chases away the elephants’

The same distribution is evident with Afrikaans coordinated complex predicatesas illustrated in examples (4).

(4) a. JanJan

sitsit

enand

jaagchase

diethe

olifanteelephants

wegaway.prt

‘Jan chases the elephants’b. . . . dat

. . . thatJanJan

diethe

olifanteelephants

sitsit

enand


‘. . . that Jan chases the elephants’

The fact that the ILV coordinated predicates and simplex verbs exhibit the samedistribution strongly suggests that the pied-piped predicate acts as a single, verbalhead. Moreover, the second position of the clause is usually reserved for verbalheads in a verb-second language like Afrikaans. This indicates that the complexpredicate which occurs in second position must also be a head.5

The second major argument for the head status of the pied-piped predicateis the fact that no non-verbal material may intrude within it. In this regard, con-sider the position of the separable particle in examples (4). It will be noted that theparticle occurs within the coordinated structure when the verbal string is in situ.

. The head-status of Complex Initials of this type is hardly controversial and has been notedby many researchers including (Den Besten 1988, 2002; Donaldson 1993; Ponelis 1993; Robbers1997).

. It might be suggested that if verb movement to T is recast in terms of remnant movement(Mahajan 2001; Müller 2004), then this argument might be vacuous. This option is discussedand rejected in Section 3.4.

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.4 (470-544)

Mark de Vos

However, when pied-piping occurs, the particle is stranded in sentence-final posi-tion. Importantly, the particle can never be pied-piped along with the coordinatedverbs (5).

(5) *JanJan

sitsit

enand


diethe

olifanteelephants

Similarly, neither high nor low adverbs can occur within the pied-piped complexpredicate (e.g. (6) and (7)) and still retain the aspectual reading. It is thus a trueverbal cluster in its pied-piped position.

(6) *JanJan

sitsit

enand

sorgvuldigcarefully

jaagchase

diethe

olifanteelephants

wegaway.prt

‘(intended) Jan chased the elephants away with care’

(7) *JanJan

sitsit

enand

waarskynlikprobably

jaagchase

diethe

olifanteelephants

wegaway.prt

‘(intended) Jan probably chased the elephants away’

These arguments strongly suggest that the pied-piped coordinated predicate is asingle head.

. Coordinated predicates in comparative perspective

Before discussing structural properties of Quirky verb-second, it is important toplace it in a comparative context. Modern Germanic languages that utilize postureverbs to indicate aspect tend to utilize a semantically ‘light’ functional head thatcan have the morphological form of a subordinator (8a) or a coordinator (8b, c)or be homophonous between the two å/og (8d).6

(8) a. Watwhat

zatsat

hijhe

teto

eteneat-inf

‘What was he eating?’ [Dutch]b. What did he sit and eat? [English]c. Wat

whatsitsit

hyhe

enand

eeteat

‘What is he eating?’ [Afrikaans]d. Hva

whatsittersit-fin

hanhe

ogand

spiser?eat-fin

‘What is he eating?’ [Norwegian]

. The Norwegian and Swedish pseudo-coordinative particles are homophonous between asubordinator and coordinator.

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.5 (544-616)


If one looks only at pseudo-coordination in Standard languages then it would seemthat there is a generalization regarding agreement:

(9) a. with subordinative markers, the aspectual verb may be finite but thelexical verb is always infinitival;

b. with coordinative markers, the aspectual verb and the lexical verb musthave the same inflectional form.7

However, a closer look at dialectal microvariation shows this to be untrue. Thefollowing examples from West-Flemish dialects show that the aspectual and lexicalverbs can have different morphological forms. For instance, in (10a) the aspectualverb has finite marking while the lexical verb is an infinitive. Thus, there is nocorrelation between the coordinator/subordinator and agreement.

(10) a. JanJan

zatsat

eena

boekbook

enand

lezenread-inf

‘Jan was reading a book’ [Izenberge: (Haslinger & Van Koppen 2003)]b. En

hestoanstand-fin

enand

zagencomplain-inf

‘He is complaining’ [Merckeghem:(Haslinger & Van Koppen 2003)]c. Dien

thejongenboy

zitsit-fin

voorzekerssurely

televisietelevision

enand

kijkenwatch-inf

‘The boy is surely watching TV’[Wulvergem: (Haslinger & Van Koppen 2003)]

These data are supported by historical evidence. The following nineteenth cen-tury example illustrates a pseudo-coordinative construction in which the aspectualverb is finite and the lexical verb is an infinitive (Ijbema 2003).

(11) a. daerthere

syshe

satsat

enand

huylencry-inf

‘she was sitting there and crying’ (Ijbema 2003)

Interestingly however, there does appear to be a correlation between the formationof complex initials and agreement. A complex initial may only contain verbs withthe same morphological form of agreement. Consider the following paradigmfrom Ijbema (2003) citing Gerritsen (1991).

(12) a. MarieMarie

zitsits

aardappelenpotatos

teto

schillenpeel-inf

[Throughout Netherlands]

b. MarieMarie

zitsits

teto

aardappelenpotatoes

schillenpeel-inf

[Friesland, Groningen, Drenthe]

. This is also the conclusion reached by Roberge (1994) based on some historical examples.

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.6 (616-684)

Mark de Vos

c. MarieMarie

zitsits

aardappelenpotatoes

enand

schiltpeels

[Zeeland]

d. MarieMarie

zitsits

enand

schiltpeels

aardappelenpotatos

[Regionally restricted in Netherlands]

e. *MarieMarie

zitsits

enand

schillenpeel-inf

aardappelenpotatos

[Not attested]

With respect to pseudo-coordinative structures, these data show that verbs canhave the same agreement marking (12c, d) or different agreement marking (10).The pair in (12d, e) are especially important because (12d) appears to be a complexinitial. This is supported by the ungrammaticality of (12e), a fact which wouldbe explained if both verbs are adjoined to T at some point in the derivation andconsequently must have the same agreement.

These comparative data lend credence to the assertion by Ponelis (1993) thatAfrikaans complex initials arose as a consequence of a loss of verbal inflection.Ponelis suggests that when the finite and non-finite forms of the verb became non-distinct, examples like (12d, e) would have been indistinguishable.8 This beginsto explain the paucity of complex initials in modern Germanic languages which,with the exception of Afrikaans, have at least some inflectional morphology. Thepreceding data show that the morphological component constrains the form ofcomplex initials. Thus, it follows that Afrikaans will be far less restricted in form-ing coordinated predicates than other Germanic languages. This also predicts thatcomplex initials are, in principle possible in Germanic languages, provided that(a) a pseudo-coordinative strategy is used and (b) that the agreement matches onboth verbs. While this strategy seems to be restricted, the data in (12d) show thatit is, in principle, possible.

. An anonymous reviewer has pointed out an alternation that supports this point. SinceAfrikaans lacks inflectional morphology, it is not possible to ascertain whether complex ini-tials have finite features on both coordinated verbs. However, the verb wees ‘be’ is one of thefew Afrikaans verbs that have distinctive finite and non-finite forms. Although it is extremelydegraded when used in an ILV context, there remains relative judgement. To the extent that thisis grammatical at all, it suggests that complex initials may derive from something like (12e).

(1) a. Waarom staan jy en ongelukkig ≺*wees/**is �why stand you and unlucky be.inf/be.fin‘Why are you being unhappy?’

b. Waarom staan en ≺*wees/**is � jy ongelukkig?why stand and be.inf/be.fin you unlucky

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.7 (684-733)


. The base structure of Afrikaans coordinated predicates

In this section, the structure motivating ILV coordinated predicates is explored.ILV coordinated predicates behave like single verbs in the sense that they can un-dergo head-movement and refer to a single event. On the other hand, the presenceof the separable particle within the coordinated structure in examples like (1) and(4b) suggests that the coordinated verbal string is not a constituent at all. Thisapparent paradox can be resolved by the following structure.

(13) vP

Subject v

v VP

ILV v

en v

ObjectParticle

V

V

]

]

In this structure, the ILV is coordinated with a phonetically empty little v. In theabsence of V-v raising, V remains in situ and occurs to the right of the object(Barbiers 2000). Thus, the lexical verb is not a constituent with the coordinatedV+en+v complex. The position for objects is to the left of V. It is also assumedthat the separable particle is adjoined to VP.9 This structure is supported by thedistributions of subjects, objects, particles and adverbs.

Given structure (13), there are potentially three fields for XP-like material.These are labelled Fields A, B and C for convenience.

(14)

↑A

ilvsit

↑B

anden

↑C

lexical verbjaag (as in example (1))

Field A would be Spec vP or higher, including AgrOP. It is the unmarked positionfor subjects, objects, adverbials and other material associated with the functionallayer. Field B corresponds to an adjunction point between two conjoined heads.Consequently, it is expected that no XP-like material could ever occur in this po-sition. Finally, Field C is equivalent to VP adjunction and Spec VP and mightpotentially host low adverbs of manner, separable verbal particles and conceiv-

. Whether or not the separable particle itself is base-generated in a still lower small clause isimmaterial for the moment. It is only necessary that the particle occurs to the left of the lexicalverb at some point in the derivation.

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.8 (733-808)

Mark de Vos

ably, some types of ‘low’ objects. These predictions are explored in the followingsubsections.

. Distribution of subjects

In the following examples, an expletive is located in subject position, forcing theindefinite subject to remain in Spec vP corresponding to Field A. Subjects canonly occur in Field A (15a). As expected, it is not possible for the subject to occurin either Fields B or C (15b, c). This is consistent with the proposed structure.

(15) a. Daarthere

salwill

altydalways

iemandsomebody

sitsit

enand

eeteat

‘There will always be somebody eating’ [In A]b. *Daar

theresalwill

altydalways

sitsit

iemandsomebody

enand

eeteat

‘There will always be somebody eating’ [*In B]c. *Wat

whatsalwill

daarthere

altydalways

sitsit

enand

iemandsomebody

eet?eat

‘There will always be somebody eating” [*In C]

. Distribution of adverbs

The fact that different adverbs systematically select different points of adjunction(Ernst 2002) provides a useful tool to disambiguate structures. Field A correspondsto a variety of positions in the functional layer and is thus the unmarked positionfor many adverbs (16a).10 Since adverbs are XPs, they cannot occur in Field B(16b). Finally, higher adverbs cannot occur in Field C, although lower adverbswhich can adjoin to VP typically can (16c). This shows that the lexical verb is not aconstituent with the ILV and the coordinator. This is congruent with the proposedstructure (13).

(16) a. Watwhat

gaango

JanJan

waarskynlik/probably

altyd/always

herhaaldelik/repeatedly

vinnig/quickly

sorgvuldig/carefully

morsigmessily

sitsit

enand

eet?eat

‘What is Jan probably/ always/ quickly/ carefully/ repeatedly going to beeating?’ [In A]

. Ernst (2002) points out that many adverbs which adjoin to VP can also adjoin to PredP,which I take to be equivalent to vP – i.e. Field A.

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.9 (808-880)


b. *Watwhat

gaango

JanJan

sitsit

waarskynlik/probably

altyd/always

herhaaldelik/repeatedly

vinnig/quickly


morsigmessily

enand

eeteat

‘What is Jan probably/ always/ quickly/ carefully/ repeatedly going to beeating?’ [*In B]

c. Watwhat

gaango

JanJan

sitsit

enand

*waarskynlik/probably

*altyd/always

*herhaaldelik/repeatedly

?vinnig/quickly


morsigmessily

eet?eat

‘What is Jan probably/ always/ quickly/ carefully/ repeatedly going to beeating?’ [Low adverbs in C]

It is interesting to note that these data also exclude a structure for the Afrikaansconstruction based on clausal subordination. The following kind of subordinativestructure would predict that higher adverbs would be able to occur in Field C.

(17) . . . dat. . . that

JanJan

diethe

olifanteelephants

[VP

[VP

sitsit

[CP

[CP

enand

. . . [VP

. . . [VP


]]]]]]

Since this prediction is falsified by (16c), the clausal subordination structure can-not be correct.

. Distribution of objects

Afrikaans objects usually move to the left of the verb, to what I take to be a specifierof vP (18a).11 However, a subset of objects can also occur in a lower position,presumably the base position for objects (18b).12 No objects can occur in Field B(18c).

(18) a. Daarthere

salwill

altydalways

iemandsomebody

boekebooks

sitsit

enand

leesread

‘There will always be somebody reading books’ [In A]b. Daar

theresalwill

altydalways

iemandsomebody

sitsit

enand

boekebooks

leesread

‘There will always be somebody reading books’ [In C]c. *Daar sal altyd iemand sit boeke en lees [*In B]

. The question of whether the object moves to a specifier of vP or Spec AgrOP does not haveany bearing on the data discussed here.

. Objects appearing in this position tend to be no longer than a single word (Donaldson1993) and so tend to be mass, generic terms or bare plurals. These tend to be the same principlesrestricting the types of non-verbal material occurring in raising-verb clusters in Afrikaans moregenerally (Robbers 1997). Although the question of why these elements are restricted in this wayis beyond the scope of this paper, answers could be found in the framework of Diesing (1997).

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.10 (880-941)

Mark de Vos

Incidentally, it is worth noting that these ‘low’ objects are probably not incorpo-rated into the lexical verb because otherwise they would be pied-piped with thecomplex, coordinated predicate. In fact, they can never be pied-piped in this way(19b). It has also been pointed out by an anonymous reviewer that it is possible tohave both stranded prepositions and objects occupying this field simultaneously.This clearly runs against an incorporation analysis.

(19) a. Waaromwhy

salwill

JanJan

sitsit

enand

boekebooks

lees?read

‘Jan will read books’b. *Waarom

whysitsit

enand

boekebooks

leesread

Jan?Jan

. Distribution of separable particles

Separable particles are XPs occurring in a VP-adjoined position as illustrated in(13). Whether or not they are base-generated in this position (as objects are) orwhether they are generated as the heads of small clauses in a still lower position isnot directly relevant at this point.

The distribution of separable particles has already been touched on in Section1. It is clear from example (1) adapted here as (20a) that a separable particle canoccur in Field C. However, the separable particle cannot occur in either Field A(20b) or Field B (20c).13

(20) a. JanJan

salwill

diethe

olifanteelephants

sitsit

enand


‘Why will Jan chase the elephants away’ [In C]b. */??Jan

Jansalwill

diethe

olifanteelephants

wegaway.prt

sitsit

enand

jaagchase

[*In A]

c. *JanJan

salwill

diethe

olifanteelephants

sitsit

wegaway.prt

enand

jaagchase

[*In B]

Incidentally, these particle distribution facts are important evidence against anapproach to Quirky verb-second based on remnant movement. Over the past sev-eral years, there has been increased interest in recasting verbal head-movementas remnant movement. Approaches such as those of Nilsen (2003) and Biberauer(2003) retain head-movement for verb movement to the head of FinP/TP, but uti-

. For some speakers (20b) is strongly ungrammatical. However, some informants claim thatexamples like (20b) are strongly dispreferred but not ungrammatical. However, for all speakers,the contrast between (20a, c) is robust. Given the contrast and the strength of the relative judge-ments, I consider (20b, c) ungrammatical. It’s ungrammaticality does not follow directly fromstructure (13) and is assumed to be an independent but important fact about this constructionin Afrikaans.

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.11 (941-1001)


lize remnant movement for verb-second itself (i.e. what was traditionally T-to-Cmovement (Den Besten 1989)). Other frameworks follow the stronger hypothesisthat all head-movement is reducible to remnant movement (Mahajan 2000, 2001;Müller 2004). In fact, it might be suggested that Afrikaans Quirky verb-secondis prima facie evidence for this kind of approach. After all, if all apparent head-movement to T did involve remnant movement, then it would be expected thatverbal strings would be pied-piped. To explore this issue more fully, consider thefollowing straw-man outline of a remnant movement derivation (21).

(21) a. Start with a vP shell: [vP ilv en object verb]b. Evacuate the vP shell: object [vP ilv en object verb]c. Move the shell to T: [vP ilv en verb] object [vP ilv en object verb]

The derivation begins with a vP shell containing an coordinated predicate. Thenon-verbal material is then evacuated from the vP. The vP is subsequently frontedto simulate verb-movement.

However, there are good reasons why Afrikaans Quirky verb-second is notderived by remnant movement (De Vos 2004). The first of these is that a deriva-tion like (21) does not conform to independently verified properties of remnantmovement in Afrikaans. It is known, that remnant movement in Germanic ingeneral (Den Besten & Webelhuth 1987), and Afrikaans in particular (Biberauer2004) does not require obligatory evacuation of non-verbal material from the vP.Such material is typically pied-piped. Consider the example in (22) of vP move-ment to Spec TP as proposed by Biberauer (2003). Crucially, in this case, remnantmovement pied pipes all non-verbal material (including the separable particle(22b)).14

(22) a. . . . dat. . . that

hyhe

diethe

olifanteelephants

(ge-)sitpst-sit

enand


hetaux

t

‘. . . that he was chasing the cattle away’b. *. . . dat

. . . thathyhe

diethe

olifanteelephants

(ge-)sitpst-sit

enand

jaagchase

hetaux

wegaway.prt

t

However, it might be argued that notwithstanding the previous argument, all non-verbal material might be extracted from vP for various reasons. For instance,arguments might be extracted for reasons to do with Case. Whatever the mer-its of this approach, there are serious problems with the extraction of separableparticles.

. In fact, this is a crucial property of this movement for Biberauer (2004) who exploits it toprovide an interesting account of the EPP. Also note that the participle prefix ge- is a preferredoption on ILVs in standard Afrikaans but is not necessarily obligatory.

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.12 (1001-1083)

Mark de Vos

Separable particles can typically scramble to the left of a verb cluster in lan-guages like Dutch and Afrikaans when a coordinating marker is not present (23a).However, the presence of an ILV construction does not allow such scrambling asillustrated in (20c), adapted here as (23b).

(23) a. Waaromwhy

souwould

JanJan

diethe

olifanteelephants

≺weg�away.prt

laatallow

≺weg� gaan?go

‘Why would Jan let the elephants go away?b. Jan

Jansalwill

diethe

olifanteelephants

≺*/??weg�away.prt

sitsit

enand

≺weg� jaagchase

This means that, in the context of ILV constructions, separable particles cannotbe scrambled outside the vP for independent reasons. However, these are preciselythe configurations that require the particle to scramble if a remnant movementapproach is to succeed (De Vos 2004). This paradoxical situation is strong evidenceagainst an approach requiring this type of remnant movement.

. Summary of distributions

This section has explored the distributions of subjects, objects, adverbs and sep-arable particles. These distributional data are tabulated here. It is quite clear thatthey are congruent with the structure proposed in (13). Thus, Field C constitutesa field of adjunction for a variety of elements.15 In contrast, Field B permits noelements to intrude. This demonstrates that the coordinator is never adjoined tothe lexical verb, but forms a constituent with the aspectual verb.

The distributional evidence also suggests that explanations in terms of clausalsubordination or remnant movement are unfeasible.

Field A B C

Subjects � No NoHigher Adverbs � No NoLow Adverbs � No �Bare Objects � No �Separable Particles No No �

. These elements are not necessarily mutually exclusive. For instance, it is conceivably possi-ble to construe an object and a stranded preposition in Field C simultaneously. Of course, onlysome syntactic entities can occur in Field C. A full discussion of why, say, definite objects cannotoccur in this field is beyond the scope of this paper. However, an answer may be lie in the direc-tion the analysis of Diesing (1997) who claims that different types of objects occur in differentpositions.

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.13 (1083-1128)


. Coordination and feature bundles

The following section will provide an analysis of Quirky verb-second in Afrikaans.However, before proceeding, since coordination plays a central role in the follow-ing discussion, it is necessary to outline a few fundamental assumptions concern-ing it. The coordinator & takes (at least) two ‘arguments’ X and Y in the following,asymmetric configuration: [& X [& Y]]. Thus where XPs are coordinated, X is aspecifier of & while Y is a complement (Johannessen 1998; Kayne 1994; Progo-vac 1998a, b; Van Koppen 2005; Zoerner 1995). In a bare-phrase-structure system(Chomsky 1995) the same configuration can apply to heads. It is known, that coor-dinative structures are different to other Spec-head-complement structures insofaras the features of the conjuncts are accessible on the mother node (Cormack &Breheney 1994; Johannessen 1998; Van Koppen 2005; Zoerner 1995).

Furthermore, coordinative structures are subject to the Law of Coordinationof Likes (LCL) (Chomsky 1957; Munn 1993; Sag et al. 1985; Schachter 1977). Ithas long been known that a curious, yet important, lexical fact about conjunction,is that in natural language, coordination almost always targets ‘like’ constituents.This property is usually referred to as the Law of Coordination of Likes in theliterature. The level of similarity may not be restricted to only syntactic features,but also extends to the semantic function or functional equivalence (Dik 1968;Haspelmath 2007; Munn 1993; Peterson 2004; Sag et al. 1985).

Coordination is also subject to the Coordinate Structure Constraint (CSC)(Ross 1967), abstracting away from the Across-the-Board (ATB) exception to it.

(24) a. Coordinate Structure Constraint: In a coordinate structure, no elementcontained in a conjunct may be moved out of that conjunct (Ross1967:89).

b. Across the Board Rule Application: In a coordinate structure, the sameconstituent may be extracted from within all the conjuncts simultane-ously (Ross 1967; Williams 1978).

The CSC is illustrated in the following example where no individual WH-itemmay be extracted from any single conjunct, but can be extracted in ATB-fashionfrom both conjuncts simultaneously.

(25) a. *Who did John see Mary and Peter observe t? [CSC]b. Who did John see t and Peter observe t? [ATB]

In its original form, the CSC was a disjunctive condition incorporating a conditionto the effect that no conjunct may be moved (Ross 1967:89). This will be referredto as the Conjunct Condition but will not play any role in the remainder of thisarticle.

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.14 (1128-1175)

Mark de Vos

These fundamental assumptions about coordination are taken to be axiomaticand ultimately a function of a deep, lexical property of coordination. Havinglaid down these assumptions, it is possible to further explore the interaction ofcoordination with verbal heads.

. Deriving the base structure for ILV constructions

In this section, the derivation of the base structure for coordinated predicates willbe outlined. The derivation of example (26) is as follows.

(26) Waaromwhy

salwill

JanJan

virfor.obj-marker

diethe

olifanteelephants

sitsit

enand

loer?look

‘Why will Jan spy on the elephants?’

First, the lexical verb and object are merged, followed by the merging of little v.

(27) v

v VP

Object V

loer

At this point, the coordinative head is merged directly to the v head, whereafterthe posture verb is merged in the same way. This yields the structure in (28). It isimportant to note that this system of direct adjunction is not particularly novel.This is precisely the same adjunction mechanism used by head-movement andexploits the similarity between merge and move.16

(28) v

v VP

Object V

loer

sit v

en v

. The present adjunction mechanism eliminates the well-known problem whereby a head-moved element does not c-command its trace. Since sit ‘sit’ has been merged directly (and nothead-moved), there is no trace that requires government. Admittedly, the question still remainsas to whether such an adjunction obeys the Extension Condition, but this is a problem faced byhead-movement approaches in general and is not unique to the system presented here.

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.15 (1175-1284)


Of course, it is also possible that the complex head sit en v is constructed first andthen merged in one fell swoop with the VP. However, such a derivation is not theonly possible one. This kind of derivation has implications which will be exploredin Section 5.3.

This derives the proposed base structure in (13). Note that the LCL is respectedinsofar as both v and the posture verb are (light) verbs. In addition, the object isnot contained within a coordinative structure, so it can be freely raised to Spec vP(or Spec AgrOP) for Case licensing. This structure accounts for the ‘restructur-ing’ properties of this construction. Importantly, however, the lexical verb and theposture verb do not form a constituent. This opens the possibility for low adverbs,particles etc. to be adjoined at VP level. This accounts for all the distributionalfacts in Section 3.

. Implications of the LCL for coordinated feature bundles

In structures like (13) and (28) where heads are coordinated, the question of whatexactly is being coordinated comes to the fore. The representation in (28) is notprecise enough since it does not necessarily distinguish between two potentialways of representing the coordination of heads. Under traditional assumptions,it is usually categories that are coordinated. However, it could just as well be thatcoordination scopes over features within feature bundles. Assume feature bundlesto be comprised of at least, categorial, formal, phonological and semantic features.These two possibilities are represented in (29) and (30), where the circle informallyrepresents the scope of coordination.

(29)

� �

(30)

� �

In (29), the entire feature bundle is coordinated with another. In (30), it is thefeatures themselves that are coordinated. Generally, these two situations are empir-ically indistinguishable. Morphologically, the feature bundle is always isomorphicwith a particular verbal form regardless of whether coordination operates over

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.16 (1284-1358)

Mark de Vos

some or all of its internal features. From a syntactic perspective, all the features arewithin the scope of coordination in both instances. Thus it would not be possi-ble to extract a feature from one conjunct without violating the CSC. However, itcould be the case that Afrikaans ILV constructions provide a context subtle enoughto distinguish between these two systems.

Consider the situation in (31) where one conjunct is a subset of the other. TheLCL ensures that only ‘like’ features are coordinated. The fact that the first fea-ture bundle has a phonological feature whereas the other does not, means that thephonological feature must remain outside the scope of coordination. This opensthe possibility for that feature to be extracted from the feature bundle withoutincurring a violation of the CSC.17

(31)

� �

. Deriving a simplex initial

Looking back at the structure in (28), it is evident that little v lacks a phono-logical feature whereas the posture verb has a phonological feature by default. Inother words, structure (28) instantiates the situation described in (31). This meansthat the phonological feature can be extracted as part of V-T movement withoutincurring a CSC violation. This derives examples like (2b) repeated here.

(32) T

T vP

VPv

sit v weg V

en v jaag

(33) Waaromwhy

sitsit

JanJan

diethe

olifanteelephants

sit enand



. The fact that one conjunct is a subset of the other is important for constraining this typeof structure; it is equivalent to underspecification of one conjunct. Were it not for this principle,any category could be coordinated with any other category with which it shared at least onefeature. This is not supported by empirical evidence.

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.17 (1358-1402)


Afrikaans ILV constructions thus provide evidence for syntactic operations beingable to manipulate features within feature bundles. In this particular instance, itis coordination that operates over features within feature bundles. Interestingly,this is not at odds with the original formulation of the CSC and LCL which wereoriginally defined as operating over syntactic entities and not categories per se.Nevertheless, the CSC and LCL can now be explicitly defined over features. I callthis subatomic syntax.

(34) a. Subatomic LCL: Coordination always coordinates ‘like’ entities. Where‘entity’ is a feature or set of features.

b. Corollary: A feature (or set of features) may only be coordinated withanother feature (of set of features) of the same type, which are madeavailable by the syntactic structure being coordinated.

(35) a. Subatomic CSC: Extraction from within any coordinated entity is disal-lowed. Where ‘entity’ is a feature or set of features.18

b. Corollary: Extraction out of a coordinative structure is disallowed if thatextraction is from within the coordinated entities themselves. Extractionis allowed if coordination does not scope over the extracted entity.

. Creating and moving a complex initial

Thus far, only half the puzzle has been solved. It is still necessary to derive struc-tures with complex initials, and more importantly, to show that such derivationsare a function of what has already been proposed. Example (2a) is repeated hereas (36).

(36) Waaromwhy

sitsit

enand

jaagchase

JanJan

diethe

olifanteelephants

sit en wegaway.prt

jaag?


The derivation begins with merging the lexical verb, object and then v as forderivation (27).

. The second part of the CSC, namely the Conjunct Constraint is presumably affected in thesame way: extraction of any single conjunct of a coordinated entity in its entirety is disallowed.

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.18 (1402-1449)

Mark de Vos

(37) v

v VP

weg V

jaag

If the lexical verb is to undergo verb-second, then it is necessary that it raises to v atan early stage of the derivation. This is a necessary condition for any theory of V-to-T movement. It is precisely this movement that sets the stage for the derivationof an example like (36).

(38) v

v VP

weg V

t

jaag v

Subsequently, the coordinative head and the posture verb are directly adjoined tothe V+v complex as described in derivations (27) and (28).

(39) v

v VP

weg V

t

sit v

en v

jaag v

Now consider the possibility raised around example (28) about whether is possibleto first generate the verbal head and then merge it directly with VP. With respectto (39) it is clearly not an option to merge the lexical verb together with v, en andsit and only subsequently merge it. It is also not possible to first merge sit en vas a complex head with VP and then raise V to form a complex V + sit en v. Inorder to do so, the lexical verb would have to raise across three heads (i.e. sit enand v) which would violate the Head Movement Constraint. Consequently, thederivation sketched in (39) is the only possible one.

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.19 (1449-1503)


Now consider the nature of the feature bundles in derivation (39). The con-junct containing the posture verb has phonological features by default. Although,little v itself does not have any phonological features, the movement of V to v haseffectively provided the second conjunct (containing V+v) with phonological ma-terial. According to the Subatomic LCL (34), all ‘like’ features must be coordinated.The feature bundles are illustrated in (40) where the circle informally representsthe scope of coordination.

(40)

� �

This means that the phonological feature resides within the scope of coordinationand cannot be extracted individually without incurring a CSC violation.19 Theonly possible outcome is for the entire coordinated head to be pied-piped to T.20

Subsequent operations ultimately derive a verb-second effect. This derives exam-ples like (2a) repeated below. Importantly, this means that head-movement of acomplex, coordinated head is not necessarily movement of a phonological featureat all, but is an operation of narrow syntax.

(41) T

T vP

Subject

V

t

v T

sit v

en

jaag

v

v

v

t VP

ObjectParticle[

[

. ATB movement of the phonological feature in both conjuncts is also excluded on the basisthat ATB cannot move all the phonological material within a conjunct as is illustrated by thefollowing example.

(1) *Who did John meet t and t?

. This tree abstracts away from object movement to Spec vP and subject movement to SpecTP.

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.20 (1503-1576)

Mark de Vos

(42) Waaromwhy

sitsit

enand

jaagchase

JanJan

diethe

olifanteelephants

sit en wegaway.prt

jaag?


At this point, analyses have been proposed that can account for the Afrikaans factsas described in the introduction. The optionality between examples (2a) and (2b)ultimately reduces to the question of whether the lexical verb moves to v or not.When V raises to v, a complex initial results; when V-v raising does not occur, asimplex initial is the outcome.

Contingent movement: Before moving on, it is necessary to note that the presentanalysis does not rule out the following example where the lexical verb has raisedto v even though subsequent movement of the coordinated predicate has notoccurred.

(43) *Waaromwhy

salwill

JanJan

olifanteelephants

[v0 sitsit

enand

jaag]chase

wegaway

jaag

It appears to be an independent property of verb-second contexts that the lexicalverb can only raise to v if it subsequently undergoes verb-second; V-v raising iscontingent on subsequent verb-second. This is illustrated by the following examplewhich shows that this effect is not a property of complex initial formation per se.

(44) *Waaromwhy

salwill

JanJan

diethe

olifanteelephants

jaagchase

wegaway

jaag

It is acknowledged that this appears to be a mysterious phenomenon for all verb-second phenomena. However, it is a question that has to be addressed within theauspices of a general theory of verb second and is beyond the scope of this paper.

. Other types of moved verbal clusters

Until now, the entire discussion has focussed exclusively on coordinated predicatesformed with ILVs. However, there are also other types of verbal clusters that canundergo V-to-T movement. These include complex initials with aspectual and re-structuring, raising verbs (i.e. Direct Linking Verbs/DLVs). It will be demonstratedthat these types of complex can also be accommodated in the proposed analysis.

Afrikaans complex initials are not restricted to ILVs with overt coordinators.A subset of aspectual verbs capable of undergoing verb-raising are Direct LinkingVerbs (DLVs). These can also occur in complex initials. Verb-raising refers to thecreation of a contiguous verbal cluster (bold) in OV languages like Dutch andAfrikaans (Evers 1976).

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.21 (1576-1647)


(45) . . . dat. . . that

JanJan

hetthe.neut

balletjeball-dim

ergenssomewhere

moetmust

latenlet-inf

vallenfall-inf

‘. . . that Jan must drop the ball somewhere’ [Dutch]

(46) . . . dat. . . that

JanJan

diethe

balball

iewerssomewhere

moetmust

laatlet

valfall

‘. . . that Jan must drop the ball somewhere’ [Afrikaans]

Afrikaans DLVs include verbs like bly ‘keep on doing’ (47), kom ‘come’(48) andlaat ‘causative’ (49).21 Most of these can also optionally alternate between sim-plex and complex initials. Importantly, however, the coordinator is not overt inthese cases.

(47) a. Waaromwhy

blystay

leesread

JanJan

diethe

boek?book

‘Why does Jan keep reading the book?’b. Waarom

whyblystay

JanJan

diethe

boekbook

lees?read

(48) a. Waaromwhy

komcome

eeteat

JanJan

bywith

ons?us

‘Why does Jan come and eat at our house?’b. Waarom

whykomcome

JanJan

bywith

onsus

eet?eat

(49) a. Diethe

heelagterfull-back

laatlet.caus

valfall

diethe

balball

‘The full-back drops the ball’ (Van Niekerk 1995:150)b. Die

theheelagterfull-back

laatlet.caus

diethe

balball

valfall

‘The full-back drops the ball’ (Van Niekerk 1995:150)

The class of DLVs in Afrikaans is not homogenous (De Vos 2005; Robbers 1997)and full justice cannot be done to the entire class here. Nevertheless, it has beenshown that many DLVs have formal properties very similar to ILVs (see De Vos(2005) for a comprehensive overview). Thus, there does not appear to be any rea-son why the same structure that was applied to ILV structures (13) should not alsobe applied to DLV structures. The only caveat is that in these cases the coordinatoris covert.22

. In addition, so called ‘control’ verbs such as probeer ‘try’ and begin ‘to begin’ can also occurin these kinds of constructions (De Vos 2005).

. Interestingly, in Nupe, it is precisely the coordination of verbal categories that requires anull coordinator. In addition, Nupe also allows for extraction from the verb cluster (Kandybow-icz 2005).

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.22 (1647-1683)

Mark de Vos

As for the non-pied-piping cases, in the absence of V-v raising, v remainsdevoid of phonological features. This means that the phonological features ofthe DLV (in this instance laat) must remain outside the scope of coordination.Consequently, they are free to move to T without incurring a violation of the CSC.

(50) T

T vP

Subject

V

val

v

laat

EN

v

v

v

VP

ObjectParticle[

[

(51) Diethe

heelagterfull-back

laatlet.caus

diethe

balball

valfall


However, when V-v raising occurs, v is lexicalized by V. Thus, both the first con-junct (containing the DLV) and the second conjunct (containing V-v) have phono-logical features. Consequently, the phonological features lie within the scope ofcoordination and the entire complex predicate must be pied piped, should move-ment be required.

(52) T

T vP

Subject

V

t

v T

laat v

EN

val

v

v

v

t VP

ObjectParticle[

[

(53) Diethe

heelagterfull-back

laatlet.caus

valfall

diethe

balball


JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.23 (1683-1733)


The proposed structure, in conjunction with the LCL and CSC can also deriveQuirky Verb Movement with DLVs.

A further prediction is that DLVs and ILVs should be able to be coordi-nated together. In natural language, coordination allows conjuncts themselves tobe coordinated. Since the proposed structures (13) and (50) explicitly utilize co-ordination, it should also be possible to coordinate DLVs and ILVs in a singlestructure.

(54) vP

SUBJ v

v VP

OBJ Vv‚1

‚2DLV1A v en

EN ILV1B

VERB

This prediction is fulfilled by examples like (55) which illustrate pied-piping ofa complex verbal predicate with both an overtly coordinated ILV and a (covertlycoordinated) DLV.23

(55) Waaromwhy

loopwalk

staanstand

enand

leesread

JanJan

diethe

boek?book

‘Why is Jan (willfully) reading the book?’ [CI]

What is interesting about this structure (54) is that it makes a prediction aboutwhat verbs can be pied-piped and which cannot. The second conjunct (subscript2) consists of v. Since it lacks phonological features, it predicts that the first con-junct (subscript 1) should be able to be pied-piped. This is demonstrated in (56).Such examples are possible but not common, according to Ponelis (1993).

(56) Waaromwhy

loopwalk

staanstand

JanJan

diethe

boekbook

enand

lees?read

‘Why is Jan (willfully) reading the book?’

. Note that DLV loop seems to have a pejorative reading in this context. This type of readingis often associated with pseudo-coordination in many languages but is not a necessary conse-quence of it. See De Vos (2005) for a discussion of semantic properties of pseudo-coordination.

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.24 (1733-1813)

Mark de Vos

In structure (54), within the first conjunct (subscript 1), there is an overt verbin each conjunct (1A and 1B respectively). There is thus phonological materialin both conjuncts. This predicts that it will not be possible to extract the DLVseparately, but that the entire first conjunct (i.e. v1) should be pied-piped. Theimpossibility of extracting the first verb individually is illustrated in (57) which isworse than either of the preceding two examples.

(57) ??Waaromwhy

loopwalk

JanJan

diethe

boekbook

staanstand

enand

lees?read

‘Why is Jan (willfully) reading the book?’

Thus, it would appear that the proposed structure makes predictions about whichheads can be excorporated from the complex predicate. These predictions turn outto be true.

. Conclusion

This paper has explored Afrikaans complex, coordinated predicates with IndirectLinking Verbs. These appear to act both as contiguous, verbal heads when theyoccur in T, but also as non-constituents when they occur in situ. Moreover, theyalso allow for optional pied-piping of the entire coordinated verbal cluster to T.This appears to be fairly unique among the Germanic verb-second languages andhitherto there was no detailed account for it.

An analysis is proposed that derives the alternations from basic assumptionsabout coordination. It is suggested that the Law of Coordination of Likes and theCoordinate Structure Constraint may operate, not only over feature bundles andcategories, but also over subsets of feature bundles. In fact, this is not really a re-formulation of the LCL and CSC so much as a strong interpretation of them. Thisis because the original constraints were not formulated with respect to categoriesper se but with respect to syntactic elements.

The proposal has implications for theories of excorporation insofar as it al-lows for limited excorporation under very specific circumstances. In addition, theanalysis also suggests that head movement can be movement of phonological fea-tures. This is a common point with the proposal of Zwart (1997) who arguesthat verbal head movement is feature movement. Importantly, however, the pro-posal also demonstrates that entire, coordinated predicate structures can undergohead movement. The implication is that head-movement is not necessarily alwaysmovement of phonological features but also has some substance in narrow syntaxcontra Chomsky (2000) and Boeckx & Stjepanovic (2001).

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.25 (1813-1966)


References

Barbiers, S. (2000). The Right Periphery in SOV Languages: English and Dutch. In P. Svenonius(Ed.), The Derivation of VO and OV (pp. 181–218). Amsterdam: John Benjamins.

Biberauer, T. (2002). Verb Second in Afrikaans: Is This a Unitary Phenomenon? Stellenboschpapers in linguistics, 34, 19–69.

Biberauer, T. (2003). Verb Second (V2) in Afrikaans: A Minimalist Investigation of Word OrderVariation. PhD thesis, St. John’s College, Cambridge.

Biberauer, T. (2004). Reconsidering the EPP and Spec-TP in Germanic. Cambridge OccasionalPapers in Linguistics, 1, 15–40.

Boeckx, C. & Stjepanovic, S. (2001). Head-Ing Toward PF. Linguistic Inquiry, 32, 345–355.Chomsky, N. (1957). Syntactic Structures. The Hague: Mouton.Chomsky, N. (2003). Bare Phrase Structure. In G. Webelhuth (Ed.), Government and Binding

theory and the Minimalist Program (pp. 383–439). Cambridge MA: Basil Blackwell.Chomsky, N. (2000). Minimalist Inquiries: The Framework. In R. Martin, D. Michaels, & J.

Uriagereka (Eds.), Step by Step (pp. 89–155). Cambridge MA: MIT Press. Reprinted.Cormack, A. & Breheney, R. (1994). Projections for Functional Categories. UCL working papers

in linguistics, 6, 35–62.De Vos, M. (2004). Verbal Compounding and Complex Initials. In D. Austin, V. Chondrogianni,

E. Daskalaki, N. Katsos, M. Mavrogiorgos, G. Newton, E. Orfanidou, M. J. Reeve, & J. Srioutai(Eds.), Proceedings of the Second Postgraduate Conference on Language Research. Cambridge,UK.

De Vos, M. (2005). The Syntax of Pseudo-Coordination in English and Afrikaans. Ph.D. thesis,Leiden. Published by Igitur Publishing http://www.lotpublications.nl/index3.html.

Den Besten, H. (1988). Universal-Grammatik und/Oder Zweitsprachenerwerb: Der FallAfrikaans. In N. Boretzky, W. Enninger, & T. Stolz (Eds.), Essener Kolloquium über Sprachkon-takt, Sprachwandel, Sprachtod (pp. 11–44). Bochum: Studienverlag Dr. N. Brockmeyer.

Den Besten, H. (1989). Studies in West Germanic Syntax. Ph.D. thesis, Katholieke UniversiteitBrabant, Tilburg, Amsterdam.

Den Besten, H. (2002). Khoekhoe Syntax and its Implications for L2 Acquisition of Dutch andAfrikaans. In R. Mesthrie, and P. Roberge (Eds.), Focus on Afrikaans Sociohistorical Linguistics.Part II, Journal of Germanic Linguistics, 14, 3–56.

Den Besten, H. & Webelhuth, G. (1987). Remnant Topicalization and the Constituent Structureof the VP in the Germanic SOV Languages. GLOW Newsletter, 18, 15–16.

Diesing, M. 1997. Yiddish VP Order and the Typology of Object Movement in Germanic. Natu-ral language and linguistic theory, 15, 369–427.

Dik, S. (1968). Coordination: Its Implications for a Theory of General Linguistics. Amsterdam:North Holland Publishing Company.

Donaldson, B. (1993). A Grammar of Afrikaans. Berlin: Mouton de Gruyter.Ernst, T. (2002). The Syntax of Adjuncts. [Cambridge Studies in Linguistics 96]. Cambridge, UK:

Cambridge University Press.Evers, A. (1976). The Transformational Cycle in Dutch Language and German Language. Ph.D.

thesis, Ann Arbor MI.Gerritsen, M. (1991). Atlas Van de Nederlandse Dialectsyntaxis (AND). Amsterdam: P. J.

Meertens Instituut.Haslinger, I. & Van Koppen, M. (2003). De Verbale Hendiadys Als Pseudocoördinatie. Taal en

Tongval, 15–16, 102–122.

JB[v.20020404] Prn:24/10/2006; 13:55 F: LA9703.tex / p.26 (1966-2107)

Mark de Vos

Haspelmath, M. (in press 2007). Coordination. In T. Shopen (Ed.), Language typology andsyntactic description II. (2nd ed.). Cambridge. Cambridge University Press.

Ijbema, A. (2003). Grammaticalization and Reanalysis in Dutch Aspectual Constructions, Ms,University of Leipzig.

Johannessen, J. (1998). Coordination. New York: OUP.Kandybowicz, J. (2005). On the Syntax of Nupe Coordination. Presented at BLS 31, 10 February

2005.Kayne, R. (1994). The Antisymmetry of Syntax. Cambridge MA: MIT Press.Kuteva, T. (1999). On sit/ stand/lie auxiliation. Linguistics, 37(2), 191–213.Mahajan, A. (2000). Eliminating head movement. GLOW Newsletter, 44, 44–45.Mahajan, A. (2001). Word Order and (Remnant) Movement. PhD dissertation, UCLA.Müller, G. (2004). Verb-second as VP first. Journal of Comparative Germanic Linguistics, 7(3),

179–234.Munn, A. B. (1993). Topics in the Syntax and Semantics of Coordinate Structures. PhD disser-

tation, University of Maryland.Nilsen, Ø. (2003). Eliminating Positions: Syntax and semantics of sentential modification. PhD

dissertation, Utrecht Institute of Linguistics OTS.Peterson, P. (2004). Coordination: Consequences of a lexical-functional account. Natural lan-

guage and linguistic theory, 22, 643–679.Ponelis, F. (1993). The Development of Afrikaans. Frankfurt am Main: Die Deutsche Bibliothek.Progovac, L. (1998). Structure for coordination (Part I). GLOT International, 3(7), 3–6.Progovac, L. (1998). Structure for Coordination (Part II). GLOT International, 3(8), 3–9.Robbers, K. (1997). Non-Finite Verbal Complements in Afrikaans. PhD dissertation, University

of Amsterdam.Roberge, P. (1994). On the origins of the Afrikaans verbal Hendiadys. Stellenbosch papers in

Linguistics, 28, 45–81.Ross, J. R. (1967). Constraints on Variables in Syntax. PhD dissertation, Massachusetts Institute

of Technology.Sag, I., Gazdar, G., Wasow, T., & Weisler, S. (1985). Coordination and how to distinguish

categories. Natural language and Linguistic theory, 3, 117–171.Schachter, P. (1977). Constraints on coordination. Language, 53, 86–103.Van Koppen, M. (2005). One Probe-Two Goals: Aspects of agreement in Dutch dialects. PhD

dissertation, University of Leiden.Van Niekerk, J. (1995). ‘n Medewerkwoordklien vir Afrikaans. South African Journal of Linguis-

tics, 13(4), 146–153.Williams, E. (1978). Across the board rule application. Linguistic Inquiry, 9, 31–43.Zoerner, E. (1995). Coordination: The syntax of & PhD dissertation, University of California,

Irvine.Zwart, J. W. (1997). Morphosyntax of Verb Movement: A minimalist approach to the syntax of

Dutch [Studies in Natural Language and Linguistic Theory]. Dordrecht: Kluwer.

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.1 (46-123)

Nominal arguments and nominal predicates

Marit JulienLund university, Sweden

It is generally assumed that whereas referential nominal phrases are at leastof DP size, predicative nominal phrases are structurally smaller. However, thispaper argues that at least in Scandinavian, predicative nominal phrases as wellas referential nominal phrases can be bigger than, equal to, or smaller than DP.Hence, the only consistent difference between nominal arguments and nominalpredicates appears to be a semantic one: predicates but not arguments get a purelyintensional interpretation. Some nominal phrases are only good as argumentsbecause their lexical content makes a purely intensional reading hard to get. It ishowever difficult, if not impossible, to connect the semantic difference betweennominal predicates and nominal arguments to any systematic difference insyntactic structure.

. Introduction

The idea that predicative nominal phrases are structurally smaller than argumentalnominal phrases, as argued by Hudson (1989), Bowers (1988, 1993a), Holmberg(1993), Longobardi (1994), Mandelbaum (1994), Zamparelli (2000) and others,has become more or less the standard view on the matter. More specifically, ac-cording to this view the argumental version of a nominal phrase contains structurethat is not present in its predicative counterpart, even when the two look similar,as is the case with the argument a cat in (1a) and the predicate a cat in (1b). Forexample, one would say that the argument is a DP while the predicate lacks the Dprojection.1

(1) a. KalleKalle

sågsaw

einindef.masc.sg

litenlittle.indef.masc.sg

katt.catMASC

‘Kalle saw a little cat.’

. The following abbreviations are used in the glosses: def = definite, indef = indefinite, inf =infinitive, masc = masculine, p = preposition, pl = plural, poss = possessive, pres = present,refl = reflexive, sg = singular.

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.2 (123-189)

Marit Julien

b. KalleKalle

erbe.pres

einindef.masc.sg

litenlittle.indef.masc.sg

katt.catMASC

‘Kalle is a little cat.’

However, it appears that in Scandinavian, full DPs, and even larger phrases, canappear not only as arguments but also as predicates. At the same time, nominalarguments as well as predicates can be structurally smaller than DPs. In short, theidea that the contrast between functioning as an argument and functioning as apredicate corresponds strictly to a contrast in syntactic category seems to find nosupport in Scandinavian.

. Bare singular nominals

It is relatively well known that although all Scandinavian varieties except Icelandichave indefinite singular determiners, countable singular nouns often appear with-out a determiner but possibly with an adjective even in Scandinavian outside ofIcelandic. Such bare singular nouns, or BSNs, can be predicates or arguments, asthe following Norwegian examples illustrate:

(2) a. HegeHege

erbe.pres

lærar.teacher

‘Hege is a teacher.’b. Hege

Hegekjøptebought

(ny)new

billett.ticket

‘Hege bought a (new) ticket.’

As Borthen (2003) discusses in much detail, the acceptability of BSNs dependson the context. One type of context that allows BSNs is the ‘conventional situationtype’, which Borthen (2003:136) defines as “a property, state or activity that occursfrequently or standardly in a given contextual frame and has particular importanceor relevance in that frame.” Being a teacher and buying a ticket are conventionalsituation types according to the definition, and consequently, the BSNs in (2a) and(2b) are both fully acceptable.

Other context types that license BSNs are the ‘comparison of types’ con-text, exemplified in (3), and contexts of acquisition, possession and production,exemplified in (4).

(3) (Ein)a

pistolpistol

erbe.pres

eita

meirmore

effektivtefficient

våpenweapon

ennthan

(ein)a

brød-kniv.bread-knife

‘A pistol is a more efficient weapon than a bread knife.’

(4) KariKari

fikkgot

(ein)a

nynew

sykkel.bike

‘Kari got a new bike.’

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.3 (189-249)


In (2), the relevant contextual frames that allow the BSNs to appear comeready-made by the culture. But even in cases where there are no such ready-madeframes, it is possible to construct new contextual frames where BSNs will be se-mantically acceptable. One way to do this is to imagine a test or contest thatconsists of several tasks. Then each task will be a conventional situation type in thatcontext, and BSNs will be allowed. For example, the following becomes acceptablein such a context:

(5) Kandidat-a-necandidate-pl-def

måttehad.to

gigive

epleapple

tilto

elefant.elephant

‘The candidates had to give (at least one) apple to (at least one) elephant.’

On the whole, there are no lexical restrictions concerning which nouns can appearas BSNs. Individual nouns differ, however, with respect to how easy it is to providea context that will make them appropriate as BSNs. For nouns that can be partof VPs that denote situations that are culturally salient, it is very easy to give BSNexamples, but for other nouns it takes more effort to find a context where they canbe part of the description of a conventional situation type. Hence, some nounsappear to appear more readily as BSNs than others, but this is only a consequenceof a difference in the availability of an appropriate context.

On the syntactic side, Borthen (2003) shows that BSNs cannot take wide scopeover quantifiers. We see this if we compare (6a) to (6b). The expression in (6a) isambiguous, such that either alle ungane ‘all the children’ takes scope over ei jakke‘a jacket’, or else ei jakke ‘a jacket’ takes scope over alle ungane ‘all the children’. Theexpression in (6b), by contrast, only allows an interpretation where alle unganetakes scope over jakke.

(6) a. All-eall-pl

ung-a-nechild-pl-def

prøvdetried

eia

jakke.jacket

(alle > ei jakke; ei jakke > alle)

‘All the children tried on a jacket.’b. Alle

all-plung-a-nechild-pl-def

prøvdetried

jakke.jacket

(alle > jakke; * jakke > alle)

‘All the children tried on some jacket or other.’

Moreover, a BSN cannot take wide scope over negation either. I demonstrate thisin (7).

(7) a. EgI

prøvdetried

ikkjenot

eia

jakke.jacket

(Neg > ei jakke; ei jakke > Neg)

‘I did not try on a jacket.’b. Eg

Iprøvdetried

ikkjenot

jakke.jacket

(Neg > jakke; * jakke > Neg)

‘I didn’t try on any jacket.’

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.4 (249-318)

Marit Julien

It is possible, as Pereltsvaig (2002) suggests, that BSNs take low scope because theydo not involve a quantifier at all. As we see from the translation of (5), a BSN doesnot imply a specified quantity, and does not refer to an individual. Either a singularor a plural entity can be pragmatically understood. Moreover, if an overt quantifieris included, the phrase no longer has the properties that characterise BSNs.

The absence of quantification in BSNs is also seen from their inability to makean activity predicate telic. Thus, whereas (8a) is fine with or without an indefinitearticle in front of lue ‘hat’, only the version with the article allows an in X time ad-verbial – see the contrast between (8b) and (8c). That is, while an object concistingof a singular noun with an article will supply the event with an end point, a BSNobject will not.

(8) a. AnneAnne

strikkaknitted

(ei)a

lue.hat

‘Anne knitted a hat.’b. Anne

Annestrikkaknitted

eia

luehat

påon

totwo

timar.hours

‘Anne knitted a hat in two hours.’c. *Anne

Annestrikkaknitted

luehat

påon

totwo

timar.hours

‘Anne knitted a hat in two hours.’

It is furthermore very difficult to get BSNs to take part in binding. Since itis hard to find relevant examples with BSNs as subjects (although BSNs can inprinciple be subjects), the binding properties of these nominals are better testedwith certain ditransitive verbs. The pattern we will use is one where a verb takesa theme argument and a location argument. The theme can then bind into thelocation phrase, as the Norwegian example in (9) shows.

(9) EgI

harhave

settput

{eina

papegøyei/papegøye-ni}parrot/parrot-def

påon

pinne-npin-def

sini.3refl.poss

‘I have placed a/the parrot on its perch.’

If we now revert to our multi-task context again, and ask if we could have a BSNas object of the infinitive in (10), it appears that an overwhelming majority ofScandinavian speakers reject this. They require the object to have a determiner, asindicated.

(10) Oppgåvetask

nummernumber

totwo

varwas

åto

setteput

*(ein)a

papegøyei

parrotpåon

pinn-enstick-def

sini.refl.poss‘Task number two was to put a parrot on its perch.’

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.5 (318-367)


The inability to take part in binding, to have individual reference, or to takehigh scope, and the obligatory absence of quantifiers suggest that ScandinavianBSNs lack the higher functional projections that we find in ordinary nominalphrases. That is, they do not have a D-projection or any projection associated withquantification. They only contain the projections associated with the noun itselfplus the adjectival layer. They can nevertheless be arguments as well as predicates,and moreover, they are subject to the same semantic licensing conditions regard-less of whether they are arguments or predicates. This shows that the syntacticcontrast between nominal arguments and nominal predicates is not as clear-cut aswe tend to think.

. Some tests for predicates

Having concluded that Scandinavian BSNs can be arguments although they aresmaller than DPs, we will now look at the argument/predicate distinction fromthe opposite angle, and ask how big a nominal predicate can be, structurally. Wethen need tests by which nominal predicates can be identified. We will apply thesetests to copular constructions with two nominals, of the general form Nominal isNominal, which can be either equative statements or predications. In an equativestatement, that is, a proposition where identity of reference between two nominalphrases is asserted or negated (see Heycock & Kroch 1998), both nominal phrasesare necessarily referential, and also, according to the standard view, of DP sizesyntactically. In a predication, on the other hand, the predicative nominal phraseis not referential, and on the view that is dominant presently, it is smaller thana DP syntactically. In any case, it turns out that the two types of nominals reactdifferently to certain syntactic operations. These operations then constitute teststhat we can use to distinguish nominal phrases that are predicates from nominalphrases that are referential.

. Embedding under ‘consider’

One property that characterises a predicative nominal phrase is that it can be thepredicate of a small clause embedded under verbs like ‘consider’ (Partee 1987;Doron 1988), and it cannot then invert with the subject (Adger & Ramchand2003). Consider the copula construction in (11a). Ein flink lærar ‘a skilful teacher’must be a predicate here, since the embedding in (11b), with ein flink lærar inpredicate position, is perfectly fine. The construction in (11c), by contrast, is se-mantically odd, as signalled by the # in front of the example. The problem is thatthe name Berit is bad as a predicate (unless it gets a very unusual interpretation).

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.6 (367-424)

Marit Julien

(11) a. BeritBerit

eris

eina

flinkskilful

lærar.teacher

‘Berit is a skilful teacher.’b. Eg

Ianserconsider

BeritBerit

somas

eina

flinkskilful

lærar.teacher

‘I consider Berit a skilful teacher.’c. #Eg

Ianserconsider

eina

flinkskilful

lærarteacher

somas

Berit.Berit

‘I consider a skilful teacher Berit.’

If we now take the construction in (12a), we see from (12b) and (12c) that it isnot good as a complement of ‘consider’, regardless of which of the two nominalphrases is put in the predicate position. This shows that there is no predicate in(12a), and that we are dealing with an equative statement.

(12) a. Denthat

mann-enman-def

eris

ChristerChrister

Platzack.Platzack

‘That man is Christer Platzack.’b. #Eg

Ianserconsider

denthat

mann-enman-def

somas

ChristerChrister

Platzack.Platzack

‘I consider that man Christer Platzack.’c. #Eg

Ianserconsider

ChristerChrister

PlatzackPlatzack

somas

denthat

mann-en.man-def

‘I consider Christer Platzack that man.’

As for the element som ‘as’, which (obligatorily) appears in front of the predicate inthe embedded small clauses shown above, it is arguably a realisation of a Pred head(see Eide & Åfarli 1999, 2001). On the assumption that all nominal predicates arecomplements of Pred heads there must also be a Pred head in (11a). The reasonwhy it is not visible here must have to do with the PredP being the complement ofthe copula. I will not go into details in this matter, however.

. Topicalisation

In Scandinavian, topicalisation is another test for predicatehood. As Pereltsvaig(2001) points out, only if the postcopular constituent is a true predicate can it un-dergo topicalisation in these languages, such that it moves to clause-initial positionwhile the subject stays in place. That is, if the predicate moves to front in a negatedcopular construction, the subject will still precede the negation. Irrelevantly, theverb will move across the subject to yield V2 order. I show in (13) that this holdsfor a copular construction with a nominal predicate, and in (14), I show the samething for a copular construction with an adjectival predicate. To emphasise the fact

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.7 (424-494)


that the claims put forward in this paper apply to Scandinavian in general, I givethese examples in Swedish.

(13) a. Berit är inte lärare. b. Lärare är Berit inte.Berit is not teacher teacher is Berit not‘Berit is not a teacher.’ ‘A teacher, Berit is not.’

(14) a. Berit är inte dum. b. Dum är Berit inte.Berit is not stupid stupid is Berit not‘Berit is not stupid.’ ‘Stupid, Berit is not.’

In an equative statement, by contrast, the two nominal phrases may switch posi-tions, so that one or the other becomes the topic. But regardless of which phraseis the topic, the other phrase will follow the negation. Thus, if we start from theequative statement in (15a) and want to make Christer Platzack the topic, we can-not do this simply by moving Christer Platzack, as in (15b); we also have to makesure that the other nominal phrase ends up following the negation, as in (15c).

(15) a. Prefekt-enprefect-def

påat

nordiskaNordic

äris

intenot

ChristerChrister

Platzack.Platzack

‘The prefect of the Nordic languages department is not Christer Platzack.’b. #Christer

ChristerPlatzackPlatzack

äris

prefekt-enprefect-def

påat

nordiskaNordic

inte.not

‘Christer Platzack, the prefect of the Nordic languages department is not.’c. Christer

ChristerPlatzackPlatzack

äris

intenot


påat

nordiska.Nordic

‘Christer Platzack is not the prefect of the Nordic languages department.’

In (15c) Christer Platzack has become the surface subject. We can conclude thatan equative statement does not allow the postcopular nominal phrase to be top-icalised. In a nominal predicative construction, on the other hand, the predicatemay well be topicalised and thereby moved across the subject.

. Pseudoclefting

As Fodor (1970) observed, and Zamparelli (2000) also points out, predicativenominals but not referential nominals can be predicates in pseudocleft construc-tions. This is true even of Scandinavian. In (16a), we see that an adjectival predicatecan be pseudoclefted, and in (16b), we see that a nominal predicate can undergothis operation. If we on the other hand pseudocleft a referential nominal phrase,as in (16c), the result is semantically odd.

(16) a. Detit

BeritBerit

är,is

äris

jätte-duktig.giant-skilful

‘What Berit is, is extremely skilful.’

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.8 (494-554)

Marit Julien

b. Detit

BeritBerit

är,is

äris

lärare.teacher

‘What Berit is, is a teacher.’c. #Det

itdenthat

därthere

mann-enman-def

är,is

äris

ChristerChrister

Platzack.Platzack

‘What that man is, is Christer Platzack.’

. Nonrestrictive relative clause

Another difference between referential and predicative nominals mentioned byZamparelli (2000) is that referential nominals but not predicative nominals cancombine with nonrestrictive relative clauses such that the relativised constituentrefers to an individual.2 Thus, while the name in (17a) is fine together with a non-restrictive relative clause, the BSN in (17b), which unmistakably is a predicate, isnot. If we add an indefinite determiner, as in (17c), thereby creating a nominalphrase that can be referential in this context, the construction is rescued.

(17) a. Denthat

mann-enman-def

äris

ChristerChrister

Platzack,Platzack

somthat

jagI

för.övrigtby.the.way

kännerknow

väl.well

‘That man is Christer Platzack, who I by the way know well.’b. *Berit

Beritäris

lärare,teacher

somthat

jagI


kännerknow

väl.well

c. BeritBerit

äris

ena

lärare,teacher

somthat

jagI


kännerknow

väl.well

‘Berit is a teacher, who I by the way know well.’

. Coordination

The last test to be presented here is coordination of postcopular phrases. Note firstthat (18), where an adjectival predicate is coordinated with a nominal predicate,is fully grammatical. (From now on, I will use Norwegian examples to representScandinavian, since it is easier for me to get them right, but that does not meanthat my conclusions hold only for Norwegian. Also note the BSN in final positionin (18).)

(18) BusterBuster

eris

eina

ivrigkeen

jegerhunter

ogand

godgood

tiltoP

åtoINF

tatake

fugl.bird

‘Buster is a keen hunter and good at catching birds.’

. But as Doron (1988) notes, nominal predicates can combine with nonrestrictive relativeclauses where a property-denoting expression is relativised.

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.9 (554-609)


Actually, the conjuncts in (18) are probably both PredPs (Bowers 1993b; Baker2003), which means that we have not coordinated constituents belonging to dif-ferent lexical categories after all. Nevertheless, not all postcopular phrases can becoordinated with equal success. For example, although the copular constructionsin (19a) and (19b) are both perfectly acceptable, coordinating the postcopularphrases of these two constructions, as in (19c), gives a bad result.

(19) a. Prefekt-enprefect-def

påat

nordiskNordic

eris

ChristerChrister

Platzack.Platzack

‘The prefect of the Nordic languages department is Christer Platzack.’b. Prefekt-en

prefect-defpåat

nordiskNordic

eris

eina

hyggelegnice

fyr.guy

‘The prefect of the Nordic languages department is a nice guy.’c. #Prefekt-en

prefect-defpåat

nordiskNordic

eris

ChristerChrister

PlatzackPlatzack

ogand

eina

hyggelegnice

fyr.guy

‘The prefect of the Nordic languages department is Christer Platzack anda nice guy.’

The reason for the oddity of (19c) is that we have coordinated a referential nom-inal phrase with a predicate. Hence, coordination can be used to show that twophrases are not of the same semantic type. Nominal phrases that can be coordi-nated with another phrase that is clearly referential must themselves be referential,while nominal phrases that can be coordinated with another phrase that is clearlya predicate – an adjectival phrase, for example – must itself be a predicate.

. Summing up the tests

Let us now sum up the tests discussed in this section. We have seen that a nominalpredicate can be in predicate position under ‘consider’, it will leave the subject infront of the negation when it is topicalised, and it will also allow pseudoclefting,but it cannot combine with a nonrestrictive relative clause. A referential nominalphrase, by contrast, cannot be in predicate position under ‘consider’, and whenit is topicalised, it will not allow another phrase to stay in the subject position.Moreover, it does not allow pseudoclefting, but it can combine with a nonrestric-tive relative clause. Finally, a nominal phrase is a predicate if it can be coordinatedwith another predicate, and it is referential if this is not possible.

Equipped with these tests, we will now turn to the question of the size ofpredicate nominals, with Scandinavian as our testing ground.

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.10 (609-654)

Marit Julien

. Definite nominal phrases

It is clear that definite nominal phrases can be predicates (as pointed out e.g. byPartee 1987). For example, if we start from a simple copula construction as the onein (20a), with a definite nominal in postcopular position, we can embed it under‘consider’ and have the definite nominal in predicate position, as in (20b), and wecan topicalise the definite nominal and have the subject stay in place, as in (20c).Further, the definite nominal phrase can be pseudoclefted, as in (20d), and it canbe coordinated with an adjectival phrase, as in (20e).

(20) a. JennyJenny

eris

vinnar-en.winner-def

‘Jenny is the winner.’b. Eg

Ianserconsider

JennyJenny

somas


‘I consider Jenny the winner.’c. Vinnar-en

winner-deferis

JennyJenny

ikkje.not

‘The winner, Jenny is not.’d. Det

itJennyJenny

er,is

eris


‘What Jenny is, is the winner.’e. Jenny

Jennyeris

yngreyounger

ennthan

degyou

menbut

likevelstill

vinnar-enwinner-def

avof

løp-et.race-def

‘Jenny is younger than you but nevertheless the winner of the race.’f. Jenny

Jennyeris

vinnar-en,winner-def

somthat

forrestenby.the.way

kjemcome.pres

her.here

‘Jenny is the winner, who by the way comes here.’

All these properties point to the same conclusion: the definite nominal vinnaren‘the winner’ can be a predicate. The fact that it can also combine with a nonrestric-tive relative clause, as in (20f), only serves to show that it can also be referential.

In themselves, these examples do not tell us much about the size of nominalpredicates, however, since the suffixed definiteness marker in Norwegian is not arealisation of D. As pointed out in Taraldsen (1990) and in many later works onScandinavian nominal phrases, when an adjective, a numeral, or a weak quantifierprecedes the definite noun, there is also normally a definiteness marker in front ofthe modifier, as illustrated in (21).3

. Some exceptions to this general pattern are discussed e.g. in Julien (2005).

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.11 (654-721)


(21) JennyJenny

eris

dendef.sg

udiskutableundisputable


‘Jenny is the undisputable winner.’

We would therefore take the preposed definiteness marker to represent D, whilethe suffixed definiteness marker represents a head much lower down in the func-tional domain of the head noun, below the surface position of adjectives. Thus,a nominal phrase consisting of nothing but a suffixed noun need not contain thehigher structural layers that we find in DPs. In fact, since even BSNs may combinewith adjectives, we cannot tell, just by looking at it, whether the nominal predicatein (20) is any bigger structurally than a BSN.

However, even the nominal phrase given in (21) can be a predicate. As Idemonstrate in (22), it can be in predicate position under ‘consider’, it can be top-icalised across the subject, it can be pseudoclefted, and it can be coordinated withan adjectival phrase.

(22) a. EgI

anserconsider

JennyJenny

somas

dendef.sg



‘I consider Jenny the undisputable winner.’b. Den

def.sgudiskutableundisputable

vinnar-enwinner-def

eris

JennyJenny

ikkje.not

‘The undisputable winner, Jenny is not.’c. Det

itJennyJenny

er,is

eris

dendef.sg



‘What Jenny is, is the undisputable winner.’d. Jenny

Jennyeris

yngreyounger

ennthan

degyou

menbut

likevelstill

dendef.sg


vinnar-en.winner-def‘Jenny is younger than you but nevertheless the undisputable winner.’

If I am right in assuming that the preposed definite determiner is a realisation ofD, the nominal predicate in (22) must be a DP. We then have another argument, inaddition to the BSN evidence, that the main distinction between predicative andreferential nominal phrases is not the absence of the DP level in the former and thepresence of the DP level in the latter. Predicate nominals can indeed be DPs. Theyare characterised by their semantics: they can get a purely intensional reading, butas will become even clearer in the following, it is difficult to connect this to anyparticular syntactic property.

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.12 (721-783)

Marit Julien

. Possessed nominal phrases

In Scandinavian, there are many ways to realise a possessor internally to the pos-sessed nominal phrase.4 In the following we will look at some of the relevantconstructions to see how they behave with respect to the predicate tests.

. Postnominal possessors

Many varieties of Scandinavian make use of constructions with postnominal pos-sessors. These possessors can be pronominal or nonpronominal, but with theexception of nonpronominal possessors in Icelandic, they all trigger a suffixeddefiniteness marker on the possessed noun. The Norwegian example in (23a)illustrates this.

(23) a. katt-encat-def.masc.sg

minmy.masc.sg

‘my cat’b. den

def.sggaml-eold-def

katt-encat-def.masc.sg

minmy.masc.sg

‘my old cat’

In the ‘double definiteness’ varieties Faroese, Norwegian, and Swedish therewill also be a preposed definiteness marker if an adjective is present.5 I showthis in (23b). Hence, when it comes to definiteness marking, the possessednominal phrases in (23) are exactly like the non-possessed nominal phrases in(20) and (21).

If we now take the nominal phrase in (23b) and run it through the predicatetests, it appears that it does not pass them. Although it is fine in postcopular po-sition, as in (24a), it cannot appear in predicate position under ‘consider’ – see(24b), it cannot be topicalised over the subject – see (24c), it cannot be pseudo-clefted – see (24d), and it cannot be coordinated with an adjectival predicate –see (24e).

(24) a. BusterBuster

eris

dendef.sg

gaml-eold-def

katt-encat-def

min.my.masc.sg

‘Buster is my old cat.’b. #Eg

Ianserconsider

BusterBuster

somas

dendef.sg

gaml-eold-def

katt-encat-def

min.my.masc.sg

‘I consider Buster my old cat.’

. For an overview, see e.g. Delsing (1998, 2003) or Julien (2005).

. Standard Swedish does not have postnominal possessors, but they are found in manySwedish dialects, in particular in the north.

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.13 (783-843)


c. #Dendef.sg

gaml-eold-def

katt-encat-def

minmy.masc.sg

eris

BusterBuster

ikkje.not

‘My old cat, Buster is not.’d. #Det

itBusterBuster

er,is

eris

dendef.sg

gaml-eold-def

katt-encat-def

min.my.masc.sg

‘What Buster is, is my old cat.’e. #Buster

Bustereris

ganskequite

smartsmart

ogand

dendef.sg

gamleold-def

kattencat-def

min.my.masc.sg

‘Buster is quite smart and my old cat.’

One might think that there is a structural reason for this result. However, it turnsout that it is possible to construct phrases that look exactly parallel to den gamlekatten min ‘my old cat’ but nevertheless pass the predicate tests. One example isden beste vennen min ‘my best friend’, which is shown in (25).

(25) a. HelgeHelge

eris

dendef.sg

best-ebest-def

venn-enfriend-def

min.my.masc.sg

‘Helge is my best friend.’b. Eg

Ianserconsider

HelgeHelge

somas

dendef.sg

best-ebest-def

venn-enfriend-def

min.my.masc.sg

‘I consider Helge my best friend.’c. Den

def.sgbest-ebest-def

venn-enfriend-def

minmy.masc.sg

eris

HelgeHelge

ikkje.not

‘My best friend, Helge is not.’d. Det

itHelgeHelge

er,is

eris

dendef.sg

best-ebest-def

venn-enfriend-def

min.my.masc.sg

‘What Helge is, is my best friend.’e. Helge

Helgeeris

snillkind

ogand

dendef.sg

best-ebest-def

venn-enfriend-def

min.my.masc.sg

‘Helge is kind and my best friend.’

Admittedly, the adjective gamle ‘old’ in (24) is in the positive grade whereas theadjective beste ‘best’ in (25) is in the superlative grade. But we have already seen, in(22), that a nominal predicate may well contain an adjective in the positive grade.Thus, there is no obvious formal difference between den beste vennen min and dengamle katten min which could be responsible for their different behaviour with re-spect to the predicate tests. The only clear difference between the two phrases is asemantic one. Whereas den gamle katten min, because of its lexical content, tendsto get an referential interpretation, den beste vennen min easily allows an inten-sional interpretation. Consequently, den beste vennen min is good as a predicatebut den gamle katten min is not.

The argument that predicate semantics is not directly connected to syntaxcan be made even stronger. Each of the adjective plus noun combinations in the

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.14 (843-906)

Marit Julien

phrases that were focused in (24) and (25) can be turned into a compound, withthe adjective as the first or nonhead part and the noun as the head. There is then nopreposed determiner, and thus no visible indication that there is a D-projection.Nevertheless, these compounds behave in the same way syntactically as the corre-sponding phrases with the adjective separated from the noun. In (26), we see thatgammelkatten min ‘my old-cat’ fails the predicate tests, but in (30), we see thatbestevennen min ‘my best-friend’ passes them.


eris

gammel-katt-enold-cat-def

min.my.masc.sg

‘Buster is my old cat.’b. #Eg

Ianserconsider

BusterBuster

somas


min.my.masc.sg

‘I consider Buster my old cat.’c. #Gammel-katt-en

old-cat-defminmy.masc.sg

eris

BusterBuster

ikkje.not

‘My old cat, Buster is not.’d. #Det

itBusterBuster

er,is

eris


min.my.masc.sg

‘What Buster is, is my old cat.’e. #Buster

Bustereris

ganskequite

smartsmart

ogand


min.my.masc.sg

‘Buster is quite smart and my old cat.’

(27) a. HelgeHelge

eris

best-e-venn-enbest-def-friend-def

min.my.masc.sg

‘Helge is my best friend.’b. Eg

Ianserconsider

HelgeHelge

somas


min.my.masc.sg

‘I consider Helge my best friend.’c. Best-e-venn-en

best-def-friend-defminmy.masc.sg

eris

HelgeHelge

ikkje.not

‘My best friend, Helge is not.’d. Det

itHelgeHelge

er,is

eris


min.my.masc.sg

‘What Helge is, is my best friend.’e. Helge

Helgeeris

snillkind

ogand


min.my.masc.sg

‘Helge is kind and my best friend.’

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.15 (906-980)


It can be argued that the possessor expressions in the possessed nominal phrasesin (26) and (27) have not moved out of the local domain of the possessed noun.6

One indication that this is true is the fact that if adjectives or other prenominal ele-ments were added, they would precede the possessed noun, just as in (24) and (25).Hence, judging from the elements that are visibly present, the nominal phrasesnow under discussion need not contain much functional material at all (althoughI would assume that they do). And again, it is not easy to point to any structuralreason for their different abilities to be predicates. Instead, it seems that lexicalcontent is more important than syntactic form in this respect.

. Prenominal possessors

In Scandinavian, possessors can also be prenominal. In some varieties, possessorsare in fact always prenominal, but even in varieties where postnominal possessorsis the unmarked option, at least pronominal possessors can precede the noun ifthey are focused. That is, all main varieties of Scandinavian have the constructionin (28), possibly with focus on the possessor.7

(28) minmy.masc.sg

(svart-e)black-def

kattcat

‘my (black) cat’

In Standard Danish and Standard Swedish, nonpronominal possessors are prenom-inal and carry the possessive -s. This is also possible in written Norwegian (espe-cially in Bokmål) and in many dialects of Norwegian and Swedish (Torp 1973;Delsing 2003). My example, in (29), is Norwegian.

(29) Kari-sKari-poss

(svart-e)black-def

kattcat

‘Kari’s (black) cat’

Another prenominal possessor construction found in Norwegian, and in the Dan-ish dialects of Jutland, is the so-called prenominal possessor doubling construc-tion, which involves a preposed (proper name or common noun) possessor and apronominal element following it. I give a Norwegian example of this in (30).

(30) KariKari

sinrefl.poss.masc.sg

(svart-e)black-def

kattcat

‘Kari’s (black) cat’

. See Julien (2005) for a much more detailed discussion of the syntax of possessed nominalsin Scandinavian.

. There may be dialects that do not have the construction, so I dare not say that allvarieties have it.

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.16 (980-1039)

Marit Julien

The prenominal possessor doubling construction is syntactically similar to theconstruction with the possessive -s (see Fiva 1987; Delsing 1993, 1998; Krause1999). The only difference between the two is that whereas the possessive -s onlyspells out a possessive feature, the pronominal element in the prenominal posses-sor doubling construction spells out other features as well.8

In all Scandinavian prenominal possessor constructions, the possessor willprecede adjectives and numerals. As illustrated in (31), the elements that canin their turn precede prenominal possessors are strong quantifiers and demon-stratives.

(31) all-eall-pl

dessethese

{min-e/my-pl/

Kari-s/Kari-poss/

KariKari

sine }refl.poss.pl

femfive

svart-eblack-pl

katt-arcat-pl

‘all these five black cats of mine/of Kari’s’

This means that prenominal possessors are located in the area where we otherwisefind the prenominal determiner in nominal phrases containing adjectives or nu-merals. In other words, Scandinavian nominal phrases with prenominal possessorsare quite large structurally.

Now consider (32), where I apply our predicate tests to the possessed nomi-nal phrase Kari sin katt ‘Kari’s cat’. We see that not only is it fine in postcopularposition, but it can also be in predicate position under ‘consider’, it can be topi-calised over the subject, it can be pseudoclefted, and it can be coordinated with anadjectival phrase. It can also combine with a nonrestrictive relative clause, whichmeans that it can be referential, but this does not in any way mean that it is not apredicate in other cases.

. In (30), the additional features are a reflexive/anaphor feature plus a set of phi-features(gender and number) that are valued by the possessee. This is what we always find in Norwegian.But in Jutlandic dialects, an alternative is to have a pronominal element that reflects the phi-features of the possessor, as in the following example from Nielsen (1986:65):

(i) ædef

konewoman

hendesher.poss

kysebonnet

‘the woman’s bonnet’

Here the possessive pronoun hendes reflects a third person singular feminine possessor, and itcan only be preceded by a possessor with matching features. The reflexive pronoun sin in (30),by contrast, requires the possessee to be masculine and singular (for those speakers that retain athree-gender system – for others it will be common gender), but it combines with any possessor,even (markedly or dialectally) with first and second person possessors:

(ii) %/?minmy.masc.sg

sinrefl.poss.masc.sg

kattcat

‘my cat’

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.17 (1039-1082)



eris

KariKari

sinrefl.poss

katt.cat

‘Buster is Kari’s cat.’b. Eg

Ianserconsider

BusterBuster

somas

KariKari

sinrefl.poss

katt.cat

‘I consider Buster Kari’s cat.’c. Kari

Karisinrefl.poss

kattcat

eris

BusterBuster

ikkje.not

‘Kari’s cat, Buster is not.’d. Det

itBusterBuster

er,is

eris

KariKari

sinrefl.poss

katt.cat

‘What Buster is, is Kari’s cat.’e. Buster

Bustereris

sjuseven

åryears

gammalold

ogand

KariKari

sinrefl.poss

katt.cat

‘Buster is seven years old and Kari’s cat.’f. Buster

Bustereris

KariKari

sinrefl.poss

katt,cat

somthat

forrestenby.the.way

sitsit.pres

derthere

borte.away

‘Buster is Kari’s cat, who by the way is sitting over there.’

It would however be possible to argue that a nominal phrase with a prenominalpossessor does not have a D-projection when it is a predicate, and that the pos-sessor is sitting in a possessor-related projection which has a D-projection over itonly when the whole possessed phrase is referential. In fact, this is what Zamparelli(2000) proposes. He further claims that when the possessor is followed by a nu-meral, the numeral occupies the position that would otherwise host the possessor.The projection of a higher layer is then necessary to make room for the possessor,and since this higher layer is the one connected to referentiality, the result is thatthe nominal phrase can no longer be a predicate. In support of his claim he pointsto the following contrast (Zamparelli 2000:136–137):

(33) a. These are Harold’s tools, and those, too, are Harold’s tools.b. *These are Harold’s four tools, and those, too, are Harold’s four tools.

As Zamparelli sees it, the phrase Harold’s tools in (33a) is a predicate, since it canbe applied to two different sets of objects (a view also held by Holmberg 1993).The phrase Harold’s four tools in (33b) is however referential, and because ofthis, it cannot be applied to more than one set at a time. And as I just have ex-plained, Zamparelli takes the numeral to be responsible for the referentiality ofthe latter phrase.

However, the (purported) ungrammaticality of (33b) can only be taken as anindication that Harold’s four tools is not a predicate that can be true of more thanone subject at a time. It does not follow that the phrase cannot be a predicate at all.We need other tests to find this out. One such test is embedding under ‘consider’,

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.18 (1082-1147)

Marit Julien

which is shown in (34a). Here it turns out that Harold’s four tools can indeed be apredicate. If the example is felt to be a little awkward, it probably has to do withthe discourse properties of the constituents, since it gets better if the possessoris emphasised, i.e. focused, so that the content of the numeral is treated as giveninformation. Moreover, in (34b) we have a perfectly acceptable nominal predicate,which includes a possessor followed by a numeral. Hence, there seems to be notruth to the claim that nominal phrases of this form cannot be predicates.

(34) a.(#)I consider these Harold’s four tools.b. I consider these Harold’s two best books.

If we now turn to a Scandinavian prenominal possessor doubling constructionwith a numeral, for example Harold sine to beste bøker ‘Harold’s two best books’,we see in (35) that it fails Zamparelli’s coordination test.

(35) #Dettethis

erbe.pres

HaroldHarold

sinerefl.poss.pl

totwo

best-ebest-pl

bøk-er,book-pl

ogand

dethey

derthere

erbe.pres

HaroldHarold

sinerefl.poss.pl

totwo

bestebest-pl

bøkerbook-pl

‘This is Harold’s two best books, and this is Harold’s two best books.’

Nevertheless, that very same phrase may well be a predicate, as shown in (36a–d),in addition to being referential, as in (36e). Hence, the reason why (35) is not wellformed is again that the predicate cannot be felicitously applied to more than onesubject, in this case one pair of books, at a time.

(36) a. Dettethis

eris

HaroldHarold

sinerefl.poss.pl

totwo

bestebest

bøker.books

‘This is Harold’s two best books.’b. Eg

Ianserconsider

dessethese

somas

HaroldHarold

sinerefl.poss.pl

totwo

bestebest

bøker.books

‘I consider these Harold’s two best books.’c. Harold

Haroldsinerefl.poss.pl

totwo

bestebest

bøkerbooks

eris

dettethis

ikkje.not

‘Harold’s two best books, this is not.’d. Det

itdettethis

er,is

eris

HaroldHarold

sinerefl.poss.pl

totwo

bestebest

bøker.books

‘What this is, is Harold’s two best books.’e. Dette

thiseris

HaroldHarold

sinerefl.poss.pl

totwo

bestebest

bøker,books

somthat

egI

harhave

lesiread

mangemany

gongar.times

‘This is Harold’s two best books, which I have read many times.’

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.19 (1147-1214)


The phrase under discussion here is quite big structurally, since it has room for anumeral, a possessive marker plus a possessor above the adjective. It is arguablya DP, with the possessor in Spec-DP, which is also the structure that Zamparelli(2000) suggests for Harold’s four tools. And note that it looks the same regardless ofwhich syntactic function it has. There is no indication that it contains less structurewhen it is a predicate than it does when it is referential, unless we take the semanticdifference itself to be such an indication.

. Strong quantifiers, demonstratives, and personal pronouns

We will now turn to the highest projections that can be found in Scandinaviannominal phrases. We have already seen, in (31), that universal quantifiers anddemonstratives are located in high positions. More precisely, although a preposeddeterminer is never required when a demonstrative is present, the determiner willfollow the demonstrative if the two co-occur. I show this in (37a). A universalquantifier will in its turn precede the demonstrative, as shown in (37b).

(37) a. dessethese

(dei)def.pl

totwo

eldsteoldest

hus-ahouse-def.pl

iin

by-entown-def

‘these (the) two oldest houses in town’b. all-e

all-pldessethese

femfive

gaml-eold-pl

hest-a-nehorse-pl-def

‘all these five old horses’

My conclusion is that the demonstrative heads a Dem projection that can be gen-erated above DP (cf. Cinque 2000; Julien 2002, 2005), and that strong quantifiersare located in a Quantifier Phrase on top of DemP (cf. Sigurðsson 1993; Giusti1997; Vangsnes 2001; Julien 2002, 2005). The higher part of a maximally expandednominal phrase is then as shown in (38).

(38) [QP Q [DemP Dem DP ]]

Now strikingly, it is possible for a nominal phrase with a universal quantifier fol-lowed by a determiner – in other words, a phrase of more than DP-size – to passthe predicate tests. The examples in (39) demonstrate this.

(39) a. Dettethis

eris

all-eall-pl

deidef.pl

interessant-einteresting-def

plate-nerecord-def.pl

mi-ne.my.pl

‘This is all my interesting records.’b. Eg

Ianserconsider

dettethis

somas

all-eall-pl

deidef.pl



mi-ne.my.pl‘I consider this all my interesting records.’

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.20 (1214-1285)

Marit Julien

c. All-eall-pl

deidef.pl



mi-nemy.pl

eris

dettethis

ikkje.not

‘All my interesting records, this is not.’d. Det

itdettethis

er,is

eris

all-eall-pl

deidef.pl



mi-ne.my.pl

‘What this is, is all my interesting records.’

If these tests mean anything at all, we must conclude that a QP can in fact be apredicate. It cannot then be true that nominal predicates are always smaller thanDPs. On the contrary, we see in (39) that the quantifier has a DP in its complement,since what follows the quantifier inside the nominal phrase has a determiner ininitial position and could function as an argument on its own:

(40) Deidef.pl



mi-nemy.pl

erare

lagrastored

påat

loft-et.attic-def

‘My interesting records are stored in the attic.’

Hence, if the semantic difference between nominal arguments and nominal pred-icates corresponds to a syntactic difference, that syntactic difference cannot bestated as the presence versus the absence of a D-projection.

But as others have already noted (see e.g. Partee 1987; Adger & Ramchand2003), nominal phrases containing demonstratives are not acceptable as predi-cates. They are necessarily referential. This holds true also of Scandinavian:

(41) a. Deidef.pl

somthat

skalshall

oppup

tilto

eksamenexam

erare

dessethese

student-a-ne.student-pl-def

‘Those who will sit for the exam are these students.’b. #Eg

Ianserconsider

deidef.pl

somthat

skalshall

oppup

tilto

eksamenexam

somas

dessethese

student-a-ne.student-pl-def‘I consider those who will sit for the exam as these students.’

c. #Dessethese

student-a-nestudent-pl-def

erare

deidef.pl

somthat

skalshall

oppup

tilto

eksamenexam

ikkje.not

‘These students, those that will sit for the exam are not.’d. #Det

itdeidef.pl

somthat

skalshall

oppup

tilto

eksamenexam

er,are

erare

dessethese


‘What those who will sit for the exam are, are these students.’e. Dei

def.plsomthat

skalshall

oppup

tilto

eksamenexam

erare

dessethese

student-a-ne,student-pl-def

somthat

forrestenby.the.way

planleggplan.pres

eina

festparty

etterpå.afterwards

‘Those who will sit for the exam are these students, who by the way areplanning a party afterwards.’

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.21 (1285-1335)


I would however suggest that this has nothing to do with a DemP being too bigstructurally to be a predicate. It is a consequence of the deictic content of thedemonstrative, which makes it incompatible with an intensional interpretation.When a demonstrative is used, the speaker will have a specific set of referents inmind, and a purely intensional interpretation is not available.

Demonstratives contrast interestingly with personal pronouns, which are infact able to pass the predicate tests. Looking at a picture, for example, one couldfelicitously utter (42a), and in that same context, the utterances in (42b), (42c),and (42d) would also be acceptable.

(42) a. Dettethis

eris

meg.me

‘This is me.’b. Ho

sheanserconsiders

dettethis

somas

meg.me

‘She considers this as me.’c. Meg

meeris

dettethis

ikkje.not

‘Me, this is not.’d. Det

itdettethis

er,is

eris

meg.me

‘What this is, is me.’

We can conclude that the pronoun meg ‘me’ is a predicate in (42). What makesthis possible must be the fact that meg here means ‘a representation of me’ ratherthan ‘me in person’. Hence, meg can be used without any specific referent be-ing picked out. In other words, personal pronouns can have a purely intensionalinterpretation, and because of this, they can be predicates.9

Let me hasten to add here that in a similar picture context, the example (12b),which was marked as semantically deviant, will be quite appropriate:

(43) EgI

anserconsider

denthat

mann-enman-def

somas

ChristerChrister

Platzack.Platzack

(picture context)

‘I consider that man Christer Platzack.’

. As Ora Matushansky has pointed out to me, personal pronouns can also be predicates incoercion contexts like the following (my example):

(i) Denthat

hatt-enhat-def

eris

heilttotally

deg!you

‘That hat is totally you!’

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.22 (1335-1393)

Marit Julien

That is, the name can take on a predicative reading if the context is right. Thisdemonstrates once again that the line between referential readings and predicativereadings is a very thin one.

. The Pred head

The preceding discussion has shown that it is very hard to find any clear and sys-tematic structural differences between nominal predicates and nominal argumentsin Scandinavian. I have mentioned, however, that there are indications that nomi-nal predicates are complements of Pred heads. In nonfinite contexts the Pred headis spelled out, most often as som ‘as’, as in (44a) and in many of the examples shownearlier, but other realisations are also possible, for example til ‘to’, shown in (44b),which has an inchoative meaning element that is missing in som.10

(44) a. EgI

anserconsider

JennyJenny

somas


‘I consider Jenny the winner.’b. Vi

wevaldechose

LiseLise

tilto

leiar.leader

‘We chose/elected Lise as (our) leader.’

In light of this, it is conceivable that there is a structural difference between nom-inal predicate constructions and constructions with referential nominal phrasesafter all, but one that has to do with the Pred head and not with the nominalphrases themselves. It might be that referential nominals are not embedded un-der Pred heads, or alternatively, that they are embedded under Pred heads of adifferent type, as proposed by Adger & Ramchand (2003).

Let us therefore look once more at some of the copular constructions that wehave found to involve a referential postcopular phrase. I repeat three of them herefor convenience:

(45) Prefekt-enprefect-def

påat

nordiskNordic

eris

ChristerChrister

Platzack.Platzack

‘The prefect of the Nordic languages department is Christer Platzack.’

(46) BusterBuster

eris


min.my.masc.sg

‘Buster is my old cat.’

. That is, the meaning difference between som and til parallels the meaning difference be-tween ‘be’ and ‘become’.

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.23 (1393-1464)


(47) Deidef.pl

somthat

skalshall

oppup

tilto

eksamenexam

erare

dessethese


‘Those that will sit for the exam are these students.’

Then we make these expressions nonfinite by embedding them under a higherverb. The higher verb cannot be of the ‘consider’ type, requiring a predication as itscomplement, but one that is able to embed an equative statement. In Norwegian,oppfatte, which means ‘understand, perceive, construe’, is such a verb. As shown inthe examples below, the nonfinite versions of the equative statements in (45)–(47)can all be embedded under oppfatte.

(48) EgI

oppfattaperceived


påat

nordiskNordic

somas

ChristerChrister

Platzack.Platzack

‘I took it that the prefect of the Nordic languages department is/was ChristerPlatzack.’

(49) Hoshe

oppfattaperceived

BusterBuster

somas


min.my.masc.sg

‘She took it that Buster is/was my old cat.’

(50) EgI

oppfattaperceived

deidef.pl

somthat

skalshall

oppup

tilto

eksamenexam

somas

dessethese

student-a-ne.student-pl-def‘I took it that those that will sit for the exam are these students.’

Now note that in each of these examples the element som shows up between thetwo nominal phrases in the complement of oppfatte. In other words, although nei-ther of the nominal phrases embedded under oppfatte in (48)–(50) are predicates,an element identical to the realisation of the Pred head in nonfinite predicationsappears even here. This fact indicates that even equative statements involve a Predhead, at least in Scandinavian. Since this Pred head is spelled out in the sameway as the Pred head in predications, we have no indication that the Pred headin equative statements is of a different type than the Pred head in predications,as proposed by Adger & Ramchand (2003). It seems instead that the only differ-ence between equative statements and predications is the referentiality of the lowernominal phrase.

. Conclusion

I have tried to show in this paper that the semantic difference between predicativenominal phrases on the one hand and referential nominal phrases on the otherhand does not necessarily correspond to a difference in syntactic structure.

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.24 (1464-1522)

Marit Julien

In Scandinavian, bare singular nominals, that is, nominal phrases consisting ofthe noun itself and possibly one or more adjectives, but lacking the projections as-sociated with quantification, can be arguments as well as predicates. These phraseshave been discussed earlier in the linguistic literature, and their ability to be argu-ments should be well known, but I have nevertheless mentioned them here, sincethey are part of the general picture that I want to present.

The question that needs to be asked is then how big a nominal predicate canbe structurally. It appears that at least in Scandinavian, nominal phrases of DP sizemay well be predicates, and moreover, so can phrases that are even bigger than DP.This is the conclusion I arrived at after using various tests that distinguish betweennominal predicates and referential nominal phrases in postcopular position.

Finally, I considered the possibility that there might be a syntactic differencebetween constructions with predicative nominals and constructions with referen-tial postcopular nominals after all, but one that is not found in the nominal phrasesthemselves but instead involves the Pred head. However, in Scandinavian the Predhead is realised in the same way in front of nominal predicates as in front of refer-ential nominals, so that there is no obvious reason to say that there are two typesof Pred heads.

On the whole, at least in Scandinavian there seems to be no visible syntacticproperty that can be connected to a structural contrast between predicative nom-inals and referential nominals. If there is a structural difference at all between thetwo types of nominals, it cannot be seen directly – it can only be postulated on thebasis of the observable differences in syntactic behaviour. But then there is also thepossibility that the difference is purely semantic: if the lexical content of a nominalphrase allows a purely intentional interpretation, that phrase can be a predicate,but if its lexical content requires an extensional reading, the phrase is necessarilyreferential.

This reasoning brings us back to Williams (1983) and Partee (1987), who ar-gued that all nominal phrases can in principle be predicates, and that if a givennominal phrase is not good as a predicate, it is a consequence of its semantics(see also Doron 1988). I think we should consider more carefully whether theseproposals are not right after all.

References

Adger, D. & Ramchand, G. (2003). Predication and equation. Linguistic Inquiry, 34, 325–359.Baker, M. C. (2003). Lexical Categories. Cambridge: CUP.Borthen, K. (2003). Norwegian Bare Singulars. PhD dissertation, NTNU.Bowers, J. (1988). Extended X-bar theory, the ECP and the left branching condition. In H. Borer

(Ed.), Proceedings of WCCFL 7 (pp. 46–62). Stanford CA: CSLI.

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.25 (1522-1635)


Bowers, J. (1993a). The syntax and semantics of nominals. Proceedings of SALT, 3, 1–30.Bowers, J. (1993b). The syntax of predication. Linguistic Inquiry, 24, 591–656.Cinque, G. (2000). On Greenberg’s Universal 20 and the Semitic DP. University of Venice Working

Papers in Linguistics, 10(2), 45–61.Delsing, L.-O. (1993). The Internal Structure of Noun Phrases in the Scandinavian Languages.

PhD dissertation, Lund University.Delsing, L.-O. (1998). Possession in Germanic. In A. Alexiadou & C. Wilder (Eds.), Possessors,

Predicates and Movement in the Determiner Phrase (pp. 87–108). Amsterdam: JohnBenjamins.

Delsing, L.-O. (2003). Syntaktisk variation i skandinaviska nominalfraser. In Ø. A. Vangsnes, A.Holmberg, & L.-O. Delsing (Eds.), Dialektsyntaktiska studier av den nordiska nominalfrasen(pp. 11–64). Oslo: Novus.

Doron, E. (1988). The semantics of predicate nominals. Linguistics, 26, 281–301.Eide, K. M. & Åfarli, T. A. (1999). The syntactic disguises of the predication operator. Studia

Linguistica, 53, 155–181.Eide, K. M. & Åfarli, T. A. (2001). Semi-lexical heads in semantically charged syntax. In N.

Corver & H. van Riemsdijk (Eds.), Semi-lexical Categories (pp. 455–473). Berlin: Moutonde Gruyter.

Fiva, T. (1987). Possessor Chains in Norwegian. Oslo: Novus.Fodor, J. D. (1970). The Linguistic Description of Opaque Contexts. PhD dissertation, MIT.Giusti, G. (1997). The categorial status of determiners. In L. Haegeman, (Ed.), The New

Comparative Syntax (pp. 95–123). London: Longman.Heycock, C. & Kroch, A. (1998). Inversion and equation in copular sentences. In A. Alexiadou,

N. Fuhrhop, U. Kleinhenz, & P. Law (Eds.), Papers in Linguistics, Vol. 10 (pp. 71–87). Berlin:ZAS.

Holmberg, A. (1993). On the structure of predicate NP. Studia Linguistica, 47, 126–138.Hudson, W. (1989). Functional categories and the saturation of noun phrases. Proceedings of

NELS, 19, 207–222.Julien, M. (2002). Determiners and word order in Scandinavian DPs. Studia Linguistica, 56,

264–314.Julien, M. (2005). Nominal Phrases from a Scandinavian Perspective. Amsterdam: John

Benjamins.Krause, C. (1999). Two notes on prenominal possessors in German. MIT Working Papers in

Linguistics, 33, 191–217.Longobardi, G. (1994). Reference and proper names: A theory of N-movement in syntax and

Logical Form. Linguistic Inquiry, 25, 609–665.Mandelbaum, D. (1994). Syntactic Conditions on Saturation. PhD disssertation, CUNY.Nielsen, Bent Jul. (1986). Om pronominet sin i jysk. Danske Folkemål, 28, 41–101.Partee, B. H. (1987). Noun phrase interpretation and type-shifting principles. In J. Groenendijk,

D. de Jongh, & M. Stokhof (Eds.), Studies in Discourse Representation Theory and the Theoryof Generalized Quantifiers (pp. 115–143). Dordrecht: Foris.

Pereltsvaig, A. (2001). On the Nature of Intra-clausal Relations. PhD dissertation, McGillUniversity.

Pereltsvaig, A. (2002). Determiner Phrase in languages with and without determiners. Posterpresented at the Scandinavian Conference of Linguistics 19: University of Tromsø, January2002.

Sigurðsson, H. Á. (1993). The structure of the Icelandic NP. Studia Linguistica, 47, 177–197.

JB[v.20020404] Prn:22/09/2006; 10:37 F: LA9704.tex / p.26 (1635-1655)

Marit Julien

Taraldsen, K. T. (1990). D-projections and N-projections in Norwegian. In J. Mascaró & M.Nespor (Eds.), Grammar in Progress (pp. 419–431). Dordrecht: Foris.

Torp, A. (1973). Om genitivomskrivinger og -s-genitiv i norsk. Maal og Minne, 1973, 124–150.Vangsnes, Ø. A. (2001). On noun phrase architecture, referentiality, and article systems. Studia

Linguistica, 55, 249–299.Williams, E. (1983). Semantic vs. syntactic categories. Linguistics and Philosophy, 6, 423–446.Zamparelli, R. (2000). Layers in the Determiner Phrase. New York NY: Garland.

JB[v.20020404] Prn:24/08/2006; 10:14 F: LA97P2.tex / p.1 (45-72)

Studies on the (pro)nominal system

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.1 (46-125)

Pronominal noun phrases, numberspecifications, and null nouns*

Dorian RoehrsIndiana University, Bloomington

Standard assumptions claim that determiner concord is subject to morphologicalagreement between the determiner and the head noun (Carstens 2000). Pronom-inal determiners (Postal 1966) provide evidence that this assumption is both toostrong and too weak. Focusing on number, it is proposed that pronominal deter-miners require not only morphological, but also semantic, agreement. Cases ofmorphological dis-agreement are argued to involve the overt noun in a Specifierposition and a null noun in the head position. As one consequence, the inventoryof null nouns is extended from countable and mass nouns (Panagiotidis 2002a)to collectives, pluralia tantum, and proper names.

. Introduction

According to standard assumptions (e.g., Carstens 2000), determiner concord inthe DP is a matter of morphological agreement such that a singular determinerco-occurs with a singular noun and a plural determiner with a plural noun:

(1) a. {diesesthis(neut/sg)

/*diese}/these(pl)

Schweinpig

(German)

‘this pig’

* Parts of this paper were presented at the 20th Comparative Germanic Syntax Workshop atTilburg University, Tilburg, the Netherlands, in June 2005. I thank the audience at that confer-ence, two reviewers, my two editors, and Leslie Gabriel for questions and comments. Further-more, I am grateful to Jan-Wouter Zwart, Olaf Koeneman, and Esther Ham for discussion ofand judgments on the Dutch examples, Anna McNay for judgments and recruiting further helpwith British English, and to Marit Julien for pointing out Josefsson, to appear, and making thepaper available to me. I thank Ashley Farris for helping me proofread this paper. All oversightsare my own.

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.2 (125-188)

Dorian Roehrs

b. {diesethese(pl)

/*dieses}/this(neut/sg)

Schweinepigs

‘these pigs’

At first sight, the same seems to apply to pronominal DPs, understood here ascombinations of “pronoun + noun”:

(2) a. {duyou(inform/sg)

/*ihr}/you(inform/pl)

Schweinpig

‘you pig’b. {ihr

you(inform/pl)/*du}/you(inform/sg)

Schweinepigs

‘you pigs’

In fact, Sauerland & Elbourne (2002:289–291) maintain that determiner con-cord is not subject to semantic agreement. This can be seen from combinationswith group nouns, as they point out, as well as from combinations with certainpluralia tantum:

(3) a. {this / *these} setb. {these / *this} pants

Among many others, Postal (1966) argues that pronouns are determiners. If this isso, then we expect that pronominal determiners pattern not only as in (1) but alsoas in (3). However, considering the counterpart of (3a) here as a first illustration,the opposite seems to hold for pronominal noun phrases:1

(4) a. *duyou(inform/sg)

verdammtesdamn

Packgang

‘you darn gang’b. ihr

you(inform/pl)verdammtesdamn

Packgang

‘you darn gang’

To be clear, there are two things to note here: first, despite morphological agree-ment in number, the example in (4a) is ungrammatical; second, in (4b) there issemantic agreement in number between the pronoun and the noun and morpho-logical dis-agreement is tolerated. In other words, the traditional assumption ofmorphological agreement between determiner and noun is both too weak and

. The grammaticality of (4a) is indicated with regard to the collective noun denoting a multi-member set. The example is grammatical when the pronoun addresses the group itself or whenthe group has only one member. I postpone the discussion of the latter cases until Section 6. Noteat this point that English is more restrictive (for some discussion, see Note 12) and translationsare only approximations.

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.3 (188-250)

Pronominal noun phrases and number specifications

too strong: it is too weak in that it would allow ungrammatical examples as in(4a) (“overgeneration”) and it is too strong in that we would not expect gram-matical cases as in (4b) (“undergeneration”). Now, if “regular” and pronominaldeterminers are to be given a uniform account, their apparent differences need tobe addressed. Furthermore and more importantly, morphological concord withinnoun phrases is a strong cross-linguistic generalization. If we are to keep this gen-eralization, then our methodology must be to find a different account for the casesinvolving morphological dis-agreement in number as in (4b) (for the discussionof mismatches in gender, see for instance Hellan 1986; Svenonius 1993; Josefssonto appear).

Focusing on German and, to some extent, on Dutch, Icelandic, English, andsome other languages, I will argue that pronominal determiners are more restric-tive than “regular” ones. While the latter only require morphological agreement,as indicated by the subscript φ in (5a), the former require not only morphologi-cal but also semantic agreement, as designated by the additional superscript g in(5b). Assuming the head position N to be filled by the null noun eN, morpholog-ically dis-agreeing nouns are argued to be in the Specifier position of a separatephrase (5c). For mnemonic purposes, I will call this phrase “DisP” standing forDis-agreement Phrase. Apparently, the option of taking DisP must be banned for“regular” determiners (5d):

(5) a. [DP dieseφ [NP

theseSchweineφ ]]pigs

b. [DP ihrφg [NP

you(pl)Schweineφg ]]pigs

c. [DP ihrφg [DisP

you(pl)Packg [NP eNφ

g ]]]gang

d. *[DP dieseφ [DisP

thesePackg [NP eNφ

g ]]]gang

To anticipate, I will argue below that “regular” and pronominal determiners arethe same with regard to morphological agreement. They differ in that pronominaldeterminers require semantically appropriate nominal predicates. Among others,these predicates must be [individualizable]. In addition to their greater semanticspecifications, pronominal determiners may not only select AgrPs (hosting adjec-tives) and NP, but also the phrase DisP. In contrast, “regular” determiners can notselect DisP. This differing lexical feature will explain the contrast between (5c) and(5d) above. I summarize the comparison in (6):

(6) diese ihrMorphological agreement φ φSemantic agreement – [individualizable]Syntactic selection AgrP, NP AgrP, NP, DisP

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.4 (250-297)

Dorian Roehrs

The conclusion that I will draw is that the different kinds of determiners arenot identical, as is well-known, but that “regular” determiners are less specifiedpronominal determiners. I will argue that agreement in semantic number is notbrought about by feature checking but by the requirement that the denotationsof the relevant elements, defined as partial functions, be compatible. As a con-sequence, semantic number is not part of syntax, which I take to be a welcomeresult. More generally, the discussion will provide evidence that there are differenttypes of null nouns, that is, besides null countable and mass nouns (cf. Panagio-tidis 2002a, b, 2003a, b), there are also null collective nouns, pluralia tantum, andproper names.

The paper is organized as follows: first, motivating the distinction betweenmorphological and semantic numbers, I group nouns and pronouns into differenttypes. Sections 3 and 4 then combine these different nouns and pronouns in var-ious ways documenting in more detail the argument that traditional assumptionsare both too strong and too weak. In Sections 5 and 6, I turn to an account of themorphological and semantic properties of these noun phrases. Section 7 discussessome extensions and consequences of the proposal. After a brief illustration ofsome further issues in Section 8, the discussion is summarized in the conclusion.

. Morphological vs. semantic number

The traditional literature distinguishes between morphological and semanticnumber (e.g., Quirk et al. 1985:316–7, 758–9). The latter is often referred to asmereology. Sauerland & Elbourne (2002:289) discuss some differences betweenthese two specifications (see also den Dikken 2001). Among others, they observethat morphological number surfaces as plural morphology on the noun phraseand can usually be made explicit through agreement with the verb (7a–b); mereol-ogy refers to the possibility of whether or not an entity is conceived of as consistingof several members and can, in certain languages, be made visible by “special”verbal inflection (7c) (the datum in (7c) is from Sauerland & Elbourne 2002: 290):

(7) a. This set is odd.b. These sets are odd.c. This set are all odd. (British English)

With this in mind, we can set out four different kinds of nouns combining mor-phological and semantic number in various ways:

(8) a. morphological number = semantic numbercountables: Schwein ‘pig’, Eimer ‘bucket’, Bande ‘gang [set]’

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.5 (297-363)


b. morphological number < semantic numbercollectives: Bande ‘gang [members]’, Gesindel ‘riff-raff ’

c. morphological number > semantic number(certain) pluralia tantum: Hosen ‘pants’, scissors

d. “transnumerals”mass nouns: Honig ‘honey’, Mist ‘manure’, Dreck ‘dirt’

Intuitively, morphological and semantic numbers coincide with countable nouns:if the noun is in the singular, there will be one salient element in the discourse;if it is in the plural, there will be several. Furthermore, while collective nouns areusually singular in morphology but plural in semantics, pluralia tantum (of therelevant type) are the reverse. Before moving on, I make the notion of semanticnumber more concrete.

Interpreting predicates here as sets (but see below), I assume that semanticnumber can be broken down into two types: (i) one group of predicates is markedby spatial integrity, such that the noun denotes a set of elements that have differentdefining subparts that, prototypically, must all be present (e.g., singular countablenouns and certain pluralia tantum), and (ii) the other group of predicates has in-ternal partitioning, such that the noun denotes a set of sets where each of thesesubsets consists of discrete, identifiable members of a similar kind (e.g., collectiveand plural countable nouns). If the first type of predicate is combined with the def-inite determiner the, then the DP denotes a unique element in the discourse; if thesecond type of predicate is combined with the definite determiner, then the DP de-notes a unique set of elements in the discourse. I will call the first type of predicate“semantically singular” and the second type “semantically plural”. Finally, typicalmass nouns (under the substance reading) denote a homogenous and shapelessamount of material, that is, stuff without spatial integrity or internal partitioning.As such, they have no semantic number and are treated as transnumerals. Table 1summarizes these combinations in different kinds of numbers (for the reason justmentioned, mass nouns are not listed here):

Table 1. Combinations of morphological and semantic number in nouns

morphologically singular morphologically plural

semantically singular Schwein ‘pig’ Hosen ‘pants’semantically plural Bande ‘gang’ Schweine ‘pigs’

A similar grouping can be established for second person pronouns.2 I assume thatverbal inflection indicates morphological number of the pronoun. Turning to the

. Note that no direct parallelism between pronoun and noun is implied here. Rather, I takepronouns to be pro-determiners and not pro-nouns (see below).

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.6 (363-424)

Dorian Roehrs

relevant inventory, besides the pronouns of informal address such as singular du‘you’ and plural ihr ‘you’, where morphological and semantic numbers coincide,there are also pronouns of formal address. The latter are of particular interest asmorphological number is independent of the semantic reference: while both Ger-man Sie and Dutch u may have singular and plural reference, Sie takes a verb inthe plural and u in the singular:3

(9) a. Sieyou(form)

{alleinealone

/alle}/all

sindare(3pl)

dafürfor.it

verantwortlich.responsible

‘You alone/all are responsible for it.’b. U

you(form){alleenalone

//

allen}all

bentare(2sg)

daarvoorfor.it

verantwoordelijkresponsible

(Dutch)

These combinations in morphological and semantic number are summarized inTable 2:

Table 2. Combinations of morphological and semantic number in second person pro-nouns

morphologically singular morphologically plural

semantically singular du / u (Dutch) Siesemantically plural u (Dutch) ihr / Sie

It is interesting to note that, while there are personal pronouns with varying refer-ence but constant morphological number (i.e., the pronouns of formal address),the reverse does not seem to exist. In other words, there are no pronouns witha specified value for semantic number that may vary in morphological number(again, Dutch varies with regard to morphological person but not number, cf.Note 3). The next two sections combine the different types of nouns with thevarious pronouns with the intention of identifying restrictions.

. Pronominal noun phrases exhibit semantic agreement

In this section, I will be working toward the generalization that the combination“pronoun + noun” is subject to agreement in semantic number. As an orga-nizing principle, I consider the two combinations involving morphological dis-agreement between the pronoun and the noun: “plural pronoun + singular noun”and “singular pronoun + plural noun”. In the process, I point out grammatical

. Dutch also allows a verb in the third person singular (cf. (16) below), which, to some speakers,sounds old-fashioned.

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.7 (424-476)


cases that are expected to be bad under traditional assumptions and ungram-matical instances that should be fine. In other words, I show that traditionalassumptions are both too strong and too weak. Importantly, none of these casesinvolves an intonational break between the pronoun and the noun or appositiveintonation. All examples here are from English, German, and Dutch (for otherlanguages, see below).

. Morphological dis-agreement: Pronoun(pl) + noun(sg)

As already pointed out in the introduction, concord between nouns and regulardeterminers is a matter of morphological, but not semantic, agreement:

(10) a. this setb. *these set

However, the apparent opposite holds for pronominal determiners.

.. Countable nouns: Semantic agreement in the singularAs discussed above, German Sie ‘you(form)’ takes a verb with plural inflection,independent of its singular or plural reference, and I assumed it to be morpho-logically plural. Surprisingly under traditional assumptions, this pronoun cancombine with a singular noun under singular reference (“sc” indicates a stylisticclash):4

(11) a. scSieyou(form)

(idiotischeidiotic

Feldmaus)field.mouse

sindare

ja schonalready

da!here

‘You idiotic field mouse are here already!’b. scSie

you(form)verrotztessnotty

Nichtsnothing

‘you snotty nothing’

Furthermore, it is also surprising that despite morphological agreement, Sie withsingular reference cannot combine with a plural noun. In other words, althoughthere is morphological concord, the following example is ungrammatical under acertain interpretation:

. Cases of this kind were, to the best of my knowledge, first noticed by Darski (1979:202).Some of the examples are taken from that paper. Notice also that, due to (mostly) identical verbalinflections, it is harder to identify relevant cases in some other Germanic languages. For instance,the English non-argumental You idiot! may present a case of morphological dis-agreement only,if we assume that the verb, and thus the pronoun, in you are {an idiot / idiots} is morphologicallyplural in both cases.

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.8 (476-533)

Dorian Roehrs

(12) *Sieyou(form)

Schweinepigs

‘you pigs’

Crucially, what the grammatical cases have in common is that the pronoun mustagree semantically but not morphologically with the noun. Consider more casesof this type.

.. Collective nouns: Semantic agreement in the pluralI assume that collective nouns consist of group nouns that can form the plural(die Banden ‘the gangs’) and those that cannot (*die Pack(s/e/n) ‘the gangs’). Rel-evant here and in contrast to countable nouns (cf. (2)), morphologically pluralpronouns can combine with morphologically singular nouns (note the idiomaticuse of Gemüse in (13c)):

(13) a. Ihryou(inform/pl)

(blödestupid

Bande)gang

seidare

ja schonalready

da!here

‘You stupid gang are here already!’b. Ihr

you(inform/pl)(dummesstupid

Pack)gang

seidare

ja schonalready

da!here

‘You stupid gang are here already!’c. Ihr

you(inform/pl)(jungesyoung

Gemüse)vegetable(s)

seidare

ja schonalready

da!here

‘You young folk are here already!’

The same holds for Sie ‘you(form)’ with plural reference:

(14) a. scSieyou(form)

blödestupid

Bandegang

(dathere

alle)all

‘you stupid gang there all’

As above, the pronoun exhibits semantic, but not morphological, agreement withthe noun. Unlike singular in Section 3.1.1, the semantic agreement here is withregard to plural.

Importantly, morphologically singular pronouns are not possible with collec-tive nouns denoting multi-member sets:

(15) a. *duyou(inform/sg)

blödestupid

Bandegang

‘you stupid gang’b. *du

you(inform/sg)dummesstupid

Packgang

‘you stupid gang’

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.9 (533-596)


c. *duyou(inform/sg)

jungesyoung

Gemüsevegetable(s)

‘you young folk’

Furthermore, as illustrated above, Dutch u takes a verb with singular inflection,independent of its reference. Thus, I took the pronoun to be morphologicallysingular. Note that, although u ‘you(form)’ with plural reference can, for someDutch speakers, be combined with these singular nouns, all speakers agree thatthe pronoun cannot have singular reference:

(16) *Uyou(form)

regeringgovernment

isis

daarvoorfor.it

verantwoordelijk.responsible

(Dutch)

‘You government are responsible for it.’

To summarize thus far, we have observed that dis-agreement in morphological,but not in semantic, number is tolerated. I turn to the second main combination.

. Morphological dis-agreement: pronoun(sg) + noun(pl)

Similar to the section above, regular determiners agree with the morphology onthe noun, independent of its semantics:

(17) a. *this pantsb. these pants

With one complication, we again find the opposite state of affairs with pronominaldeterminers.

.. Pluralia tantum: Semantic agreement in the singularIn order to construct a relevant example, the noun must fulfill a number of con-ditions: first, while having singular “reference”, it must exhibit plural morphology;second, it must take a verb in the plural with a regular overt or covert determiner.English smarty pants in (18a) is a promising candidate in that the noun can only“refer” to a single person but shows plural morphology. However, there is someindication that it has been reanalyzed as morphologically singular: first, the verbshows singular inflection (18b); second, pants in this idiomatic use can be plu-ralized with concomitant plural verb inflection (18c–d) (the data are due to RexA. Sprouse):

(18) a. you smarty pantsb. Smarty pants {is / *are} coming.c. you smarty pantsesd. Some smarty pantses are coming.

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.10 (596-651)

Dorian Roehrs

Considering the data in (18b–d), you smarty pants is not a relevant example and sofar, I have not come across an appropriate plurale tantum that can co-occur with asingular pronoun such as German du or Dutch u with singular reference.5 We willsee below, however, that there is evidence for a null plurale tantum.

.. Countable nouns: Semantic agreement in the pluralAs discussed above, I assume that Dutch u is morphologically singular. Concen-trating on its plural reference, u can combine with a plural noun:

(19) a. Uyou(form)

(taalkundigen)linguists

bentis

fantastisch.wonderful

(Dutch)

‘You linguists are wonderful.’b. u

you(form){studentenstudents

//

lerarenteachers

//

docenten}lecturers

‘you students/teachers/lecturers’

Again, as in the cases above, there is semantic agreement despite a morphologicalmismatch. As expected by now, morphological agreement is not enough to licensethese constructions: although speakers vary in their judgments from marginal toungrammatical when the pronoun with singular reference is combined with asingular noun, they all agree that it cannot have plural reference in this case:

(20) *uyou(form)

{studentstudent

//

leraarteacher

//

docentlecturer

//

mazzelaarlucky.person

//

geluksvogel}lucky.person

‘you student/teacher/lecture/ lucky guy / lucky person’

.. Interim summarySummarizing the individual cases, note that semantic agreement is a necessarycondition for licensing pronominal DPs but morphological agreement is not:

(21) Summary of Number Agreement

(i) Pronouns and nouns must agree in semantic number:a. *Sie(sg) + countable noun in the pluralb. *du(sg) + collective nouns in the singularc. *u(sg) + collective noun in the singulard. *u(pl) + countable noun in the singular

(ii) Pronouns and nouns agree in morphological number with the exceptionof:a. Sie(pl) + countable noun in the singularb. ihr(pl) + collective noun in the singular

. This also means that we cannot test whether or not the combinations of ihr or Sie with pluralreference + plurale tantum are ungrammatical, as we would expect.

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.11 (651-711)


c. Sie(pl) + collective noun in the singulard. u(sg) + countable noun in the plural

Comparing regular and pronominal determiners, this leads to a first generaliza-tion:

(22) Generalization #1Regular determiners show morphological agreement and pronominal deter-miners semantic agreement with nouns.6

Before I turn to the proposal, consider mass nouns.

. Pronominal noun phrases contain “individualizable”and concrete nouns

Mass nouns on the substance reading are not possible unless they are licensedby a classifier, a semi-suffix involving a proper name, or agentive derivationalmorphology. Compare (23a) to (23b):7

(23) a. *{duyou(inform/sg)

//

Sieyou(form)

//

ihr}you(inform/pl)

Mistmanure

‘you manure’

. The following exclamation is an apparent counterexample where a semantically singularpronoun is combined with a semantically plural noun:

(i) Ach,PRT,

duyou(inform/sg)

grünegreen

Neune!nine

‘Wow!’

However, I believe this is not a true counterexample for two reasons: (i) unlike the regular nounNeun ‘nine’, there is an additional ending (-e), and (ii) this example seems to be an idiom as du‘you’ is not used to address someone.

. As a reviewer points out, this does not seem to be true for all mass nouns:

(i) Duyou

Dreckdirt

‘you dirt’

In certain respects, this state of affairs is similar to English:

(ii) a. you *(sweet bit of) honeyb. you (piece of) slime

For present purposes, I will concentrate on the cases that are ungrammatical.

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.12 (711-779)

Dorian Roehrs

b. duyou(inform/sg)

{*(Haufen)(heap of)

Schleimslime

//

SchleimbertSlimebert

//

Schleimer}slimer

‘you piece of slime/slime bucket/slime’

Recalling that mass nouns have no spatial integrity or inner partitioning, we con-clude that the nouns in the pronominal DPs must be “individualizable”.

Abstract mass nouns do not combine with second person pronouns either(24a). However, the examples improve if made “individualizable” as well as moreconcrete (24b):

(24) a. *{duyou(inform/sg)

//

Sieyou(form)

//

ihr}you(inform/pl)

Liebelove

‘you love’b. du

you(inform/sg)Dummheitstupidity

*(in Person)personified

‘you stupidity personified’

In other words, beside “individualizable”, the nouns must also be concrete.Considering that mass nouns denote a homogenous and shapeless amount of

material, they have no semantic number and, as a consequence, do not fall un-der the generalization in (22). For now, I state a second generalization, but I willcollapse the two below:

(25) Generalization #2The noun must be “individualizable” and concrete.

Finally, mass nouns, especially edible and drinkable substances, are “ambiguous”between a substance and an item interpretation (26) (for recent discussion ofmass/count coercion, see Wiese & Maling 2005). If this is true, then mass nounson the item reading are expected to be possible. This expectation is borne out (27):

(26) a. Dasthis

istis

{Eiegg

//

Gold}.gold

‘This is egg/gold.’b. Das

thisistis

einan

{Eiegg

//

*Gold}.gold

‘This is an egg/gold.’

(27) a. Duyou(inform/sg)

Eiegg

‘you egg’b. Du

you(inform/sg)Gold*(stück)gold.piece

‘you piece of gold’

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.13 (779-857)


Note also that the noun does not have to be [+animate] but has to denote someemotive property (cf. the exclamation of surprise Ach, du dickes Ei! ‘Oh, you thickegg’ or the disparaging vocative Eierkopf ! ‘egg.head’). Table 3 summarizes therelevant judgments of the individual cases (rows two through four contain themorphologically plural pronouns and the last three rows the morphologically sin-gular ones; number indications in parentheses after the pronouns denote semanticnumber; exclamation points indicate the cases surprising under the traditionalassumption of morphological agreement only; sc stands for stylistic clash):

Table 3. Summary of the judgments of the different combinations of pronoun and noun

singular plurale tantum plural collective masscountable (Section 3.2.1) countable (Note 1) (Note 7)

Sie ‘you(sg)’ sc (!) – * (!) * *ihr ‘you(pl)’ * – √ √ (!) *

Sie ‘you(pl)’ * – sc sc (!) *du ‘you(sg)’ √ – * * (!) * (!)

u ‘you(sg)’ * (!) – * * (!) * (!)u ‘you(pl)’ * (!) – √ (!) * (!) * (!)

To reiterate, the traditional assumption of morphological agreement between de-terminer and noun is both too weak and too strong: it is too weak in that it wouldallow ungrammatical examples such as *du Bande ‘you(sg) gang’ (“overgenera-tion”); it is too strong in that we would not expect grammatical cases such as ihrBande ‘you(pl) gang’ (“undergeneration”).

. The proposal

One finding of the above discussion was that agreement with nouns is morpho-logical with regular determiners and semantic with pronominal determiners. If wemake the strongest claim that determiners in the broad sense are to be a homoge-nous class (cf. Postal 1966), then some way or another, the lexical specificationsof regular determiners need to contain some statement with regard to semanticnumber and pronominal ones with regard to morphological number. Further-more, in order to keep the overriding generalization of determiner concord insidethe DP, our methodology must be to find a different account for the “exceptional”,morphologically dis-agreeing nouns.

In its simplest form, the main proposal is that pronominal determiners aremore restrictive, that is, more specified, than regular determiners: the former re-quire not only morphological, but also semantic, agreement with the head noun.In order to explain this difference, we will proceed in a stepwise fashion. Starting

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.14 (857-898)

Dorian Roehrs

with the “exceptional” cases, I will propose that the morphologically dis-agreeingnominal is in a Specifier position and that the head of the matrix DP containingthat Specifier is filled by a null noun (Section 5.1). Given two independent nounphrases, there are two concord domains and the dis-agreement is only apparent.After discussing some potential alternatives (Section 5.2), I propose that semanticdis-agreement between the two nominals is ruled out by Predicate Modification(Section 5.3). This operation introduces a form of conjunction between the nom-inal in the Specifier and the null noun, guaranteeing their semantic compatibility.Next, employing feature checking and partial functions, Section 6 explains howmorphological and semantic agreement between the pronoun and the head nounis brought about. After extending the discussion in Section 7, some remainingissues are taken up in Section 8.

. Accounting for morphological dis-agreement

If pronominal determiners agree in both morphological and semantic num-ber with the head noun, then grammatical instances of (morphological) dis-agreement (cf. (21ii)) require a different account. Two syntactic solutions presentthemselves: (i) the “exceptional” nouns are located in a Specifier position, or (ii)they are right-adjoined.

.. “Exceptions” as SpecifierThe following structural analysis is inspired by Giusti (2002:80), who allows dis-agreement in gender with an element in a Specifier position:

(28) a. lathe

miamy(fem)

amicafriend(fem)

medicodoctor(masc)

(Italian)

‘my doctor friend’b. [DP la mia [ amicai [ medico [NP ti ]]]]

Similar to Giusti, I propose that the noun and its adjectival modifier are in a Spec-ifier position. To be concrete, I assume that this position is Spec,DisP, Dis being amnemonic for dis-agreement. Furthermore, I assume with Grimshaw (1991) thatall noun phrases, including pronominal ones, have a noun as the head of theirextended projection. This head can be overt, as in you linguists (e.g., Postal 1966;Roehrs 2005 and references cited therein), or covert, as in [DP you [NP eN ]] (for ex-tensive discussion, see Panagiotidis 2002a, b, 2003a, b). Under these assumptions,the example in (29a) can be analyzed as in (29b):

(29) a. ihryou(inform/pl)

blödestupid

Bandegang


JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.15 (898-943)


b. DP

Dihr

DisP

[ ]blöde Bande Dis’

Dis XPeN

.. “Exceptions” as AdjunctsAlternatively, the second part of the noun phrase could be right-adjoined to alower phrase:

(30) DP

Dihr

XP

[ ]blöde BandeXPeN

Note that both analyses (Specifier vs. adjunct) base-generate two independentnoun phrases: a matrix, argumental DP and an embedded predicate noun phrase.Given two nouns, there are two independent concord domains and morphologicaldis-agreement is only apparent. Deciding between these two possibilities, it is im-portant to note that there can only be one embedded nominal of this sort. In orderto rule out the occurrence of a second dis-agreeing noun phrase, I propose that thepronoun may select the phrase DisP, whose Specifier hosts this nominal (see alsoSection 8; note that the adjunct analysis in (30) seems to be available for “indefi-nite pronoun + modifier” constructions such as etwas Schönes ‘something nice’,which allow several modifiers; see Roehrs 2006). Arguing for two independentnoun phrases makes a prediction.

It is well-known that adjectives without preceding (overt) determiners musthave a strong ending in German: frische*(s) Gemüse ‘fresh vegetables’, nette(*n)Mädchen ‘nice girls’ (for discussion, see Roehrs in progress, Chapter 4). Now, ifthe dis-agreeing noun phrase is in a Specifier, then the locally unpreceded ad-jective contained in it should only have a strong ending. This is exactly what wefind: Bhatt (1990:154–155) observes that adjectives in plural pronominal DPs canusually have both a strong and a weak ending (31a); however, the cases involvingmorphological dis-agreement only allow a strong adjectival ending (31b):

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.16 (943-992)

Dorian Roehrs

(31) a. ihryou(nom)

{dumme/dummen}stupid(strong/weak)

Idiotenidiots

‘you stupid idiots’b. ihr

you(nom){dummes/*dumme}stupid(strong/weak)

Packgang


To be clear, besides morphological dis-agreement in number, these cases are alsoexceptional in that they only allow a strong ending on the adjective. Under the cur-rent analysis, which involves two morphologically independent noun phrases, thisis expected.8 Before I turn to an account of the semantic properties of the elementin Spec,DisP, it is insightful to consider briefly some other potential analyses andsome reason to believe why these are probably incorrect.

. Alternative analyses

First, one could stick with the traditional assumption that all determiners agreewith their noun only morphologically. As above, we could analyze the grammaticalinstances of (morphological) dis-agreement as involving Specifier positions. How-ever, under these assumptions, it would be unclear how to rule out ungrammatical,but morphological agreeing, noun phrases, which should involve complementa-tion (e.g., *Sie(sg) Schweine ‘you pigs’, *du Bande ‘you gang’). In other words,ungrammatical examples would be predicted to be good and reference to semanticagreement would be needed after all to rule these cases out.

Second, one could argue that pronominal determiners are fundamentally dif-ferent from regular determiners in that they only require semantic agreement.In this scenario, there would be no exceptions. However, this would allow mor-phological dis-agreement between determiners and head nouns, and we would beforced to give up the overriding generalization involving morphological concordinside the DP. Note in this respect that pronouns do agree with their nouns overtlyin some languages (for Japanese, see Furuya 2004).

Third, one could claim that the verb does not agree with the pronoun mor-phologically, as assumed so far, but semantically. Assuming that pronouns are atype of noun and undergo N-to-D movement (e.g., Cardinaletti 1994; Rauh 2004),one could hypothesize that agreement is similar to that triggered by certain groupnouns in some languages (cf. Sauerland & Elbourne 2002):

. It is sometimes proposed that swear words do (optionally) not have phi-features (e.g., Doetjes& Rooryck 2003:290–291). If this were so, then we could put the overt noun in the head positionand there would be no morphological dis-agreement. However, under these assumptions, it isunclear how the agreement features of the adjective are specified.

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.17 (992-1058)


(32) This set {is / are all} odd. (British English)

However, while semantic agreement in (32) is “optional”, the pronouns of for-mal address (German Sie and Dutch u) always trigger certain verbal inflection,independent of their semantic reference. Besides raising doubts about the N-to-Dmovement analysis, there is morphological agreement between the pronoun andthe verb in person (and gender in some languages) and the question would arisewhy this is not so for number.

Finally, one could assume the presence of a null numeral classifier in the con-struction under discussion. In a different context, Wiese & Maling (2005:27) arguethat German has (semantically present) null numeral classifiers in Restaurant Talk.One could suggest then that this is also the case here, as schematized in (33b):

(33) a. Zweitwo

(Glas)glass

Wein,wine

bitte.please

‘Two wines, please.’b. pronoun + classifier + noun

However, the following cases raise strong doubts about the general assumptionof a null classifier: first, classifiers are not needed for countable nouns; in otherwords, Sie Schwein ‘you pig’ and u taalkundigen ‘you linguists’ would still presentinstances of morphological dis-agreement. Second, classifiers occur with numer-als, which, similarly to pronouns, have semantic number but, unlike them, musthave an overt classifier (drei *(Haufen) Gesindel ‘three heaps of riff-raff ’). Thisraises the question of why the classifier must be overt with numerals but not withpronouns. Third, assuming a null classifier, all mass nouns should be grammatical,contrary to the facts (du / Sie / ihr (*Mist) ‘you manure’; cf. Note 7). None of theproblems mentioned in this section arise if morphologically dis-agreeing nounsare located in Spec,DisP, as proposed above. I turn to an account of the semanticrestrictions on the element in this Specifier.

. Accounting for the semantic properties of Spec,DisP

I noted above that the noun phrase in Spec,DisP is not a referring element (uniqueindividual/unique set of individuals), but a predicate that must agree in semanticnumber and be “individualizable” and concrete. To make this part of the discus-sion more concrete, Sie Nichtsnutz ‘you good-for-nothing’, under my assumptions,is analyzed as [ Sie [ Nichtsnutz [ eN ]]]. I follow Panagiotidis (2003a:425) in thateN is a trivial nominal predicate of type <e,t> that takes any individual in thediscourse and yields a ‘true’ value for it. For expository purposes, however, I as-sume with Rauh (2004) that eN is a semantically light noun denoting the propertyPERSON. Consider the semantic calculation.

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.18 (1058-1120)

Dorian Roehrs

In the framework of Heim & Kratzer (1998), the relevant lexical items aredefined in the following, simplified way (in Section 6, determiners will be definedas partial functions):

(34) [[Sie]] = λf: f ∈ D<e,t> . the unique y (formally) addressed such that f(y) = 1.[[Nichtsnutz]] = λz: z ∈ D<e> . z is a good-for-nothing.[[eN]] = λw: w ∈ D<e> . w is a person.

Skipping some later steps, the semantic calculation proceeds as follows:9

(35) [[Sie Nichtsnutz eN]] =

a. [[Sie]] ([[Nichtsnutz eN]]) =b. [[Sie]] (λv: v ∈ D<e> . [[Nichtsnutz]] (v) and [[eN]] (v) = 1) =c. the unique y (formally) addressed such that y is a good-for-nothing and

y is a person

The crucial step is Predicate Modification in (35b), which introduces a form ofconjunction. Assuming that conjunction can only combine two elements that arethe same semantically, this semantic mechanism guarantees that the predicate inSpec,DisP and the head noun eN are semantically the same while at the same timeallowing for morphological dis-agreement (Section 5.1). To be precise, this nullpredicate noun is, under my assumption, a plurale tantum as it must be singularsemantically, agreeing with Nichtsnutz ‘good-for-nothing’ in Spec,DisP, but pluralmorphologically, agreeing with Sie ‘you’ in D (for an inventory of null nouns, seeSection 7). I turn to (part of) the second generalization.

The relevance of the property [individualizable] was motivated by the factthat second person pronouns cannot freely combine with mass nouns on the sub-stance reading. Recall from above that nouns with semantic number have spatialintegrity or inner partitioning. Intuitively, what these types of nouns have in com-mon is that they denote a set that consists of members that are discrete and can beidentified, that is, they are “individualizable”. We can now collapse the two gener-alizations, saying that semantic number implies [individualizable] and vice versa(for the feature “concrete”, see below):

(36) Generalization (revised)While regular determiners show only morphological agreement with thenoun, pronominal determiners require both morphological and semanticagreement where: semantic number ↔ [individualizable]

. I will concentrate on the restrictive reading of the nominal predicate here:

(i) Sieyou

Nichtsnutzgood-for-nothing

kommencome

mitalong

undand

Sieyou

Witzboldjoker

bleibenstay

hier.here

‘You good-for-nothing come along and you joker stay here.’

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.19 (1120-1172)


For pronominal DPs, this “rules out” mass nouns in the head position directly andin Spec,DisP indirectly (by Predicate Modification with eN). So far, I have not ad-dressed the question if agreement in morphological and semantic number mightbe mediated by a functional phrase. Employing NumP, I now turn to agreementbetween the (lower) part of the nominal and the pronoun.

. Possible values on Num and DP-external “special” agreement

First I entertain the idea that both morphological and semantic number are partof syntax. However, I will argue that semantic number can not only be eliminatedfrom the syntax but, in fact, must be. To this end, I discuss cases of “special” se-mantic agreement between the DP and the verb. I conclude that it is not the overtDP that agrees with the verb but a null pronoun. Having reduced DP-external “se-mantic” to regular morphological agreement, I make semantic agreement withinthe DP more precise by employing partial functions.

. Semantic number is not feature checking

Above, I argued that pronominal determiners and head nouns are subject to mor-phological and semantic agreement. This raises the question of how this agreementis brought about. In particular, making the widely held assumption of an interme-diate Number Phrase (NumP), as first postulated by Ritter (1991), I consider thepossibility that the head Num may be marked not only for morphological, but alsofor semantic, number. This is not a novel idea.

In order to account for the semantic agreement between the verb and certaingroup nouns in British English (37a), Sauerland & Elbourne (2002:291) pro-pose that nouns have interpretable number and mereology features (37b) and thefunctional node T has the uninterpretable counterparts:

(37) a. This set {is / are all} odd. (British English)b. set: [number: sg; mereology: pl]

Glossing over the details of the account, they assume that (regular) determinersonly agree with (morphological) number of the noun while verbs, anaphors, and(DP-external) pronouns may agree with either number or mereology of the noun.If pronouns are determiners, we need to extend Sauerland & Elbourne’s analysis.Employing NumP, I will flesh out this broader account first for DP-internal andthen DP-external agreement.

Developing ideas of Delfitto & Schroten (1991), Panagiotidis (2002a:58,2003a:421) proposes that Num may take on the following specifications: Plural,Singular, and Zero. Simplified, if Numpl combines with a noun, it brings about

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.20 (1172-1236)

Dorian Roehrs

a plural countable noun; if Numsg combines with a noun, it derives a singularcountable noun; and if NumØ combines with a noun, it results in a mass noun.As a consequence, he argues that the categorization of nouns into countable andmass type elements is a function of the specification on Num.

With the above discussion in mind, suppose that the head Num is freely spec-ified in the Numeration with regard to both morphological and semantic number,where the values of α and β may range independently over Plural (+pl), Singular(–pl), and Zero (0pl):

(38) Num [αmorph; β sem]

Nouns such as Ei ‘egg’ are merged unspecified as in (39a) and second person pro-nouns such as singular du and plural ihr can only occur in the contexts in (39b)and (39c), respectively:10

(39) a. Ei [γmorph; δ sem] (where γ = δ)b. du [εmorph; ζ sem] (where ε = ζ = –pl)c. ihr [εmorph; ζ sem] (where ε = ζ = +pl)

Assuming some checking/valuing mechanism between Num, on the one hand, andthe determiner and the noun, on the other, the morphological value for γ and εare checked by α and the semantic value for δ and ζ by β. As a consequence, γ = ε =α and δ = ζ = β. If morphological and semantic number coincide (α = β on Num),then γ, ε, δ, and ζ all have the same value as α and β.

To illustrate, a second person pronoun and the noun Ei ‘egg’ are spelled out asfollows: the specification Plural (+pl) will bring about ihr Eier ‘you eggs’, Singular(–pl) will result in du Ei ‘you egg’, and Zero (0pl) will give the noun Ei ‘egg’, thesubstance, and a mass determiner. For the cases discussed so far, it seems thatfeature checking/valuing can bring about the relevant agreement. In what follows,I show that feature checking of semantic number can not only be eliminated but,in fact, must be.

Above, I did not discuss the options involving the Zero (0pl) specification onNum:

(40) a. Num [+pl morph; 0pl sem]b. Num [–pl morph; 0pl sem]c. Num [0pl morph; +pl sem]d. Num [0pl morph; –pl sem]

To the extent that I am aware, these combinations are not instantiated. Supposethen that a Zero specification for semantic number entails a Zero specification for

. For arguments that person is independent of number, see Panagiotidis (2002a:22–25).

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.21 (1236-1295)


morphological number and vice versa or, alternatively, that a non-Zero specifica-tion for one type of number entails a non-Zero specification for the other:

(41) a. 0pl morph ↔ 0pl semb. ±pl morph ↔ ±pl sem

Both entailments essentially state the same: a Zero specification for morphologi-cal number can only be combined with a Zero specification for semantic numberand a non-Zero specification for morphological number can only combine witha non-Zero specification for semantic number (where the latter does not excludedifferent values for Singular and Plural, see below). While both rule out the fournon-instantiated options, the alternative entailment is unaffected. For instance,(41b) does not rule out the combination [0pl morph; 0pl sem]. Thus, mass nounsand mass (pronominal) determiners are not excluded but, due to feature check-ing, they cannot combine with non-mass (pronominal) determiners or non-massnouns. However, unlike (41a), the entailment in (41b) has the advantage that wecan express in a direct way a connection between morphological number and thetwo original generalizations collapsed at the end of Section 5.3, where ±pl semnow equals semantic number:

(42) ±pl morph ↔ ±pl sem ↔ [individualizable]

Suppose that this entailment is part of Universal Grammar, perhaps parameter-ized for classifier languages.11 If a learner has evidence for morphological numberor knows that a nominal predicate is [individualizable], then she can infer thatthe predicate has semantic number. For the cases discussed above, this allows usto eliminate semantic number from the syntax. Let us see if we can go furtherthan that.

Thus far, we have only looked at the cases where the value for ±pl coincidedfor morphological and semantic number. Next, I turn to the remaining two op-tions showing that feature checking of semantic number not only can but must beeliminated from the system. In fact, DP-external agreement with the verb, whichmotivated semantic number features in British English, provides an argumentagainst it if extended to pronominal DPs.

If pronominal determiners are syntactically similar to regular determiners,as assumed all along, then we might expect that there may be cases of “special”agreement à la British English also with pronominal DPs. In order to test this, the

. Most recently, Watanabe (2006) argues that the difference between classifier languages likeChinese and Japanese and non-classifier languages like English and German lies in the mor-phological realization of Num (his #). He argues that NumP is always present. If so, then (42)may also hold in classifier languages once the specification for morphological number has beenappropriately modified.

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.22 (1295-1347)

Dorian Roehrs

pronoun and the noun must be semantically compatible while, at the same time,they must be different in some number specification. Recall now that in isolation,Dutch u always takes a verb with singular inflection, independent of its reference.Although not all speakers allow (43a), one might expect that those who do also al-low plural semantic agreement triggered by the noun. Furthermore, German Sie byitself always triggers plural inflection on the verb. With a morphologically singularnoun as part of the DP, one might expect singular inflection on the verb triggeredby the morphology of the noun. Crucially, neither semantic nor morphologicalagreement triggered by the noun is possible:

(43) a. Uyou(form)

regeringgovernment

{is/*zijn}is/are

daarvoorfor.it

verantwoordelijk.responsible

(Dutch)

‘You government are responsible for it.’b. Sie

you(form)blödestupid

Bandegang

{sind/*ist}are/is

dafürfor.it

verantwortlich.responsible

‘You stupid gang are responsible for it.’

As a point of departure, special morphological agreement as in (43b) is the easiestto rule out: unlike in the Dutch example, I argued that the collective noun in Ger-man is not in the head position but in Spec,DisP. In other words, the overt nounin German is not the lexical head of the pronominal DP and is thus not expectedto take part in morphological concord or agreement relations of the matrix DP. Iturn to special semantic agreement.

I assume that, in contrast to the ungrammaticality of (37) in American En-glish, the ungrammaticality of (43a) in Dutch is not just a language-specific featurebut is, to the extent that I am aware, a general property of languages. In otherwords, making the strongest claim that the phenomenon in British English doesnot exist with pronominal determiners, its absence calls for a principled account.Our goal is to state the relevant properties of both regular and pronominal deter-miners as similarly as possible (cf. Postal 1966), while still allowing DP-externalsemantic agreement with regular but not pronominal determiners.

As discussed above, the pronouns of formal address in German and Dutchcan be both semantically singular and plural, but are morphologically specified:plural in German and singular in Dutch. In other words, they can only occur inthe following contexts:

(44) a. Sie [+pl morph; ζ sem]b. u [–pl morph; ζ sem]

Furthermore, unlike nouns such as Ei ‘egg’, other nouns seem to be fully speci-fied for number. For instance, pluralia tantum with singular meaning (e.g., Hosen‘pants’) and collectives that can only occur in the singular (e.g., Pack ‘gang’) haveopposite specifications and can occur only in the following contexts:

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.23 (1347-1456)


(45) a. plurale tantum [+pl morph; –pl sem]b. collective [–pl morph; +pl sem]

If Num happens to have the same specifications as in (45a), then this generalcontext illustrates German Sie under singular reference, that is, Sie takes a (null)plurale tantum as its complement. The context in (45b) exemplifies Dutch u underplural reference, that is, u has a collective noun as a complement. With differ-ing values for number, this is the case where special agreement could arise. Inorder to find an answer why it cannot with pronouns, we need to compare thespecifications of pronominal determiners to those of regular ones:

(46) a. these [+pl morph; ζ sem]b. this [εmorph; ζ sem]

Note that, under these assumptions, regular determiners are very similar to pro-nouns in the way their number specifications are stated: on the one hand, ab-stracting away from a person feature and a statement about a formal situation,these is identical to Sie; on the other, this is almost the same as u with the differ-ence that this can combine with mass nouns. Although this syntactic similarity isa desideratum, as implied by Postal (1966), it raises the question as to how to ruleout special semantic agreement with pronominal DPs since both this and u areunspecified for semantic number: in each case, the collective noun, mediated byNum, could specify the semantic number on the determiner allowing “optional”external agreement in the plural on the verb.

I will (gently) take the bull by the horns and tentatively claim that seman-tic agreement in these cases is not brought about by feature checking with theovert noun.12 Rather, I propose that all number agreement of this kind is mor-

. There is another indication that feature checking with the noun is probably incorrect. Itis well-known that unlike German, English disallows singular pronominal DPs in argumentposition (ia). Interestingly, although collective nouns are (marginally) allowed in British English(ib), an issue originally brought to my attention by Erika Troseth (p.c.), number agreementbetween the pronominal DP and the verb has to be triggered by the plural pronoun rather thanthe morphologically singular noun (ic):

(i) a. *You idiot are driving me crazy.b. ?You committee are driving me crazy. (British English)c. Us committee {??are / *am / *is} frustrated with the way things are going.

If true, then neither mereology nor number features on the noun may trigger DP-external agree-ment directly. (There is a caveat here: (ib) presumably involves morphological dis-agreement,cf. Note 4. If so, Sauerland & Elbourne could suggest that, similar to my proposal for German(43b), the noun is in Spec,DisP and DP-external agreement would not be expected. However,with Spec,DisP at least in principle available, their proposal is becoming more similar to thepresent analysis.) Note in passing that the analysis of English is not entirely straightforward.

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.24 (1456-1473)

Dorian Roehrs

phological, that is, the checking/valuing of morpho-syntactic features is mediatedby Num.13 To this end, I suggest that DP-external “semantic” agreement may, infact, be morphological agreement that involves a null matrix pronominal DP anda semantically agreeing noun phrase in Spec,DisP:14

(47) a. [DP pron [DisP this set [NumP Num [NP eN ]]]]b. (*)[DP pron [DisP u regering [NumP Num [NP eN ]]]]

Importantly, for cases similar to (47a), there is evidence for overt pronouns inSwedish (see the data in Josefsson, to appear) and Romanian (Giusti 2002:76).In order to derive the difference between the embedded regular DP in (47a),which triggers apparent semantic agreement, and the embedded pronominal DPin (47b), which does not, I suggest that pronominal DPs cannot be embedded

Besides the difference pointed out for (ia), English usually also requires an additional modifierwith collective nouns (there seems to be more variation with regard to (iic) in British English;the judgments are provided in their average):

(ii) a. you gang *(of killers)b. you group *(of idiots)c. us committee ??/*(of teachers)

To speculate on (ii), English presumably has a different structural account than German. Due toconsiderations of space, I will not discuss these cross-linguistic differences here.

. A reviewer points out that this is consistent with the fact that in Distributive Morphology,functional heads have only morphological features throughout the syntactic derivation.

. Den Dikken (2001) also argues for a DP containing a null pronoun (for more general dis-cussion, see his paper). Unlike here, however, he tentatively proposes that the overt material isin apposition (pp. 38–39). Importantly, both analyses seem to face the problem that the overtmaterial can be indefinite (den Dikken 2001:29 Fn. 13):

(i) [ pron [ Any committee worth their salt ]] are going to look into that.

Sauerland & Elbourne (2002:295 Fn. 13) point out that the presence of pro(n) cannot be justifiedhere since it is not referential and it cannot be bound or E-type. While I have no solution to thisproblem myself, note that, in order to bring about person agreement, a pronominal elementmust presumably also be present in Spanish, for both a definite and an indefinite DP (the dataare from Lyons 1999:144):

(ii) a. {Losthe

//

Algunos}some

estudiantesstudents

trabajamoswork(1pl)

mucho.much

(Spanish)

‘We students / A number of students (including the speaker) work hard.’b. [NP pron [DisP los / algunos estudiantes [NP eN ]]]

Although unclear at this point, the indefinite cases in (i) and (ii) might be related. If so, then theassumption of a null pronoun may be justified in (i) after all.

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.25 (1473-1538)


in another pronominal DP. Below, I will try to connect this restrictiveness ofpronouns to the restrictions discussed in the first part of this paper.

More generally, if this discussion is on the right track, then we have an argu-ment that semantic agreement and morphological concord inside the DP shouldnot be accounted for by uniform feature checking, as in Sauerland & Elbourne(2002), but in different ways. (That part of the account is semantico-conceptualbecomes clear when we consider the fact that the noun must be concrete, whichwas part of the second generalization. This property is presumably not broughtabout by feature checking.) In order to explain the higher degree of restrictive-ness of pronominal determiners, both internal and external to the DP, I proposethat semantic number is part of the denotations of these determiners and nominalpredicates. In what follows, I briefly sketch an account for both restrictions.

. Ruling out DP-internal semantic dis-agreement

I propose that DP-internal dis-agreement can be ruled out by the requirement thatthe pronominal functor and its nominal argument be semantically compatible. Tomake this idea concrete, I will formulate the denotations of some of the elements aspartial functions. To illustrate, as discussed above, semantically singular pronounscannot be combined with semantically plural nouns (48a). Assuming that the plu-ral noun is derived by attaching a plural marker, here illustrated by the default -s,to the noun, this example can potentially be analyzed in two ways: the pluralizednoun can be in the matrix DP (48b) or in Spec,DisP (48c), where the latter has anull collective noun in NP and a null singular marker in NumP:

(48) a. *duyou(sg)

Nichtsnutz-egood-for-nothing-s

b. [DP du [NumP -s [NP Nichtsnutz ]]]c. [DP du [DisP [NumP -s Nichtsnutz ] Dis [NumP Ø [NP eN ]]]]

The representation in (48b) is ruled out by morphological (and semantic) dis-agreement. The one in (48c), however, cannot be ruled out by recourse to amorphological mismatch as the singular pronoun agrees with the singularized,null collective noun. Rather, this analysis must be ruled out semantically.

Consider the following lexical entries for the singular pronoun (type <<e,t>,e>), the plural marker -s (<<e,t>,<e,t>>), the nominal predicate Nichtsnutz(<e,t>), and the null nominal predicate eN (<e,t>):

(49) [[du]] = λf: f ∈ D<e,t> and there is exactly one x ∈ C such that f(x) = 1.the unique y ∈ C (informally) addressed such that f(y) = 1,where C is a contextually salient subset of D.

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.26 (1538-1609)

Dorian Roehrs

[[-s]] = λf: f ∈ D<e,t> [λv: v ∈ D<e> and the number of x ∈ C is greater thanone such that f(x) = 1. f(v) = 1, where C is a contextually salientsubset of D].

[[Nichtsnutz]] = λz: z ∈ D<e> . z is a good-for-nothing.[[eN]] = λw: w ∈ D<e> . w is a group.

Some remarks are in order here. Both the pronoun and -s are defined as partialfunctions, that is, functions that have presuppositional conditions on their do-main of application. In other words, they only apply to a subset of their potentialdomain. If these conditions are not met, then this leads to presupposition failureand the element receives no semantic value (for more details, see Heim & Kratzer1998:75). Unlike the pronoun, the number marker is a partial identity function,that is, it maps its argument to itself provided that certain presuppositional condi-tions are met. As can be checked in (49), these conditions check for the number ofelements in the contextually salient domain. This will be the key in the account tofollow. Let us first look at the relevant tree structure in terms of the semantic types.

We start at the bottom. Putting aside the interaction between the null sin-gular marker and the null collective noun (but see below), the semantic type ofeN is passed up to Dis’. Functional Application applies to -s and the NP, con-taining Nichtsnutz. The resulting [-s Nichtsnutz] combines with Dis’ by PredicateModification. Finally, DisP and du combine by Functional Application:

(50) DP<e>

du<<e,t>,e> DisP<e,t>

Dis’<e,t>

Dis

NumP<e,t>

NumP<e,t>-s<<e,t>,<e,t>> NPNichtsnutz< >e,t

Ø NPeN<e,t>

(Functional Application)

(Predicate Modification)

(Functional Application)

There is no general type mismatch and the expression is, in principle, inter-pretable. This is a welcome result as we argued for the possibility of this type ofstructure in (35), now updated as [DP Sie [DisP [NumP Ø Nichtsnutz ] Dis [NumP ØeN ]]] ‘you good-for-nothing’. Note, however, that the two partial functions havecontradictory presuppositional conditions: while the pronoun is only defined ifthere is exactly one relevant element in the domain, the plural marker is only de-fined if there is more than one. Applied to the same contextually salient domain,this will lead to a necessary presupposition failure for one of the elements. Conse-

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.27 (1609-1662)


quently, the example in (48a) has no semantic value.15 In (50), I ignored the nullsingular marker and the null collective noun. I now turn to the discussion of anovert counterpart.

So far, we have focused on cases where a semantically singular pronoun cannotcombine with a semantically plural noun. However, as briefly mentioned in Note1, singular pronouns can take plural nouns under certain conditions, namely if thepronoun refers to the group itself (“collective” reading), as in (51a), or the grouphas only one member (“distributive” reading), as in (51b):16

(51) a. Duyou

blödestupid

Regierung!government

‘you stupid government’b. u

youéénmansregeringone.man.s.government

(Dutch)

‘you one-man government’

To be clear, these cases involve neither morphological nor semantic dis-agreement.The question arises as to how to account for the different interpretations. On myassumptions, these data have the pronoun in DP, a null singular marker in NumP,and the noun in NP:

(52) [DP pron [NumP Ø [NP collective noun ]]]

Without going into too much detail, I assume that the denotation of collectivenouns is defined in such a way that the relevant (sub)sets must typically have sev-eral members but may, in the marked case, have just one member. The partialidentity function of the singular marker delimits the collective predicate to one rel-evant set in the discourse (there is, in principle, no restriction on the number of themembers in this set). Now, if the pronoun addresses the set itself, the “collective”interpretation is brought about, independent of whether or not the set has one ormore members. If the pronoun addresses the members of the set, this set can onlyhave one member as the pronoun and the collective noun must be semanticallycompatible, as discussed above. In this scenario, the “distributive” reading results.

. As pointed out by Heim & Kratzer (1998:81–82), a presuppositional account raises in-teresting issues. Note in this respect that number mismatches with countable nouns are notstrictly impossible. For instance, the grammaticality status of *ihr Nichtsnutz ‘you(pl) good-for-nothing’ improves if we imagine an alien with one body but several heads. In this scenario,addressing the individual “heads”, we could refer to the entire being. If so, the acceptablility ofthese cases seems to have to do with physical facts rather than with mismatches of the semantictypes in the tree representation. This might be taken as an indication that the presuppositionalaccount is on the right track. (Cf. Heim & Kratzer 1998:244–245, who define partial identityfunctions of type <e,e> for pronouns.)

. I thank the two reviewers for discussion and examples.

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.28 (1662-1713)

Dorian Roehrs

Finally, as illustrated above, regular determiners do not exhibit these kindsof restrictions. As just discussed, the number of entities is delimited by the par-tial identity function of the corresponding number marker. However, unlike thepronoun, the regular determiner is defined as a partial function that requires thegeneral supremum, the largest relevant quantity, be “picked out” in the domain.In other words, definite determiners are “totality extractors” that work equallyfor singular count, plural count, or mass nouns (cf. Sharvy 1980). In the lattersense, regular determiners are less specified than pronominal ones, which “pickout” either singleton or multi-member totalities.

. Ruling out DP-external semantic dis-agreement

DP-external semantic agreement with a verb is ruled out by the semantic incom-patibility of the null pronoun in D and the overt pronoun in Spec,DisP in (47b),repeated here in simplified form as (53):

(53) (*)[DP pron [DisP u regering [NP eN ]]]

I will assume here that the null pronoun “picks out” a different set of peoplethan the second person pronoun. If this is so, then these are two different, non-intersecting sets within one referring DP. I assume that this leads to semanticincompatibility.

To sum up, I proposed that morphological number is part of syntax. Imple-mented by checking/valuing features, Num mediates concord inside the DP and averb may agree with a possible null pronoun. In contrast, semantic number is partof semantics, instantiated here by defining the relevant denotation as partial func-tions. I conclude that morphological and semantic numbers are to be dissociatedfrom one another.17 Let us cast our net wider.

. In contrast to the semantic derivation proposed in the text, a reviewer makes the intriguingsuggestion that pronominal DPs, as in (ia), involve predication parallel to clauses, as in (ib).Furthermore, s/he points out that, while the nominal example only allows singular reference ofthe pronoun, the clausal datum allows both singular and plural reference:

(i) a. uyou

studentstudent

(Dutch)

‘you student’#‘you students’

b. Uyou

bentare

student.student

‘you are a student’‘you are students’

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.29 (1713-1764)


. Pronouns of other morphological person and null nouns

Above, I proposed that pronouns and nouns agree morphologically and seman-tically. This makes a number of predictions: (i) most pronouns have coincidingnumber values. If these pronouns may, under certain conditions, take on a differ-ent semantic reference, then these noun phrases should also allow a morpholog-ically dis-agreeing, but semantically agreeing, noun in Spec,DisP. And (ii), if thecombination “pronoun + noun” apparently dis-agrees in semantic number, thisconstruction must have a different interpretation from that of a determiner and ahead noun. I consider each of these predictions in turn using pronouns of differentmorphological person. At the end, I provide an inventory of null nouns.

. Some non-canonical cases

In some non-canonical uses, pronouns have different referential properties. Forinstance, plural pronouns may refer to a single person. This is not only possiblewith second person ‘royal’ you but also with first person ‘nursely’ we (Harley & Rit-ter 2002:507). As expected, while external agreement remains plural, the pronouncan combine with a singular noun:

(54) a. Ihryou

Nichtsnutzgood-for-nothing

könntcan(2pl)

nichtnot

derthe

Königking

sein.be

‘You good-for-nothing can not be the king.’b. Wie

howfühlenfeel(1pl)

wirwe

Suffkoppbooze.head

unsrefl.

dennPRT

heute?today

‘How are we lush feeling today?’

These cases are similar to German Sie and nothing else needs to be said. I turnto other non-canonical uses where a third person pronoun combines with anovert noun.

It is well-known that a number of Scandinavian dialects allow a proper nameto be preceded by a pronoun, that is, a preproprial article (for discussion, seeDelsing 1993:54–55; Roehrs 2005). As pointed out by Josefsson (1999:732, 740),Icelandic pronouns and nouns agree morphologically, as shown in (55a). Interest-ingly, this language also has male names that decline like weak feminine nouns. Ifgender is similar to number in the relevant sense, then we also expect semantic,but not morphological, agreement to hold. This is indeed so, as shown in (55b):

Similar facts hold in German. Considering the traditional assumption that declarative clausesdenote truth values (type <t>) and referential noun phrases individuals (type <e>), I arguein work in progress that clausal predication of the type in (ib) should not be extended to thenominal case in (ia). For a possible morphological reflex of this difference, see Note 20 below.

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.30 (1764-1827)

Dorian Roehrs

(55) a. húnshe

AnnaAnna

(Icelandic)

‘Anna’b. {hann

he//

*hún}she

SturlaSturla(fem)

‘Sturla’

Besides morphological dis-agreement in gender with proper names, Old Icelandicalso seems to tolerate dis-agreement in morphological number. To set the stage,German allows an optional expletive article with a singular proper name but notwith two coordinated ones. Compare (56a) and (56b). However, this is possible inOld Icelandic with pronouns, as illustrated in (56c):

(56) a. (der)the

PeterPeter

‘Peter’b. *die

the(pl)PeterPeter

undand

MartinMartin

‘Peter and Martin’c. þeir

they(masc/nom)GísliGísli(masc/nom)

okand

VésteinnVésteinn(masc/nom)

‘Gísli and Vésteinn’

Note that, while the two coordinated proper names agree with the pronoun “se-mantically”, they do not agree with the pronoun morphologically in an obviousand straightforward way.

I follow Panagiotidis’ work in that (morphological) gender is marked on thehead noun but spelled out on the pronoun. If this is so, then the noun phrase hannSturla ‘Sturla’ must contain a null masculine head noun. As assumed above, thedis-agreeing noun is in Spec,DisP. I assume the same for number dis-agreementbetween þeir and Gísli ok Vésteinn:

(57) a. [DP hann [DisP Sturla [NP eN ]]]b. [DP þeir [DisP Gísli ok Vésteinn [NP eN ]]]

Positing a null head noun in these cases, we must be careful not to allow a sub-stantive reading of the pronoun in D as this would give rise to a non-existentadjunction interpretation (i.e., he, Sturla). I propose that these null nouns arenull proper names, that is, covert equivalents of nouns such as Peter and Cubsin the Cubs. These null proper names semantically agree with the overt material in

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.31 (1827-1879)


Spec,DisP and morphologically with the expletive pronoun in D.18 Finally, I turnto non-expletive uses of the pronoun in Old Icelandic.

Thus far, I have argued that the pronoun and the noun must agree not onlymorphologically but also semantically. If this is true, then grammatical cases in-volving apparent semantic dis-agreement must have a different interpretation.Again, Old Icelandic seems to provide evidence for this: while the pronoun and thenoun in (58a) agree both morphologically and semantically, there is a mismatchin both number and gender in (58b).

(58) a. þeirthey(masc/nom)

brœðrbrother(masc/nom/pl)

‘the brothers’b. þau

they(neut/nom)GísliGísli(masc/nom)

farago(pl)

‘Gísli and the other(s) travel’

Importantly, the plural pronoun in the first case has the interpretation of a def-inite determiner but the one in the second case is usually translated as “and theother(s)”. I propose that the former involves complementation and the latter sometype of asyndetic co-ordination:19

(59) a. [DP þeir [NP brœðr ]]b. [[DP þau [NP eN ]] [DP Gísli ]]

Let me take stock with regard to the different overt and covert nouns in the DP.

. Note that, although the semantic relation between the null proper name and the expletivepronoun does not involve agreement of the type seen so far, it also does not involve semanticdis-agreement.

. The structure in (59b) raises the question as to how the surprising singular reference of theplural pronoun, translated as “and the other”, can be derived. One may claim that this is due tosome non-canonical use similar to you and we as discussed in the text. However, if it turns outthat this is not tenable, one might propose that the coordination is actually inside Spec,DisP.In order to avoid type mismatch between the proper name (<e>) and the other conjunct, thiselement must presumably be a null pronominal DP (<e>):

(i) [DP þau [DisP pron eN & Gísli [NP eN ]]]

In order to derive both the singular and the plural interpretation, the null pronominal DP mustbe either singular or plural in reference. If this is on the right track – note the similarity of (i) to(57b), then semantic agreement would also hold in this case and the translation of the pronounas “and the other(s)” is somewhat misleading.

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.32 (1879-1977)

Dorian Roehrs

. The inventory of null nouns

I have argued that morphological and semantic agreement in number have differ-ent accounts. Simplified, I proposed that, while the pronoun and the null nouneN agree morphologically (φ), the pronoun, the noun in Spec,DisP, and eN mustagree in semantic number (g):

(60) [DP ihrφg [DisP

you(pl)Packg [NP eNφ

g ]]]gang

We have also seen that Spec,DisP may contain quite different types of overt nouns(column two in Table 4 below). If the noun in Spec,DisP and eN in matrix com-plement position agree semantically and the pronoun and eN agree semanticallyand morphologically, then we have evidence for different types of eN: besides nullcountable nouns in the singular and plural (Panagiotidis 2002a, b, 2003a, b), thereare also null pluralia tantum, collectives and singular and plural proper names(column three in Table 4 below; recall that we did not find a relevant overt pluraletantum, the case shown in row five):

Table 4. Inventory of the different types of null predicate nouns

D overt Spec,DisP null N

Sie(sg), wir(sg) singular countable noun plurale tantumihr, Sie(pl) (singular) collective noun plural countableu(pl) plural countable noun (singular) collectivedu, u(sg) plurale tantum singular countablehann feminine proper name singular proper nameþeir two conjoined proper names plural proper name

Note that in the context of our discussion, we have not come across any evidencefor null mass nouns (but see Panagiotidis 2002a:57). This is due to the fact thatmass nouns are not [individualizable]. Furthermore, if semantic agreement be-tween the nouns in Spec,DisP and eN under N is a necessary condition, then therecannot be combinations of a singular countable noun with a plural countablenoun or a (multi-member) collective noun with a (one-member) plurale tan-tum.20 Finally, it is often assumed that covert elements have overt manifestations orcounterparts. With the exception of relevant overt pluralia tantum, pronouns can

. This restricted semantic combinatoriability manifests itself slightly differently in the clause.Similar to nominals, semantically similar predicates can also be combined:

(i) a. They are nice guys and a good band.b. We are a good team but also good individual workers.

However, different countable nouns can also be (marginally) coordinated:

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.33 (1977-2019)


indeed be combined with overt countable and collective nouns as well as propernames, as documented above. In other words, overt and covert nouns are parallelin the relevant sense and I argue that, in the cases where the pronoun seems to beintransitive, it is in fact a transitive determiner co-occurring with eN under N.

. Further issues

In this section, I discuss two remaining issues.

. Semantic agreement is not a sufficient condition

Although a necessary condition, semantic agreement is not sufficient to licensethese pronominal DPs in German. Note in this respect that the examples in(61a–b) add other cases to the well-known type of data in (61c), where only anemotive/affected noun is grammatical:


{Bande/gang/

??Gruppe}group

‘you gang/group’b. Wie

howfühlenfeel

wirwe

{Suffkopp/booze.head/

?(?)Kranker }sick(masc/sg)

unsrefl.

dennprt

heute?today

‘How are we lush / sick one feeling today?’c. du

you(inform/sg){Idiot/idiot/

??Linguist}linguist

‘you idiot/linguist’

For constructions such as in (61c), Rauh (2004) provides a pragmatic accountwithin Gricean conversational maxims. Abstracting away from the details, heraccount can be extended to other cases not discussed by her: while the singularnon-emotive noun Linguist is still awkward in (62a) (although there is no stylisticclash), the plural counterpart is better:

(ii) a. Siethey

sindare

beideboth

Arztdoctor

undand

Professor(??en).professor(s)

‘They are both doctors and professors.’b. Sie

theysindare

beideboth

Ärztedoctors

undand

Professor??(en).professors

This presumably has to do with the fact that a plural pronoun can be combined not only with aplural, but also a bare singular, noun: Sie sind beide {Ärzte / Arzt} ‘They both are doctors.’ Thisis usually not possible in the pronominal DP: *ihr Arzt ‘you(pl) doctor’.

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.34 (2019-2081)

Dorian Roehrs

(62) a. Sieyou(form)

{scIdiot/idiot/

??Linguist}linguist

‘you idiot/linguist’b.(?)Sie

you(form)Linguistenlinguists

‘you linguists’

Some speakers judge (62a) and (62b) comparatively worse but also perceive acontrast between the two (something similar has been reported by my Dutch con-sultants, see discussions surrounding (19) and (20)). Returning to (61a–b), it isunclear at this point if these cases underlie the same conditions as the one in (61c),that is, if they improve under certain conditions, as extensively discussed by Rauh(2004). If it turns out that they do, then Rauh’s account does not cover these newcases: for instance, the pronoun in (61a) has plural reference.

. Ruling out overgeneration due to DisP

The occurrence of DisP is subject to conditions. If Spec,DisP is filled, there cannotbe another adjectival modifier in the matrix DP, either preceding the Specifier (orfollowing it). Note here that this is not an isolated case. In Roehrs (2006), I arguethat “indefinite pronoun + modifier” constructions contain two null nouns whereonly the lower one tolerates modifiers (also Leu 2005). Compare (63a) to (63b).The corresponding analyses are provided in (64):


(*dumme(n))stupid(st/wk)

jungesyoung(st)

Gemüsevegetable(s)

‘you stupid young folks’b. jemand

somebody(masc)(*Süßer)sweet(masc)

Nettesnice(neut)

‘somebody nice’

(64) a. [DP ihr [AgrP dumme(n) [DisP junges Gemüse [NP eN ]]]]b. [DP jemand [AgrP Süßer [NP eN ]] [ Nettes eN ]]

Although these cases are not identical, it is interesting to note that Spec,AgrP host-ing the one adjective seems to be in complementary distribution with the phrasecontaining the other. Having ruled out recursive dis-agreeing nouns by selectionalrestrictions (see Section 5.1.2), I tentatively propose that pronominal determin-

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.35 (2081-2127)


ers can only select one type of phrase containing adjectives.21 The assumption ofselection also seems to be needed to rule out some other cases.

Second, if Spec,DisP is available, this raises the issue of why regular determin-ers cannot have a morphologically dis-agreeing noun in that position. In otherwords, considering the fact that regular determiners seem to be unspecified for se-mantic number (cf. (65a) and (66a)) and bearing in mind that there are differenttypes of null nouns, why are the cases in the (b)- examples ungrammatical (seealso German in (56b))?

(65) a. [DP this [NP set / pig]]b. *[DP this [DisP pants / pigs [NP eN ]]]

(66) a. [DP these [NP pants / pigs ]]b. *[DP these [DisP set / pig [NP eN ]]]

Note first that this does not seem to have anything to do with the licensing andidentification of null nouns, as these DPs are grammatical without a Specifier: [DP

this / these [NP eN ]]. In fact, if Panagiotidis (2002a) is right, then licensing andidentification mechanisms do not exist. Furthermore, we have seen evidence thatthird person pronouns in Icelandic can license such a Specifier. As above, I suggestthat this restriction has to do with selectional properties. In other words, I assumethat regular, non-pronominal determiners cannot select DisP. I will leave openhere if this selection can be reduced to other properties of the extended projectionof the noun, as discussed in Panagiotidis (2002a). Note incidentally that assumingright-adjunction of the dis-agreeing noun would not straightforwardly rule thesecases out as adjunction is typically somewhat less restricted and does not interferewith selectional relations.

Third, with Spec,DisP available, one may wonder why noun phrases wherethe pronoun and the noun fully agree cannot have the overt noun in this position:(*) [NP du [DisP Schwein [NP eN ]]] ‘you pig’. In other words, we need an accountthat forces the projection of a matrix DP only. In a different context, Panagiotidis(2003a:417, 427–428) argues that learners must have some evidence to assumethe presence of null elements. Note now that this type of evidence is not given

. With regard to certain analyses, a similar claim could also be made for some regular deter-miners. For instance, Hendrick (1990:254) proposes that adjectives in English modified by adegree word move to a higher Specifier position. However, Felber & Roehrs (2004) point outthat these constructions do not allow a second adjective:

(i) a. how tall a (*German) manb. [DP [ how tall ]i a [ ti [AgrP German [NP man ]]]]

To the extent that this type of proposal is correct, non-pronominal determiners may also haveselectional restrictions making them, in turn, more similar to pronominal ones.

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.36 (2127-2184)

Dorian Roehrs

here as the pronoun and the noun do agree in all respects. Assuming some typeof economy of representation (e.g., the Minimal Structure Principle, see Boškovic1997:25–29, 37–39), I suggest that the learners must project a less complex DP.

Finally, it is not clear why the noun in complement position cannot be overtwhen Spec,DisP is filled: *ihr dumme Bande Idioten ‘you stupid gang idiots’. Al-though I have no definite answer at this point, note that the account shouldprobably not be too “deep” if Giusti’s (2002) analysis of Italian la mia amica medico’my doctor friend’ is correct.

. Conclusion

Demonstrating that the traditional assumption of determiner concord is both tooweak and too strong, I proposed that pronominal determiners must agree withtheir noun in both morphological and semantic number. I showed that the differ-ent kinds of determiners are, unsurprisingly, not identical: while they are the samewith regard to morphological requirements, they differ in their semantic denota-tions and in their selectional properties. In a sense, regular determiners are lessspecified pronominal determiners:

(67) Determiner regular pronominal (non-mass)Morphological agreement φ φPartial functions supremum one/multi-member supremumSyntactic selection AgrP, NP AgrP, NP, DisP

The specifications in (67) accounted for the different agreement phenomena inand outside of the DP. A welcome result of the discussion was that agreement insemantic number is not brought about by feature checking but by the requirementthat the relevant partial functions be compatible.

More generally, arguing that morphologically dis-agreeing nouns are inSpec,DisP, the analysis provided more evidence for the existence of null nouns,in particular, null collectives, pluralia tantum, and proper names. Furthermore, Iproposed that morphological and semantic numbers are to be dissociated fromone another, a conclusion that Rullmann (2004) seems to have reached for “in-transitive” pronouns on independent grounds. While a number of question haveto remain open here, I hope that the present discussion will inspire further inves-tigation of the syntax and semantics of pronominal DPs.

References

Bhatt, C. (1990). Die syntaktische Struktur der Nominalphrase im Deutschen. Tübingen: Narr.

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.37 (2184-2297)


Boškovic, Ž. (1997). The Syntax of Nonfinite Complementation: An economy approach.Cambridge MA: The MIT Press.

Cardinaletti, A. (1994). On the internal structure of pronominal DPs. The Linguistic Review, 11,195–219.

Carstens, V. (2000). Concord in minimalist theory. Linguistic Inquiry, 31(2), 319–355.Darski, J. (1979). Die Adjektivdeklination im Deutschen. Sprachwissenschaft, 4, 190–205.Delfitto, D. & Schroten, J. (1991). Bare plurals and the number affix in DP. Probus, 3(2), 155–185.Delsing, L.-O. (1993). The Internal Structure of Noun Phrases in the Scandinavian Languages:

A Comparative Study. PhD dissertation, University of Lund.Dikken, M. den (2001). Pluringulars, pronouns and quirky agreement. The Linguistic Review,

18, 19–41.Doetjes, J. & Rooryck, J. (2003). Generalizing over quantitative and qualitative constructions. In

M. Coene & Y. D’hulst (Eds.), From NP to DP. Volume 1: The syntax and semantics of nounphrases (pp. 277–295). Amsterdam: John Benjamins.

Felber, S. & Roehrs, D. (2004). * So weird a baffling construction. Snippets, 8, 7–9.Furuya, K. (2004). Us linguists. MIT Working Papers in Linguistics, 46, 227–242.Giusti, G. (2002). The functional structure of noun phrases. A bare phrase structure approach.

In G. Cinque (Ed.), Functional Structure in DP and IP: The cartography of syntactic structures(pp. 54–90). Oxford: OUP.

Grimshaw, J. (1991). Extended Projections. Ms, Brandeis University.Harley, H. & Ritter, E. (2002). Person and number in pronouns: A feature-geometric analysis.

Language, 78(3), 482–526.Heim, I. & Kratzer, A. (1998). Semantics in Generative Grammar. Malden MA: Blackwell.Hellan, L. (1986). The headedness of NPs in Norwegian. In P. Muysken & H. van Riemsdijk

(Eds.), Features and Projections (pp. 89–122). Dordrecht: Foris.Hendrick, R. (1990). Operator movement within NP. In A. L. Halpern (Ed.), The Proceedings of

the Ninth West Coast Conference on Formal Linguistics (pp. 249–261).Josefsson, G. (1999). On the semantics and syntax of Scandinavian pronouns and object shift.

In H. van Riemsdijk (Ed.), Clitics in the Languages of Europe (pp. 731–757). Berlin: Moutonde Gruyter.

Josefsson, G. (to appear). Semantic and grammatical genders in Swedish – independent butinteracting dimensions. Lingua.

Leu, T. (2005). Something invisible in English. UPenn Working Papers in Linguistics, 11(1), 143–155.

Lyons, C. (1999). Definiteness. Cambridge: CUP.Panagiotidis, P. (2002a). Pronouns, Clitics and Empty Nouns: ‘Pronominality’ and licensing in

syntax. Amsterdam: John Benjamins.Panagiotidis, P. (2002b). Pronominal nouns. In H. J. Simon & H. Wiese (Eds.), Pronouns –

Grammar and Representation (pp. 183–203). Amsterdam: John Benjamins.Panagiotidis, P. (2003a). Empty nouns. Natural Language & Linguistic Theory, 21(2), 381–432.Panagiotidis, P. (2003b). One, empty nouns, and θ-assignment. Linguistic Inquiry, 34(2), 281–

292Postal, P. M. (1966). On the so-called pronouns in English. In F. Dinneen (Ed.), Nineteenth

Monograph on Language and Linguistics (pp. 177–206). Washington DC: GeorgetownUniversity Press.

Quirk, R., Greenbaum, S., Leech, G., & Svartvik, J. (1985). A Comprehensive Grammar of theEnglish Language. New York NY: Longman.

JB[v.20020404] Prn:22/09/2006; 11:57 F: LA9705.tex / p.38 (2297-2351)

Dorian Roehrs

Rauh, G. (2004). Warum Linguist in ich/du Linguist kein Schimpfwort sein muß. LinguistischeBerichte, 197, 77–105.

Ritter, E. (1991). Two functional categories in noun phrases: Evidence from Modern Hebrew.In S. D. Rothstein (Ed.), Perspectives on Phrase Structure: Heads and licensing [Syntax andSemantics 25] (pp. 37–62). San Diego CA: Academic Press.

Roehrs, D. (2005). Pronouns are determiners after all. In M. den Dikken & C. Tortora (Eds.),The Function of Function Words and Functional Categories (pp. 251–285). Amsterdam: JohnBenjamins.

Roehrs, D. (2006). Something post-pre-nominal in German and beyond. Paper presented at the21st Comparative Germanic Syntax Workshop, University of California at Santa Cruz.

Roehrs, D. (in progress). The morpho-syntax of the Germanic noun phrase: Determiners Moveinto the determiner phrase. PhD dissertation, Indiana University.

Rullmann, H. (2004). First and second person pronouns as bound variables. Linguistic Inquiry,35(1), 159–168.

Sauerland, U. & Elbourne, P. (2002). Total reconstruction, PF movement, and derivational order.Linguistic Inquiry, 33(2), 283–319.

Sharvy, R. (1980). A more general theory of definite descriptions. The Philosophical Review,89(4), 607–624.

Svenonius, P. (1993). Selection, adjunction, and concord in the DP. Studia Linguistica, 47(2),198–220.

Watanabe, A. (2006). Functional projections of nominals in Japanese: Syntax of classifiers.Natural Language & Linguistic Theory, 24(1), 241–306.

Wiese, H. & Maling, J. (2005). Beers, kaffi, and schnaps: Different grammatical options forrestaurant talk coercions in three Germanic languages. Journal of Germanic Linguistics,17(1), 1–38.

JB[v.20020404] Prn:24/10/2006; 15:03 F: LA9706.tex / p.1 (47-126)

Toward a syntactic theoryof number neutralisation

The Dutch pronouns je ‘you’ and ze ‘them’*

Gertjan PostmaMeertens Institute, Amsterdam and Nijmegen University

Dutch has two weak pronouns je (‘you’) and ze (‘she/they’ or ‘her/them’), whichshow number neutralisation in function of the syntactic context. Je can be sin-gular or plural only when it is bound, while ze is either singular or plural whenit is subject but not when it is object (in some dialects). This is accounted for byKayne’s syntactic theory of number neutralisation: the plural reading, as well asits syntactic sensitivity, comes about by an abstract distributor, DIST, which mustbe bound. Extending Kayne’s theory, it is proposed that DIST must be bound byan argumental position (je) or by an non-argumental position (ze). The theory isapplied to two diachronic issues: 1. the transition of sg/pl Middle Dutch hem(‘him/them’) to sg Modern Dutch hem (‘him’) and 2. the introduction of theEnglish pronoun ‘they’, which was needed because of loss of neutralisation inOld-English hio (‘she/they’).

. Introduction

Language, and morphology in particular, can be studied in its syntagmatic and inits paradigmatic aspects (Saussure 1910 [1993]). Ever since Baker (1985[1988]),the syntagmatic aspects of morphology has been subject to reduction to syn-tax proper. It raised the more general question whether syntax and morphologyrepresent distinct modules of grammar, as traditional grammarians assume, orthat syntax and (the syntagmatic part of) morphology share their basic formal

* I would like to thank the audience of the TIN-conference, Utrecht 2005 and the audience ofthe Germanic Comparative Syntax Workshop, Tilburg 2005, and two anonymous reviewers fortheir helpful comments on a preliminary version of this paper.

JB[v.20020404] Prn:24/10/2006; 15:03 F: LA9706.tex / p.2 (126-178)

Gertjan Postma

properties, such as hierarchical ordering, binary branching, anti-symmetry, etc. sothat there is no firm basis to hypothesize two distinct modules. The introductionof functional projections has boosted the integration of morphology and syntaxfurther, especially after the work of Pollock (1989) and others, who show a sys-tematic connection between inflectional morphology and syntactic operations.Later developments have put the connection between morphological features andsyntactic operations on a more abstract footing e.g. in Chomsky’s minimalism,who introduces abstract formal features on morphemes as the trigger of syntacticmovements.

The paradigmatic aspect of morphology has resisted integration into syntaxmuch more. But also here, proposals can be reported that argue for a syntactic mo-tivation for paradigmatic structure, e.g. patterns in defectiveness and allomorphy(Postma 1993; Bobaljik 2004; Barbiers 2005). Kayne (2000) contains an intriguingsuggestion in the direction of a syntactic theory of paradigmatic feature neutral-isation. In the present study we will apply these ideas to two weak pronouns inDutch and extend the theory slightly. It will provide us with the first contours of atheory of feature neutralisation.

. Number neutralisation in Dutch je ‘you’

The Dutch weak object pronoun, oblique pronoun, and possessive pronoun je‘you’ is both singular and plural. In traditional terms: je exhibits number neu-tralisation.1 This is illustrated in (1)–(3).

Position number(1) a. Jij

you.sgzagsaw

jeyou

inin

dethe

spiegelmirror

object sg

‘you saw yourself in the mirror’b. Jullie

You.plzagensaw.pl

jeyou

inin

dethe

spiegelmirror

object pl

‘you saw yourselves in the mirror’

(2) a. Jijyou.sg

kuntcan

datthat

naastnext-to

jeyou

neerzettendown put

oblique sg

‘you can put that down next to you’b. Jullie

you.plkunnencan.pl

datthat

naastnext-to

jeyou

neerzettendown put

oblique pl

‘you can put that down next to you’

. In very special registers in Dutch, je displays forms of person neutralization. For a discussion,cf. Bennis (2003).

JB[v.20020404] Prn:24/10/2006; 15:03 F: LA9706.tex / p.3 (178-240)

Syntactic theory of number neutralisation

(3) a. Jijyou.sg

kuntcan

jeyour

boekbook

bijin

hetthe

examenexam

gebruikenuse

poss sg

‘you can use your book at the exam’b. Jullie

You.plkunnencan.pl

jeyour

boekenbooks

bijin

hetthe

examenexam

gebruikenuse

poss pl

‘you can use your books in the exam’

Significantly, this property of je is dependent on the syntactic context: only if je isa bound variable can it be both singular and plural. In pronominal use, only thesingular reading is retained.

(4) a. IkI

zagsaw

jeyou

inin

dethe

spiegelmirror

sg/*pl

b. IkI

kancan

datthat

naastnext-to

jeyou

neerzettendown put

sg/*pl

c. IkI

kancan

jeyour

boekbook

bijin

hetthe

examenexam

gebruikenuse

sg/*pl

To get a plural reading in (4a–c), using the complex plural form jullie ‘youguys’ isthe only option.

(5) a. Ik zag jul-lie in de spiegelb. Ik kan dat naast jul-lie neerzettenc. Ik kan jul-lie boek bij het examen gebruiken

The traditional way to handle neutralisation phenomena is to assume two dis-tinct lexical forms je with the same phonological matrix, an anaphoric pronounje which is underspecified for number ([±plural]), and a pronominal pronoun jewhich is specified as singular ([–plural]. We may call this the lexicalist approach.Obviously, the lexical approach works in a technical sense. But it fails in providinga restrictive theory of natural language. By storing the relevant information in thelexicon, we tacitly assume that it is an arbitrary quirk of Dutch: the facts mighthave been the other way around, with the pronominal use of je being underspec-ified and the anaphoric use singular. Or even quirkier: the anaphoric use couldbe plural only. However, there are various arguments against neutralisation beingsubject to this type of arbitrariness: morphological, comparative, theoretical, anddiachronic. These support the idea that neutralisation is a dynamic process. In fact,the pronoun is inherently singular but it is, under particular circumstances, usedas a plural. I will review these arguments in Section 2. In Section 3, I will presentand discuss a proposal by Kayne (2000) to handle these cases. In Section 4, I willapply this theory to another case of number neutralisation in Dutch: the pronounze. It will lead us to modify and extend the theory. In Section 5, I will discuss andtest the proposal. We will finish with some remaining problems and conclusions.

JB[v.20020404] Prn:24/10/2006; 15:03 F: LA9706.tex / p.4 (240-303)

Gertjan Postma

. Arguments against the lexicalist approach

. A morphological argument

In the previous paragraph I discussed a pronoun with number neutralisation. Itwas suggested that the Dutch pronoun je is singular, while its plural use comesabout by syntactic means. The first argument is morphological. The pronoun jeseems to be part of the singular paradigm. The singular paradigm has both heavyand weak pronominals, as can be seen in (6). The plural paradigm, on the otherhand, has only strong forms, unless the pronoun can also be used as a singular.

(6) singular plural1 mij me 1 ons2 jou je 2 jullie3 hem ‘m

haar ze 3 hen/hun (ze)het ‘t

It seems that the plural use is paradigmatically parasitic on the singular, ratherthan the other way around. This argument is in fact taken from Kayne (2000), whodeveloped the argument while discussing Italian. This brings in the comparativeargument.

. A comparative argument

Kayne (2000) shows a similar state of affairs holds in the Romance languages. Indiscussing the behaviour of se in Italian, French, and other Romance dialects, heargues that se has the shape and behaviour of the singular paradigm. The pronounis inherently singular, but can be used in some contexts as a plural.

(7) singular plural1. me 1. ci2. te 2. vi3. se 3. se

The singularity is not lifted but in contexts where se is seemingly plural, the contextdistributes over the singular.2

We come back to the details of the proposal in Section 3. What is important inthe present argument is that semantically plural morphemes can be morpholog-

. As we will see later on, the plural use does not come about by deletion of the singular feature,but by distributing over it syntactically, very much as inalienable constructions, exemplified in(i), where the singular noun phrase ‘een lange nek is distributed over, in this case by the pluralnoun phrase ‘die giraffen’. Distributing over singular typically involves variable binding.

JB[v.20020404] Prn:24/10/2006; 15:03 F: LA9706.tex / p.5 (303-347)


ically singular. This is not limited to Dutch but is observed in various languages.Plural morphemes can never be used as singulars. The fact that the languages un-der consideration behave the same in taking the singular as the basis can be statedin the lexicon but the lexicon cannot provide a principled account for it.

. Theoretical considerations

The neutralisation effects under discussion occur in bound contexts. This is notan accident. According to the Theory of Reflexivity (Reinhart & Reuland 1993),anaphoric use of a pronoun implies that it is used as a dependent in a chain.According to Reinhart en Reuland, the referent in the head of a chain must bereferential [+R], while pronouns in a dependent position should be referentiallydefective [–R]. They link the [–R] feature to an under-specification for features,particularly number (cf. Reuland 2000).3 This shows a theoretical relation betweennumber neutralisation and the anaphoricity in the paradigm of (1)–(3). A disad-vantage of the framework is the stipulation of the relation. Though theoreticallyplausible, it does not follow from basic principles nor does it provide us with amechanism.

. Diachronic evidence

There is diachronic evidence that the link between anaphoric use and numberneutralisation is not accidental. Middle Dutch did not have a reflexive pronoun. Itused the ordinary pronoun hem ‘him’ in 3rd person reflexive use. So the sentencein (8) encodes a reflexive context.

(8) NuNow

keertturns

hemhimrefl

daerthere

toeto

mijnmy

zinmind

(Middle Dutch)

‘My mind turns itself to it’

Importantly, the Middle Dutch pronoun hem can also be used as a plural (‘them’).Hem displays number neutralisation. When the plural use of hem was lost (from1400 onwards), it could not be used as a reflexive anymore. A gap in the paradigmwas created, which was filled by Eastern forms, such as sick and sich. Using a cor-

(i) diethose

giraffengirafs

hebbenhave

eena

langelong

nekneck

those girafs λx [ x has a long neck]

. Another attested underspecification that causes anaphoricity is [±oblique] in those languagesthat have such a feature, such as Frisian. For a discussion, cf. Hoekstra (1994), Reuland (2000),Postma (2006).

JB[v.20020404] Prn:24/10/2006; 15:03 F: LA9706.tex / p.6 (347-396)

Gertjan Postma

pus of verdicts4 in the province of Drenthe, Postma (2004) shows that the twoprocesses are correlated. Over a period of one century (roughly 1400–1500) thetwo changes are proceed in tandem.

(9) a. Rise of 'zich' in 'zich vermeten / to commit oneselves'

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

1400-27 1427-72 1472-1516

øsick+sichhem+om

b.

When sick/sich had completely taken over hem’s position as a reflexive, hem couldonly be used as a singular. This shows us two things. First, anaphoricity is a neces-sary condition for number neutralisation.5 Secondly, when number neutralisationdisappears, the pronoun retreats to its singular meaning. Apparently, the Old-Germanic singular form hem/ihm had only temporarily been used as a plural. Ifhem in Middle Dutch plural contexts were a true plural, one would expect vari-ation in the outcome when the neutralisation was lifted. But in all dialects ofDutch that underwent the change, the outcome was singular. Traditionally, onewould say that the singular is the unmarked form. Marked forms have morpho-syntactic complexity (Kusters 2003). In the next section we will discuss a proposalthat implements this notion of complexity in a syntactic way.

. The corpus consists of 5000 verdicts of almost every year, 225000 words in total.

. In fact, it appears that number neutralization is a necessary and sufficient condition foranaphoricity for 3rd person pronouns. According to many researchers only 3rd person pronounscan be specified for number (Postal 1972; Polettta 2005 and many others).

JB[v.20020404] Prn:24/10/2006; 15:03 F: LA9706.tex / p.7 (396-444)


. Kayne’s syntactic theory of number neutralisation

It is useful to put the plural use of singular pronouns in the perspective of a pro-posal made in Kayne (2000), who shows that Italian sé is part of the singularparadigm. Despite its inherent singularity, it can be used as a plural, cf. (10b).

(10) a. Il‘the

ragazzoboy

hahas

parlatospoken

diabout

séhimself ’

b. ?I‘the

ragazziboys

hannohave

parlatospoken

diof

séthemselves’

The plural reading is more marked than the singular reading. Kayne suggests thatsé acquires plural readings by an abstract distributor, DIST.

(11) I ragazzi hanno parlato DIST di sé

Potentially, the markedness of the plural construction can be accounted for by thepresence of DIST. By means of DIST, the subject distributes over the object in away floating quantifiers do. For that reason, DIST must have a kind of anaphoricrelation with the subject.6 Kayne mentions two additional arguments that a syn-tactic object is the mediating factor. First, number neutralization is sensitive to thesyntactic configuration. This is explained by the fact that DIST occupies a syn-tactic position, i.e. some configurations allow for insertion of a distributor whileother configurations do not. Prepositional constructions seem to provide a slot intheir specifiers. Secondly, DIST has syntactic properties, such as the requirementto be locally bound by a plural antecedent. In this way, the special behaviour of theplural reading with respect to long-distance anaphora can be accounted for.

(12) a. ?Il‘the

ragazzoboy

mihas

haconvinced

convintome

ato

parlarespeak

diabout

séhimself ’

b. *I‘the

ragazziboys

mihave

hannoconvinced

convintome

ato

parlarespeak

diabout

séthemselves’

The ungrammaticality of (12b) follows from the fact that there is no possible slotfor DIST. The various possibilities drawn in (13) lead to violation of the localityconditions.

(13) I ragazzii mik hanno (*DISTi) convinto a PROk parlare (*DISTi) di séi

The first occurrence of DIST violates locality with respect to SE, the second oc-currence violates locality between DIST and its antecedent. Kayne’s theory can be

. DIST can be compared with binominal each (Beghelli & Stowell 1997; Postma 2000), recip-rocal each, which all involve two theta-positions and have A-bar and A anaphoric dependencies,cf. Aoun (1985).

JB[v.20020404] Prn:24/10/2006; 15:03 F: LA9706.tex / p.8 (444-513)

Gertjan Postma

considered as a syntagmatic account of morphological neutralisation, instead ofthe traditional paradigmatic approach.

This theory is straightforwardly applicable to the Dutch data listed above. Thesingular reading of je is well-formed, whether it is anaphoric or not. The pluralreading is only grammatical when je is anaphoric. By assumption, the plural read-ing is established by a distributor. The anaphoricity of the plural reading derivesfrom the fact that DIST must be bound.7

(14) a. JullieYou.pl

zagen DISTsaw.pl

jeyou

inin

dethe

spiegelmirror

‘you saw yourselves in the mirror’b. *Jan

Johnzag DISTsaw

jeyou.pl

inin

dethe

spiegelmirror

We conclude that the syntactic approach to number neutralisation has several ad-vantages. It captures the fact that plural forms are more complex, it makes thecorrect link between binding effects in the plural reading, in Italian sé and in Dutchje, and it clarifies the diachronic issue of Middle Dutch plural hem. A potential ob-jection is that the precise conditions on DIST are distinct in the two languages.According to Kayne (2000), DIST in Italian can be inserted before prepositions,probably in the specifier of PP, while we must allow Dutch DIST to occur morefreely as an adjunct. However, such differences are to be expected, as DIST is asyntactic object and the languages under consideration are syntactically different.In the optimal case, such differences will be derivable from independently estab-lished properties of the languages, for instance, they could be linked to differencesin the distribution of other distributors such as EACH. Obviously this is a researchprogram that exceeds the scope of this paper. As a first beginning, though, wewill develop in the next section one dimension of this complicated field, wherewe describe the behaviour of Dutch ze, a pronoun that is etymologically related toItalian SE.

. Number neutralisation in Dutch 3rd person pronouns

. Middle Dutch ‘hem’

Before we can understand the position of ze in Dutch, it is enlightening to re-turn to the discussion of 3.4 on the Middle Dutch hem. As we have observed in3.4, Middle Dutch hem displayed number neutralisation, and could therefore be

. In fact, the Dutch data are even ‘nicer’ than those from Italian, since singular sé in Italian isanaphoric as well, which must be stated independently.

JB[v.20020404] Prn:24/10/2006; 15:03 F: LA9706.tex / p.9 (513-557)


used anaphorically. However, hem could be used as a plural in non-reflexive con-texts as well. If we assume that this reading comes about through mediation ofDIST, which is anaphoric, the question arises what the antecedent of DIST mightbe in the pronominal reading of hem. It is, of course, unattractive to assume anon-anaphoric DIST, since that would destroy the correlation between anaphoraand number neutralisation. A plausible solution is that DIST is always anaphoricbut that it can select an antecedent in an argument position as well as in a non-argument position, e.g. referential features in Comp. Extending Rizzi’s criterialapproach to languages that have a grammaticalized topic structure, such as thefull V2 languages, we may assume that there are topic features in CP that induceverb placement by some Topic criterion (Zwart 1993; Rizzi 2004). Let us assumethat such topic features are in C◦. So, hem is anaphoric to the features in C, but isnot reflexive, as there is no argumental co-indexation. We will denote this A-bardistributor that has an A-bar dependency as DIST’ (DIST-bar).8 We arrive at astructure in (15) for non-reflexive use of hem.

(15)

In this case, hem behaves as a kind of topic or discourse pronoun that picks up itsreferent in its first anteceding CP. We will postpone a discussion of the predictionsand the tests to Section 4.3. Let us now turn to the fact that Middle Dutch lost theoption of hem being in the scope of DIST or DIST’. We then predict that there willbe three empty slots in the paradigm: 3rd sg and pl reflexive, and 3rd plural in non-reflexive context. Hem recourses to its singular non-reflexive reading. As we haveseen in Section 2.4 the empty slots in the reflexive paradigm were filled throughborrowing of sick/sich. The filling of the empty slot in the plural paradigm wasmore problematic. In the written standard language, the object pronouns hen/hun‘them’ emerged but they are uncomfortable until the present-day and are virtuallyabsent from the spoken language (Uit den Boogaard 1975). Most of the time, adiscourse pronoun9 is used suppletively: die ‘them’, which may remain in situ withinanimate referents, but fronts with animates.10

(16) a. IkI

hebhave

diethem

gezienseen

(books / *?people)

b. Diethem

hebhave

ikI

gezienseen

(books / people

. Why it is the case that 2nd person je takes DIST while 3rd person pronouns take DIST’ orDIST and ze only takes DIST’ remains a stipulation that abides explanation.

. For properties of discourse pronouns, cf. Postma (1984), Reinhart (1983).

. The in situ use with animates is possible with a pejorative reading.

JB[v.20020404] Prn:24/10/2006; 15:03 F: LA9706.tex / p.10 (557-611)

Gertjan Postma

This deictic pronoun die has not yet developed into a true pronoun (this is whathappened with English them). Deictic pronouns move to COMP before or afterspell-out, but pronouns obtain their interpretation in situ. The difference betweenpersonal pronouns and discourse pronouns can be traced by various tests, suchas coordination, reflexivity, disjoint reference, oblique context test. For a moreextended discussion of the various tests, I refer to Postma (2004).

(17) a. IkI

zagsaw

jouyou

enand

hen/*diethem

coordination test

b. ZijThey

zagensaw

(?)hunzelf/*diezelfthemselves

reflexivisation test

c. ZijThey

zeidensaid

datthat

ikI

hen/*die moest helpenmust help them

disjoint reference

d. ZijThey

zorgdentook

voorcare

hen/*die11

for them

oblique test

In oblique contexts, discourse pronouns have a special suppletive form, the so-called pronominal adverbs,12 R-pronouns in the generative literature (Van Riems-dijk 1978). All these restrictions immediately follow from the assumption that dis-course pronouns move to CP at some point of the derivation. Under a movementanalysis, the ungrammaticalities in (17) are explained: the coordination constraintis violated in (17a), the structure in (17b) is a case of strong cross-over, as is (17c).The block on prepositional contexts (17d) follows from the absence of prepositionstranding in Dutch.

. Number neutralisation in Modern Dutch ze

.. Ze as an object pronoun (them)In this section we will discuss number neutralisation with the pronoun ze‘her/them’, as in (18).

(18) IkI

zagsaw

zeher/them

There are two caveats to be made here. Although the singular use of ze is a fea-ture of standard Dutch according to all descriptive grammars, the actual use isvery much limited to the South. In (19a), I give an impression of the area of theactual neutralisation. It is an impression based on data found in the literature(De Schutter 1987) and from informants. A second caveat is that not the entire

. With very strong deictic focus the construction becomes fine. Perhaps a deictic focus featurepercolates to the PP and moves the PP at LF.

. The so-called “voornaamwoordelijke bijwoorden” (pronominal adverbs).

JB[v.20020404] Prn:24/10/2006; 15:03 F: LA9706.tex / p.11 (611-653)


Dutch area allows for the prepositional use of ze. In (19b), the dots indicate thearea where prepositional use is attested. The map is created using data extractedfrom the RND,13 which were entered into the map-drawing program of the SANDdatabase.14

(19) a.

b.

If we inspect the two maps, we conclude that the dialects that have neutralisation(the South), do not allow for prepositional use. The reverse is not true, as there aredialects in the Northeast that do not allow for either. In other words, there is animplicational relation. It is rendered in (20) for further reference.

(20) A dialect δ has neutralisation in ze → δ does not allow for [P ze]

A theory of neutralisation should provide an answer for why this is the case. Itmust be noted that a block on the prepositional use of ze is unique in the Dutch

. Series of Dutch Dialect Atlases, cf. RND in the references.

. Only those dialects are drawn whose geographical number in the RND coincided witha number of the SAND atlas (cf. references). The statistics is in fact much better than themap suggests.

JB[v.20020404] Prn:24/10/2006; 15:03 F: LA9706.tex / p.12 (653-698)

Gertjan Postma

language. No other pronoun in Dutch displays this curious behaviour.15 Instead ofintroducing an entirely new opposition between objective Case and oblique Case,it is attractive to pursue a syntactic line.

As we have seen in Section 4.1, the slot of plural hem became available whenhem retreated to its singular base. The gap was partly filled by discourse pronouns,but another new form was object ze. In most of the grammars, ze is treated as apersonal pronoun, but evidence for this is not very strong. Let us apply our tests ofSection 4.1. First, it cannot be used in co-ordinations (21), it cannot enter reflex-ivity (cf. (22)) not even in dialects that allow pronouns to do so, it displays disjointreference effects (see Section 4.3), and finally it does not occur in prepositionalcontexts (23).

(21) a. Hijhe

zagsaw

eena

fotopicture

vanof

jouyou

enand

mezelfmyself

b. Hijhe

zagsaw

eena

fotopicture

vanof

jouyou

enand

*zezelfher-self

(22) a. ?ZijThey

zagensaw

hunzelfthemself

b. *ZijThey

zagensaw

zezelfthem-self

(23) a. Zij‘They

zorgdentook

voorcare

hen/*ze16

for them’

I conclude that ze does not behave as a personal pronoun. Rather, it has propertiesthat remind us of discourse pronouns, such as die, cf. (17a–d). On the other hand,we have the result that ze is part of the paradigm of me/je/ze. Apparently, thereis a specific additional property of ze that disqualifies it as a personal pronoun.In view of our discussion on the nature of feature neutralisation, an explanationemerges. As we saw, DIST’ is anaphoric to CP. DIST’[ze] will therefore have an ab-

. The neuter pronoun het ‘it’ cannot be used after prepositions either. The pronoun het sharesthis property with unstressed neuter pronouns in general, such as alles ‘everything’, niets ‘noth-ing’, dat ‘that’, etc. It might be that the theory developed in this paper is applicable to neuters.It is far from clear whether neuter pronouns are singular or plural. They pass various tests onsingularity and plurality.

(i) DitThis

enand

datthat

moet/*moetenmust.sg

verkochtsold

wordenbe

(agreement test – singularity)

(ii) DatThat

versterktreinforces

elkaareach other

(reciprocity test – plurality)

Further research is needed.

. This sentence is grammatical in the Dutch area without neutralisation, cf. (19b).

JB[v.20020404] Prn:24/10/2006; 15:03 F: LA9706.tex / p.13 (698-752)


stract relation to CP, which we may think of as a LF-movement relation. It is DIST’that disqualifies ze as a personal pronoun. The movement analysis predicts thegrammaticality judgements in (21)–(23), parallel to the data in (17). Furthermore,the assumption of a syntactically active DIST’ immediately explains the implica-tional relation of (20): if there is neutralisation, ze moves. If it moves, it cannot beprepositional as Dutch does not have preposition stranding.

.. ze as a subject pronounIn the previous paragraph we have seen that there is number neutralisation in theDutch object pronoun ze. This comes about by an anaphoric distributor (DIST’),which is anaphoric to topic features in CP. The distributor turns ze (a singularpersonal pronoun) into a discourse pronoun. This number neutralisation is onlypresent in the southern dialects. The situation as to subject ze (‘she/they’) is muchsimpler. All dialects display the neutralisation. Moreover, the weak pronoun hasits emphatic counterpart zij (‘she/they’) with identical properties.

(24) a. Ze/zijShe

gaatgoes.sg

naarto

AmsterdamAmsterdam

b. Ze/zijThey

gaango.pl

naarto

AmsterdamAmsterdam

We could of course copy the reasoning from the previous paragraph, but thiswould leave unexplained why there is no language variation with respect to neu-tralisation in the subject pronouns. This lack of variation extends to the Germandialects (sie = ‘she/they’). Pronouns that are not etymologically related, such asFrisian hja (‘she/they’), equally display number neutralisation. This absence ofvariation, language internally (all pronouns comply) and cross-linguistically (allcontinental Germanic variants comply) asks for an explanation. Does it correlatewith another property? We would like to suggest that it correlates with a specifictype asymmetric V2: Dutch, Frisian and German at the one hand, and Swedish,Yiddish and Icelandic at the other hand. The definition needs some care, as variousdemarcation line s are possible. As it has been argued in the literature that Germanand Frisian have embedded V2 under bridge verbs (Reis 1997; De Haan & Weer-man 1986), a property that has similarity with Mainland Scandinavian (Vikner1994), which display inversion with embedded topicalisation under bridge verbsonly. However, the similarities are superficial. They disappear if we confine our-selves to integrated embedded clauses, i.e. to subordinated clauses with an overtcomplementizer that can have dependencies with quantifiers in the main clause. Asargued in De Haan (2001), Frisian embedded clauses with V2 resist such relations.Embedded V2 structures are in fact coordinated structures with dat as a coordina-tor. Its distinct status is confirmed by the fact that complementizer argeement inembedded V2 is blocked.

JB[v.20020404] Prn:24/10/2006; 15:03 F: LA9706.tex / p.14 (752-804)

Gertjan Postma

(25) a. Doyou

sjochstlook.2sg

saso

minbad

datthat

doyou

soestshould.2sg

marPRT

opto

bêdbed

b. *Doyou

sjochstlook.2sg

saso

minbad

datstthat.2sg

doyou

soestshould.2sg

marPRT

opto

bêdbed

c. Doyou

sjochstlook.2sg

saso

minbad

datstthat.2sg

doyou

marPRT

opto bed

soestshould.2sg

‘you look so bad that you better go to bed’

As we want to investigate the anaphoric relation between subject pronouns andfeatures in C, this restriction to true embedding seems natural. In (26), I renderthe correlation in a table.

(26) Language symmetric V217 number neutralisation(m.sg/f.sg//m+f.pl)

Dutch no yes (hij/zij//zij)German no yes (er/sie//sie)Frisian no yes (hy/hja//hja)Surselvan no yes (ej//ej)

Swedish yes no (han/hon//de)Yiddish yes no (er/si//sii) (cf. Katz 1987)English yes no (he/she//they)Icelandic yes no (hann/hun//their/thaer)

Even the Swiss Rhaeto-Romance dialects (e.g. Surselvan), which have a Romancebase, display number neutralisation to the extent they do not have gender (ej‘ambient it’, ‘arbitrary they’, Haiman 1990). This tells us two things: number neu-tralisation seems to be blocked by overt gender marking, and secondly, numberneutralisation in subjects does not seem a lexical property of any specific rootbut seems to be linked to a common parameter setting of continental Germanic.Let us see if the Theory of Neutralisation gives us a clue what this parametersetting might be.

As we see from the data in (24), the situation with neutralised subjects is dif-ferent from the situation with objects: neutralisation in subjects does not give riseto ambiguity because of the overt singular-plural opposition in the verbal agree-ment. It seems that verbal agreement somehow facilitates number neutralisationin Dutch, Frisian and German, but not for instance in English or Icelandic. It doesso because of some property of common continental Germanic. Within a theorywith an anaphoric DIST’, the suggestion is imminent that agreement can functionas an antecedent to DIST’ in continental Germanic. Put differently: in languages

. Evaluation of main clauses versus integrated embedded clauses with overt complementizer.

JB[v.20020404] Prn:24/10/2006; 15:03 F: LA9706.tex / p.15 (804-856)


in which C acts as a probe (Van Koppen 2005), the agreement features in C canfunction as an antecedent to DIST’. The structure is given in (27).

(27) [CP spec Cagr, α◦ [IP [DIST’α [ze]] tV. . .

In main clauses the inflected verb is always in C in V2 languages. So, the solu-tion in (27) is straightforward for main clauses.18 However, number neutralisationalso occurs in embedded clauses, cf. (28). In embedded clauses the finite verb isclause final.

(28) dat ze naar huis gaat/gaan

1. that she to home goes2. that they to home go

There is ample evidence, however, that C is a probe both for main and embeddedclauses (Van Koppen 2005), which shows up as the well-known and well-studiedphenomenon of complementizer agreement (Zwart 1993; Hoekstra 1997), a prop-erty of many dialects of the asymmetric V2 languages. An example is given in (29),which is Dedemsvaarts, a dialect from the North-eastern part of the Netherlands.

(29) a. dattethat.plC

wiwe

(. . . )(. . . )

speultplay.plI

(Low-Saxon, Van Haeringen 1962)

b. DanThen

speuleplay.plC

wiwe

In (29) we have a Low-Saxon dialect that has two verbal present tense paradigms,a clause final paradigm in -t and a V2 paradigm in -e. Significantly, the inflec-tional plural morpheme in inversion contexts, -e, also shows up as inflection onthe complementizer. This shows that inversion contexts are CP contexts. For em-bedded clauses we therefore assume the same structure as in (26), although theagreement features in C remain in some dialects without spell-out. The true cor-relation of number neutralisation in the subject pronoun is therefore with abstractphi-features in C rather than overt. C with abstract phi-features act as a probe forphi-features in I and causes that these languages display overt I-to-C, which derivesthe correlation in (25).

. Discourse properties of ‘ze’

In the previous sections, we have shown that the 3rd person pronoun ze behaveslike a discourse pronoun with respect to various tests. We attributed this behaviourto the abstract distributor DIST’, which is CP oriented. We have left one essential

. DIST[je] cannot take the AGR features in C as an antecedent, since DIST has only A-dependencies in the case of je.

JB[v.20020404] Prn:24/10/2006; 15:03 F: LA9706.tex / p.16 (856-920)

Gertjan Postma

test for later discussion: disjoint reference. It is now the moment to verify to whatextent ze indeed complies with principle C. As ze is a pronoun, we can only studydisjoint reference by means of bi-clausal structures. If we take standard bi-clausalstructures, however, our expectation is not borne out. Consider (30).

(30) De jongensi dachten dat ik *die/zei bedroog.‘The boys thought.pl that I them cheated’

While a true discourse pronoun, such as die ‘that’, indeed displays disjoint refer-ence with a c-commanding antecedent, ze can be bound. It must be kept in mind,though, that a true discourse pronoun moves to the highest CP, and causes a cross-over effect with the subject of the main clause (de jongens). DIST’ on the otherhand, has a relation with the first c-commanding CP. The structures are in (31).

(31) a. (cross-over)

b. (no cross-over)

This explains the asymmetry between die and ze with respect to disjoint refer-ence. If this line of reasoning is on the right track, decisive data will be bi-clausalstructures without intermediate CP. Typical structures are given in (32).19 Theseare AcI constructions, where the embedded subject receives Accusative case fromthe matrix verb. This is evidence that no embedded CP is present. Let us lookat the data:

(32) a. Mijnmy

kindereni

childrenvindenbelieve

IP[mijme

*diei/*zei/hun/heni

themtetoo

weiniglittle poket

zakgeldmoney

geven]give.inf

‘My children think I give them too little pocket money’b. De

thejongensi

boyshoordenheared.pl

IP[dethe

directeurdirector

*diei/*zei/hun/heni

them

bespotten]mock.inf‘The boys heard the director mock them’

We see that in such structures the object pronoun ze, in fact [DIST’ [ze]], behaveson a par with die rather than with true pronouns without number neutralisation,such as hen/hun. This is clear evidence that ze has a special relation with CP. Thestructure of (32b) is in (33).

(33)

. We have tested these judgements with a group of 8 informants.

JB[v.20020404] Prn:24/10/2006; 15:03 F: LA9706.tex / p.17 (920-982)


Both die and ze cause a cross-over effect by moving past the co-indexed sub-ject, while hun/hen do not move. This is evidence that a pronoun with numberneutralisation (ze) behaves syntactically distinct from pronouns without numberneutralisation, such as hun/hen. A minimal pair is given in (34), where we havetwo sentences with the same meaning, one finite, the other infinitival.

(34) a. Dethe

jongensi

boysvroegenasked.pl

mijme

ofif

ikI

zei/heni/huni

themwildewanted.sg

helpenhelp.inf

b. DeThe

jongensi

boysvroegenasked

mijme

*diei/*?zei/heni/huni

themteto

helpenhelp

‘The boys asked me (if I wanted) to help them’

Once again, the disjoint reference effect shows up with ze, but not with hen/hun.Moreover, ze does display the effect in function of the absence/presence of an em-bedded complementizer. We can take this as independent evidence that numbergeneralisation is established in the syntax.

. An application: The rise of English ‘they’

In Section 4.2.2, we established a correlation between number neutralisation insubject pronouns and asymmetric V2. The relation between an alleged mor-phological property and a well-established syntactic property (V2) can be ex-plained using Kayne’s abstract distributor, which is A-anaphoric (DIST) or A-baranaphoric (DIST’). In this section, we apply it to older stages. We will see thatthe correlation between the two grammatical properties sheds new light on animportant language change in the history of English.

In (35) I give the correlation from Section 4.2.2 for older language stages.

(35) Language symmetric V2 number neutralisation

Old-Saxon no yes (he/siu//siu)Old-Frisian no yes (hy/hia//hia)Old-HGerman no yes (er/siu//siu)Old Kentish no? yes? (he/hio//hio)

Anglo-Saxon no? yes? (he/{heo, hio}//{hie, hio})Old-Norse yes no (hann/hon//their/thaer)Gothic yes no (is/si//eis/ijos)

The correlation parallels the one in (25). Only Anglo-Saxon is a bit problematic,since the number of distinct 3rd person forms is huge. The introductory grammarsof Old-English do not claim number neutralisation, but if one considers the textsand consult more advanced grammars one gets doubts. The CHEAL (1907) writes:

JB[v.20020404] Prn:24/10/2006; 15:03 F: LA9706.tex / p.18 (982-1038)

Gertjan Postma

The forms of the Old English pronouns of the third person, in all dialects, were,in several instances, curiously near to being alike in pronunciation. The mascu-line nominative h[char] was not very different from the feminine nominative andaccusative h[char]o (also h[char]e, h[char]), and this closely resembled the pluralnominative and accusative h[char]e or h[char]. (With [char]=i and its variants)

This is illustrated in (36).

(36) Ond eghwylc thara aerfewearda the aefter him to thaem londe foe, thonneageofen hio tha ilcan elmessan to Cristes-cirican for Aelfredes sawle‘And whoever of the heirs (that) succeeds to the land after him, (then) theygive the same charity to the Christian church for Alfred’s soul’.

(Kentish Charters 23c42, AD 871-886)

The situation is even more extreme in Old-Kentish. Sweet (1908: lxvi) notes thatin Kentish texts hio ‘she’ was used as a singular fem, as a plural ‘they’ and evenin object position. In this respect, Kentish displays a pronominal subject systemthat is close to German, where sie (‘she, they, her, them’) is used as a singularand a plural, both in subject and in object position. Not accidentally, this Ken-tish dialect is most conservative with respect to the V2 constraint (Kroch & Taylor2000).20 This shows us three things. Once again, it shows us that neutralisation inthe pronominal system is not tied to particular roots: both German sie and Kentishhio display a similar system. Secondly, it suggests that Kentish represents the orig-inal situation: number neutralisation plus asymmetric V2. Third, it suggests thatthere might have been an (internally triggered?) change in the pronominal sys-tem in the 9th c. long before the invasions by the Normans. The specialisation ofsingular and plurals 3rd person had implications for the asymmetric V2 property.

However, whether this change was language internal or not, it is clear that thechange was boosted by the introduction of the Nordic form they. The new systemhe/she//they (from 1300 onward) established a clear loss of neutralisation with theconsequence that the language started to be in the category of Icelandic, MainlandScandinavian, etc. From this perspective the borrowing of they and the changes ofasymmetric V2, properties that occurred under language contact with the Nordicinvaders, are not completely independent.

. Conclusions

We have shown that number neutralisation is not a lexical paradigmatic prop-erty, but is established by syntactic means. There is an abstract distributor, DIST,

. “Except in Kentish, a particularly archaic southern dialect, we find by the mid-fourteenthcentury that the V2 constraint is clearly being lost” (Kroch & Taylor 2000).

JB[v.20020404] Prn:24/10/2006; 15:03 F: LA9706.tex / p.19 (1038-1134)


with syntactic properties that are responsible for number neutralisation. DIST isanaphoric, and can take argumental antecedents (Romance se, Dutch object je)and non-argumental antecedents (features in CP). A non-argumental antecedentis active when a pronoun resides in subject position. In that position, C◦ can bindDIST provided that C◦ acts as a probe for agreement features, which typically isthe case in asymmetric V2 languages.

References

CHEAL – Ward and Trent, et al. (Eds.). The Cambridge History of English and American Litera-ture. New York: G.P. Putnam’s Sons, 1907–1921.

MAND – Morfologische Atlas van de Nederlandse Dialecten – Goeman, Taeldeman, Van Reenendialect transscripties, 1980–1995. Meertensinstituut. Amsterdam.

RND – Blancquaert e.a. (1931). Reeks Nederlandse Dialectatlassen. Text en Kaarten.SAND – Syntactische Atlas van de Nederlandse Dialecten – S. Barbiers, et al. Meertensinstituut

2005.Aoun, Joseph (1985). A Grammar of Anaphora. Cambridge MA: The MIT Press.Baker, J. (1985[1988]). Incorporation. Chicago IL: Chicago University Press.Barbiers, S. (2005). Variation in the morphosyntax of ONE. Journal of Comparative Germanic

Linguistics, 8(3), 159–183.Beghelli, F. & Stowell, T. (1997). Distributivity and negation: The syntax of each and every. In A.

Szabolcsi (Ed.), Ways of Taking Scope. Dordrecht: Kluwer.Bennis, H. (2003). Pronoms de la deuxième personne en neérlandais; Contrastes en forme et en

interprétation. Journal of the British Institute in Paris, 33/34, 10–21.Bobaljik, J. D. (2004). Paradigms (optimal and otherwise): A case for scepticism. Ms. UConn.Haan, G. de & Weerman, F. (1986). Finiteness and verb fronting in Frisian. In H. Haider & M.

Prinzhorn (Eds.), Verb Second Phenomena in Germanic Languages (pp. 77–110). Dordrecht:Foris.

Haan, G. de (2001). More is going on upstairs than downstairs: Embedded root phenomena inWest Frisian. Journal of Comparative Germanic Syntax, 4, 3–38.

Haeringen, C. B. van (1962). Gramarie. Assen: Van Gorcum.Haiman, J. (1990). Rhaeto-Romance. In M. Harris & N. Vincent (Eds.), The Romance Languages.

London: Routledge.Hoekstra, E. (1997). Vervoegde voegwoorden in de Nederlandse dialecten: Een aantal

generalisaties. In E. Hoekstra & C. Smits (Eds.), Vervoegde voegwoorden (pp. 6–30).Amsterdam: P.J. Meertensinstituut.

Hoekstra, J. (1994). Pronouns and case: On the distribution of Frisian harren and se ‘them’.Leuvense Bijdragen, VL, 83, 47–65.

Katz, D. (1987). Grammar of the Yiddish Language. London: Duckworth.Kayne, R. S. (2000). Person morphemes and reflexives in Italian, French and related languages.

In Richard S. Kayne, Parameters and Universals [Oxford studies in comparative syntax] (pp.131–162). Oxford: OUP.

Koppen, M. van (2005). One Probe – Two Goals: Aspects of agreement in Dutch dialects. PhDdissertation, Leiden University.

JB[v.20020404] Prn:24/10/2006; 15:03 F: LA9706.tex / p.20 (1134-1241)

Gertjan Postma

Kroch, A. & Taylor, A. (2000). Verb-object order in early Middle English. In S. Pintzuk et al.(Eds.), Diachronic Syntax. Models and mechanisms (pp. 132–163). Oxford: OUP.

Kusters, W. (2003). Linguistic Complexity – The Influence of Social Change on Verbal Inflection.PhD dissertation, University of Leiden.

Platzack, C. (1989). The role of AGR and finiteness in Germanic VO languages. Working Papersin Scandinavian Syntax, 43, 51–76.

Pollock, J-Y. (1989). Verb movement, universal grammar and the structure of IP. LinguisticInquiry, 20, 265–424.

Poletto, C. & Benincà, P. (2005). The third dimension of person features. Talk presented at theWorkshop Variation in Inflection, December 19–20, 2005, University of Amsterdam.

Postal, P. (1969). On the so-called ’pronouns’ in English. In D. Reibel & S. Shane (Eds.), ModernStudies in English (pp. 201–224). Englewood Cliffs NJ: Prentice-Hall.

Postma, G. J. (1984). The Dutch pronoun diens, distribution and reference properties. Linguisticin the Netherlands, 1984, 147–157.

Postma, G. J. (1993). The syntax of the morphological defectivity of BE. HIL Manuscripts, 3,31–67.

Postma, G. J. (2000). Distributive universal quantification and aspect in Brazilian Portuguese. InJ. Costa (Ed.), Portuguese Syntax – New Comparative Studies (pp. 241–265). Oxford: OUP.

Postma, G. J. (2004). Structurele tendensen in de opkomst van het reflexief pronomen ‘zich’ inhet 15de-eeuwse Drenthe en de theorie van reflexiviteit. Nederlandse Taalkunde, 9, 144–168.

Postma, G. J. (2006). Language contact and linguistic complexity – the rise of the reflexivepronoun ‘zich’ in a 15th century Netherlands’ border dialect. In D. Jonas & S. Andersen(Eds.) Proceedings of DIGS-8. Oxford: OUP.

Reinhart, T. & Reuland, E. (1993). Reflexivity. Linguistic Inquiry, 24, 657–720.Reinhart, T. (1983). Coreference and bound anaphora. Linguistics and Philosophy, 6, 47–88.Reis, M. (1997). Zum syntaktischen Status unselbstständiger Verbzweit-Sätze. In C. Dürscheid,

K. Ramers, & M. Schwarz (Eds.), Syntax im Fokus (pp. 121–144).Reuland, E. (2000). Primitives of binding. Linguistic Inquiry, 32, 439–492.Riemsdijk, H. van (1978). A Case Study in Markedness: The binding nature of prepositional

phrases. Dordrecht: Foris.Rizzi, L. (2004). On the Form of Chains: Criterial positions and ECP effects. Ms. University of

Sienna.Saussure, F. de (1910[1993]). Course in General Linguistics (ed. by C. Bally & A. Sechehaye with

the collaboration of Albert Riedlinger; translated from the French and annotated by R.Harris). Oxford: Pergamon Press.

Schutter, G. de (1987). Morfologische categorieën van Persoonspronomina in Nederlandsedialecten. In Morfologie Themanummer Taal & Tongval, 43–53.

Sweet, H. (1908). An Anglo-Saxon Reader. Oxford: Clarendon Press.Vikner, S. (1994). Finite verb movement in Scandinavian embedded clauses. In N. Hornstein &

D. Lightfoot (Eds.), Verb Movement (pp. 117–147). Cambridge: CUP.Uit den Boogaart, P. C. (1975). Woordfrequenties in Geschreven en Gesproken Nederlands.

Utrecht: Oosthoek, Scheltema & Holkema.Zwart, J.-W. (1993). Dutch Syntax. PhD dissertation, University of Groningen.

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.1 (48-132)

Long relativization in Zurich Germanas resumptive prolepsis

Martin SalzmannLeiden University, The Netherlands

Long relativization in Zurich German is a construction with paradoxical proper-ties. Some properties (reconstruction effects) suggest movement out of the com-plement clause, whereas others show that the complement clause is a barrier.This paradox is resolved by assuming a tough-movement style analysis: Operatormovement in the complement clause derives a predicate and licenses an extraargument, the proleptic object. This in turn is A’-moved in the matrix clause anddeleted under identity with the external head. The predication analysis proposedhere makes an alternative strategy for reconstruction available and accounts formany of the puzzling properties of the construction.

. Introduction: Relativization in Zurich German1

Restrictive relatives in Zurich German (ZG) are interesting for three reasons. First,ZG (and Southern Alemannic dialects more generally) stand out among relativesin German(ic) in that they use resumptive pronouns (ignoring Yiddish). Second,the distribution of resumptive pronouns in ZG yields a pattern that is crosslinguis-tically interesting (although not unique). Third, resumptive pronouns in ZG onlyoccur in relativization, but not in wh-movement or topicalization.2

In this section, I will first discuss general properties of restrictive relatives inZG and then the distribution of resumptive pronouns.

. The research reported on here is presented in much more detail in Chapter 4 of Salzmann(2006b).

. They also occur in comparatives, but to a much more limited degree, cf. Salzmann (2006a,2006b).

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.2 (132-199)

Martin Salzmann

. General form of restrictive relatives in Zurich German

ZG relatives are postnominal and head external, which is little surprising for aGermanic language. More interestingly, there are no relative pronouns (except forcertain adverbial relations like the reason why and the manner how), but insteadan invariant complementizer wo is used to introduce relative clauses.3 The useof an invariant complementizer is an inconspicuous property of many varieties ofGerman, bare wo is used in all Alemannic dialects and also found in many Bavarianand Upper Franconian dialects, cf. Fleischer (2003:227). In certain grammaticalrelations, a resumptive pronoun appears instead of a gap.4 Those resumptives areformally identical to weak pronouns and tend to occur relatively high in the clause,arguably in the Wackernagel position.5

. Distribution of resumptive pronouns: Local relativization

In local relativization, resumptive pronouns are only found in oblique relations,i.e. with datives, possessors and complements of prepositions, but crucially notwith subjects and direct objects (Weber 1964; van Riemsdijk 1989):6, 7

. The relative complementizer appears as won before vowels.

. The term “resumptive pronoun” is used inconsistently in the literature. I use it as a purelydescriptive term for elements that look like personal pronouns and are A’-bound, i.e. have anantecedent in an A’-position (whether resumptive pronouns are also found in A-chains is un-clear). Whether the pronoun participates in a movement or a binding dependency and in case ofmovement is just the spell-out of a trace or rather a true pronoun are issues of implementationI will discuss below.

. The resumptive element is not always a weak pronoun. In local relativization, strong pro-nouns and demonstratives are possible as well; in long relativization, epithets are found; adiscussion of these cases is beyond the scope of this paper, but cf. Salzmann (2006b) for detailedanalysis.

. Things are more complex when it comes to datives. Resumptives are systematically foundfor animate indirect objects (even though some speakers have started dropping the resumptivein recent years). With inanimates and unaccusative verbs with dative > nominative order, re-sumptives are awkward, often leading to ungrammaticality. Puzzlingly, many of those cases donot improve if the resumptive is omitted. See Salzmann (2006b) for full discussion.

. ZG – and Swiss dialects in general – is merely a spoken language and has no strict or-thography. In my transcription, I follow basically the spelling guidelines of Dieth (1938), andDieth & Schmid-Cadalbert (1986) respectively. Dieth’s (1938:13) key principle “schreibe so, wiedu sprichst, wie du es hörst und empfindest” ‘write like you speak, like you hear and feel’ hasbeen widely adopted. This is particularly true for vowel and consonant lengths. I chose, how-ever, not to use diacritics as proposed in the above-mentioned guidelines. This is because suchphonetically close transcriptions are not necessary for the purposes pursued here. Moreover,

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.3 (199-221)

Long relativization in Zurich German as resumptive prolepsis

(1) a. dthe

Frau,woman

woC

(*si)(she)

immeralways

ztoo

spaatlate

chuntcomes

‘the woman who is always late’ (subject)b. s

theChuchichäschtli,kitchen.cupboard

wonC

iI

(*s)it

vor.churzemrecently

kchauftbought

hanhave.1s

‘the kitchen cupboard I recently bought’ (direct object)c. de

theBueb,boy

woC

merwe

*(em)(he.dat)

esa

Velobike

versprochepromised

händhave.1p

‘the boy we promised a bike’ (indirect object)d. d

theFrau,woman

wonC

iI

vonfrom

*(ere)(she.dat)

esa

Buechbook

überchoogot

hanhave.1s

‘the woman from whom I got a book’ (P-object)

. Distribution of resumptive pronouns: Long-distance relativization

Once we look at long-distance relativization, we find resumptives for all relations,even subjects and direct objects:

(2) a. dthe

Frau,woman

woC

tyou

gsäitsaid

häsch,have.2s

dassthat

*(si)she

känno

Fründboyfriend

häthas

‘the woman who you said has no boyfriend’ (subject)b. s

theBild,picture

woC

tyou

gsäitsaid

häsch,have.2s

dassthat

*(es)it

dethe

PeterPeter

wettwants

verchauffeto.sell‘the picture that you said Peter wants to sell’ (direct object)

many of the sounds at issue are in (near-)allophonic variation with sounds that correspond tographemes present in the ordinary alphabet. This is particularly true for the – virtually nondis-tinctive – lengths of the palatal and the velar fricatives as well as for some vowel qualities. Someproblems arise with the several e-sounds: The letter <e> is used for [e], as well as for Schwa,which exclusively appears in reduced syllables predominantly at the end of the word, and for [7]while <ä> exclusively corresponds to [æ]. Again, there is little need to use diacritics to distin-guish [e] from [7] in my data set especially since the contrast is neutralized in certain contexts.The only word where I explicitly mark the vowel quality is (g)gëë ‘give(n)’. Here, <ë> corre-sponds to [7]. In some cases, I intentionally deviate from the spelling guidelines by using formssimilar to Standard German spelling in order to facilitate comprehension. A case in point isthe complementizer <dass> which phonetically would be transcribed as [das] in Zurich Ger-man. I chose the Standard German spelling to distinguish it from the Standard German relativepronoun <das>.

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.4 (221-288)

Martin Salzmann

. Long relativization as aboutness relatives: Van Riemsdijk (to appear)

This asymmetry between local and long relativization is somewhat surprising. Thelocal pattern can be explained straightforwardly by the assumption that obliquecase needs to be realized, a condition operative in many languages that use resump-tive pronouns, cf. e.g. Pesetsky (1998). Bayer et al. (2001) have pointed out that thisconstraint is operative in Standard German (though in other areas), and Salzmann(2006a) discusses matching effects that clearly show that the local ZG pattern isbest understood as a means to guarantee the recoverability of oblique case.

The long-distance pattern, however, is surprising given the facts from localrelativization. One would expect the transparent local pattern, for which there isindependent evidence in other parts of the grammar, to be found in all aspectsof relativization. The asymmetry clearly suggest that something else must be re-sponsible for the appearance of resumptive pronouns in long relativization.8 Thisis, I believe, the starting point for a reanalysis of long relativization in ZG by vanRiemsdijk (to appear). I will introduce this approach in the next section.

. Locative relatives and aboutness relatives: Adverbial wo

Next to resumptive relatives, there is one type of relative where a gap appearsin both local and long relativization, namely locative relatives. They are also in-troduced by wo. The same form is used for so-called aboutness relatives (vanRiemsdijk to appear) which have a vague locative meaning and express corollarycircumstances (similar to English expressions like with this weather):

(3) a. sthe

Huus,house

woC

tyou

gsäitsaid

häsch,have.2s

dassthat

dethe

PeterPeter

__ wontlives

‘the house where you said Peter lives’b. es

aWätter,weather

woC

sit

sichself

__ nödnot

loont,is.worthwhile

dethe

Raaselawn

zto

määjemow

‘a weather where there is no point in mowing the lawn’

Van Riemsdijk (to appear) assumes that in both cases there is a phrasal relativeadverb wo ‘where’ next to the relative complementizer wo. It moves to Spec, CPand is eventually deleted under haplology with the complementizer:

. In certain languages, e.g. Irish, resumptives are only barred from the matrix subject position.For those cases, there have been proposals (e.g. McCloskey 1990) that make reference to thenotion A’-disjointness, basically the A’-version of Principle B. A very different explanation isfound in Boeckx (2003). The Zurich German pattern is also found in Welsh (Rouveret 2002).Unfortunately, the syntax of Welsh resumptive relatives is different in relevant respects so thatcomparison does not provide new insights.

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.5 (288-362)


(4) DPi [cp [xp wo]i Cwo [xp wo]i . . . ]

Since this is a normal case of phrasal A’-movement, a resumptive pronoun is notexpected: The lower copy is deleted due to normal deletion of the lowest chain linkand the upper copy is exceptionally deleted by haplology.

. Long relativization as aboutness relativization

Now comes the crucial step: Van Riemsdijk (to appear) proposes that long rela-tivization actually involves aboutness relativization in the matrix clause. If I in-terpret him correctly (the paper is not very explicit on this point), the resumptivepronoun we find in the complement clause is simply a bound pronoun linked to itsantecedent by construal and not movement (van Riemsdijk speaks of an “apparentvariable”):9

(5) the mani [cp [xp wo] Cwo I [xpwo] think [cp hei . . . ]]

. Advantages

This proposal has a number of important advantages. First, the appearance of apronoun is predicted because movement is not involved, and since ZG is not apro-drop language, an overt pronoun is necessary. Second, ZG can be argued toinstantiate a more abstract version of an alternative strategy to long relativiza-tion in both German and Dutch (Section 6, Salzmann 2006b) whereby an ‘of ’-XPconstituent appears in the matrix clause:

(6) derthe

Mann,man

vonof

demi

who.datichI

glaube,believe.1s

dassthat

eri

heintelligentintelligent

istis

‘the man who I believe is intelligent’

. An anonymous reviewer has correctly pointed out that the indexing employed in this exam-ple and elsewhere in the paper is strictly speaking incorrect: The external head is only an NP andtherefore not a referring term so that it cannot be directly co-indexed with a pronoun. Rather,the pronoun is co-indexed with an element in the matrix clause, arguably wo under van Riems-dijk’s analysis (see Subsection 3.1.1), or [Op + a representation of the external head] under theanalysis proposed in Section 4. The NP part of the operator phrase in the matrix clause is thenrelated to/predicated of the external head. Under van Riemsdijk’s proposal, the resumptive is ar-guably re-interpreted as a variable bound pronoun (a pronoun bound by an operator (wo) in anA-position that undergoes further A’-movement). Despite these complications, I will retain theindexing for reasons of legibility, especially in the context of reconstruction effects. In Salzmann(2006b) I discuss these issues and employ a different notation system.

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.6 (362-427)

Martin Salzmann

Third, there is a base construction: The constituent corresponding to wo inboth long relativization and aboutness relatives is realized as bi+DP ‘at’+DP if itremains in-situ:

(7) a. esa

Wätter,weather

woC

sit

sichself

__ nödnot

loont,is.worthwhile

dethe

Raaselawn

zto

määjemow

‘a weather where there is no point in mowing the lawn’b. Es

itloontbe.worthwhile

sichself

biat

demthis

Wätterweather

nöd,not

dethe

Raaselawn

zto

määje.mow

‘With this weather, there is no point in mowing the lawn.’

(8) a. esa

Mäitli,girl

woC

merone

säit,says

dassthat

esit

gernlikes.to

isin.the

Kinomovie

gaatgoes

‘a girl who one says likes to go to the movies’b. Mer

onesäitsays

biat

demthis

Mäitli,girl

dassthat

esshe

gernlikes.to

isin.the

Kinomovie

gaat.goes

‘One says about this girl that she likes to go to the movies.’

Fourth, a fact not discussed in van Riemsdijk (to appear), if the bi-phrase is wh-moved across another wh-phrase, we do not get a superiority violation. This wouldbe unexpected if that constituent were extracted from the complement clause be-cause ZG, like Standard German, shows long distance superiority effects, cf. (9a).However, since only two matrix wh-phrases compete, the absence of superiorityeffects is predicted (9b):

(9) a. *Welemwhich.dat

Schüelerpupil

glaubtthinks

welewhich

Leerer,teacher

dassthat

merone

__ söttshould

äisone

aat

dthe

Ooreears

gëë?give

lit.: ‘Which pupil does which teacher think that one should give a box onthe ears?’

b. Biat

welemwhich.dat

Schüelerstudent

glaubtthinks

welewhich

Leerer,teacher

dassthat

merone

emhe.dat

söttshould

äisone

aat

dthe

Ooreears

gëë?give

Fifth, “long relativization” is insensitive to locality (the island appears in angledbrackets):

(10) dethe

Autor,author

woC

dthe

MarieMary

< jedesevery

Buechbook

list,reads

wonC

erhe

schriibtwrites

>

lit.: ‘the author that Mary reads every book he writes’

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.7 (427-483)


The alleged resumptive appears inside a relative clause. Since ZG obeys local-ity constraints on movement elsewhere as in the following example with wh-movement10

(11) *Werwho

listreads

dthe

MarieMary

< jedesevery

Buech,book

woC

__ schriibtwrites

>?

lit.: ‘Who does Mary read every book that writes?’

this would be unexpected under a movement account. It follows naturally,however, under a pure construal relationship as proposed by van Riemsdijk(to appear).

Although I believe that the basic idea of this analysis is correct, there are anumber of problems. They are discussed in the following section.

. Problems of van Riemsdijk’s (to appear) proposal

. There seems to be a copy of the external head inside the relative

The first type of problem concerns the nature of wo. If “long relativization” reallyinvolves movement of wo in the matrix clause there is no relative clause inter-nal representation of the external head. As a consequence, we do not expect anyreflections of this for processes like binding. However, this is exactly what we find.

.. Strong crossover effectsIf a matrix subject is co-indexed with a resumptive in the complement clauseungrammaticality results:

(12) *dethe

Maai,man

wonC

eri

hetänkt,thinks

dassthat

eni

himniemertno.one

gern.hätlikes

lit.: ‘the mani whoi hei thinks no one likes’

I would like to argue that the ungrammaticality is due to a Strong Crossover (SCO)effect, i.e. that there is movement of a coreferential element across the matrixsubject as in the following example from local relativization:

(13) *dethe

Maai,man

wonC

eri

he__ gern.hät

likeslit.: ‘the mani whoi hei likes’

. The example does not improve if the gap is replaced by a resumptive pronoun, cf. Salzmann(2006b).

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.8 (483-551)

Martin Salzmann

Here, a direct object is A’-moved across a coreferential pronominal subject. Im-portantly, there has to be a relative clause internal representation of the externalhead (i.e. an empty operator in D + Maa as in Bhatt (2002) (cf. 4.1), or at least arelative pronoun with the same phi-features as the external head) to explain theungrammaticality of these examples. However, if according to van Riemsdijk (toappear) only phrasal (local) wo moves, this does not follow:

(14) *dethe

Maai, [cp

man[wo]j

wowonC

eri

he[wo]j tänkt,

thinksdassthat

eni

himniemertno.one

gern.hät ]likes


Since wo – being a locative form – neither has phi-features nor contains a repre-sentation of the external head, the SCO effect is unexpected.

One of the reviewers suggested assigning wo the index “i” instead. This seemsindeed a reasonable assumption: It is clear that wo somehow has to be related tothe external head and to the resumptive pronoun. Assigning it the same index asthe two would take care of this, thereby yielding the right result for the SCO effects:

(15) *dethe

[Maa]i, [CP

man[wo]i

wowonC

eri [wo]i

hetänkt,thinks

dassthat

eni

himniemertno.one

gern.hät ]likes


In a sense, wo would be the overt counterpart of the empty operator, which isalso compatible with any kind of external head, any value for animacy, gender,number and person. I think there are two reasons why such an approach is stillproblematic: First, overt relative pronouns are normally not so flexible, they areonly compatible with certain antecedents even if they are invariant (cf. e.g. Dutchprepositional waar). Second, adopting an operator movement analysis for relativeclauses, i.e. the traditional Head External Analysis, is at odds with recent work thathas shown it to be quite undesirable, especially when it comes to reconstructioneffects, cf. e.g. Bhatt (2002). This aspect is important in the light of the followingsubsection.

.. Reconstruction into the matrix clauseAn even stronger argument for the relative clause internal representation of theexternal head comes from reconstruction effects. The following example showsthat material contained inside the external head can be bound by elements insidethe matrix clause. The following example illustrates this for Principle A:11

. As opposed to English, anaphors in ZG cannot be used logophorically so that the concernvoiced in Safir (1999) and Bhatt (2002) does not apply. Anaphors are subject to Principle A inZG and require a c-commanding antecedent to be licensed. For anaphor binding I have made

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.9 (551-625)


(16) sthe

äinzigeonly

Grüchtrumor

überabout

siichi,self

woC

dethe

Peteri

Peterfindt,finds

dassthat

esit

unggrächtunfair

ischis‘the rumor about himselfi that Peteri thinks is unfair’

Given standard assumptions about the treatment of reconstruction in the Min-imalist Program (e.g. Fox 1999; Bhatt 2002), a full copy of the external head isnecessary inside the matrix clause. This implies that movement of bare wo is in-sufficient. I assume for the sake of the argument that wo is the antecedent of theresumptive pronoun and that it is co-indexed with the external head as under thetraditional Head External Analysis of relative clauses:

(17) sthe

[Grüchtrumor

überabout

siichi]j, [cp

self[wo]j

wowoC

dethe

Peteri

Peter[wo]j

wofindt,finds

dassthat

esj

it

unggrächtunfair

isch ]is

. Reconstruction into the embedded clause

The second type of problem concerns the absence of an A’-dependency. If I readhim correctly, van Riemsdijk (to appear) assumes that the resumptive pronoun inthe embedded clause is not part of an A’-dependency, i.e. that it is not a resump-tive pronoun at all. That is probably why he refers to it as an “alleged variable”.However, there is evidence that it does participate in an A’-dependency.

Reconstruction effects turn out to be more pervasive in ZG relatives. Materialcontained inside the external head can also be bound by elements inside the em-bedded clause. The following triple illustrates this for Principle A, variable bindingand idiom formation:12, 13

sure that binding cannot be due to a coreferential implicit PRO by using ‘rumor’ where this isruled out, cf. Bianchi (1999:117f.).

. The expression e Reed schwinge lit. ‘swing a speech’ means ‘give a speech’. As pointed outin de Vries (2002:78f.) the types of expressions that can be used in relativization all involve NPsthat more or less retain their meaning in these idiomatic expressions/collocations. Completelyopaque idioms cannot be used in relativization.

. I do not deal with reconstruction for scope and the interpretation of superlative adjectives(Bhatt 2002; Heycock 2003) here because a full discussion of the (intriguing) complicationsfound in that domain is beyond the scope of this paper. See Salzmann (2006b) for detailedanalysis.

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.10 (625-666)

Martin Salzmann

(18) a. sthe

[Bildpicture

voof

siichi]j,self

woC

tyou

gsäitsaid

häsch,have.2s

dassthat

dethe

Peteri

Petersj

it

wettwants

verchauffeto.sell

‘the picture of himselfi that you said Peteri wants to sell’b. De

the[Abschnittperiod

voof

simi

hisLäbe]j ,life

wonC

iI

glaub,think

dassthat

niemerti

no.onedrj-überit-about

redt,talks

ischis

dthe

Pubertät.puberty

‘The period of hisi live that I think no onei talks about is puberty.’c. D

the[Reed]j,speech

wonC

iI

gsäitsaid

han,have

dassthat

erhe

sij

itgeschteryesterday

gschwungeswung

hät,has

häthas

merme

gfale.pleased

‘I liked the speech I said he gave yesterday.’

The resumptive pronouns indicate the reconstruction sites.14 Again, given stan-dard assumptions about reconstruction in recent versions of the Principles andParameters framework, a copy of the external head has to be present inside thecomplement clause. That this is incompatible with the phrasal wo that van Riems-dijk (to appear) postulates was shown in the previous subsection. Even if we grantthat reconstruction is also possible via co-indexing between external head andrelative pronoun (as in traditional analyses of relative clauses), this will not be suf-ficient for the case at hand because wo does not originate in the embedded clauseand therefore never occupies the position where the content of the external headis interpreted in the examples above. The only possible way out for van Riemsdijkseems to be to assume that reconstruction is in principle independent of move-ment and that the content of the external head can somehow be copied into theposition of the resumptive pronoun. While I cannot fully discuss reconstructionin non-movement contexts here, it seems to be generally agreed upon that recon-struction under a pure binding relationship requires at least an A’-dependency. Anon-movement A-dependency, which is what I take van Riemsdijk to be postulat-ing between wo and the “resumptive” in long ZG relativization, however, normallydoes not show reconstruction effects: Control, an A-dependency which undertraditional assumptions does not involve movement, does not show reconstruc-

. In (18b) the resumptive is an R-pronoun that appears whenever the antecedent is inani-mate, neuter and governed by a preposition (see Salzmann 2006b for a more precise statement).

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.11 (666-721)


tion effects while raising (which does involve A-movement) does, as the followingcontrast shows. Only the raising example is ambiguous (cf. e.g. Fox 1999):15, 16

(19) a. Someone from New York is likely to win in the lottery.∃ > likely; likely > ∃

b. Someone from New York tried/promised to win in the lottery.∃ > tried; *tried > ∃

. Obligatoriness of the resumptive

Under van Riemsdijk’s approach, the aboutness wo is an adjunct that is indepen-dently (semantically) licensed. One would expect the same to hold for the wo inlong relativization. Interestingly, however, more seems to be necessary to licensewo: “long relativization” requires a resumptive pronoun in the embedded clause:

(20) esa

[Resultaat]j,result

wonC

iI

glaub,believe

dassthat

dethe

HansJohn

zfridesatisfied

ischis

*(dej-mit)it-with

‘a result that I believe John is satisfied with’

The obligatoriness of the resumptive (and the preposition) cannot be related toselectional properties of the adjective in the embedded clause because it allows itsargument to be dropped:

(21) Dethe

Chefboss

ischis

zfridesatisfied

(de-mit).it-with

‘The boss is satisfied with it.’

Omitting the PP-complement of the adjective does not affect the interpretation;there can still be an implication that the boss is satisfied with something. This op-tionality is not found in long-distance relativization in (20). This is unexpected ifwo is independently licensed. Matrix clause adjuncts (and arguments except thoseof Control verbs) normally do not have to be resumed in the embedded clause. Thefollowing illustrates this for an aboutness adjunct (the construction has a Teutonicflavor, corresponding examples in Standard German are frequently found on theinternet):

. Since wo undergoes A’-movement in the matrix clause, we get a Parasitic Gap-like configu-ration. To the extent that Parasitic Gaps exist at all in Standard German (cf. Kathol 2001), theydo not allow for reconstruction.

. Another argument against a binding approach is anaphor binding in intermediate posi-tions to be discussed in 4.2.

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.12 (721-770)

Martin Salzmann

(22) Dethe

Parteipresidäntparty.leader

häthas

bezüglichconcerning

emthe

Waalresultatelection.result

bemerkt,remarked

dassthat

merone

sit

ganzquite

offesichtlichobviously

nödnot

gschaftmanaged

hät,has

dethe

Wäälervoter

voof

dethe

äigeneown

Ideeideas

zto

überzüüge.convince

‘The party leader remarked concerning the election result that one has obvi-ously not managed to convince the voter of one’s ideas.’

It seems unlikely that wo has quantificational properties (so that (20) without aresumptive would be a case of vacuous quantification) given the fact that no suchobligatory binding is necessary in locative and aboutness relatives, cf. (3).17 Rather,it suggests that something else is necessary to license the wo-constituent in long-distance relativization.

The problems reviewed in this section point towards a different implemen-tation of van Riemsdijk’s (to appear) proposal. This is what I attempt in thenext section.

. Proposal: Long relativization as resumptive prolepsis

The previous section has shown that in order to get the right interpretation, weneed a relative clause internal representation of the external head in both thematrix and the embedded clause. While the matrix representation is quite straight-forward, the representation inside the embedded clause requires more machinerybecause (as we will see) the embedded clause is an island. I will propose an anal-ysis reminiscent of tough-movement where there is operator movement inside thecomplement clause. This movement licenses an extra argument, the bi+DP con-stituent, which I will refer to as the proleptic object/constituent. This constituentis not directly related to the resumptive pronoun in the embedded clause. Rather,it is related via ellipsis to the operator in Spec, CP of the complement clause thatbinds the resumptive, thereby making an alternative strategy for reconstructionavailable. The bi+DP constituent undergoes A’-movement in the matrix clauseand can be deleted in the operator position, partially due to the locative origin ofthe relative complementizer. The presence of a resumptive pronoun follows froma constraint that requires specific chains to be phonetically realized in ZG. Theentire structure represents what I call “resumptive prolepsis”.

. Put more carefully, wo is an operator that binds the variable it leaves in the matrix clause,but it does not require an additional variable such as the resumptive pronoun.

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.13 (770-820)


. The derivation in the matrix clause: Against a phrasal wo

In this first subsection, I will argue that there is no phrasal wo in “long relativiza-tion” and aboutness relatives and possibly even in locative relatives.

.. C-wo makes deletion of locative Ps recoverableThe SCO effects and the reconstruction effects into the matrix clause discussedin 3.1 show that there has to be a representation of the external head inside thematrix clause. I propose that instead of bare wo, we find a full copy of the externalhead, governed by the preposition bi ‘at’ which we find in the base construction, cf.(8b). The entire PP undergoes A’-movement to Spec, CP of the matrix clause. Thecopy of the external head is PF-deleted under identity with it;18 the preposition isalso PF-deleted because the complementizer wo, due to its locative origin, makesthe deletion of locative and aboutness prepositions recoverable. As a consequence,there is no phonetic realization of the bi-PP:19, 20

(23) dethe

Maai, [cp

man[biat

OpOp

Maai]j

manwonC

iI

[biat

x Maai]j

mantänke,think

dassthat

eri

he

intelligäntintelligent

ischis

‘the man who I think is intelligent’

. To be more specific, I assume a Matching Analysis (e.g. Sauerland 1998, 2003; Citko 2001)for ZG relative clauses. I cannot discuss the full range of facts that motivate this analysis for ZGfor reasons of space. Very briefly, it has been pointed out in Heck (2005) that the Head RaisingAnalysis is confronted with serious problems when applied to Standard German (among otherthings, it leads to wrong case assignment on the external head and violates the CED); theseobjections apply to ZG as well. At the same time, a pure Head External Analysis that only has anempty operator inside the relative clause cannot capture the reconstruction effects (Bhatt 2002).The Matching analysis avoids the problems of the Head Raising Analysis and manages to handlethe reconstruction effects (Bhatt 2002; Citko 2001). Ffurthermore, as we will see in 4.4.1 below,it allows a straightforward explanation of the non-reconstruction for Principle C.

. The external D is never reconstructed in relative clauses (cf. Bianchi 1999; Bhatt 2002). Forreasons of simplicity, I represent the relative operator as an empty operator. It might just as wellbe a relative pronoun that is deleted, but in the absence of any evidence for this, I will stick tothe more innocuous choice.

. The following representations encode both LF and PF. Outline indicates PF-deletion.strikethrough LF-deletion. The restriction of the operator is LF-deleted and only retained inthe base position, in accordance with the Preference Principle (Chomsky 1995). The copy left bythe operator itself is converted into a variable.

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.14 (820-914)

Martin Salzmann

This assumption yields the desired result for the SCO effects and the reconstruc-tion data:21

(24) a. *dethe

[Maai],man

[cp [biat

OpOp

Maai]j

manwonC

eri

he[biat

x Maai]j

mantänkt,thinks

dassthat

eni

himkäneno.one

gern.hät ]likes

lit.: ‘the mani whoi hei thinks no one likes’b. s

the[Bildpicture

voof

siichi]k,self

[cp [biat

OpOp

[Bildpicture

voof

siichi]k]j

selfwoC

dethe

Peteri

Peter[biat

x Bildpicture

voof

siichi]j

selffindt,finds

dassthat

esk

itguetgood

gglungeturned.out

ischis

‘the picture of himselfi that Peteri thinks turned out good’

In both cases, there is a representation of the external head inside the relativeclause, which explains why we get SCO effects and why reconstruction is possible.The following triple illustrates schematically to what extent the complementizerwo makes the PF-deletion of prepositions recoverable in normal relatives wherethe relativized constituent originates in an oblique position:

(25) a. the city [cp [in the city]i wo I have [in the city]i livedb. the weather [cp [bi the weather]i wo one [bi the weather]i should stay homec. the man [cp [with the man]i wo I have [with him]i talked

In (25a–b) locative and aboutness prepositions are recoverable. (25c) shows thatwith other, semantically more specific prepositional relations, deletion is not re-coverable. As a consequence, the preposition has to be realized (together with aresumptive pronoun).22

.. Alleged phrasal wo does not pattern with other adverbial relativesThere is further evidence that the alleged phrasal wo does not exist: It can be shownthat locative relatives fail to pattern with other adverbial relatives which employ a

. The external head is LF-deleted when it contains material that is not licensed there such asanaphors, bound variables or idiomatic NPs, elements which I refer to as elements with a posi-tive licensing requirement. Elements which are licensed in the external head (e.g. R-expressions)need not be deleted. A more explicit version of the deletion system assumed here is found inSalzmann (2006b).

. One may wonder why the preposition is realized in the base position and not upstairs. SinceI have assumed that PF-deletion of the constituent in Spec, CP is obligatory, the lower copy isthe only possible chain link for the preposition to be realized. The presence of the resumptivefollows either from the assumption that oblique case needs to be realized or the ban againstpreposition stranding in ZG, cf. Salzmann (2006a, 2006b).

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.15 (914-980)


phrasal relative adverb that is not PF-deleted. Those adverbial relatives allow the Cposition to be filled with the declarative complementizer dass ‘that’:

(26) a. dethe

Grund,reason

werumwhy

dassthat

dethe

PeterPeter

ztoo

spaatlate

choocome

ischis

‘the reason why Peter came late’b. D

theArt,way

wiehow

dassthat

dethe

PeterPeter

sthe

Probleemproblem

gglööstsolved

hät,has

häthas

mime

beiidruckt.impressed‘The way Peter solved the problem impressed me.’

Interestingly, with locative wo a declarative complementizer is much worse:

(27) Dethe

Ort,place

wowhere

(??dass)that

erhe

wont,lives

willwants

erhe

niemertemnobody.dat

verraate.tell

‘He does not want to reveal to anybody the place where he lives.’

Crucially, when we look at the phrasal wh-adverb wo, we find no such restriction.This suggests that the deviance of (27) is not a property of phrasal wo as such:

(28) IchI

wäissknow

nöd,not

wowhere

dassthat

erhe

wont.lives

‘I do not know where he lives.’

I conclude from this that there is no phrasal relative adverb wo. One major advan-tage of the approach advanced here is that there is only one wo in relativization,namely the complementizer wo.23, 24

. Admittedly, the deviance of (27) also follows, if one assumes – as van Riemsdijk (to appear)does – that the C- position is occupied by the complementizer wo. There is simply no spacefor dass. The question is then why C-wo does not occur in the other adverbial relatives. Ananonymous reviewer suggests extending the constraint against two wo in the left periphery totwo “w-words”. This would rule out werum wo and wie wo, but crucially not werum dass andwie dass. That is certainly a possibility. It implies that both wo and dass are in principle possiblecomplementizers in relativization. But this immediately raises a further question: Why is dassnot an option in resumptive relatives? Furthermore, this constraint may run into difficultieswith multiple wh-questions where on an LF-movement analysis multiple wh-phrases occupyspecifier positions of C. It seems therefore preferable to me at this point to keep resumptive andadverbial relatives separate.

. Henk van Riemsdijk (p.c.) has pointed out to me that the relative modifying reason is alsodifferent in that the wh-word can be dropped. This might suggest that there are independentreasons for the different behavior. Josef Bayer (p.c.) has suggested in the same context that thereason relative clause is actually a wh-complement. These facts might admittedly weaken theargument made in the text. However, they leave (26b) unexplained. I will leave this for furtherresearch.

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.16 (980-1046)

Martin Salzmann

. Reconstruction into an opaque domain?

The most intricate aspect of “long relativization” in ZG are arguably the data thatshow reconstruction into the embedded clause (18). Since there is a base con-struction with the aboutness constituent in the matrix clause, a direct movementrelationship from the embedded clause is unlikely. In fact, the complement CPeven turns out to be an island for extraction. The following triple first illustratesthe base sentence with an aboutness constituent (29a); (29b) shows argument ex-traction from the aboutness construction and (29c) shows argument extractionfrom a normal complement clause:

(29) a. IchI

hoffehope

bimat.the

Leererteacher

Müller,Müller

dassthat

erhe

emthe.dat

HansliJohn

ea

guetigood

Nootegrade

git.gives

‘I hope about teacher Müller that he will give little John a good grade.’b. ??[Welem

which.datSchüeler]i

studenthoffschhope.2s

bimat.the

Leererteacher

Müller,Müller

dassthat

erhe

ti ea

guetigood

Nootegrade

git?gives

lit.: ‘Which student do you hope about teacher Müller that he will give agood grade?’

c. [Welemwhich.dat

Schüeler]i

studenthoffsch,hope.2s

dassthat

dethe

Leererteacher

MüllerM.

ti ea

guetigood

Nootegrade

git?gives

‘Which student do you hope that teacher Müller will give a good grade?’

The following pair contrasts adjunct extraction from the aboutness construction(30a) with adjunct extraction from a normal complement clause (30b):

(30) a. *Werumi

whyglaubschthink.2s

bimat.the

Peter,Peter

dassthat

erhe

dthe

MarieMary

ti wettwants

hüraate?marry

‘Why do you think about Peter that he wants to marry Mary?’b. Werumi

whyglaubsch,think.2s

dassthat

dethe

PeterPeter

dthe

MarieMary

ti wettwants

hüraate?to.marry

‘Why do you think that Peter wants to marry Mary?’

Argument extraction is strongly degraded in the aboutness construction; adjunctextraction is impossible. (30a) only has a matrix construal. No such restrictionsobtain with extraction from normal complement clauses. Argument and adjunctextraction are both fine, (30b) allows both matrix and embedded construal.

This constitutes a paradox: There is reconstruction into a domain from whichextraction is impossible. One possible way out would be to assume that recon-

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.17 (1046-1121)


struction is done via binding (perhaps Chain Binding along the lines of Barss1986). However, this does not work because there is reconstruction into interme-diate positions:25

(31) sthe

[Bildpicture

voof

siichi/j]k,self

wonC

iI

glaube,think.1s

[cp dassthat

dethe

Peteri

Peterz.Unrechtwrongly

tänktthinks

[cp dassthat

dthe

Mariej

Marysk

itlässigcool

findtfinds

]]

lit.: ‘the picture of himi-/herselfj that I believe Peteri wrongly thinks that Maryj

likes’

The reflexive can be bound by different subjects. Given our assumptions aboutreconstruction, this implies that there must be an additional representation ofthe external head inside the relative clause. In addition to the copy in the baseposition, there must be at least another one in the intermediate Spec, CP posi-tion so that binding by Peter is possible. The copies are indicated in the followingrepresentation:

(32) sthe

[Bildpicture

voof

siichi/j]k,self

wonC

iI

glaubethink.1s

[cp dassthat

dethe

Peteri

Peterz.Unrechtwrongly

tänkt,thinks

[cp Bildpicture

voof

siichi/j

selfdassthat

dthe

Mariej

MaryBildpicture

voof

siichi/j

selfsk

it

lässigcool

findt ]]finds

lit.: ‘the picture of himi-/herselfi that I believe Peteri wrongly thinks that Maryj

likes’

The external head can be interpreted in the lowest chain link or in the intermediateSpec, CP position. The latter possibility automatically rules out binding because anoccurrence in that position can only result from movement. Once we need severalcopies inside the complement CP, we are effectively dealing with successive-cyclicmovement.

Still, the paradox remains: We have reconstruction into an opaque domain;within that domain, we have evidence for successive cyclic movement up to thehighest Spec, CP, but it is unclear what happens thereafter. The following sectionprovides the first part of the answer.

. Reconstruction into intermediate positions for anaphor binding has been disputed forStandard German, cf. Kiss (2003). I do not share this judgment, neither for Standard Germannor for ZG even though intermediate binding indeed seems less straightforward than in English.See Salzmann (2006b) for detailed discussion.

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.18 (1121-1167)

Martin Salzmann

. Movement in the complement CP derives a predicate

I adopt the conclusion from the previous section that there is successive-cyclic A’-movement in the complement clause. This movement turns the CP into an opensentence as in tough-movement (Cinque 1990; Den Dikken/Mulder 1992). TheCP is then merged with the matrix verb whereby a complex predicate is formed.26

This predicate is still unsaturated. It is the proleptic object, more precisely the DPwithin the PP, that saturates it. Operator movement can therefore be thought tolicense an extra argument. This is illustrated in the following figure:

(33)

[ P[ ] V [ V]CP CPDP DP DPi i i

subject predicate

operator movement

predication

Crucially, the aboutness constituent is only licensed if there is operator move-ment in the complement. The preposition bi is needed to case-mark the DP. Thechoice of bi arguably follows from the semantics. Just like benefactive adjuncts aregoverned by for in English, aboutness phrases are governed by bi in ZG. For con-creteness’ sake I assume that the proleptic constituent is adjoined to VP. Finally,the whole complex plus the little v, which inherits the external theta-role of theverb, is predicated of the syntactic matrix subject.

This gives us the right configuration: First, the verb and the complementclause form a constituent to the exclusion of the proleptic object; this is shownin the following asymmetry in VP-topicalization:

(34) a. [Gglaubt,believed

dassthat

erhe


isch]j

ishanhave.1s

iI

bimat.the

PeterPeter

schoprt

immeralways

tj.

‘I have always believed that Peter is intelligent.’b. *[Bim

at.thePeterPeter

gglaubt]j

believedhanhave.1s

iI

schoprt

immeralways

tj, dassthat

erhe


isch.is

. The complex predicate is formed in the semantics or rather follows from the semanticinterpretation of the syntactic structure. There is no syntactic complex predicate formation in-volving incorporation or the like. A detailed exposition of the semantics is beyond the scopeof this paper; see den Dikken & Mulder (1992) or Rezac (2004) for implementations of tough-movement which in slightly modified form can be extended to the proleptic construction.

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.19 (1167-1226)


Second, the proleptic object can be shown to be base-generated below the syntacticsubject. In the following example, I have made sure that the subject remains inthe vP by using an (negative) indefinite and a modal particle that marks the vPboundary (Diesing 1992). In that position, the subject can bind a pronoun insidethe proleptic object:

(35) dassthat

dochprt

känei

no.onebiat

simi

hisSoonson

würdwould

glaube,believe

dassthat

erhe

ena

Verbrächercriminal

ischis‘that no onei would believe of hisi son that he is a criminal’

Third, predication is known to be subject to a c-command requirement. The fol-lowing example shows that the proleptic DP within the PP can c-command out ofit because it licenses a Negative Polarity Item in the complement clause:

(36) IchI

glaubbelieve.1s

biat

kämno

Holländer,Dutchman

dassthat

erhe

au.numeeven

äinone

EuroEuro

würdwould

verschwände.spill‘I believe of no Dutchman that he would spill even one euro.’

C-command out of a PP is not infrequent and found for example in the followingsentence:

(37) John thinks of Bill as silly.

I take this to be sufficient evidence that there is the right c-command rela-tionship for predication to be possible in the proleptic construction. The nextsubsection shows that this captures most of the major properties of “long rela-tivization” in ZG.

.. AdvantagesFirst, the operator movement approach explains the obligatoriness of the resump-tive pronoun discussed and its coindexation with the aboutness-DP (inside thebi-PP) in (20) under the assumption that the pronoun marks the tail of the Op-chain and thereby the variable. The absence of a resumptive implies absence ofoperator movement so that the proleptic object cannot be licensed.27 Second, it ex-plains the opacity of the CP-complement, as discussed in 4.2: operator movementcreates a weak island. Third, it makes an alternative strategy for reconstructionavailable, under the assumption that what moves inside the complement CP is ac-tually an operator with a full copy of the aboutness constituent. I will elaborate

. I will discuss in 4.5 why the chain link has to be overt.

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.20 (1226-1296)

Martin Salzmann

on this in 4.4. Fourth, it is directly compatible with the absence of long-distancesuperiority discussed in (9). The proleptic object is base-generated in the matrixclause so that it can be freely reordered with respect to another wh-phrase in thematrix clause, as German does not show any short-distance superiority effects.The following subsection discusses strong parallels with tough-movement.

.. Parallels with tough-movementI mentioned above that my approach is very similar to what has been proposed (bysome) for tough-movement. This subsection shows that tough-movement behaveslike the proleptic construction in a number of crucial ways.

First, the tough-subject is not independently licensed. As with resumptive pro-lepsis, it requires operator movement in the complement clause to be licensed. Inthe absence of a gap, ungrammaticality results:

(38) *This book is tough for the students to pass the exam. (Cinque 1990:153)

Second, the tough-clause is an island for extraction. The following examples il-lustrate this for argument and adjunct extraction (Rezac 2004:19, his (51a) and(50a)):28, 29

(39) a. *[Which violin]2 is that sonata1 hard to imagine you playing __1 on t2?b. *[How intelligent]2 is John1 easy to think of __1 as t2 ?

Third, there is reconstruction of the tough-subject, mediated by operator move-ment (den Dikken/Mulder 1992:310n8):30

(40) b. [Pictures of himselfi nude] are tough for me to think that any mani wouldlike __.

a. [Pictures of hisi wife nude] are tough for me to think hat any mani wouldshow his friends __ .

The correlations are striking and support the proposal advanced here. In 4.4.2 Iwill discuss further parallels between the two constructions.

. There are certain complications with argument extraction because there is a strong con-trast between nested and crossing dependencies. With another level of embedding as in the text,the result is straightforward. See Rezac (2004) for insightful discussion.

. Since the tough-subject is not directly extracted from the infinitival CP, I indicate theposition where it is interpreted via underline instead of trace notation.

. These examples clearly show that reconstruction can go below the experiencer, contrary towhat is (wrongly) claimed in Rezac (2004).

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.21 (1296-1337)


. The link between the operator in the complement and the prolepticobject: ellipsis

I have argued so far that operator movement instantiates an alternative strategyfor reconstruction. However, it still needs to be explained how to get a full copy ofthe proleptic object, i.e. the DP within the PP, inside the complement CP, which,as we have seen, is an island. I argue that this comes about via ellipsis, a mecha-nism that has gained some popularity in the analysis of certain A’-movement typessuch as relativization (Sauerland 1998; Citko 2001) and comparatives (Lechner1999). I mentioned in note 18 that I assume a Matching Analysis for ZG relativeclauses: The copy of the external head in Spec, CP is deleted under identity withit. The same, I argue, happens to the operator in Spec, CP of the complement inresumptive prolepsis.

.. Arguments for ellipsis in resumptive prolepsisThere are two major motivations for an ellipsis analysis: First, there are often case-mismatches between the proleptic object and the operator in Spec, CP (the sameholds for relative clauses, of course): the proleptic object is assigned dative by thepreposition bi whereas the operator can receive any case. Ellipsis has been shownto be able to handle such mismatches systematically, as in the following sluicingexample where we have accusative in the antecedent and nominative in the elidedIP (Jeroen van Craenenbroeck p.c.):

(41) They told me to go, but I didn’t know when I should go

Second, we find Vehicle Change effects, another peculiar phenomenon associatedwith ellipsis, first discussed in Fiengo & May (1994). It (originally) describes amismatch in VP-ellipsis: An R-expression in the antecedent can correspond to apronoun in the elided material:

(42) a. *John likes Maryi and shei does <like heri >, toob. John likes Maryi, and shei knows that I do <like heri >, too

(42a) is ungrammatical despite Vehicle Change because of a Principle B violation.(42b), however, where one level of embedding is added, is grammatical. Cru-cially, it has been observed that the same correspondence seems to be possiblebetween the external head of relatives and its representation in Spec, CP. Severalresearchers have pointed out that there seems to be no reconstruction for PrincipleC in restrictive relatives (Munn 1994; Sauerland 1998; Safir 1999; Citko 2001) asopposed to wh-movement. This is not only true of English relatives but also of ZGrelative clauses:

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.22 (1337-1424)

Martin Salzmann

(43) a. sthe

[Fottipicture

vomof.the

Peteri]j,Peter

wonC

eri

hetj am

thebeschtebest

findtfinds

‘the picture of Peteri that hei likes best’b. *[Weles

whichFottipicture

vomof.the

Peteri]j

Peterfindtfinds

erj

hetj am

thebeschte?best

lit.: ‘Which picture of Peteri does hei like best?’

The issue is actually more complex than I can do justice to. Some speakers arepuzzled by examples like (43a) when first confronted with them. The coreferenceis more easy to get if the subject is slightly stressed. This arguably has to do withthe somewhat exceptional anaphoric relation in this case, the antecedent not beingprominent enough (Bianchi 2004). Once this is taken into account, the sentencesare fine for practically all speakers, and there is a clear contrast between relativiza-tion and wh-movement.31 The influence of Vehicle Change is illustrated in thefollowing simplified LF-representation:

(44) sthe

[Fottipicture

vomof.the

Peteri]j,

Peter[cp [Op Fotti vo

pictureimi]j

ofwonhe.dat

eri

C he

[x Fottipicture

voof

imi]j

he.datamthe

beschtebest

findt]finds

This implies that the binding relationship is the same as in the following sentence:

(45) Eri

Hefindtfinds

[dasthat

Fottipicture

voof

imi]him

amthe

beschte.best

‘He likes this picture of him best.’

In ZG (and also Standard German, cf. Kiss 2003) pronouns and reflexives are moreor less in free variation in picture nouns, and this is exactly why Vehicle Changeleads to an alleviation of Principle C effects.

The crucial observation is that the same lack of reconstruction for Principle Cis found with the proleptic construction as well:

(46) sthe

[Fottipicture

vomof.the

Peteri]j,Peter

wonC

iI

tj glaub,think

dassthat

eri

hesj

itamthe

beschtebest

findtfinds

‘the picture of Peteri that I think hei likes best’

. There seem to be cases where material contained in wh-moved arguments fails to recon-struct, see Heycock (1995), Romero (1998), Fox (1999) and Safir (1999) for discussion. I tendto find the corresponding (Standard/Zurich) German examples relatively bad, though. Fischer(2004) discusses antireconstruction for Principle C in Standard German and arrives and par-tially different generalizations. For my purposes, it is sufficient to assume that wh-movementand relativization differ systematically with respect to reconstruction for Principle C, and thisseems to be generally agreed upon (Safir 1999; Sauerland 2003). See Salzmann (2006b) fordetailed discussion of this issue.

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.23 (1424-1504)


However, this does not yet provide evidence that there is ellipsis between the pro-leptic object and the operator in Spec, CP because the Vehicle Change effect couldalso be due to Vehicle Change in the matrix clause where the external head is alsorelated to the operator in Spec, CP via ellipsis (under the Matching Analysis). For-tunately, this ambiguity can be avoided: The proleptic construction is also possiblewith wh-movement or topicalization in the matrix clause where the possibility ofVehicle Change in the matrix clause can be ruled out. Crucially, we find the sameobviation of Principle C effects:

(47) a. [Biat

[demthis

Fottipicture

vomof.the

Peteri]k]j

Peterglaubthink.1s

iI

tj sofort,immediately

dassthat

eri

hesk

itguetgood

findt.finds

lit.: ‘This picture of Peteri, I immediately believe that hei likes.’b. [Bi

at[welemwhich

Fottipicture

vomof

Peteri]k]j

Peterglaubschthink.2s

tj, dassthat

eri

hesk

itguetgood

findt?findslit.: ‘Which picture of Peteri do you think that hei likes?’

Here, the lack of Principle C effects can only be due to Vehicle Change between theproleptic object and the DP in Spec, CP of the embedded clause as shown in thefollowing LF-representations:

(48) a. [Biat

[demthis

Fottipicture

vomof.the

Peteri]k]j

Peterglaubthink.1s

iI

[Biat

[x Fottipicture

vomof.the

Peteri]k]j

Petersofort,at.once

[cp [Op Fottipicture

voof

imi]k

himdassthat

eri

he[x Fotti

picturevoof

imi]k

himsk

itguetgood

findt]finds

b. [Biat

[welemwhich

Fottipicture

vomof.the

Peteri]k]j

Peterglaubsch,think.2s

[Biat

[x Fottipicture

vomof.the

Peteri]k]j

Peter[cp [Op Fotti

picturevoof

imi]k

himdassthat

eri

he[x Fotti

picturevoof

imi]k

himsk

it

guetgood

findt?finds

The argument for Vehicle Change can be strengthened even more: there are casesin German, where a pronoun cannot serve as a coreferential element inside a

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.24 (1504-1577)

Martin Salzmann

picture noun. Instead, a reflexive is needed. These cases involve semi-idiomaticexpressions and collocations such as the following:32

(49) a. Eri

hehäthas

ea

unschmäichelhaftiunflattering

Mäinigopinion

voon

*imi/siichi.him/self

‘Hei has an unflattering opinion of *himi/himselfi.’b. Eri

hehäthas

esa

Porträtportrait

voof

*imi/siichi

him/selfggmaalet.painted

‘Hei painted a portrait of *himi/himselfi.’

Crucially, once we test reconstruction for Principle C with such expressions (innormal restrictive relatives), the result is ungrammatical:

(50) a. *Dthe

[Mäinigopinion

vomof.the

Peteri]j,Peter

wonC

eri

hetj hät,

hasischis

unschmäichelhaftunflattering

lit.: ‘The opinion of Peteri that hei has is unflattering.’b. *S

the[Porträtportrait

vomof.the

Peteri]j,Peter

wonC

eri

hetj ggmaalet

paintedhät,has

ischis

unvortäilhaft.unfavorablelit.: ‘The portrait of Peteri that hei painted is unfavorable.’

This follows under an ellipsis approach because the Vehicle-Changed structurescorrespond to the base structures with pronouns in (49) as illustrated in thefollowing representations:

(51) a. *Dthe

[Mäinigopinion

vomof.the

Peteri]j,Peter

[cp [Op Mäinigopinion

voof

imi]j

himwonC

eri

he

[x Mäinigopinion

voof

imi]j

himhät,has

ischis

unschmäichelhaft].unflattering

lit.: ‘The opinion of Peteri that hei has is unflattering.’b. *S

the[Porträtportrait

vomof.the

Peteri]j,Peter

[cp [Op Porträtportrait

voof

imi]j

himwonC

eri

he

[x Porträtportrait

voof

imi]j

himggmaaletpainted

hät,has

ischis

unvortäilhaft.]unfavorable

lit.: ‘The portrait of Peteri that hei painted is unfavorable.’

The same is found with the proleptic construction (illustrated with topicalizationin the matrix clause to rule out interfering Vehicle Change in the matrix clause):

. The lack of free variation has been attributed to a coreferential implicit PRO inside thepicture NP, cf. Reinhard & Reuland (1993:685).

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.25 (1577-1635)


(52) a. *[Biat

[derethis

Mäinigopinion

vomof.the

Peteri]k]j

Peterglaubthink.1s

iI

nöd tj,not

dassthat

eri

hesik

it

hät.haslit.: ‘This opinion of Peteri I do not believe that hei has.’

b. *[Biat

[demthis

Porträtportrait

vomof

Peteri]k]j

Peterglaubbelieve.1s

iI

nöd,not

dassthat

eri

hesk

it

ggmaaletpainted

hät.has

Lit.: ‘This portrait of Peteri I do not believe that hei painted.’

The parallelism between normal relativization and resumptive Prolepsis is strikingand lends strong support to the ellipsis approach. See Salzmann (2006b) for moredetailed discussion.

.. Ellipsis in tough-movementThe parallels with tough-movement extend to the side-effects of ellipsis as well,thereby lending further support to the proposal advanced here and providing anew handle on reconstruction in tough-movement.

First, it was noted in Wilder (1991:123) that apart from mismatches in struc-tural case (the tough-subject bears nominative case while the gap is assignedaccusative case), there are more drastic mismatches as in the following pair:

(53) a. [For him to be top of the class] is hard to believe __.b. *I cannot believe for him to be top of the class.

The grammaticality of (53a) is surprising if there is a movement relationship be-tween the tough-subject and the gap, given the ungrammaticality of (53b). Here,the difference is no longer one of structural case (nominative vs. accusative), butbetween a PP and a DP. As with the proleptic construction, such a mismatch canbe handled by ellipsis only.

Second, we find the same absence of reconstruction for Principle C (Munn1994:403):

(54) [Pictures of Johni] are hard for himi to like __.

This follows if there is Vehicle Change between the tough-subject and the operatorin Spec, CP of the infinitival clause:33

(55) Pictures of Johni are [cp [pictures of himi]j hard for himi to like [pictures ofhimi]j.

. English allows coreferential pronouns inside picture NPs just like German, cf. Reinhard &Reuland (1993:661).

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.26 (1635-1719)

Martin Salzmann

Interestingly, we also find a class of exceptions (idiom-like expressions etc.) werethere seems to be reconstruction for Principle C:

(56) *[Pictures of Johni] are hard for himi to take __.

Crucially, the ungrammaticality follows under a Vehicle Change approach as wellbecause the Vehicle-Changed basis is bad: the expression take a picture only allowsreflexives just like the German cases in (49):34

(57) Johni took pictures of *himi/himselfi.

.. SummaryThe ellipsis approach advocated here nicely captures crucial properties of bothtough-movement and resumptive prolepsis and unifies the two constructions inrelevant respects. The derivation in resumptive prolepsis is schematically shownbelow:35

(58)

[ P[ ] V [ V]CP CPDP DP DPi i i

subject predicate

operator movement

predication

ellipsis

. Why a resumptive?

So far, one crucial asymmetry between the proleptic construction and tough-movement has not been addressed: The presence/absence of a (resumptive) pro-noun in the complement clause.

. Cf. Munn (1994:402) for a different interpretation of these facts.

. One of the reviewers has correctly pointed out that it is not innocuous to assume that aPP-internal DP can be the antecedent of a DP in ellipsis. Since DP ellipsis is rather rare in thefirst place, it is somewhat difficult to provide independent evidence for the ellipsis operationproposed here. Apart from syntactic amalgamations, which have been claimed to involve DP-ellipsis (Lakoff 1974), the only instance of DP-ellipsis in ZG I can think of is topic drop, i.e.ellipsis of a topical element in the prefield (Spec, CP). Topic drop is indeed possible if the an-tecedent is governed by a grammatical preposition, just like bi in the proleptic construction (thedropped topic appears in outline):

(i) Häschhave.2s

aat

dethe

Mantelcoat

tänkt?thought

– Ja,yes,

[dethe

Mantel]coat

hanhave

iI

debii.with.me

‘Did you think of the coat? Yes, I have it with me.’

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.27 (1719-1761)


The fact that we find resumptive pronouns across the board in ZG resump-tive prolepsis is surprising in itself because ZG does not require resumptives forsubjects and objects in local relativization, cf. 1.2. Nor does it require resumptivepronouns in wh-movement or topicalization. In a first step, it is reasonable to as-sume a constraint that bars phonetic realization of more then one chain link in(ZG) A’-movement (cf. Merchant 2004 for a similar idea). This correctly rules outa resumptive in those cases where the operator/highest chain link is overt, as in wh-movement, free relatives and topicalization. The next step is then to explain whyin certain cases a chain link must be overt.36 Cases like (1c–d) can be handled bythe requirement to spell out oblique case (Bayer et al. 2001; Salzmann 2006a), butthis still leaves subject and direct object resumptives in long-distance relativizationunexplained.

The solution I would like to propose is based on the specificity of the chain.Cinque (152f.) has argued convincingly that operator movement chains are spe-cific. This certainly holds for the proleptic construction which requires D-linkedproleptic objects (cf. Salzmann 2006b). Bianchi (2004) in turn has pointed out thatresumption is crosslinguistically most frequent with and often limited to specificchains. Consequently, there must be a spell-out constraint in ZG that requires thelexicalization of specific chain links.

In ZG there are three types of A’-dependencies without overt operator: Localrelative clauses, comparatives and resumptive prolepsis constructions. Only one ofthem, the proleptic construction, features a specific chain and resumptives acrossthe board. Comparatives abstract over degrees and therefore certainly do not in-volve a specific/referential dependency; the same holds for restrictive relatives,where what is left behind is simply a (nonreferential/nonspecific) variable.37

This gives us the desired result. It does not yet explain why spelling out thetrace is illicit in tough-movement, which, of course, also involves a specific chain,cf. Cinque (1990:152f.). I believe that there is nothing particularly insightful to sayabout this because languages can differ in partially arbitrary ways when it comes to

. I have assumed without argument that a pronoun can be the spell-out of a full copy. Whythis should be so is generally poorly understood. Spelling out a full copy is arguably ruled outfor reasons having to do with linearization, cf. Nunes (2001). Spelling out only a pronoun couldbe a consequence of the features left behind after copying as in van Koppen (2004). A Big-DPapproach as e.g. in Boeckx (2003) (which was proposed for the equivalent in Standard Germanin Salzmann 2005) is a possibility as well. Its major drawback for the data at hand is the fact thatit is no longer easy to state the incompatibility of overt operators (wh, top) and resumptives. SeeSalzmann (2006b) for extensive discussion of the issue spell-out vs. Big-DP.

. It remains somewhat unclear how local non-restrictive relatives in ZG fit in because theyhave been argued to instantiate specific chains yet show the same spell-out possibilities asrestrictives. I leave this for future research.

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.28 (1761-1822)

Martin Salzmann

the spell-out options of chains.38 On a general level, ZG is a language that in prin-ciple allows resumptive pronouns whereas English is not. All we find in Englishare intrusive pronouns, which occur in islands to repair otherwise illicit extrac-tions, cf. Chao & Sells (1983). Since the extraction site of tough- movement is atransparent domain, a resumptive is not expected.

In ZG resumptive pronouns occur in specific chains without overt operator inall positions (resumptive prolepsis), in non-specific chains without overt operatoronly in oblique positions to realize oblique case (local relativization, compara-tives). In English, resumptives only (marginally) occur inside islands to repairlocality violations.

. A remaining issue: Locality and reconstruction

The approach sketched here is confronted with one serious problem that wasfirst mentioned in (10), the insensitivity of “long relativization” to locality con-straints, a property that follows straightforwardly from van Riemsdijk’s (to appear)binding approach.

Since other types of A’-movement in ZG are sensitive to those constraints, theinsensitivity of resumptive prolepsis comes as a surprise. One initial possibilitycan be ruled out: A base-generation approach with the operator binding the pro-noun (what Aoun et al. 2001 refer to as true resumption) cannot be correct becauseone finds reconstruction effects for both anaphors and bound pronouns even intostrong islands, as the following examples show (islands appear in angled brackets):

(59) a. Sthe

[Bildpicture

voof

siichi]j,self

woC

alleveryone

lached,laughs

<if

wänn dethe

Peteri

Petersj

it

zäiget>,shows

ischis

iin

dethe

Stube.lounge

lit.: ‘The picture of himselfi that everyone laughs when Peteri shows it, isin the lounge.’

b. Dasthat

ischis

sthe

[Buechbook

überabout

siichi]j,self

wonC

II

find,find.1s

dassthat

d <the

Art,way

wiehow

dethe

Peteri

Petersj

itvermarktet>,promotes

gruusigdisgusting

isch.is

. For instance, nobody asks why Hebrew employs resumptives in relativization. Rather, re-sumption is an option in the language and is employed in some parts of the A’-syntax. Once weaccept the fact that resumption is an option in ZG (and Standard German), the resumptive pro-lepsis analysis can be fruitfully extended to other constructions like Copy-Raising. See Salzmann(2006b) for discussion.

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.29 (1822-1880)


lit.: ‘This is the book about himselfi that I think that the way Peteri

promotes it is disgusting.’c. De

the[Abschnittperiod

voof

simi

hisLäbe]j ,life

wonC

iI

glaub,think.1s

dassthat

< dthe

Erfaarige,experiences

woC

jedeevery

Buebi

boydej-biit-with

macht,>makes

seervery

unterschidlichdifferent

sind,are

ischis

dthe

Pubertät.puberty

lit.: ‘The period of hisi life that I think that the experiences that every boyi

makes during it, are very different, is puberty.’

Even more spectacularly, there is evidence for reconstruction into intermediatepositions:

(60) sthe

[Buechbook

überabout

siichi/k]j,self

wonC

iI

glaub,think.1s

dassthat

dethe

Hansi

John< d

theArt,way

wiehow

dethe

Peterk

Petersj

itvermarktet, >promotes

gruusigdisgusting

findt>, . . .finds

lit.: ‘the book about himselfi/k that I think that Johni finds the way Peterk

promotes it disgusting’

This example is ambiguous and suggests that there are several copies of the externalhead inside the complement clause, one of them in a CNPC island. This impliesthat there is successive-cyclic movement out of the island. Even if there is a way ofdoing reconstruction without movement, cases like (60) will remain unaccountedfor and call for a movement approach.

A precise explanation of why movement out of islands is possible in this case isbeyond the scope of this paper (but see Salzmann 2006b for detailed discussion).I will simply offer a tentative idea: It seems necessary to attribute the possibilityto void locality constraints to the resumptive. I would like to suggest that an overtpronoun repairs an otherwise illicit chain (just like PF-deletion in sluicing canrepair deficient chains, cf. Merchant 2001). This implies that (at least some aspectsof) locality are checked at PF. The fact that it is normally the lowest copy thatis spelled out appears problematic because this is not always the offending copy.Quite often, the offending copy is higher up, the position from which movementout of the island takes place (i.e. normally a Spec, CP position). In other words, aresumptive does not repair a particular copy, but the entire chain. The fact that thelowest copy is chosen arguably follows from some principle that favors structureswhich are treated identically by both the PF and the LF interface (similar to thenotion “Minimize Mismatch” in Bobaljik 2002): The lowest copy is always relevantfor theta-role assignment and in many cases for the interpretations of restrictionsof quantifiers.

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.30 (1880-1947)

Martin Salzmann

Importantly, the possibility to save island violations with a resumptive is re-stricted by the constraint introduced above that prevents spelling out more thanone chain link. Accordingly, island violations can only be overcome in resump-tive prolepsis and comparatives (see Salzmann 2006b for discussion of the latter),but not in wh-movement or topicalization (see Merchant 2004 for similar reason-ing).39

. Resumptive prolepsis in Standard German and Dutch

I mentioned in 2.3 that long relativization in ZG instantiates a more abstract ver-sion of a construction that functions as an alternative to long-distance relativiza-tion in both Standard German and Dutch. In that construction, the proleptic ob-ject (in form of a relative pronoun) is governed by the preposition von/van ‘of ’ andin German sometimes by bei ‘at’.40 The of -PP is base-generated in the matrix clauseand undergoes short A’-movement. There is no locative relative complementizerso that deletion of the preposition would be irrecoverable and as a consequence itis retained. In the complement clause, we find a resumptive, as in ZG:41

(61) a. eina

Maler,painter

vonof

demi

who.datichI

glaube,think

dassthat

PetraPetra

ihni

himmaglikes

‘a painter who I think that Petra likes’

. I should point out that the facts discussed here are also directly compatible with Boeckx’(2003) model where island-insensitivity is a side-effect of resumption. A detailed evaluation ofthat complex approach is unfortunately beyond the scope of this paper. See Salzmann (2006b)for discussion.

. It is not fully clear what governs the distribution of von and bei in German. They tend to bein complementary distribution, bei e.g. being the preference with reflexives, but due to a lot ofspeaker variation, the picture is somewhat murky. See Salzmann (2006b) for detailed discussion.

. In German, regular long-distance relativization is unacceptable for most speakers. InDutch, the situation is less clear. For many speakers, both constructions are equally availablewhile for others there is a certain preference for the proleptic construction. There is some evi-dence (Lühr 1988:79) that prescriptive pressure in the 18th and 19th century gave rise to thisconstruction in German. Long A’-movement (referred to as Satzverschränkung ‘sentence inter-leaving’) was considered illogical in those sources. Nowadays, the construction is pretty muchgrammaticalized and extremely frequent. It is surprising in this light that ZG uses a construc-tion that was forced by prescriptivists since it is well-known that dialects are normally immuneto such pressure. It is indeed the case that in other German dialects long relativization (with orwithout relative pronouns and without resumptives) is unproblematic, cf. e.g. Swabian or Hes-sian German (Schmitt 2005). Interestingly, however, wo-relativization of the ZG type is morewidespread than is usually thought. In less formal registers of Standard German, it is a frequentalbeit stigmatized alternative to the of -version of the proleptic construction:

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.31 (1947-2002)


b. hetthe

boekbook

waari-vanwhich-of

ikI

denkthink

datthat

PietPeter

heti

itleukcool

vindtfinds

‘the book I think Peter likes’

Since a full discussion of the German and Dutch facts is beyond the scope of thispaper (but see Chapter 3 of Salzmann 2006b) I will simply illustrate the two cru-cial paradoxical properties of the construction: opacity of the CP-complement andreconstruction into that complement. The first pair illustrates reconstruction forvariable binding and Principle A:

(62) a. Diethe

[Periodeperiod

seinesi

his.genLebens]j ,life.gen

vonof

derj

whichichI

glaube,believe

dassthat

keineri

no.one

gernelikes.to

darj-anabout-it

denkt,thinks

istis

diethe

Pubertät.puberty

‘The period of hisi life I think no onei likes to remember is puberty.’b. das

the[Bildpicture

vonof

sichi]j,self

vonof

demj

whichichI

glaube,believe

dassthat

Peteri

Peteresj

itmaglikes

‘the picture of himselfi that I think Peteri likes’

The following pair illustrates argument and adjunct extraction from the comple-ment CP:

(63) a. IchI

glaubebelieve

vonof

Hans,John

dassthat

erhe

jedemevery.dat

Schülerstudent

gutegood

Notengrades

gibt.gives

‘I believe of John that he gives every student good grades.’b. ??Welchem

whichSchüler1

studentglaubstthink

duyou

vonof

Hans,John

dassthat

erhe

t1 gutegood

Notengrades

gibt?gives‘Which student do you think John gives good grades?’

c. *Warum1

whyglaubstthink

duyou

vonof

Hans,John

dassthat

erhe

t1 PeterPeter

gutegood

Notengrades

gibt?gives

‘Why do you think John gives Peter good grades?’

(i) Binam

jetztnow

inin

AmerikaAmerica

[jemand]somebody

aufon

derthe

Spur,trace

wowo

ichI

vermute,suspect

dassthat

erhe

ebenfallsalso

mitwith

meinermy

Familiefamily

verwandtrelated

seinbe

könnte.could

‘I am now tracing someone in America who I suspect could be related to my family.’www.wer-weiss-was.de/theme49/article767487.html

This implies that the wo-strategy is probably used in the entire German speaking area and canbe considered a generally unmarked strategy for long-distance relativization. Its appearance inZG is therefore not problematic: ZG simply uses an unmarked option.

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.32 (2002-2067)

Martin Salzmann

. Conclusion

Long relativization in ZG is a particularly interesting construction because ithas paradoxical properties. There are reconstruction effects into the complementclause, but at the same time, there is clear evidence that the complement clauseis a barrier and that the proleptic object originates in the matrix clause. The re-sumptive prolepsis approach presented here manages to reconcile these conflictingproperties. Operator movement in the complement clause turns the complementinto a predicate and licenses an extra argument, the proleptic object. This predi-cation analysis makes an alternative reconstruction strategy available as in tough-movement and accounts for the opacity of the complement. The link between theproleptic object and the operator in the complement clause is an ellipsis operation.Together with concomitant Vehicle Change effects this nicely explains the intricateCondition C pattern in both the proleptic construction and in tough-movement.

On a more theoretical level, this approach suggests a straightforward wayof handling such exceptional and hitherto ill-understood cases of reconstructionwithin a theory that makes crucial use of full copies of the antecedent. It unifies re-sumptive prolepsis with tough-movement in crucial respects and thereby providesa fresh look at the latter.

Acknowledgments

I would like the thank the audience at CGSW 20 for helpful discussion, especiallyJosef Bayer, Marcel den Dikken, Henk van Riemsdijk, and Jan-Wouter Zwart. Thedevelopment of many of the ideas presented here has greatly benefited from dis-cussions with Rajesh Bhatt, Lisa Cheng, Henk van Riemsdijk, Johan Rooryck andKathrin Würth. I am very much indebted to Jürg Fleischer, Kathrin Würth, andTobias Zimmermann for providing detailed judgments. Finally, I would like tothank the two anonymous reviewers whose corrections and suggestions have leadto substantial improvement of the paper. All remaining errors are mine.

References

Aoun, J., Choueiri, L., & Hornstein, N. (2001). Resumption, movement, and derivationaleconomy. Linguistic Inquiry, 32, 371–403.

Barss, A. (1986). Chains and Anaphoric Dependencies. PhD dissertation, MIT.Bayer, J., Bader, M., & Meng, M. (2001). Morphological underspecification meets oblique Case:

Syntactic and processing effects in German. Lingua, 111, 465–514.Bhatt, R. (2002). The raising analysis of relative clauses: Evidence from adjectival modification.

Natural Language Semantics, 40, 43–90.

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.33 (2067-2203)


Bianchi, V. (1999). Consequences of Antisymmetry. Berlin: Mouton de Gruyter.Bianchi, V. (2004). Resumptive relatives and LF chains. In L. Rizzi (Ed.), The Cartography of

Syntactic Structures. Volume 2 (pp. 76–114). Oxford: OUP.Bobaljik, J. D. (2002). A-chains at the PF-interface: Copies and covert movement. Natural

Language and Linguistic Theory, 20, 197–267.Boeckx, C. (2003). Islands and Chains. Resumption as stranding. Amsterdam: John Benjamins.Chao, W. & Sells, P. (1983). On the interpretation of resumptive pronouns. NELS, 13, 47–61.Chomsky, N. (1995). The Minimalist Program. Cambridge MA: The MIT Press.Cinque, G. (1990). A’-dependencies. Cambridge MA: The MIT Press.Citko, B. (2001). Deletion under identity in relative clauses. NELS, 31, 131–145.Dieth, E. (1938). Schwyzertütschi Dialäktschrift. Leitfaden. Zürich: Orell Füssli Verlag.Dieth, E. & Schmid-Cadalbert, C. (1986). Schwyzertütschi Dialäktschrift. Dieth-Schreibung. (2.

Aufl. revised and edited by Christian Schmid-Cadalbert). Aarau: Sauerländer.Dikken, M. den & Mulder, R. (1992). Tough parasitic gaps. NELS, 22, 303–317.Diesing, M. (1992). Indefinites. Cambridge MA: The MIT Press.Fiengo, R. & May, R. (1994). Indices and Identity. Cambridge MA: The MIT Press.Fischer, S. (2004). Towards an Optimal Theory of Reflexivization. PhD dissertation, University

of Tübingen.Fleischer, J. (2003). A typology of relative clauses in German dialects. In B. Kortmann (Ed.),

Dialectology meets Typology. Dialect grammar from a cross-linguistic perspective (pp. 211–243). Berlin: Mouton de Gruyter.

Fox, D. (1999). Reconstruction, binding theory, and the interpretation of chains. LinguisticInquiry, 30, 157–196.

Heck, F. (2005). Gegen Kopfanhebung in deutschen Relativsätzen. Handout: GGS Tübingen.Heycock, C. (1995). Asymmetries in reconstruction. Linguistic Inquiry, 26, 547–570.Heycock, C. (2003). On the interaction of adjectival modifiers and relative clauses. Ms.:

University of Edinburgh.Kathol, A. (2001). On the nonexistance of true parasitic gaps in standard German. In P. W.

Culicover & P. M. Postal (Eds.), Parasitic Gaps (pp. 315–338). Cambridge MA: The MITPress.

Kiss, T. (2003). Die Genese der Ausnahmemetapher. Ms., University of Bochum.Lakoff, G. (1974). Syntactic amalgams. Papers from the 10th Meeting of the Chicago Linguistic

Society.Lechner, W. (1999). Comparatives and DP-Structure. PhD dissertation, University of Massa-

chusetts, Amherst.Lühr, R. (1988). Zur Satzverschränkung im heutigen Deutsch. GAGL, 29, 74–87.McCloskey, J. (1990). Resumptive pronouns, A’-binding, and levels of representation in Irish.

In R. Hendrick (Ed.), the Syntax of Modern Celtic Languages (pp. 199–248). New York NY:Academic Press.

Merchant, J. (2001). Variable Island Repair under Ellipsis. Ms., University of Chicago.Merchant, J. (2004). Resumptivity and non-movement. Studies in Greek Linguistics, 24, 471–481.Munn, A. (1994). A minimalist account of reconstruction asymmetries. NELS, 24, 397–410.Nunes, J. (2001). Sideward movement. Linguistic Inquiry, 32, 303–344.Pesetsky, D. (1998). Some optimality principles of sentence pronunciation. In P. Barbosa, D. Fox,

M. McGinnis, & D. Pesetsky (Eds.), Is the Best Good Enough? Optimality and competition insyntax (pp. 337–383). Cambridge MA: The MIT Press.

Rezac, M. (2004). On tough movement. lingBuzz/000045.Reinhart, T. & Reuland, E. (1993). Reflexivity. Linguistic Inquiry, 24, 657–720.

JB[v.20020404] Prn:24/10/2006; 14:30 F: LA9707.tex / p.34 (2203-2266)

Martin Salzmann

Riemsdijk, H. van (1989). Swiss relatives. In D. Jaspers, W. Klooster, Y. Putseys, & P. Seuren(Eds.), Sentential Complementation and the Lexicon (pp. 343–354). Dordrecht: Foris.

Riemsdijk, H. van (to appear). Identity avoidance. In R. Freidin, C. Otero, & Maria-LuisaZubizarreta (Eds.), Festschrift for Jean Roger Vergnaud. Cambridge MA: The MIT Press.

Romero, M. (1998). Focus and Reconstruction Effects in WH-Phrases. PhD dissertation,University of Massachusetts.

Rouveret, A. (2002). How are resumptive pronouns linked to the Periphery? In P. Pica & J.Rooryck (Eds.), Linguistic Variation Yearbook, Volume 2 (pp. 123–184). Amsterdam: JohnBenjamins.

Safir, K. (1999). Vehicle change and reconstruction in A’-chains. Linguistic Inquiry, 30, 587–620.Salzmann, M. (2005). Long A’-movement in resumptive disguise. Resumptive Prolepsis

constructions in German and Dutch. Talk given at GLOW 2005, Geneva.Salzmann, M. (2006a). Resumptive pronouns and matching effects in Zurich German relative

clauses as distributed deletion. Leiden Papers in Linguistics, 3(1), 17–50.Salzmann, M. (2006b). Resumptive Prolepsis. A study in indirect A’-dependencies [LOT Disser-

tation Series 136]. Utrecht: LOT.Sauerland, U. (1998). The Meaning of Chains. PhD dissertation, MIT.Sauerland, U. (2003). Unpronounced heads in relative clauses. In K. Schwabe & S.

Winkler (Eds.), The Interfaces: Deriving and interpreting omitted structures (pp. 205–226).Amsterdam: John Benjamins.

Schmitt, V. (2005). Hessische Relativsätze. Handout GGS 2005, Tübingen.Vries, M. de (2002). The Syntax of Relativization. Utrecht: LOT.Weber, A. (1964). Zürichdeutsche Grammatik: Ein Wegweiser zur Guten Mundart. Zürich:

Schweizer Spiegel Verlag.Wilder, C. (1991). Tough movement constructions. Linguistische Berichte, 132, 115–132.

JB[v.20020404] Prn:24/08/2006; 12:00 F: LA97P3.tex / p.1 (47-74)

Historical studies

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.1 (53-172)

Auxiliary selection and counterfactualityin the history of English and Germanic*

Thomas McFadden and Artemis AlexiadouUniversität Stuttgart

The retreat of be as perfect auxiliary in the history of English is examined. Cor-pus data are presented showing that the initial advance of have was most closelyconnected to a restriction against be in past counterfactuals. Other factors whichhave been reported to favor the spread of have are either dependent on the coun-terfactual effect, or significantly weaker in comparison. It is argued that the effectcan be traced to the semantics of the be perfect, which denoted resultativity rath-er than anteriority proper. Related data from other older Germanic and Romancelanguages are presented, and finally implications for existing theories of auxiliaryselection stemming from the findings presented are discussed.

. Introduction

In earlier stages of its history, English used both have and be as auxiliaries toform the perfect.1 In Old English and early Middle English (henceforth OE andME), the choice between the two was determined primarily by the properties ofthe main predicate, much as in the other older Germanic languages, as well asmodern German, Dutch and Italian. This is exemplified by the two ME examples

* We would like to thank Jonny Butler, Dave Embick, Susann Fischer, Sabine Iatridou, TonyKroch, Florian Schäfer and the audiences at PLC 29, WCCFL 24 and CGSW 20 as well as twoanonymous reviewers for helpful comments and suggestions on previous versions of the workpresented here. Special thanks are due to Florian Janner for his indispensable assistance with theresearch. This work was funded by DFG grant no. AL 554/3-1, awarded to the second author.

. We write have and be in small caps rather than italics to make clear that we are talking herenot about the specific forms from a particular stage of English, but about the perfect auxiliariesmore generally. I.e. have is a cover term for Old English habban, Modern English have andModern German haben and their various finite forms, while be is a cover term for Old Englishbeon/wesan, Modern English be, and Modern German sein and their finite forms, etc.

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.2 (172-235)


in 1.2 In example (1a), the non-agentive, change-of-state verb fall takes be, likeGerman fallen. By contrast, in (1b), the agentive activity verb fight takes have, likeGerman kämpfen.

(1) a. aswhen

hathey

þreothree

werenwere

ifolenfallen

onslepe. . .asleep. . .

‘When the three of them had fallen asleep. . . ’ (CMANCRIW-2,II.272.440)b. . . . huanne

. . . whenhihe

heþhas

welwel

yuoŠtefought

‘. . . when he has fought well’ (CMAYENBI,252.2315)

In the course of the ME period, have first began to show up with verbs thatpreviously only took be. According to previous studies, have was favored at thistime especially in various modal and irrealis contexts, past perfects, infinitive andprogressive perfects, negatives, and iterative and durative contexts (see especiallyRydén & Brorström 1987; Kytö 1997).3 During the Early Modern English period(henceforth EModE), be was increasingly restricted to the most common intransi-tives come and go, before being pushed out here as well over the course of the 18thand 19th centuries. By around 1900, the modern situation was reached, wherehave is the universal perfect auxiliary, and be appears only in relics.4

This development raises a number of questions, both historical and theoreti-cal. First, why did have start spreading at the expense of be in the first place? Inother words, what made English different from German, Dutch and Italian, wherebe has been retained in full force up to the present? Second, what is the relevanceof the factors identified by previous researchers which we noted above? Which ofthem really had the strongest effects, how are they related to each other, and whydid they favor have in the first place? Third, why did the change take on the or-

. The data for this chapter come from the York-Toronto-Helsinki Parsed Corpus of Old EnglishProse (Taylor et al. 2003), the Penn-Helsinki Parsed Corpus of Middle English, 2nd edition (Kroch& Taylor 1999) and the Penn-Helsinki Parsed Corpus of Early Modern English (Kroch et al. 2005)The final line of each example gives the sentence ID as it appears in the original corpus file.

. Notable earlier discussions of the change include Hoffmann (1934), Fridén (1948), Johannis-son (1958), Mustanoja (1960) Traugott (1972), Zimmermann (1973), Kakietek (1976). Most ofthe have-favoring factors mentioned in the text were already identified by one or more of thesescholars. However, we will mostly discuss the details reported by (Rydén & Brorström 1987) andKytö (1997) because their studies were based on modern techniques of corpus-based research.Thus their claims are based on explicitly reported and replicable numbers.

. E.g., be is retained in fixed phrases from the (archaic/archaizing) Christian liturgy like Christis risen. Its productive occurrence with gone cannot be interpreted as a true holdover of its useas a perfect auxiliary. In this usage, gone is a (lexicalized) stative adjective, not a real perfect par-ticiple. Evidence for this is that be is not possible with gone in unambiguously eventive contexts(They *are/have gone to the store) and of course the fact that it is only go that behaves this way.

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.3 (235-281)

Auxiliary selection and counterfactuality in older English

der of 800 years to go to completion? Was it really a single, monolithic change, orare we dealing with a series of potentially independent developments which con-spired to erode away the use of be? Fourth, what implications does this changehave for general theories of auxiliary selection? Can we integrate the diachronicdata from English into accounts designed to deal with synchronic data from Ger-man, Dutch and Italian, or will the English data force us to revise or even abandonsuch accounts?

The first question above will serve as our point of departure. We will look atthe first appearance of have with the verb come – one of its earliest clearly iden-tifiable advances onto territory previously held by be – and find that it relates toa ban on be in past counterfactuals. Our examination of this counterfactual effectwill lead us to at least partial answers to the other questions just posed. We willpropose an account for the effect based on a difference in the semantic status ofbe and have, framed in terms of Iatridou’s (2000) Exclusion theory of counterfac-tuals, and show how it relates to other factors that influenced auxiliary selection.We will present evidence on the chronology of the changes involved, showing thatthe spread of have was in fact not a single change and offering a proposal for whythe counterfactual effect became relevant when it did. Finally, we will present datafrom other languages that show similar effects and lay out the problems that sucheffects present for existing theories of auxiliary selection.

. The first appearances of have with come

For a number of reasons, come is an ideal verb to focus on when investigating theloss of be as a perfect auxiliary in English. First, among the verbs that can take bein the perfect, it is by far the most frequent, which means that we can get largeenough numbers to allow basic statistical analysis.5 Second, clauses containing thepast participle of come with a form of be are unambiguous perfects in all peri-ods. Since come has no transitive uses, there are no formally identical passives, andthere is no uncertainty like we find with he is gone, as come never develops a lexi-calized adjectival reading. Third, and perhaps most importantly, come selected becategorically in the earliest historical periods of English. This means that we canidentify as an innovation the point in time when come first starts showing up withauxiliary have. Indeed, this is one of the clearest indicators we have for when theloss of be in favor of have first got underway. By examining the earliest examples

. For example, in our ME corpus, of the 676 perfects with verbs that take be at least once, 273have come. Next in frequency are go with 107 and fall with 39.

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.4 (281-355)


Table 1. Auxiliary selection with come

OE ME EModE1150– 1250– 1350– 1420– 1500– 1570– 1640–1250 1350 1420 1500 1569 1639 1710

BE 93 64 11 97 75 138 200 72HAVE 0 1 0 14 11 20 32 25Total 93 65 11 111 86 158 232 97% HAVE 0% 1.5% 0% 12.6% 12.8% 12.7% 13.8% 25.8%

of come with have, we may then be able to get an idea of why the change began inthe first place.

Table 1 shows the incidence of the two auxiliaries with come throughout thetime covered by the three corpora.6 As alluded to above, be is obligatory with comein OE, and nearly so through the first half of ME. have first appears in significantnumbers quite suddenly in the third ME period (1350–1420). Its frequency jumpsimmediately to around 13%, but then stays at essentially that same level for severalcenturies, well into EModE. In the third and final period of the EModE corpus(1640–1710), have again becomes more common, but even still has only about a26% share.

Some observations are possible purely on the basis of these numbers. First, itlooks like we can put the beginning of our change in the 3rd period of ME. This iswhere have first shows up with come in significant numbers, and this is where wewill want to first focus our attention, to see if there is anything notable about theexamples with have.7 Second, we have the beginnings of an answer to the thirdquestion posed above about why the loss of be took so long to complete. Appar-ently, it was not a single gradual change, but rather a series of discrete changes,each increasing the frequency of have by a small amount. Specifically, somethinghappened around 1350 that first made have possible with come. After this, thingswere stable again for a few hundred years, before something else happened around1650 causing another jump in the use of have. We do not yet have reliable data forthe period after 1710 (Late Modern English), so we cannot say whether the subse-quent development was a single gradual rise in the frequency of have or a seriesof further discrete changes. In any case, we have confirmation that – at least with

. In the table, the ME and EModE data are grouped according to the periods distinguished bythe corpora. The three periods distinguished by the OE corpus are collapsed, since there is nodevelopment to be seen there.

. Actually, it may be that the change began in the second period. However, the M2 portionof the corpus is quite small in general, and perfects with come are particularly rare (note thefluctuation from 71 examples in M1, down to 11 examples in M2 and back up to 116 examplesin M3), so we do not have enough data to tell.

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.5 (355-444)


Table 2. Perfects of come with HAVE by modality

ME EModE1350–1420 1420–1500 1500–1569 1570–1639 1644–1710

Counterfactuals 5 9 7 9 7Modals 4 1 2 8 4Other 5 1 11 15 14Total 14 11 20 32 25% Ctf/Modal 64.3% 90.9% 45% 53.1% 44%

come – the replacement of be by have was still far from complete at the beginningof the 18th century.

What was it, then, that caused have to first start showing up with come inperiod M3? As noted above, several factors have been identified by previous re-searchers as favoring the use of have over be, but the first examples of come withhave mostly fall into a single category. Of the 14 clauses in question, 9 are in whatwe might call modal contexts. 5 are counterfactuals, like (2a), while 4 have overtmodal auxiliaries above a non-finite form of have, like (2b) and (2c).8

(2) a. Andand

ifif

þowyou

hadesthad

comecome

betyme,timely

hehe

hadehad

yhadehad

þethe

maistremaster

‘And if you had come in time, he would have prevailed.’(CMBRUT3,227.4102)

b. . . . she shulde nouŠt haue comen in his sight bi his wille‘. . . she would not have come into his sight by his will.’(CMBRUT3,115.3483)

c. . . . syþ þei myton liŠtly haue come to blysse‘since they might easily have come to bliss’(CMWYCSER,303.1386)

Indeed, the frequency of counterfactuals and modals among perfects of come withhave remains remarkably high throughout the time covered by our corpora, asshown in Table 2. Each column of the table splits up the perfects of come with

. Both Arnim von Stechow and an anonymous reviewer have pointed out that, from a semanticpoint of view, in some of these examples the modal seems to scope under the perfect auxiliary.Thus we would have in effect a perfect of the modal, not a perfect of come, in which case theappearance of auxiliary have would be unremarkable. However, the morphosyntax of thesesentences clearly places the modal above the perfect auxiliary, and a process which would derivethis from the opposite underlying configuration would violate widely-assumed syntactic prin-ciples (see Condoravdi 2002; Stowell 2004: for discussion of essentially the same data in ModernEnglish). What exactly is going on here to derive the apparent mismatch is not entirely clear, butbarring a convincing analysis to the contrary, we will continue to assume that examples of thiskind involve a perfect of come, and are thus of interest.

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.6 (444-510)


have for a period into three categories: those with counterfactual semantics, thosewith a modal and all others, and indicates what percentage the counterfactual andmodal sentence together make of the total. As we will see below, counterfactualand modal perfects are nowhere near this common overall, so it would seem thatthere is a connection between these contexts and the appearance of have. Whatis at first somewhat baffling is why it is not until the third period of ME that westart finding come with have in such contexts. If modality favored the use of haveat this time, why didn’t it do so earlier? This is a question that we will need toaddress if we want to have an explanation for the start of the loss of be in theseenvironments.

A surprisingly clear answer is available. Perfects with counterfactual meaningor a modal above the auxiliary turn out to have been extremely rare in early ME,as shown in Table 3.9 The top row of each column of the table gives the total num-ber of intransitive perfects. The next two rows then indicate how many of thesehave an overt modal and how many have a counterfactual interpretation, alongwith in each case what percent of the total this represents. With come specifically,perfects with modals and counterfactuals are even more rare early on, as shown inTable 4.10 So the reason why we don’t find examples like would have come or if hehad come before around 1350 is that there are very few examples at that time ofmodal and counterfactual perfects at all.11 In other words, we do not have wouldhave come replacing earlier would be come. Indeed, we will see below that the latterwas never actually possible. Rather, it is an innovation here that perfects are being

Table 3. Modal and counterfactual pexfects with all intransitives

1150–1250 1250–1350 1350–1420 1420–1500

Intrans. perfects 294 145 794 565Modals 3 (1%) 7 (4.8%) 54 (6.8%) 66 (11.7%)Counterfactuals 5 (1.7%) 7 (4.8%) 91(11.5%) 83 (14.7%)

. Our investigation of OE on this point is currently in progress. There are no perfects ofcome with modals in the OE corpus, and we have not yet found such examples with any otherintransitive verbs. See below for discussion of the counterfactuals.

. This lag seems to be due at least in part again to the small number of texts from the M2period and the low occurrence of come in the few texts that we have.

. Modals were far more common outside the perfect. Counting clauses of all types, the rateof occurrence of modals for the four periods of ME was 1150–1250 10.0%; 1250–1350 12.5%;1350–1420 11.5%; 1420–1500 11.7%. The differences between these numbers and the corre-sponding numbers for perfect clauses in Table 3 are statistically significant for the first threeperiods (M1: χ2 26.4, p < .001, M2: χ2 7.7, p < .01, M3: χ2 17.0, p < .001), but not the fourth, bywhich time the modal perfect was fully established.

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.7 (510-623)


Table 4. Modal and counterfactual perfects with come

1150–1250 1250–1350 1350–1420 1420–1500

Perfects of come 65 11 111 86Modals 0 (0%) 0 (0%) 4 (3.6%) 1 (1.2%)Counterfactuals 2 (3.1%) 0 (0%) 5 (4.5%) 9 (10.4%)

used in modal contexts at all. In OE and early ME, modal and counterfac-tual clauses were expressed with simple past subjunctive forms, potentially withmodals, but without any perfect morphosyntax (see e.g. Mitchell 1985:85), as inthe following OE example (from Molencki 2000):

(3) acbut

hitit

wærewere

totoo

hrædlicquick

gifif

hehe

þathen

onon

cildcradolechild-cradle

acwealdkilled

wurdewere

‘but it would have been too early if he had been killed in his cradle then’(ÆCHom i.82.28).

This change is just one part of the general expansion of the English auxiliary systemand the spread of the perfect into new semantic contexts (see e.g. Traugott 1972;Warner 1993; Elsness 1997 for discussion of these developments).

This provides us with the beginnings of an understanding of our change. In thefirst half of ME, perfects first start being used in counterfactual and modal clauses,as part of the general expansion of the auxiliary system. At the same time, we seeauxiliary have showing up with come, a verb that previously appeared only withbe. A comparison of Tables 2 and 3 shows that these two innovations are related bymore than just relative chronology. The frequency of counterfactuals and modalsamong perfects of come with auxiliary have is far higher than it is among otherperfect intransitives: between 1350 and 1420, 64.3% of have + come perfects arecounterfactuals or modals, compared with only 18.3% of other intransitive per-fects; between 1420 and 1500 the frequencies are 90.9% and 26.4%.12 Thus we canhypothesize that this initial spread of have was in fact due (at least in part) to theincreasing use of perfects in modal contexts. As noted in the Introduction, the factthat modals and counterfactuals favored have in early English has been reportedby previous researchers (see Fridén 1948; Mustanoja 1960; Traugott 1972; Rydén &Brorström 1987; Kytö 1997; Lipson 1999). However, the tight relationship betweenthe first appearance of such contexts in the perfect and the very first advances ofhave has not to our knowledge been made explicit until now.

. The percentages reported here for ‘other intransitive perfects’ are slightly lower than whatone would get by adding the percentages for modals and counterfactuals in Table 3, because thelatter are for all intransitive perfects, while the former exclude the perfects of come with have,since it is their behavior we are comparing. The differences are again statistically significant: forM3 χ2 = 20.2, p < 0.001; for M4 χ2 = 24.1, p < 0.001.

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.8 (623-663)


. Isolating the counterfactual effect

In order to further develop our tentative explanation for the first stage of the lossof be, we need a better understanding of the effect that modal contexts had on aux-iliary selection. To begin with, we need to more precisely characterize what ‘modalcontext’ means. To this point, we have been using this term to refer collectivelyto what look like two distinct types of perfect clauses – those with counterfactualsemantics and those with an overt modal above the perfect auxiliary. We have keptthe two clause types separate because they are identified in the corpus on the ba-sis of different criteria, but have treated them together because they seem to havesomething in common. However, if we want to figure out how and why these fac-tors really affect auxiliary selection, we will have to figure out what unifies themand distinguishes them from others which one might expect to also fall into thecategory of ‘modal’.

To begin with, let us clarify exactly what we mean by ‘counterfactual’. Intothis category we place only those clauses where the implication is clearly that theproposition being considered does not (or did not) hold. This includes the fol-lowing types, exemplified by the sentences in (12b):13 counterfactual conditionals,both the antecedent clause (4a) and the consequent clause (4b); clauses which haveessentially the function of the consequent of a counterfactual conditional, but haveno conditional antecedent ((4c), where else essentially means something like if Iwere not satisfied); and counterfactual wishes (4d).

(4) a. and if they had come sooner, they could haue holpen them.‘and if they had come sooner, they could have helped them.’(GIFFORD-E2-P2,G3V.246)

b. he had never come to himself . . . if he had not met with this allay‘he would never have come to himself . . . if he had not met with this dis-traction’(BEHN-E3-H,189.165)

c. I am satisfy’d with every thing that pleases you; else I had not come to Townat all.‘I am satisfied with everything that pleases you; otherwise I wouldn’t havecome to town at all.’(VANBR-E3-H,32.10,11)

. The examples are taken from EModE because it is easier to understand than ME and is thusbetter suited for demonstration purposes. For consistency, all the examples have come as themain verb, with the exception of (4d), where an example with go is used because there are nocounterfactual wishes with come in the EModE corpus.

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.9 (663-788)


d. And he . . . will wish he had with the poore peoples children gon barefoot.‘And he . . . will wish he had gone barefoot with the poor people’s chil-dren.’ (LOCKE-E3-P1,35.46)

Now, as it turns out, a division between counterfactual clauses and clauses withmodals is unwarranted for ME. Consider again the sentences exemplifying perfectswith modals above, repeated here in (5).

(5) a. . . . she shulde nouŠt haue comen in his sight bi his wille‘. . . she would not have come into sight by his will.’(CMBRUT3,115.3483)

b. . . . syþ þei myton liŠtly haue come to blysse‘since they might easily have come to bliss’(CMWYCSER,303.1386)

Sentence (5a) is talking about a situation where the woman in question has comeinto the man in question’s sight, thus the statement that she would not have doneso by his will is a counterfactual. Similarly, sentence (5b) is about the damned, i.e.people who have not come to bliss, but could have easily done so if they had be-haved properly. Again, it is a counterfactual. Indeed, all of the clauses containingmodals above intransitive perfects in our ME corpus turn out to have counterfac-tual semantics like this. In other words, at this stage we don’t yet find exampleslike He must have come to London or She may have gone home, where the modal ex-presses some epistemic meaning and is not counterfactual. It is not entirely clearwhy this should be.14 In any case, this allows us to hypothesize that what favoredhave was specifically counterfactuality, and not an ill-defined modality categorythat subsumes counterfactuals and clauses with modals. If this is correct, thenonce non-counterfactual perfects with modals start to show up in the language,we might expect them to be able to take auxiliary be. Such sentences do beginto appear (though still in small numbers) in EModE, and as predicted, we findauxiliary be with a few of them:15

(6) a. Your Mother, you know, will be gone to Church.(FARQUHAR-E3-P2,24.173)

b. My Spouse will be got to the Ale-house with his Scoundrels.(FARQUHAR-E3-P2,24.174)

. We can speculate that this is just another facet of the process by which the modern auxiliarysystem was gradually constructed. At first no modals were possible above the perfect, then onlya restricted type which expressed little more than counterfactuality, and only later the full arrayof epistemic modals.

. Note that we cannot analyze (6a) as an instance of the non-perfect use of be gone familiarfrom Modern English due to the presence of the goal PP to Church.

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.10 (788-807)


Table 5. ME perfect auxiliary selection by modality

BE HAVE % BE

Counterfactuals 3 183 1.6%Modals 0 130 0All other intransitives 540 942 36.4%

Table 6. EModE perfect auxiliary selection by modality

BE HAVE % BE

Non-modal ctf 3 344 0.9%Modal ctf 1 246 0.4%All other intransitives 986 2362 29.5%

Now that we have identified counterfactuality as the relevant factor, we can con-sider just how strong its effect was. In order to do this, we will have to reverse theperspective we took on the data in the last section. Rather than looking at perfectswith have and figuring out how many of them are counterfactual, we must re-strict our attention to the counterfactual perfects and see how the two auxiliariesare distributed among them. The results of this investigation are quite striking.Table 5 covers all ME intransitive perfects – i.e. not just those with come, but withall intransitive verbs – comparing the frequency of the two auxiliaries in coun-terfactual environments with their frequency elsewhere.16 What we find is that beis extremely rare with counterfactual perfects, whether an overt modal is presentor not. It should be noted that an effect this categorical is remarkable for ME,which is otherwise known for showing variation, particularly in areas where thereis change in progress. The effect remains just as strong through EModE, as shownin Table 6.17 The counterfactual effect is thus not just a contributing factor in theselection of perfect auxiliaries, it is the determining factor in those clauses where itis at work.

Recall now that previous researchers have listed several other factors alongwith counterfactuals and modals as favoring the spread of have at the expense ofbe. While much of the relevant work – in particular that of Rydén & Brorström(1987) and Kytö (1997) – achieves a high quality of description, providing de-tailed statistical data on the various influencing factors, we would argue that itfails to explain why the various factors are relevant and how they are related to

. We continue to list clauses with modals separately from the other counterfactuals because, asnoted above, the categories are formally distinct and thus were identified by different methods.

. In Section 4 we will argue that most of the apparent counterexamples here are actuallypresent counterfactuals, and that the counterfactual effect is in fact properly restricted to (thefar more common) past counterfactuals.

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.11 (807-903)


each other. We propose to address this issue by approaching the other factors fromthe perspective of the counterfactual effect. It is reasonable to choose this effect asthe pivot precisely because it is the only one which is anywhere near categorical.Proceeding in this way turns out to be productive, since some of the other factorsare in fact related to counterfactuality. In particular, the higher frequency of havein certain contexts is due at least in part to the fact that these contexts tend toco-occur with counterfactuality.

This is most clearly the case for the pluperfect. Kytö (1997), for example,claims that “[t]he past perfect, which highlights the perfectivity of action, pavedthe way for the rise of have. . . From early on, the use of have is more commonin past perfect than in present perfect constructions” [52f.]. Indeed, the numbersfrom her study of corpora covering the time from Late ME up to the present indi-cate just this, with have showing up in 55% of the past perfect clauses comparedto 47% in the present perfect.18 However, there is the potential here for interfer-ence from the counterfactual effect. Formally speaking, all counterfactual perfectsthat do not involve a modal are in fact pluperfects, with the past participle below apast form of the perfect auxiliary. That is, only If I had gone can be a counterfactualconditional. Something like If I have gone may be a conditional perfect, but can-not be counterfactual.19 Given our finding that counterfactual perfects always takehave, they will skew the overall frequency of have with formal pluperfects. Sincepresent perfects are never counterfactual, what we really want to know is, how dothey compare with non-counterfactual pluperfects? As the examples in 7 show, wedo find come with both be and have in this context.

(7) a. For his tyme was not come to dyen at þe Pasc þat he hadde ordeynot‘For his time had not come to die at the passover that he had ordained.’(CMWYCSER,I,414.3405)

b. For also thei hadden comun to the feeste dai‘For they had also come to the feast day.’(CMNTEST,IV,40.334)

Table 7. ME non-counterfactual present and past perfects

BE HAVE % BE

Pluperfects 331 424 43.8%Present perfects 189 469 28.7%

. The difference in the percentages may seem small, but given the large number of examples(N = 2130), it is highly statistically significant: χ2 = 18.5, p < 0.001.

. Of course, If I went can be a (present) counterfactual conditional as well, but is irrelevant tothe discussion since it is not a perfect. See section 4 for further discussion.

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.12 (903-977)


Table 8. EModE non-counterfactual present and past perfects

BE HAVE % BE

Pluperfects 364 805 31.1%Present perfects 504 1267 28.5%

Table 9. Non-counterfactual perfects, negative vs. non-negative

BE HAVE % BE

ME Negative 12 50 19.4%Non-negative 528 942 35.9%

EModE Negative 58 124 31.9%Non-negative 928 2238 29.3%

The relative frequency of the two auxiliaries with intransitives in non-counterfac-tual clauses in ME is given in Table 7. As it turns out, the pluperfect actuallydisfavors have once we exclude counterfactuals, and the difference is statisticallysignificant (χ2 = 34.5, p < .001). In EModE, the frequency of have is still nominallylower with non-counterfactual pluperfects than with non-counterfactual presentperfects, as shown in Table 8. In this case, however, the difference is not statisticallysignificant (χ2 = 2.43, p < .20). In Section 4 we will make a suggestion as to whybe should have appeared more often in the past than in the present perfect. Whatis relevant for the moment is that it is not correct that the pluperfect itself favoredthe use of have, at least not in the corpora at our disposal.

Negation has also been claimed to favor have, with Kytö (1997) reporting that68% of negative perfects take have, compared to 53% of affirmatives. Again, how-ever, we have to be on the lookout for interference from the counterfactual effect,since negation is common with counterfactuals of the type If Jones hadn’t X hewouldn’t have Y. Indeed, in our ME corpus, we find that 37.4% of negative clausesare counterfactuals, compared to only 15.9% of non-negative clauses. In EModE,32.6% of negatives are counterfactuals, compared to 13.8% of non-negatives.20 Ex-cluding the counterfactuals from the negatives we get the numbers in Table 9. Astatistically significant difference between negative and non-negative contexts re-mains in ME (χ2 = 7.2, p < .01), but in EModE they are essentially identical (χ2 =.54). At least in ME, then, negation does seem to have favored have, though notnearly as strongly as counterfactuality.

Another category which has been claimed to strongly favor have is the perfectinfinitive. Of course, one of the main contexts where perfects show up with non-finite forms of the auxiliary is below modals. In order to avoid interference from

. These differences are again statistically significant. For ME, χ2 = 30.3, p < .001. For EModE,χ2 = 69.6, p < .001.

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.13 (977-1042)


Table 10. EModE perfect auxiliary selection with sirnple infinitives

BE HAVE % BE

Simple infinitives 20 127 13.6%All other intransitives 970 2825 25.6%

the counterfactuality effect, we must exclude these and restrict our attention toexamples like those in (8):

(8) a. to take grete sham & conscyence whan we rede them to haue doon sozelously in goddys cause‘. . . to take great shame and conscience when we read that they have actedso zealously in God’s cause’(CMFITZJA,B1V.108)

b. to make vnable prelatis eithir curatis in the chirche of God, is to haue cometo the hi+gest degree of trespasis‘. . . to make people who are incompetent prelates or curates in the churchof God is to have come to the highest degree of trespasses.(CMPURVEY,I,32.1568)

An independent preference for have remains here as well. Kytö (1997) reports thatin her corpora, have shows up with 87% of the infinitives compared to only 51%of all the other perfects. In our ME corpus, we found a total of 30 non-modal per-fect infinitives, and only one of these has auxiliary be. There thus seems to be astrong effect (though be is not completely ruled out), but with such low numberswe cannot say much more than that. In EModE, simple infinitive perfects becomefar more common, yielding numbers large enough to allow more confident in-terpretation. Again, we find a clear tendency to favor have, as shown in Table 10,though not as drastic as that in ME. The difference between infinitives and all otherintransitives here is statistically significant (χ2 = 10.8, p < .01). Nonetheless, weagain do not have anything like the categorical effect found with counterfactuals.21

. A possible explanation for the numbers here comes from a peculiar type of perfect infinitivefound in older English that is unfamiliar to the Modern language, where the perfect seems to beshowing up in the embedded non-finite clause due to something like a sequence of tense effect:

i. for he was commaundyd to have londyd at Calys by the kynge‘For he was commanded to land at Calais by the king.’(CMGREGOR,206.1781)

A full 17 of the 30 infinitives we have found in ME are of this type. Under the analysis we willpropose in Section 4, it is unsurprising that the be perfect would be inappropriate here, since wewill claim that it does not encode a past semantics in the way that the have perfect does.

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.14 (1042-1113)


Table 11. EModE perfect auxiliary selection with pxesent participle auxiliary

BE HAVE % BE

Progressives 50 87 38.5%All other intransitives 940 2865 24.7%

Finally, Rydén & Brorström (1987) have reported that have was favored also inperfects where the auxiliary is in the form of a present participle, as in the examplesin (9).

(9) a. he approved extremely of your having come away(DRUMMOND-E3-P1,2.4,201.37)

b. and at night being come to the Towne, I found good ordinary Countrey en-tertainment(JOTAYLOR-E2-H,1,128.C2.9)

Here there is no formal connection to counterfactuals, so interference of the kindwe found with modals and infinitives shouldn’t be an issue. However, at leastwithin our corpora, the present participle form seems to favor be, not have. Wefound no examples of this kind in the ME corpus, but there were 137 in EModE,the numbers for which are given in Table 11. Again, the difference shown here isstatistically significant (χ2 = 9.8, p < .01). Kytö (1997) also finds a minor prefer-ence for be during EModE, but in later periods this disappears. Since Rydén &Brorström (1987) also made their claim on the basis of data from Late ModernEnglish, this may partly explain the discrepancy with our findings. In any case,whatever effects do show up here are relatively minor.

To sum up then, there are indeed several factors that correlate with higherfrequencies of have. However, the counterfactual effect stands clearly apart fromthe others, being the only one that is essentially categorical. Negative and infinitiveperfects do tend to use have more than affirmative and finite ones do, but theseare tendencies, not hard and fast rules. Other factors that have been claimed tohave such an effect, like the past perfect and the progressive, turn out not to doso at all, once interference from the counterfactual effect is removed. Our strategy,then, will be to first attempt to explain the counterfactual effect, and then to see ifwhat we learn from that can shed light on the other factors.

. Towards an explanation

Three central points emerge from the discussion thus far which must inform thesearch for a convincing explanation of the counterfactual effect. First, the effect isessentially categorical. This points toward a solution in terms of sharply-defined

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.15 (1113-1184)


syntactic or semantic categories. Second, counterfactuality is of a different typethan the other (primarily lexical) factors involved in auxiliary selection. In struc-tural/scopal terms, things like argument structure, agentivity and lexical aspectare encoded fairly low, presumably within the VP/νP region, while counterfac-tuality is presumably encoded fairly high, probably somewhere in the IP region.Furthermore, counterfactuality overrides these other factors. That is, a counter-factual perfect will take auxiliary have, no matter what the main verb is. Thus itseffect would seem to be operating independently, on a different level than normalselection. Third, languages like German, Italian and Dutch show no sign of theeffect. So whatever we suggest to account for older English must be able to handlethis variation, and should ultimately be relatable to other ways in which (Middle)English perfects differ from perfects in these other languages.

With these points in mind, we would like to propose an analysis that makesuse of Iatridou’s (2000) theory of counterfactuals. Iatridou’s point of departureis the fact that counterfactuality is marked by the same morphology that is usedto encode past tense in languages like English and Greek. Thus for example insentence (10a), the past form had encodes counterfactuality, not a temporal pastinterpretation. That is, the if clause is about having (or not having) a car now, notabout having a car in the past.

(10) a. If she had a car, we could drive to Vegas.b. If she had had a car, we would have driven to Vegas.

If we want to talk about having a car in the past, we need a second layer of pastmorphology, resulting in a pluperfect, as in (10b). In order to account for thesedata, Iatridou proposes that “past” morphology is not directly tied to past seman-tics. Rather, it spells out what she calls an Exclusion feature (ExclF), and this ExclFhas the more abstract semantics given in (11). It encodes an exclusion relationshipbetween some aspect x of the topic and the same aspect x of the utterance. This xcan vary over times and possible worlds, yielding the two instantiations of (11) in(12):

(11) T(x) excludes C(x)

(12) a. The topic time excludes the utterance time.b. The topic worlds exclude the utterance world.

When x ranges over times, we get the past tense interpretation in (12a). That is, thetime interval(s) that are being talked about in the utterance do not include the timeat which the utterance is made. Iatridou argues that this results in a past, becausea future interpretation is unavailable for independent reasons. On the other hand,when x ranges over possible worlds, we get the counterfactual interpretation in(12b). In other words, the world in which the utterance is made is not included inthe set of possible worlds being talked about.

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.16 (1184-1224)


What is relevant for us is that this ExclF which can yield either a past or acounterfactual can only come from the finite tense marking, not from perfect mor-phology. A present perfect form like If she has had a car cannot be used to conveycounterfactuality. This goes for older English as well as the modern language: wehave found no formal present perfects with counterfactual meaning in the cor-pora. This means that in a past counterfactual pluperfect like If she had had a car,it must be the perfect morphology which contributes the temporal past meaning,since the finite tense-marking is handling counterfactuality. Now, since the per-fect is periphrastic, we can ask which morpho-syntactic portion of it is actuallyresponsible for this anteriority, the auxiliary or the past participle. Iatridou et al.(2003) have argued that in Modern English it is the auxiliary, since the participlehas no such past meaning when it occurs independently, e.g. in the passive. Thisalso holds for the ME and EModE passive, thus it is reasonable to assume that inthe older stages of the language it was the auxiliary as well. Now, if the temporalmeaning of the perfect is localized in the auxiliary, it would not be that surprisingfor different auxiliaries to have different temporal properties, i.e. for perfects withbe to be different from those with have in this respect. We would like to proposethat this was in fact the case, and that it is this difference which was responsible forthe counterfactual effect and other restrictions on the older English be perfect.

Consider that the historical source of the be perfect is a resultative participlepredicated of the subject, under a copula (see e.g. Traugott 1972:93, among manyothers). The anteriority in such a construction originally comes by implicationfrom what it means to have a resultative state. I.e. it comes from the meaning of theparticiple, not from any temporal semantic features on the auxiliary.22 Of course,a be + participle construction can grammaticalize and become something otherthan the sum of its original parts. In the familiar modern European languageslike German and French such structures are clearly no longer simple resultativestative constructions, but have come to have more general perfect or even simplepast semantics. German examples like (13a) and the first conjunct of (13b), e.g.,cannot be interpreted as describing resultant states:

. One might object that the same is true of the historical source of the have perfect, and thatwe thus cannot explain the different behavior with have and be in this way. However, whilethis is true for very early stages in the development of the have perfect, it has more clearly un-dergone subsequent grammaticalization away from its historical source than the be perfect has.A sentence like I have worked with an unergative verb cannot be interpreted along the lines ofthe presumed ancestor of the have perfect, something like I have the can opened. On the otherhand, be perfects like I am come, at least at the relevant stages of English, could still be inter-preted compositionally as composed of a resultative stative participle plus copula. See Mustanoja(1960:500) for some related remarks.

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.17 (1224-1279)


(13) a. ErHe

istis

zehnten

Jahreyears

imin.the

Auslandoutland

geblieben.stayed

‘He stayed abroad for ten years.’b. Er

heistis

gegangengone

undand

dannthen

gleichimmediately

wiederagain

zurückgekommen.back.come

‘He left and then came right back.’

There is evidence, however, that the be perfect in ME had not undergone thisdevelopment.

First, a factor noted by other researchers to favor have is the presence of it-erative or durative semantics. We have waited to discuss it until now because itonly makes sense in the light of the idea that the be perfect is necessarily resulta-tive. Iteratives and duratives are about the eventuality expressed by the verb, notits resultant state, so we can expect that they will be incompatible with the be per-fect.23 Second, recall that be shows up at a higher rate in pluperfects than it doesin present perfects in ME, once we abstract away from the counterfactual effect.Let us now consider those data the other way around. Given a particular perfectauxiliary, what is the frequency of the pluperfect vs. the present perfect? Whereasonly 47.5% of perfects with have show a past tense form of the auxiliary (N=893),a full 63.7% of those with be do (N=520). We have not investigated these data inenough detail to say with certainty what is going on here, but the difference canbe explained if the have and be perfects differ in whether or not they introducean anteriority relation. We can imagine that in instances where the pastness of aneventuality had to be made explicit, the simple be perfect did not suffice and hadto be augmented with additional past morphology contributing an ExclF. Withthe have perfect this was not necessary, since have itself could contribute such afeature.

If this assessment of the two perfect auxiliaries is correct for the relevant pe-riod of ME, then the counterfactual effect can be explained. The be + participlestructure simply contains no specification for pastness. The resultativity of theparticiple is sufficient to supply an implication of anteriority in certain contexts,but the construction will not be appropriate in instances where a real past is re-quired. This is of course exactly the situation in a past counterfactual. Consideragain the relevant clause of ex. (2a), repeated as (14):

(14) And if þow hadest come betyme. . .

The finite past tense morphology on hadest supplies the ExclF, which contributescounterfactuality to the interpretation. The past meaning is then contributed by

. See McFadden & Alexiadou (2006) for detailed discussion of the semantics of the olderEnglish be perfect and its interaction with iterativity and durativity.

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.18 (1279-1342)


have itself, and all is well. Consider, however, what would happen with a parallelexample with be. Here we would have the ExclF supplied by the past morphology,but no true past meaning contributed from below. If the ExclF is interpreted tem-porally, this yields the past of a be perfect, i.e. a resultative state in the past, withno counterfactual meaning, as in example (1a) above and (15) below:

(15) And whan nyght was comyn, þe lordes & ladies wente to bedde‘And when night had come, the lords and ladies went to bed.’(CMBRUT3,3.52)

In principle, the ExclF of the past form of be should also be interpretable as acounterfactual instead of a past if our analysis is correct. This would yield thecounterfactual of a be perfect rather than the past of a be perfect. Given our claimsabout the be perfect, this should mean something along the lines of “if you were(now) in the result state of coming”, which is not the same thing as the true pastcounterfactual “if you had come.” Of course, utterances with such a semanticswould only be appropriate under fairly marked circumstances, so we do not ex-pect them to be very common, but they do seem to exist. In particular, most of theseven examples of be with counterfactual perfects reported in Tables 5 and 6 canbe interpreted in just this way. Consider e.g. those in (16):24

(16) a. and this is to singnefie the certeynte of profecie, whos bifalling of tyme tocomynge is so certeyn, as if it were passid now‘and this is to signify the certainty of prophecy, the happening of whichin time to come is as certain as if it had already happened now.’(CMPURVEY,I,55.2214)

b. The Fellow looks as if he were broke out of Bedlam.‘The fellow looks as though he had broken out of Bedlam’ (the infamousLondon psychiatric hospital) (FARQUHAR-E3-H,60.477)

c. yf he had your sowle I wene he shold be gone.‘If he had your soul, I think he would have/be gone.’(MERRYTAL-E1-P1,10.128)

. Note that in the 1st and 3rd persons singular, be has a distinct past subjunctive form werewhich contrasts with the past indicative was, and that it is this subjunctive form which showsup in counterfactuals in older English (as well as conservative varieties of Modern English).See Iatridou (2000) for arguments on the basis of cross-linguistic evidence that it is still the pastmorphology that contributes the counterfactual semantics in such cases, not the subjunctive. Allthat is really important for us is that (at least at this stage of the language) a past subjunctive formcannot by itself contribute the equivalent of two ExclF features, i.e. contribute both past andcounterfactual interpretation. Perfect morphosyntax under the past subjunctive morphology isrequired to contribute the past semantics.

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.19 (1342-1389)


The correct interpretation of these sentences is not certain, but present counter-factual readings are plausible or even likely. In the ME sentence (16a), the adverbnow suggests a present state rather than a past eventuality. In the EModE (16b), thepresent tense in the main clause points to a present counterfactual interpretationof the embedded clause. Finally, in (16c), the antecedent clause yf he had your sowlelooks like a present counterfactual, since it is formally a simple past rather than apluperfect, thus we expect the consequent clause to be a present counterfactual aswell.

. Some cross-linguistic notes

As noted above, none of the modern European languages that have formed thebasis for theoretical discussion of perfect auxiliary selection have been reportedto show anything like the counterfactual effect. The choice between have and bethere seems to depend only on factors related to argument structure, telicity andother things determined within the νP. Higher clausal properties from the tenseand mood area are irrelevant. However, older English is not unique in showingsensitivity to such things. Indeed, there is mounting evidence that this sort of in-teraction – while perhaps not the norm – is not uncommon. In this section we willbriefly discuss the examples we have found of things like the counterfactual effectin other languages and discuss how such data bear on our analysis.

A preference for have in modal contexts in older Germanic languages otherthan English was already noted by Kern (1912) and Johannisson (1958). Shan-non (1995), largely following Kern, discusses the effects of modality on auxiliaryselection in Middle Dutch and Middle Low German, noting for Middle Dutch“a strong, though by no means absolute tendency for mutative verbs, which ofcourse are otherwise normally conjugated with be in the perfect, to take have inirrealis contexts” [p. 138]. Note especially example (17b), where the verb ‘fall’ ap-pears with be in the realis matrix clause, but have in the counterfactual (modal)embedded clause:

(17) a. haddihad he

hemthem

oecalso

nietnot

ontlopen,escaped,

sithey

haddent. . .had. . .

‘Had he also not escaped from them, they would have. . . ’b. veel

manyludenpeople

sijnare

ghevallenfallen

. . . die

. . . whonietnot

ghevallenfallen

soudenwould

hebbenhave

dan. . .but‘Many people have fallen . . . who would not have fallen, but. . . ’

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.20 (1389-1449)


Johannisson (1958:108) identifies “the subjunctive expressing unreality” as a keyfactor favoring have with verbs that otherwise took be in Old Swedish (though,interestingly enough, not in Old West Norse).

A similar pattern is reported by Ledgeway (2003) for 14th and 15th centuryNeapolitan. At that time, be was the rule with unaccusatives and certain types ofreflexives, but was frequently replaced by have in modal contexts. Like English,Neapolitan ended up completely losing be as a perfect auxiliary (unlike standardItalian), and Ledgeway argues that the modal effect was the first step on the way tothat change. Note, on the other hand, that Dutch did not ultimately lose be, butrather lost the counterfactual effect. So it seems that such an effect can combinewith other changes to lead to the loss of be, but need not necessarily do so. Thisis consistent with the pattern shown in Table 1, where the appearance of have incounterfactuals correlates with just one of the two discrete jumps in the frequencyof have with come.

The fact that the counterfactual effect has been found in a number of older Eu-ropean languages, all spoken approximately 600–700 years ago, is noteworthy, andmay lend some additional support to the analysis we have proposed. A reasonableinterpretation of the situation, assuming that it is not an accident, is that the coun-terfactual effect is a product of a fairly early stage in the grammaticalization of thebe perfect.25 In fact, this is essentially what we have been claiming. The counter-factual effect results because the ME and EModE be perfect remains at least fairlyclose to its resultative origins and does not develop true past semantics the waythe have perfect does. The other European languages whose be perfects have thesame historical source would be expected to go through a similar stage, thoughthey may differ in their subsequent development. In Modern German, Dutch andItalian, the be perfect has clearly developed further into a true perfect or even sim-ple past, functioning as a full analog to the have perfect. Thus there is no problemwith using auxiliary be in past counterfactuals, as in the German example in (18)(modeled on (14)):

(18) Wennif

duyou

pünktlichtimely

gekommencome

wärest. . .were

‘If you had come on time. . . ’

We can hypothesize that Middle Dutch, Middle Low German, Old Swedish and15th century Neapolitan, like ME and EModE, had not (yet) reached this stage inthe development of the be perfect. It remained an essentially resultative construc-

. It is not really problematic that we are dealing with two Germanic languages and one Ro-mance. While it is true that these languages did not inherit the periphrastic perfect from acommon ancestor, it is well known that the constructions have developed largely in parallelin them, presumably due at least in part to contact.

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.21 (1449-1522)


tion, and thus was inappropriate for past counterfactuals. In ongoing research weare investigating whether there is independent evidence to support this idea.26

. Problems for other theories of auxiliary selection

To this point we have offered little discussion of other theories of auxiliary selec-tion. This is because most of them were formed without the older English facts inmind, and they simply are not built to deal with them. The well-known accountsof the perfect auxiliaries are phrased primarily in terms of argument structure re-lations, lexical semantics and (low) aspect, because as noted these are the factorsthat are relevant to selection in German, Dutch, Italian and French. We would not,however, expect any of these things to be affected by counterfactuality. It is unlikelythat a theory of auxiliary selection couched in such terms could satisfactorily coverMiddle and Early Modern English without extensive modifications. In this sectionwe will briefly discuss some of the most influential theories and the issues that thecounterfactual effect presents for them.

Perhaps the most popular analysis of auxiliary selection among generativesyntacticians, associated with Burzio (1986) and many others, is that choice be-tween be and have in languages like Italian, German and Dutch depends onthe underlying position or grammatical function of the subject. be is selected byunaccusatives, whose subject is an underlying internal argument, while have is se-lected by unergatives and transitives, whose subject is an external argument. Kayne(1993) proposes to explain this difference in selection in terms of the presence orabsence of a P head which is required to introduce the participial structure, butonly when there is an external argument. The auxiliary verb is always underlyinglybe, but when the P is present, it incorporates into be, yielding have. Theories ofthis kind seem to work reasonably well for the modern European languages, andthey get part of the story for older English – i.e. they distinguish more or lesscorrectly between the verbs that always take have (transitives and unergatives)and those that can at least sometimes take be (the unaccusatives). However, theyhave no way to deal with counterfactuality effect, because this has nothing to dowith argument structure. In particular, it is hard to see how putting come under acounterfactual would turn it into an unergative.

A more traditional account frequently offered specifically for the changes inauxiliary selection in the history of English is that be was eliminated due to pres-

. It has recently come to our attention that counterfactual contexts allow the use of havewith verbs that otherwise require be in certain spoken varieties of Modern Dutch and Norwe-gian. Unfortunately, this seems to have been little noted in the literature, and the details remainuncertain. We are currently investigating the phenomenon with native-speaker informants.

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.22 (1522-1593)


sure to avoid ambiguity (see e.g. Traugott 1972; Zimmermann 1973; Rydén &Brorström 1987). In particular, clauses with be + past participle were potentiallyambiguous between a perfect and a passive, while have + past participle was un-ambiguously a perfect. Thus – so the reasoning goes – speakers increasingly usedhave with verbs that had previously taken be in order to avoid confusion. Insupport of this, the comparison with German is noted, where there is a separateauxiliary werden ‘become’ for the eventive passive, and sein ‘be’ has been retainedas a perfect auxiliary. There are serious problems with this theory, however. First,languages like Italian and French seem to have no problem with using be as anauxiliary in both the perfect and the passive. Second, only transitives regularlyform passives, while only intransitives could take be as a perfect auxiliary. Thusthe only way that ambiguity of the proposed kind can arise is with verbs that haveboth transitive and intransitive uses, which are not distinguished morphologically.While this kind of alternation is reasonably common in Modern English, it wasrare in the relevant older stages of the language. Consider that in our reading ofthe ME corpus, we found only 9 clauses to be ambiguous in this way, comparedto 543 clear intransitives with auxiliary be (1.7%). Finally, even if the ambiguity-avoidance theory could be used as a (partial) explanation for the loss of perfectauxiliary be, it is again completely unhelpful for the specific pattern with coun-terfactuals. Counterfactual clauses should be no more prone to ambiguity thannon-counterfactual ones, so there’s no reason why they should so completely fa-vor have. Inasmuch as the data we have presented indicate an important role forthe counterfactual effect in the loss of auxiliary be, such theories would thus be atbest incomplete.

Sorace (2000) takes a different approach, proposing that auxiliary selection issensitive to a hierarchy of semantic verb classes. Verbs tend more or less stronglyto select have or be depending on where they fall on the hierarchy. The verbsat one end – non-motional controlled process verbs like work – most strongly se-lect have, while those at the other end – change of location verbs like arrive – moststrongly select be. Furthermore, languages can vary in where on the hierarchy theydraw the line between selecting have and selecting be. This approach providesa means to capture cross-linguistic variation and change in a formal descriptiveframework, something that is notoriously problematic for unaccusativity-basedtheories. However, it provides no real explanation for why a given type of verbshould behave one way and not another, and it gives no clue as to why languagesshould vary and change along the scale of the hierarchy. Furthermore, Sorace’s the-ory has basically the same problem with the counterfactual effect as unaccusativitytheories. Since her hierarchy is based on the semantics of the main predicate, thereis no reason to expect things like counterfactuality to affect auxiliary selection. I.e.putting a modal above come won’t convert it from a change of location to, say,an existence of state verb.

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.23 (1593-1615)


The only theory of which we are aware which specifically addresses the coun-terfactual effect is that proposed by Shannon (1995).27 Shannon proposes that beis most strongly selected by clauses that approximate what he calls a mutative in-transitive prototype, which is defined in terms of a cluster of semantic properties.Like in the theories already mentioned, properties of the eventuality like telicityand agentivity are relevant here, but Shannon claims that higher level propertiesof the utterance go into defining the prototype as well. In particular, the mutativeintransitive prototype is defined as a positive assertion about a non-agentive even-tuality denoting a change of state or place in the sole nominal argument (mutativeis a cover-term for change of state or place used in the older literature on auxil-iary selection). Things like irrealis mood and negation move a clause away fromthe mutative prototype by canceling the assertion that the change has taken place.Since this prototype is what triggers selection of be, these factors can thus havethe effect of favoring have. Shannon’s prototype-based theory suffers from manyof the same deficiencies as Sorace’s hierarchy-based theory. No real explanation isoffered for why selection of be should correlate with the mutative prototype. Fur-thermore, while it can accommodate cross-linguistic variation, it does not explainit. It is simply claimed that languages vary in how close an eventuality must beto the mutative prototype to trigger selection of be. There is no attempt to relatethe different behavior of, say, Italian and Middle Dutch to independently observ-able differences in the languages. Even on a descriptive level, the older Englishdata discussed here may be problematic for Shannon’s theory. As we have seen,counterfactuality on its own was enough to rule out be without consideration oftelicity, agentivity or anything else. Putting counterfactuality on the same level asthese other properties in the definition of a single prototype fails to reflect thisasymmetry, and saying that the prototype was hyper-sensitive to counterfactualityin Middle and Early Modern English would just be a restatement of the facts.28

Unlike most of these theories, the account we proposed in Section 4 can handlethe special properties of auxiliary selection in older English. Furthermore, it dis-tinguishes itself from Shannon (1995)’s theory in that it relates these properties toother characteristics of the perfect in the language. Specifically, older English had a

. Most of the other works which discuss the effect are essentially descriptive – i.e. they are con-cerned primarily with documenting the changes that happened in English, and only secondarily,if at all, with offering a cross-linguistically valid theoretical interpretation of those changes. Oth-ers are theoretically oriented (e.g. Traugott 1972; Lipson 1999), but are more concerned withother phenomena and do not offer an explanation for the counterfactual effect.

. To be fair, it should be noted that Shannon was concerned primarily with data from Mid-dle Dutch and Middle High German, and in those languages the counterfactual effect wasapparently not categorical. His account is thus descriptively accurate for the languages it wasspecifically designed to deal with.

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.24 (1615-1666)


counterfactual effect while Modern German does not, because the older English beperfect was still a stative resultative construction, while the modern German oneis a full-fledged temporal perfect. Clearly, our account is not intended to replaceother theories discussed here completely. We have not presented a theory of thegeneral distribution of be and have for older English, let alone cross-linguistically,but have simply offered an explanation for the effect of counterfactuality on thechoice of auxiliary. A complete account of the alternation between be and havein a language like ME will require a combination of our account of the counter-factual effect with some (perhaps heavily modified) version of one of the theoriesdiscussed in this section. As pointed out above, the counterfactual effect seemsto be something that operates independent from and on top of the general pat-terns of auxiliary selection in the language, so such a modular approach seemsappropriate.

. Conclusion and open questions

In this chapter we have argued that the initial retreat of be as a perfect auxiliaryin English was the tied to the rise of the counterfactual effect. We have shown thatauxiliary be was categorically incompatible with past counterfactual semantics,and that the first appearance of have with come correlates with the first appear-ance of counterfactuals in the perfect. Other syntactic and semantic factors whichhave previously been claimed to favor have have been shown to either be at-tributable indirectly to the counterfactual effect, or to be significantly weaker. Wehave proposed an analysis of the effect based on the proposal that the older En-glish be perfect was in fact still a transparent resultative stative construction, andin this way we were able to account for some of the additional facts which arenot obviously related to counterfactuality. Phenomena similar to the counterfac-tual effect in related languages were then discussed, which lend some support tothe diachronic aspect of our analysis. Finally, we have discussed the implicationsthat the counterfactual effect has for familiar theories of auxiliary selection, boththose based on languages like German, Dutch and Italian and those which haveattempted to take the older English data into account.

Of course, a number of empirical and theoretical questions are raised by ourfindings and analysis which we have not yet addressed. Many of these concernthe development of English after the period that we have focused on here, andare the subject of our ongoing corpus research. Most importantly, what is the na-ture of and explanation for the second jump in the frequency of have with come,which came at the end of the EModE period? Was this what led to the ultimatedisappearance of be? Furthermore, how did the counterfactual effect fare in LateModern English? Was there a period when the be perfect developed into a true

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.25 (1666-1745)


parallel of the have perfect before it disappeared, or did it remain restricted to re-sultative contexts? Even for the ME and EModE periods that we covered, questionsremain regarding the influence of infinitives and negation on auxiliary selection.Will these reduce to interactions with the semantics of the be perfect too, or issomething else going on? Cross-linguistic issues arise as well. If our analysis of thedifference between ME, Middle Dutch and Old Neapolitan on the one hand andthe modern languages on the other hand is correct, then we should also find evi-dence for the counterfactual effect in earlier stages of German and French. Finally,as hinted at in the previous section, the account we have developed here for thecounterfactual effect needs to be embedded in a general theory of auxiliary selec-tion with which it is consistent, and which ideally can account for the subsequentstages in the loss of be as a perfect auxiliary.

References

Burzio, L. (1986). Italian Syntax. Boston: Reidel.Condoravdi, C. (2002). Temporal interpretation of modals: Modals for the present and for

the past. In D. Beaver, L. C. Martínez, B. Clark, & S. Kaufmann (Eds.), The Constructionof Meaning (pp. 59–87). Stanford CA: CSLI Publications.

Elsness, J. (1997). The Perfect and the Preterite in Contemporary and Earlier English. Berlin:Mouton de Gruyter.

Fridén, G. (1948). Studies on the Tenses of the English Verb from Chaucer to Shakespeare, withSpecial Reference to the Late Sixteenth Century. Cambridge MA: Harvard University.

Hoffmann, G. (1934). Die Entwicklung des umschriebenen Perfektums im Altenglischen undFrühmittelenglischen. PhD dissertation, Universität Breslau.

Iatridou, S. (2000). The grammatical ingredients of counterfactuality. Linguistic Inquiry, 31,231–270.

Iatridou, S., Anagnostopoulou, E., & Pancheva, R. (2003). Observations about the form andmeaning of the perfect. In A. Alexiadou, M. Rathert, & A. von Stechow (Eds.), Perfect Explo-rations (pp. 153–204). Berlin: Mouton de Gruyter.

Johannisson, T. (1958). On the be and have constructions with mutative verbs. Studia Linguis-tica, 12, 106–118.

Kakietek, P. (1976). The perfect auxiliaries in the language of Shakespeare. Studia Anglica Pos-naniensa, 8, 45–53.

Kayne, R. (1993). Toward a modular theory of auxiliary selection. Studia Linguistica, 47, 381–405.

Kern, J. H. (1912). De met het Participium Praeteriti Omschreven Werkwoordsvormen in ’t Neder-lands. Amsterdam: Johannes Mueller.

Kroch, A., Santorini, B., & Delfs, L. (2005). Penn-Helsinki parsed corpus of Early Modern English.Philadelphia PA: University of Pennsylvania.

Kroch, A. & Taylor, A. (1999). Penn-Helsinki parsed corpus of Middle English (2nd ed.). Philadel-phia PA: University of Pennsylvania.

JB[v.20020404] Prn:24/10/2006; 14:34 F: LA9708.tex / p.26 (1745-1828)


Kytö, M. (1997). Be/have + past participle: The choice of the auxiliary with intransitives fromLate Middle to Modern English. In M. Rissanen, M. Kytö, & K. Heikkonen (Eds.), Englishin Transition: Corpus-based studies in linguistic variation and genre styles (pp. 19–85). Berlin:Mouton de Gruyter.

Ledgeway, A. (2003). The distribution of the perfective auxiliary avere in Early Neapolitan: Splitintransitives conditioned by modal factors. Archivo Glottologico Italiano, 88, 29–71.

Lipson, M. (1999). The loss of auxiliary selection in English. University of Pennsylvania WorkingPapers in Linguistics, 6(2), 49–61.

McFadden, T. & Alexiadou, A. (2006). Pieces of the perfect in German and Older English. InD. Baumer & M. Scanlon (Eds.), Proceedings of the 25th West Coast Conference on FormalLinguistics.

Mitchell, B. (1985). Old English Syntax. Oxford: Clarendon Press.Molencki, R. (2000). Parallelism vs. asymmetry: The case of English counterfactual conditionals.

In O. Fischer, A. Rosenbach, & D. Stein (Eds.), Pathways of Change: Grammaticalization inEnglish (pp. 311–328). Amsterdam: John Benjamins.

Mustanoja, T. (1960). A Middle English Syntax. Helsinki: Société Néophilologique.Rydén, M. & Brorström, S. (1987). The Be/Have Variation with Intransitives in English. Stock-

holm: Almqvist & Wiksell International.Shannon, T. (1995). Toward a cognitive explanation of perfect auxiliary variation: Some modal

and aspectual effects in the history of Germanic. American Journal of Germanic Linguistics &Literatures, 7, 129–163.

Sorace, A. (2000). Gradients in auxiliary selection with intransitive verbs. Language, 76, 859–890.

Stowell, T. (2004). Tense and modals. In J. Guéron & J. Lecarme (Eds.), The syntax of Time (pp.621–635). Cambridge MA: The MIT Press.

Taylor, A., Warner, A., Pintzuk, S., & Beths, F. (2003). York-Toronto-Helsinki parsed corpus ofOld English prose. University of York.

Traugott, E. (1972). The History of English Syntax. New York NY: Holt, Rinehart and Winston,Inc.

Warner, A. (1993). English Auxiliaries. Cambridge: CUP.Zimmermann, R. (1973). Structural change in the English auxiliary system: On the replacement

of be by have. Folia Linguistica, 6, 107–117.

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.1 (47-117)

The loss of residual “head-final” orders andremnant fronting in Late Middle English*

Causes and consequences

Mary Theresa Biberauer and Ian Gareth RobertsLinguistics Department, Cambridge University

Since Canale (1978), “head-final” structures surfacing after the Early MiddleEnglish period have generally been ascribed to the operation of a “special” ratherthan a productive “head-final” grammar (cf. i.a. Kroch & Taylor 2000). Biberauer& Roberts/B&R (2005) propose a different analysis in terms of which all MiddleEnglish (ME) word-order patterns are the output of a single, optionality-per-mitting grammar, with attested word-order changes ultimately being the conse-quence of loss of this optionality. Specifically, B&R propose that ME “head-final”orders reflect the continuing availability of vP-raising, alongside DP-raising, asa means of satisfying T’s EPP-requirements. This paper shows how vP-raising(= DP-raising + pied-piping) can account for the occurrence of various previ-ously unrelated “special” structures in ME, including Stylistic Fronting and Verb(Projection) Raising, and also how the loss of these orders and the correspondingrise of expletives and obligatory subject-raising can be understood as relatedconsequences of the loss of optional vP-raising.

* We would like to thank Ans van Kemenade for the stimulating discussion that initially led usto consider more closely the nature of the empirical evidence that would support the MiddleEnglish-related proposals in Biberauer & Roberts (2005); and also the audiences at DIGS VIIIand CGSW 20, as well as the participants in the Diachronic Dialogue group who attended theseminar held at Radboud Universiteit Nijmegen in December 2004 and, in particular, two ex-tremely thorough CGSW reviewers, whose comments forced us to consider some of our claimsmore carefully; and, finally, our thanks also go to Jutta and Laszlo for their forbearance. We ac-knowledge the financial support of the Arts and Humanities Research Council of Great Britain(AR14458).

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.2 (117-185)


. Introduction

The primary empirical focus of this paper is the residual “head-final” orders foundin Middle English (ME). The usual chronology for the general change from OV toVO in English situates it in Early ME (Canale 1978; van Kemenade 1987; Light-foot 1991; Roberts 1997a; Kroch & Taylor 1994; Fischer et al. 2000), but as variousauthors have pointed out, orders which are indicative of some kind of persistingOV grammar are found, albeit at rather low frequency and somewhat disguisedby other factors, until the 15th century (see Fischer et al. 2000:177 for a sum-mary and references). Here we will propose an analysis of these orders whichsupports the novel proposal in Biberauer & Roberts (2005; henceforth: B&R) thatthe loss of residual “head-final” orders is related to the introduction of obliga-tory clause-internal expletives. The reason for this is that both developments resultfrom the loss of vP-movement to SpecTP and its replacement by DP-movement tothat position.

The orders we will look at are so-called Stylistic Fronting (Styl-F), Verb Rais-ing, i.e. SVAux sequences, and what has been analysed as Verb Projection Raising(VPR), i.e. AuxOV sequences. Following and developing the proposals in B&R,we propose new analyses of these orders. We also integrate the observations andanalysis of van der Wurff (1997, 1999) and Ingham (2002, 2003) regarding thelast attested OV orders with non-pronominal DPs. Furthermore, we show howthe changes that we propose for Late ME created some of the preconditions forthe well-known development of a syntactically distinct class of modal auxiliariesin the 16th century (Lightfoot 1979; Roberts 1985, 1993; Warner 1997; Roberts &Roussou 2003; Biberauer & Roberts 2006a, b).

The paper is organised as follows. In §2, we outline our theoretical assump-tions, in particular those relating to the different ways in which EPP-features canbe satisfied (see Biberauer 2003; Richards & Biberauer 2004, 2005; Biberauer &Richards 2005, 2006) and show how these ideas give rise to the novel account ofword-order change in ME presented in B&R. In §3, we summarise the prevailingview of the chronology of the loss of “head-final” orders in ME, basing our dis-cussion mainly on Kroch & Taylor (2000). We then present our analyses of Styl-Fand of V(P)R. Finally, in §4, we look at some of the consequences of the proposedanalysis of V(P)R, showing how van der Wurff and Ingham’s results can be in-corporated, and sketch the connection between these Late ME changes and thedevelopments affecting modals in Early Modern English (ENE).

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.3 (185-219)

Loss of residual “head-final” orders in Late Middle English

. EPP-satisfaction and pied-piping

The technical notion central to the account of word-order change proposed byB&R is that of pied-piping. More specifically, they adopt the analysis of pied-piping put forward in Biberauer & Richards (2005, 2006) and Richards & Bib-erauer (2004, 2005). This analysis exploits the distinction between two relatedproperties a given head may have in the version of the theory of movement andchecking/agreement of features proposed in Chomsky (2001, 2004), namely thatof being a Probe on the one hand and that of being associated with one or moreEPP-features on the other. In terms of this theory, a head may bear an (active)uninterpretable/unvalued feature (e.g. a φ-feature) which functions as a Probe,necessitating the location of a matching Goal, i.e. a category bearing an inter-pretable/valued counterpart of the probing feature since the unvalued feature ofthe Probe must be eliminated from the derivation in order for convergence (well-formedness) to be possible. The operation facilitating this feature-elimination isassumed to be Agree. Agree holds between a Probe P and a Goal G under thefollowing three conditions:

(1) a. P must (asymmetrically) c-command G;b. P and G must be non-distinct in features; andc. there must be no Goal G’ �= G such that P c-commands G’,

G’ c-commands G and G does not c-command G’.

Agree relationships can be successfully set up without the need for movement: aslong as P and G meet the conditions outlined above, they can enter into an Agreerelation, eliminating unvalued features in both the Probe and the Goal, whichremains in situ. The theory of feature-checking (or, more accurately, feature-valuing/Agree) that we adopt therefore departs from earlier Minimalist proposals(cf. Chomsky 1995) in terms of which movement and the creation of a very local(Spec-head/head-head) configuration was regarded as a prerequisite for feature-checking.

The Agree-based theory that we adopt does not, however, rule out the possibil-ity that feature-checking (Agree) and movement may coincide: wherever a Probe-bearing head is associated with an EPP-feature, convergence is in fact only possibleif the creation of the appropriate Agree relation is accompanied by movement ofthe Goal-bearing category. The most important characteristic of this system forB&R’s purposes is that there is nothing which prevents a Goal G from being prop-erly contained inside a category which is moved in order to satisfy the Probe’sEPP-feature. This option must be admitted in order to allow for standard cases ofpied-piping, as when the object of a preposition is questioned or relativised in alanguage such as French which disallows preposition-stranding, see (2):

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.4 (219-346)


(2) ATo

quiwhom

as-tuhave.you

parlé?spoken?

“ Who(m) have you spoken to?”

We can schematise this situation as in (3):

(3) whPROBE . . . [PP whGOAL . . . ] . . .

Here PP moves to satisfy the Probe’s EPP-feature, but the Goal is the wh-featureon qui which is properly contained in PP.

Pied-piping under wh-movement is obligatory in French, as is well known,while English famously allows both options, pied-piping and stranding. It is thusclear that Universal Grammar allows variation as to whether it is simply the Goalthat moves or whether a larger category is required to move in order to satisfy theProbe’s EPP-feature.

More generally, pied-piping involves the abstract configuration in (4) and Uni-versal Grammar allows crosslinguistic variation as to whether ZGOAL moves to Xor the larger category YP containing ZGOAL moves to SpecXP:

(4) . . . XPROBE . . . [YP . . . ZGOAL . . . ] . . .

Biberauer (2003), Richards & Biberauer (2004, 2005) and Biberauer & Richards(2005, 2006) exploit this option to account for aspects of word-order variation inGermanic and to provide a unified analysis of T-related EPP-satisfaction in thislanguage family. Specifically, they propose that, in terms of the schema in (4),X may be T, YP may be vP and Z an element with D-features (either a subject-DP/expletive or nominal morphology on the verb; see below) since T is assumed toprobe for a D-bearing Goal. This means that vP-movement may take place wherethe Goal is in fact a D-element (Z in (4)) probed by T (X in (4)). In other words, Twith a D-oriented Probe may in fact attract a vP. The D-features of the Goal con-tained in vP satisfy the active uninterpretable formal feature (i.e. the D-feature) ofT, while vP-movement (i.e. pied-piping) satisfies T’s EPP-feature.

On this basis, Richards & Biberauer (2005) construct a four-way typology ofways of satisfying T’s EPP- and D-features based on the two parameters of thesource of the D-feature and the size of the category containing or bearing the D-feature targeted by the Probe’s EPP-feature. The source of the D-feature may beeither the verb morphology, in languages where this morphology is sufficiently“rich” (cf. Borer 1986; Barbosa 1995; Alexiadou & Anagnostopoulou 1998), or, inlanguages with more impoverished verbal morphology, the DP contents of (outer)SpecvP (see below). The size of the category containing or bearing the D-featuremay either correspond to that of the Goal (the finite verb or the DP subject) or tothat of the maximal category containing the Goal (vP). Table 1, from Richards &Biberauer (2005), illustrates the typology:

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.5 (346-346)


Table 1. Typology of modes of (T-related) EPP-satisfaction

Source: [D] on Vf Source: [D] in outer SpecvP

Size: [–pied-piping] Head raising Spec raisingSize: [+pied-piping] Head pied-piping Spec pied-piping

The first row of Table 1 indicates two modes of EPP-satisfaction that have beenmuch discussed in recent literature, namely that exhibited by Modern English-type as opposed to so-called null subject languages. In the former case, T’s EPP-requirements are satisfied by DP-raising from SpecvP (i.e. spec-raising), whichgives rise to the consistent presence of a DP in subject position, including oblig-atory overt expletives in the relevant contexts. The latter case is instantiated bycanonical null-subject languages such as Greek or Italian on the type of analysis ofnull subjects which assumes that the subject argument is represented by the “rich”verbal morphology (see the references given above): in this case, V-to-T movement(i.e. head raising) results in EPP-satisfaction. The second row of Table 1 presentstwo new modes of EPP-satisfaction put forward in Richards & Biberauer (2005)which, however, differ minimally from those that are already established in theliterature. Thus so-called head pied-piping languages are those which have “rich”verbal inflection and therefore target the D-feature located on the finite verb, butwhich additionally pied-pipe the category (vP) containing the D-bearing head.German is an example of this type of language. The option which is of most rele-vance in the present context is the spec pied-piping type, i.e. that where the “source”of the D-features is, as in Modern English, the DP located in SpecTP, while the[+pied-piping] setting of the “size” parameter requires movement of the categorycontaining this DP, i.e. vP. Richards & Biberauer (2005) argue that this particu-lar combination of parameter settings, uniquely among the possible combinationsrepresented in Table 1, allows two distinct operations to satisfy T’s featural require-ments: either movement of the DP in the specifier of vP (i.e. DP-movement), orvP-movement. They argue that these two options both represent pied-piping ofcategories of different sizes containing the [D] Goal, and that both options areavailable because there are no system-internal or input-based reasons for speakersof spec pied-piping languages to rule out DP-raising as an alternative to vP-raising.As far as the spec pied-piping grammar is concerned, DP- and vP-raising are thusequally allowable modes of T-related EPP-satisfaction. According to B&R, this is sobecause pied-piping can be seen as an instruction to the grammar to move a cat-egory larger than the nominal or verbal head which actually constitutes the Goalof the Agree relation, with no precise specification as to the “size” of that category;all that matters is that an XP of some type, rather than just the D-bearing head,ultimately satisfies T’s EPP-requirements. Both DP-raising and vP-raising there-fore count as instances of pied-piping, although they instantiate slightly different

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.6 (346-401)


species of pied-piping: with DP-raising, it is arguably the case that the features forwhich the Probe searches are instantiated in distributed and partially overlappingfashion on the various elements making up the DP. Consider, for example, the wayin which φ-features are very clearly seen to be distributed across the various lexicalitems making up DPs in languages with rich nominal agreement – cf. the GermanDP in (5) in this connection:

(5) Der kleine Junge – “the small boy”nom nomMasc Masc MascSing Sing Sing

If we assume (finite) T to be in search of a complete set of φ-features (cf. Chomsky2000 et seq.), it is clear that just raising Junge will not suffice as this would amountto raising only part of the Goal; exclusively raising the D-element der might seemsufficient in this case, but the fact that it is not obligatory for D-elements to beinstantiated in German nominals may be relevant here. We leave this interestingmatter aside for further research into the rather poorly understood nature of pied-piping, noting only that DP-raising differs from vP-raising in respect of whetheror not every instance of this operation will pied-pipe non-Goal (i.e. “extra”) fea-tures – with vP-pied-piping, this is mostly the case; with DP-raising, it will oftennot be (although cf. DPs associated with relative and complement clauses, etc.).For our purposes, what is important is that it is not incoherent to view DP-raisingas an operation which “qualifies” as both [+pied-piping] and [–pied-piping]. Specpied-piping grammars may thus admit optionality in respect of how T’s EPP-feature is satisfied; and spec pied-piping languages alone have a choice of waysof satisfying T’s features since (a) the non-pied-piping languages by definition donot have a pied-piping option and (b) in head pied-piping languages, movementof the DP in SpecvP would not amount to movement of the Goal (or of a categorycontaining the Goal), and so is not possible, while movement of just the head-Goalwould result in surface strings not attested in the input and, as such, does not con-stitute the kind of operation conservative acquirers would be expected to postulate(cf. Biberauer & Richards 2006, Note 14).

B&R propose extending the domain within which this analysis has been saidto hold, and applying it also to the case where, in terms of (4), X is v, YP is VPand Z is, as in the previous case, an element with D-features. For the same reasonas in the case just described, this means that a v with D-features may attract aVP into its specifier in order to Agree with the D-element contained in the VPand in order to satisfy its EPP-feature. Again, as in the previous case, wherever thetargeted D-element is a DP, two options present themselves for the satisfaction ofv’s EPP-feature: VP-raising or DP-raising, i.e. pied-piping or exclusive movementof the Goal (“stranding”). B&R thus postulate an essential parallelism between

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.7 (401-461)


the options available for the satisfaction of T’s and v’s D-features: in both cases,the system with the [+pied-piping] setting of the size parameter permits eitherpied-piping of respectively vP and VP or, alternatively, DP-movement.

B&R’s central proposals are, first, that Old English (OE) was initially a uni-formly spec pied-piping language in the sense just described. As such, it allowedoptional pied-piping wherever T and v probed a phrasal D-element, thus givingrise in the TP-domain to a choice between subject DP-movement to SpecTP or vP-movement to this position, and, in the vP-domain, to either object DP-movementto SpecvP or VP-movement to this position. B&R argue that this optional pied-piping gave rise to much of the attested word-order variation in OE. Their analysistherefore provides an account for much of what was previously seen as theoreti-cally inexplicable variation in grammars (i.e. “true optionality”) or grammars incompetition (Kroch 1989; Pintzuk 1991) in terms of a single grammar, one whichfreely allows both pied-piping and stranding movements to achieve the satisfactionof its EPP-requirements. In terms of this analysis surface head-final orders repre-sent one kind of derived order, alongside various kinds of head-initial and mixedorders (see Kayne 1994; Zwart 1997). Henceforth, the term head-final (withoutscare quotes) should thus be taken to refer to surface head-final order.

B&R’s second proposal is that the loss of the optionality and its replace-ment with just the non-pied-piping/stranding variant, i.e. subject DP-movementto SpecTP and object DP-movement to SpecvP (which was later lost), under-lies much of the word-order change observed in the ME period (cf. Biberauer &Roberts 2006a, c for more detailed consideration of the non-syntactic factors thatled to the respective pied-piping options being – at different stages in English’shistory – insufficiently robustly attested for an optionality-permitting grammarto be postulated by acquirers, and the role that Berwick’s (1985) Subset Principleappears to have played in determining the loss of the relevant pied-piping op-tions and the consequent move to a “smaller” grammar.). Focusing specifically onthe TP-domain, the proposal is that a combination of system-internal and exter-nal (i.e. input) considerations led to the vP-movement (i.e. spec pied-piping) thatwas optionally available as a means of satisfying T’s EPP-feature (by pied-pipinga D-element into SpecTP) becoming unavailable as a means of EPP-satisfactionby Late ME. DP-movement (i.e. spec-raising) therefore became the only mech-anism via which this feature could be satisfied, with the result that it becamecrucial that SpecvP always be filled by a raisable nominal. Assuming expletives tobe merged in SpecvP (cf. Richards & Biberauer 2005), one of the consequences ofthis development would be that expletives became obligatory at the point at whichvP-raising was lost, whereas they would only optionally have been present prior tothis change as there was always an alternative source of D-features present in vP(see Richards & Biberauer 2005; Biberauer & Richards 2006; Biberauer & Roberts2005, 2006a, c for more detailed discussion). So we see that B&R’s analysis relates

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.8 (461-512)


the rise of TP-expletives to word-order change: assuming auxiliaries are merged inor moved to T and non-finite verbs remain inside vP, vP-movement will give riseto V-Aux orders in non-V2 (i.e. non-root) contexts until such time as DP-raising,and thus also obligatory TP-expletives, became established as the sole means ofsatisfying T’s EPP-feature. B&R thus predict that non-root V-Aux and a varietyof other apparent reflexes of a head-final grammar and additionally also “anoma-lous” phenomena like Stylistic Fronting will decrease and ultimately disappear asTP-expletives emerge and ultimately become obligatory (see §3.1 below).

Although interesting and non-obvious, B&R’s prediction does not conformto the standard view of the chronology of syntactic changes in ME: it is usuallysaid that V-Aux orders disappear in Early ME, whilst it is clear that TP-expletivesonly consistently appear from around 1400, becoming obligatory in the mid-15thcentury (cf. Haeberli 1999a; Williams 2000; Ingham 2001). In the next section,we present the standard view of the status in ME of V-Aux orders and head-finalorders more generally and then propose an alternative account of the changes inME that leads to a reconsideration of the data that is in fact consistent with B&R’sconclusions.

. The loss of residual head-final orders in Middle English

In work on ME word-order change it is usually assumed that OV orders disappearin Early ME, despite the fact that examples of this order do occur – albeit at in-creasingly low frequencies – until Late ME. Since it has influentially been arguedthat the principal VP-related word-order changes took place in the 12th century(see Canale 1978; van Kemenade 1987; Lightfoot 1991), the residual head-finalorders of Late ME have generally been analysed as resulting from some specialprocess(es). In this section, we will argue two things. First, we show that one ac-knowledged case of “special syntax” in ME, Stylistic Fronting, in fact represents astraightforward case of vP-fronting to SpecTP of the kind introduced in the previ-ous section which therefore requires no further stipulations. We will in fact suggestthat reflexes of vP-fronting are generally more common in later ME than has usu-ally been thought. Second, we will focus on structures that are traditionally viewedas biclausal, namely those involving modals, and show how these also played a rolein shoring up the ME acquirer’s analysis of ME as a system permitting both DP-and vP-raising in order to satisfy T’s EPP-requirements.

. Stylistic fronting

A good recent example of the standard line of reasoning on residual head-finalorders is Kroch & Taylor (2000) who point out, in a discussion of a group of

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.9 (512-573)


13th-century prose texts, that “superficially INFL-final [i.e. Aux-final – MTB/IGR]clauses often have another possible analysis; that is, they can also be analysed asinstances of stylistic fronting” (138). Stylistic Fronting (Styl-F) is an operationfirst observed in Icelandic by Maling (1990) which involves the fronting of a par-ticiple, adverb or negation. The principal condition on Styl-F is that there mustbe a subject-gap. Styl-F is also subject to an Accessibility Hierarchy which statesthat negation takes precedence over adverbs which in turn take precedence overparticiples and other verbal elements. The following example illustrates Styl-F inIcelandic:

(6) HonumHim

mættimight

standastand

áon

sama,same

hvaðwhat

sagtsaid

væriwas

umabout

hannhim

“It might be all the same to him what was said about him.”(Maling (1990)’s (5), Kroch & Taylor’s (7), p. 139)

This is an example of a subject relative; hence the subject-gap condition on Styl-Fis met by wh-movement. The participle sagt has been fronted to a position whichimmediately precedes the inflected auxiliary. Note that other VP-internal mate-rial – here the PP um hann (“about him”) – remains in its normal position, giventhat Icelandic is a VO language.

Here are some examples of Styl-F in ME, from Kroch & Taylor (2000:139) andTrips (2002:306):

(7) a. auricheevery

manneman

ðethat

i-boreŠ ensaved

scalshall

bienbe

“every man who shall be saved”(CMVICESi, 63.695; Kroch & Taylor’s (8), p. 139)

b. andand

hehe

besohtesought

atof

godeGode

þatthat

nahtnot

neneg

scoldeshould

reininrain

“And he asked of God that it should not rain.”(CMVICESi,143.1787; Kroch & Taylor’s (9), p. 139)

c. wiþþwith

allall

þattthat

lacsacrifice

þattthat

offreddoffered

wasswas

biforennbefore

CristessChrist’s

comecoming

“with all the sacrifice that was made before Christ’s coming”(Ormulum I.55.525; Trips’s (123), p. 306)

The subordinate clause in (7a) is a subject relative and so the subject-gap conditionis met in the same way as in (6). Here the adjectival participle i-boreŠen has beenfronted to a position immediately preceding scal. We treat premodals like scal asrestructuring verbs in ME, i.e. potential triggers of Verb Raising (VR) and VerbProjection Raising (VPR), in line with the standard treatment of these verbs inContinental West Germanic languages (cf. Evers 1975; Haegeman & van Riemsdijk1986; Rutten 1991; Hinterhölzl 1999, etc.; see §3.2). Following Lightfoot (1979),

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.10 (573-628)


Roberts (1985) and Warner (1993), we assume that these elements only developedtheir characteristic Modern English auxiliary properties after the ME period (see§4.3). The non-finite auxiliary bien follows scal; this is the typical position forinfinitives in the complement of restructuring verbs. In (7b), the subject gap isconstituted by an apparently “missing” expletive and the fronted element is thenegative adverb naht (“not”), while the gap in the passivised relative in (7c) issimilar to that in (7a), with the passive participle offredd (“offered”) representingthe fronted element in this case.

There have been several proposals concerning the nature of Styl-F. An influ-ential recent proposal is that proposed in Holmberg (2000), where it is argued thatStyl-F is an operation which places the phonological matrix of an arbitrary con-stituent in SpecIP. This operation is subject to a locality condition which requiresthe nearest available phonological matrix to move to SpecIP. The Accessibility Hi-erarchy originally noticed by Maling (1990) can then be derived, assuming thatnegation is structurally closer to I than vP-adverbs which in turn are closer thanVP-internal verbal material (see Holmberg 2000:462–463 for details).

Kroch & Taylor (137ff.) propose reducing surface V-Aux orders to cases ofStyl-F. They observe (138) that treating all clauses with V-Aux order and a sub-ject gap as cases of Styl-F has the consequence that the number of “true” V-Auxorders, i.e. where there is no subject gap and therefore in Kroch & Taylor’s termsless possibility of a Styl-F analysis, in the texts they look at is considerably reduced:from 212 to 171, i.e. by just under 20%. Furthermore, they follow Platzack (1988)in considering that Styl-F may also be allowed in clauses with subject pronouns,assuming that pronouns in ME create a subject gap by cliticising to C. This furtherreduces the total number of “true” V-Aux orders to 40, i.e. to roughly 19% of theoriginal total of 212.

Kroch & Taylor’s figures are, however, suspect on several grounds. First, themain reduction (from 80% of surface V-Aux orders to 19%) depends on the as-sumption that subject pronouns undergo cliticisation to C. However, this putativesubject-cliticisation operation is dubious for several reasons (Falk 1993 raises someof these objections as well). First, it is unclear whether subject cliticisation appliesin narrow syntax or in PF; if it applies in PF then it does not create the subject gapin the syntax which Styl-F requires (cf. Ackema & Neeleman 2003, who argue thatthe exactly comparable subject-cliticisation operation in Dutch is a PF-operation).Second, subject cliticisation would involve right-adjunction of the subject pro-noun to C, an operation which violates Kayne’s Linear Correspondence Axiom(LCA), and, as such, may even be suspect as a PF-operation (see i.a. Moro 2000;and Chomsky 2001:37ff., on the idea that the LCA may be a PF linearisation re-quirement). Third, we are unaware of any current spoken language in which Styl-Fis facilitated by subject cliticisation; Icelandic, notably, seems to lack subject cliti-cisation. Fourth, counting clauses with subject pronouns as involving Styl-F skews

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.11 (628-691)


the data since subject pronouns are, for discourse-informational reasons, muchmore frequent than non-pronominal subjects in most kinds of connected writtentext, and so the incidence of putative “Styl-F” is artificially inflated by Kroch &Taylor’s relaxing of the subject-gap requirement. We therefore suggest that Kroch& Taylor’s conclusion that only 19% of surface V-Aux orders really are V-Aux or-ders is not warranted. Instead, if we count the examples with subject pronouns asnot involving Styl-F, then the number of V-Aux clauses in the texts Kroch & Taylorlooked at returns to 171, a not insignificant number.

What we would now like to argue is that all instances of surface V-Aux or-dering, including the apparently superficially “clear” cases of Styl-F and otherinstances of surface head-final order, should, in fact, be viewed as exponents of asingle, formerly fully productive, but in ME ever more residually active operationgenerating surface head-final order. Following B&R, we propose that putative casesof Styl-F and head-final ordering more generally in fact involve vP-movement toSpecTP. In these terms, the TP inside the relative clause in an example like (7c) hasthe structure given in (8):1

(8) [ [ t ] [ [ wass ] tTP P Op T’ T Pv voffredd t biforenn Cristess comeOp ]]

This structure requires several comments. The most important aspect of it forthe purposes of this paper is that vP, containing the string tOp offredd, has raisedfrom its first-merged position following [T wass ] to SpecTP. This operation takesplace in order to satisfy T’s D-oriented EPP-feature. As we saw in the previoussection, B&R argue that in ME the pied-piping option for satisfying T’s EPP-requirements was still available. In the case under consideration, the D-featureis borne by the passive participle offredd, which we, following Baker, Johnson &Roberts (1989), assume to contain the “absorbed” logical subject (cf. also Richards& Biberauer 2005).

The PP biforenn Cristess come was also a constituent of VP (and therefore ofvP). However, it does not appear before the auxiliary in the surface string because

. The examples in (6a, b) are more complex than (6c) in that they involve “verb-raising” in thesense of Evers (1975), as seen from the order finite verb – non-finite verb in clause-final position.For a detailed analysis of this order in terms of the general assumptions made here, see B&R, pp.11ff; see also §3.2. What we would like to highlight here, however, is that the analysis that wepropose for modals – namely that they are verbs which select a very specific type of complement,a restructuring TP – allows us to account for the fact that vP-raising will not, as one mightinitially expect in cases like (6a, b), result in the non-finite verb undergoing movement intoSpecTP along with any non-verbal element undergoing Styl-F. As discussed in more detail in§3.2, the infinitive in modal complements raises to embedded T and is consequently unavailablefor fronting into the matrix clause, with the consequence that it is not present in the vP thatraises to SpecTP in Styl-F structures.

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.12 (691-743)


at this stage the pied-piping option was no longer available for v. The object isextracted under relativisation, which we have indicated by tOp; the leftmost occur-rence of this symbol marks its successive-cyclic movement through SpecvP.2 ThePP therefore remains within the VP throughout the vP phase of the derivation,and it surfaces in final position owing to the effects of the Phase ImpenetrabilityCondition (PIC; see Chomsky 2000:108)):

(9) In a phase α with head H, the domain of H (i.e. its complement – MTB/IGR)is not accessible to operations outside α, only H and its edge are accessible tosuch operations.

For our purposes, the relevant phases α are CP and vP. The heads H are thereforeC and v. Chomsky suggests that the PIC holds because the derivation proceedsby phases, with material in the domain of the head of each phase being trans-ferred to the PF and LF interfaces at the end of that phase. Such material therebybecomes inaccessible to the formal mechanisms of the syntactic component. Theconsequence of this for (9) is that the PP biforenn Cristess come is transferred tothe PF-interface when the derivation completes the vP phase and, as such, is notspelled out as part of the moved vP, but in the final position it occupied at the endof the vP phase. We have indicated this in (8) by presenting the content of the PPin outline font (the same is true for the trace/copy of the constituent extracted un-der relativisation, although obviously in this case there is no visible effect on wordorder; see Note 2).

The structure in (8) is straightforwardly derivable on B&R’s assumptionsabout the movement-triggering features associated with T: it is simply the outputof a derivation in which T’s EPP-feature has been satisfied by vP-raising (i.e. specpied-piping). This analysis generalises to all cases which Kroch & Taylor identifyas Styl-F, and additionally also provides a non-stipulative account of the pronoun-containing structures that they are obliged to analyse as involving the dubiouscliticisation operation discussed above: in our terms, V-Aux orders with pronom-inal subjects are derived in exactly the same way as the more conventional Styl-Fcases just discussed, namely via vP-raising with vP in this case containing thesubject pronoun in addition to the non-finite verb.

A further advantage of the approach advocated here is that it facilitates a verysimple analysis of V-Aux structures that are very evidently not amenable to a Styl-F analysis, i.e. of the V-Aux structures that Kroch & Taylor acknowledge to be thebona fide output of a head-final grammar. Consider (10) in this connection:

. Note that nothing here hinges on the assumption of a null-operator rather than a raisinganalysis of relatives. We have placed the rightmost occurrence of tOp in the “leaked” part of theVP, along with the PP biforenn Cristess come – see the discussion of the surface position of thisPP in the text to follow.

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.13 (743-797)


(10) er þannebefore

þethat

heueneheaven

oðeror

eorðeearth

shapencreated

werewere

“before heaven or earth were created”(Trinity Homilies, 133.1776; Kroch & Taylor’s (4a), p. 137)

Examples like this clearly do not feature a subject gap, and the overt subject is not apronoun, but a coordinated DP, i.e. a subject that cannot plausibly be said to havecliticised to C in order to create a subject gap. As such, this type of V-Aux examplewould not seem to be amenable to a Styl-F analysis and would therefore have tobe viewed as the output of a “genuine” head-final grammar. On our approach,by contrast, the occurrence of structures like (10) is completely straightforwardand would be expected to fall out from the same (single) grammar that gives riseto “Styl-F” structures such as those in (7). The representation of the embeddedclause in (10) is shown in (11):

(11) [CP þe [TP [vP heuene oðer eorðe shapen theuene oðer eorðe] [T’ [T were] tvP ]]]

As noted above, this kind of example must, in Kroch & Taylor’s terms, be treatedas generated by the head-final grammar competing with the head-initial grammarthat is active in other cases. But the existence of a head-final grammar means thatat least some of the Styl-F cases are in fact indeterminate for language acquirers,since they could be treated either as the output of a head-final system (cf. our pro-posal, which could, of course, be translated into more traditional head-final terms)or as the “special” output of a head-initial system (cf. the usual analysis of Styl-Fin Icelandic). Since we do not assume competing grammars, our approach doesnot allow this indeterminacy. On our analysis, the logical-object DP heuene oðereorðe in (11) raises under passivisation to SpecvP. Following and updating Baker,Johnson & Roberts’ (1989:222) proposal, we take it that passive v is not in factdefective,3 but rather has a D-feature and therefore may have a D-oriented EPPfeature which triggers movement of this DP into its specifier (this feature is sat-isfied by successive-cyclic movement of tOp in (8)). The entire vP is then frontedto SpecTP, giving the observed surface order. This is thus a further instance ofpied-piping satisfying T’s EPP-feature. It must be the case that different featuresof the moved DP Agree with v and T’s D-features; perhaps, given the facts of Ro-mance past-participle agreement (Kayne 2000:Chapters 2 and 3), Agree involvesgender and number features at the v-level and person at the T-level (cf. also Chom-sky 2000:125). This would thus be a case of partial feature-matching between the

. Note that the assumption we make here regarding v’s defectiveness should not be interpretedas indicative of our commitment to the idea that passive v is universally non-defective. See Biber-auer & D’Alessandro (2006) for consideration of data that may signal the existence of parametricvariation in respect of this point.

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.14 (797-888)


Probes T and v and the Goal DP heuene oðer eorðe. In a Germanic language likeME, v’s φ-features have no visible morphological consequences, but we proposehere that it may in fact be possible to observe their syntactic effects in exampleslike (10).

We thus have a straightforward general account for V-Aux orders in ME, whichdoes away with Styl-F as a separate option, reducing it to another case of (remnant)vP-movement to SpecTP. We have also shown that this view entails that “genuine”(i.e. OE-like, systematically derived) V-Aux structures were more frequent in MEthan is usually thought.

Our analysis also allows us to account for a number of other fairly well-knownanomalies in ME word order. First, we can account for what is often seen as Styl-F of the negative element noht/naht (also spelt in various other ways). This isillustrated in (7b) above. In this example, vP containing naht and (possibly) aquasi-expletive null subject DP has moved to SpecTP of the clause þat naht nescolde reinin. As mentioned in Note 1, this is a verb-raising construction withthe infinitive reinin moved to the T-position of the infinitival TP complement toscolde. We also find examples of putative Styl-F of negation where there is no aux-iliary of any kind present, showing that the phenomenon is quite independent ofauxiliaries:

(12) Thairwithat-this

hehe

nochtnot

growitshrunk

“At this he did not shrink (in fear)”(c1448: Richard Holland The Buke of the Howlat, 7; cited in Gray 1985:152;Roberts 1993:252)

Second, various kinds of adverbs can be pied-piped to SpecTP along with vP.4 In asubordinate clause, this gives rise to the order complementiser – (subject) – adverb –verb, as illustrated in (13):

. An anonymous reviewer raises the question whether our analysis would not lead us to ex-pect vP-adjuncts to surface in pre-subject position, contra the most common pattern in earlierEnglish and Germanic more generally (cf. Haeberli 1999b, 2000 for discussion of the excep-tions). We assume that Cinque (1999)’s “lower” adverbs are merged below the subject (or, moreaccurately, the external argument; cf. the fact that non-external argument subjects frequentlysurfaced in “low” positions until Late ME) and that they therefore do not constitute a prob-lem for our approach. As regards higher adverbs of the subject- and discourse-oriented type,we are happy to allow for the possibility that these are indeed merged higher than the externalargument, but that the external argument generally (but not always; see below) surfaces higherthan adverbs of this kind owing to the presence of an optional EPP-feature which drives subject-movement to the highest specifier of vP. Since this is an optional EPP-feature, we would, inaccordance with Chomsky’s (2001:34) constraint on optional features and operations, expect itto trigger an “extra” interpretive effect (cf. the detailed discussion in Biberauer & Richards 2006).

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.15 (888-919)


(13) a. Forfor

manymany

areare

thatthat

nevernever

kanecan

haldehold

thethe

ordyreorder

ofof

lufelove

“For there are many that can never keep the commandment of love”(14th century: Rolle The Bee and the Stork, 20–21; cited in Mossé 1968;Kroch 1989:227; Roberts 1993:254)

b. þatthat

hehe

neurenever

maremore

sculdeshould

cumancome

utout

“that he shouldn’t come out any more”(ChronE (Plummer) 1140.48; Fischer et al. 2000:243)

These orders have often been seen as problematic for the widely held and oth-erwise quite well-motivated idea that ME was like Modern French in havingV-to-T movement in finite clauses (see Roberts 1985, 1993; Pollock 1989). InModern French, orders of this kind are impossible (whether the subject is pronom-inal or not; see Pollock 1989; Kayne 2000). If they are derived by vP-movementin ME prior to roughly 1450, then, as long as French is analysed as a “spec-raising” language in the sense defined in §2, we see why ME and French differin this respect.

Third, Haeberli (1999a:409ff.) points out that until approximately 1450,ME allowed adverbs to intervene between the complementiser and the subject,as in (13):

(14) a. Andand

ofof

thysethese

twotwo

folowethfollows

asas

aa

corelarycorollary

yethe

thirdethird

trouthtruth

“And the third truth follows as a corollary from these two”(Fitzja, B6R.220; Haeberli’s (25d), p. 410)

This would indeed seem to be the case if one considers “Diesing effects” of the kind that havebeen said to arise in languages like German, which we would assume to employ vP-fronting:in these languages, subjects are differently interpreted depending on their location relative tohigher adverbs such as ja doch (“after all”) and wahrscheinlich (“probably”). For us, this wouldfollow straightforwardly if a post-adverbial subject is simply in its First Merge position, whereasa pre-adverbial subject has undergone optional EPP-driven movement to the edge of vP, whichthereby brings with it a specific interpretive effect (we leave aside here the vexed issue of pre-cisely what kind of interpretive effect results). The fact that our approach requires us to assumethat discourse-related adverbs are merged inside vP would not seem to us to be particularlyproblematic since (a) this assumption has also been made by other linguists working on whatwe would view as vP-raising languages (cf. Müller 2004) and (b) more traditional analyses alsohave to account for the fact that discourse adverbials standardly surface below the subject in em-bedded clauses, despite the fact that the subject is usually thought to be located in SpecTP, withthe relevant adverbs therefore having to occupy a lower position. See Biberauer (2006) for moredetailed discussion of this matter.

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.16 (919-1002)


b. andand

sooso

isis

geytyet

þethe

termeperiod

ofof

fiftyfifty

gereyears

wyttwith

usus

þethe

geryear

ofof

gracsegrace“And so the period of fifty years is for us the year of grace”(Siege, 70.17; Haeberli’s (25e), p. 410)

Again, we can treat the adverb as contained within the vP which undergoes move-ment to SpecTP in these cases. The fact that the frequency of this order declinesfairly sharply around 1450 (see Tables 10 and 11 in Haeberli 1999a:409–410; andHaeberli 2000) once again fits in with the analysis proposed here, namely that thiswas the point at which vP-raising was lost (see below). As Haeberli (1999b) showsin his survey of this order across a range of Germanic languages, the order in (14)is possible just in those languages which Richards & Biberauer (2005) identify as“spec pied-piping” languages as defined in §2.

Fourth, the existence of unambiguously in-situ subjects and sentences lack-ing overt expletives in SpecTP provide further evidence for vP-raising to SpecTP.These phenomena are illustrated in (15):

(15) a. Andand

inin

þisthis

tymetime

werewere

sentsent

writtiswrits

þorowoutethroughout

þethe

londcountry

“And in this time, writs were sent throughout the country”(Capgrave Chronicles 213.72; Haeberli’s (29a), p. 420)

b. Andand

inin

alleall

thethe

worldworld

isis

nono

grettergreater

treson . . .treason . . .

“And in all the world there is no great treason . . .”(Prologues, Caxton 12.15; Haeberli’s (23a), p. 406)

Finally, it is worth pointing out that very many simple sentences in ME would havebeen compatible with a vP-fronting analysis. This is true of V2 sentences generally.Thus, for example, any sentence with the order Adverb – Verb – Subject – Objectwas amenable to either of the two analyses given in (16):

(16) a. [CP Adv [C V ] [TP [vP Su [v’ [v (V+v) ] ] [T (V+v)] [VP (V) Ob ] ]]b. [CP Adv [C V ] [TP Su [T (V+v)] [vP (Su) [v’ [v (V+v) ] [VP (V) Ob ] ]]]

(We do not indicate the trace of the Adverb, as its position depends on the typeof adverb, which is not essential for the point being made here). These structureswill always be ambiguous since there will never be any overt material to be sent toSpellout in T, given the T-to-C movement operation operative in the derivation ofV2-clauses. When V2 was lost (c1450, according to van Kemenade 1987; Haeberli1999a; Fischer et al. 2000), this ambiguity disappeared, which played a crucial rolein the loss of vP-fronting (this idea is further developed in Biberauer & Roberts2006a, b; and Biberauer 2006).

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.17 (1002-1033)


As noted in the introduction to this paper, B&R’s analysis predicts a connec-tion between the loss of surface head-final-type orders and the development ofobligatory overt expletives in SpecTP, as well as obligatory subject-raising to thisposition: as soon as vP-raising is lost as a possible means of satisfying T’s EPP-requirements, subject-/expletive-raising remains as the only available mechanismto achieve EPP-satisfaction. We know from the work of researchers like Breivik(1990), Haeberli (1999a), Williams (2000) and Ingham (2001) that expletives werestill very commonly omitted during the Early ME period, particularly in clause-internal position (cf. Biberauer 2004). In terms of B&R’s analysis, this possibilitycorrelates directly with the availability of vP-raising, a mechanism which would, aswe have seen, also be expected to systematically give rise to surface head-final or-derings as well as the orders and ambiguities that we have discussed in this section.We have now seen that there is a range of phenomena present throughout the MEperiod up until roughly 1450 which could support the postulation of a grammarwith vP-movement, despite the relatively low frequency of V-Aux and OV order.Additionally, there are a number of high-frequency structures (e.g. V2 clauses)which are ambiguous in respect of the manner in which T’s EPP-requirements aremet. Taken together, therefore, we conclude that B&R’s optionality-based proposalis indeed empirically supported.

B&R’s analysis also predicts that surface head-final orderings and Styl-F phe-nomena will no longer surface once expletives and, more generally, the kind of“raising-to-subject” phenomena characteristic of Modern English, become oblig-atory (i.e. mid-15th century), and that the disappearance of these orders is con-nected to the loss of V2 (cf. Biberauer & Roberts 2006a, b). This prediction isborne out, not just in Modern English but, significantly, also in other Germaniclanguages which have previously been said to have permitted either V-Aux order-ings or Styl-F or both and now systematically require DP-raising to SpecTP (cf.the Mainland Scandinavian languages, as discussed in i.a. Holmberg & Platzack1995; see also Biberauer 2004, 2006): in all cases, it is possible to connect theunavailability of V-Aux ordering and/or Styl-F to the fact that these languagespresently require DP-raising and only DP-raising (i.e. spec-raising) to satisfy T’sEPP-requirements. In all cases, V2 has survived in matrix clauses, thereby guar-anteeing the SpecTP-related ambiguity noted above; crucially, however, all ofthese languages underwent ME-style loss of V2 in embedded clauses, with theconsequence that the relevant ambiguities were no longer attested in embeddedcontexts: it became clear that embedded T’s EPP-requirements are satisfied viaDP-raising. Significantly, this embedded change appears to have affected the anal-ysis of matrix clauses, with the establishment of an exclusively spec-/DP-raisingEPP-satisfaction mechanism in that context “biasing” the grammar generally infavour of DP-raising, i.e. the loss of the vP-fronting option. If this is correct, thisdevelopment would seem to argue against Lightfoot’s (1991) Degree Zero Learn-

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.18 (1033-1093)


ability proposals (cf. also §4.2 for discussion of a further instance of change whichappears to have spread from embedded to matrix contexts).

. Verb (Projection) Raising alternations

In this section, we look at so-called Verb (Projection) Raising (V(P)R) triggered bypremodals, the forerunners of the modern English modals which are traditionallythought to have started out as elements selecting a clausal complement, i.e. to haveoccurred in biclausal structures. As first pointed out in van Kemenade (1987:55ff.),modal, causative and perception verbs were optional V(P)R (i.e. restructuring)triggers in OE, a state of affairs that entailed that the infinitival Vs selected bythese verbs could follow their selectors, as illustrated in (17):5

(17) a. . . . þewho

æfreever

onin

gefeohtebattle

hishis

handahands

woldewould

afylandefile

“. . . whoever would defile his hands in battle”(Ælfric’s Lives of Saints 25.858; Pintzuk’s (62), p. 102)

b. . . . þætthat

hithey

mihtoncould

swaso

bealdliceboldly

GodesGod’s

geleafanfaith

bodianpreach

“. . . that they could preach God’s faith so boldly”(ÆCHom I, 16.232.23; cited in Fischer et al. 2000:156)

Although it is generally agreed that the OE (and ME) modals were main verbs,modal-containing examples consistently feature in discussions about the locationof Infl (medial or final) and, importantly in the context of the present discussion,of Aux relative to V. We will follow this practice here as we assume that the OE andME modals (exceptionally among the class of finite verbs; see §4.3) underwentraising to T and we also do not exclude the possibility that some of these verbsmay in fact be merged in the T-domain (see §2 above and §4 below). What wewould like to show now is what role these verbs played in the loss of vP-raisingand the associated change in the nature of SpecTP in ME. We concentrate on thiscase to the exclusion of aspectual auxiliaries with participial complements sinceit is less clear that the latter consistently involved biclausal structures during theME period.

. Here we leave aside for future research the question of why the OE premodals appear toexhibit optionality in respect of whether they selected “full” vs restructuring complements(cf. the fact, pointed out to us by an anonymous reviewer, that 3456/8649 [i.e. 39.96%] OEsubordinate clauses taken from Haeberli & Pintzuk 2004 featured V-Aux order rather thanrestructuring Aux-V).

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.19 (1093-1117)


We consider a VR structure of the kind illustrated in (17a) by way of illustra-tion:6

(18) TP

vP2 T’

T vP2

vP1

DP-Subj v’

v’

v VP

TPINFVR

T’vP1

T vP1

V+v T

where the vP labelled vP1 for expository convenience has the following internalstructure (bracketed elements are those which have undergone movement out ofvP1):

(18’) vP1

VP

DPPRO

v’

v’

(V+v) (VP)(V) DP-Obj

We are assuming that the complement of a restructuring verb is a TP (cf. i.a.Wurmbrand 2001; and Lee-Schoenfeld 2005 for arguments in favour of the well-

. Here we indicate the subject of the infinitive as PRO. We do this largely for convenience,remaining on the one hand agnostic regarding the correct analysis of control (see Hornstein1999; Manzini & Roussou 2000), and on other hand not wishing to imply that restructuringpredicates are never raising predicates.

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.20 (1117-1214)


established idea that restructuring complements are “smaller” than other clausalcomplements, and Roberts 1997a:412 for arguments that such complements arelarger than VP in OE). In the context of the theoretical framework we are assuminghere, the specific assumption is that restructuring complements are TPs headed bya “defective” T, i.e. one that is not selected by C (cf. B&R 14ff., Chomsky 2004,2005). For our present purposes, this idea has the important consequence that thematerial in the restructuring complement is not sent to Spellout prior to mergerof VR, the way material in the clausal complements of non-restructuring verbsis (owing to the PIC; cf. (9) above). This accounts for the “clause union” effectscommonly associated with restructuring structures. Let us see how our analysis ofV(P)R works in more detail.

The derivation of the VR order in OE (17a) proceeds by the following steps.First, V moves to v inside the vP of the embedded clause. We take this to be astandard and possibly universal operation (cf. Marantz 1997; Chomsky 2004:122);it certainly holds for all periods of earlier English (cf. Zwart 2005 on the non-adjacency of V-Obj in earlier English). Second, the remnant VP moves to SpecvP(cf. the fact that not just the object, his handa, but additionally also a loca-tive/situative PP, on gefeohte, which may reasonably be thought of as having beenmerged inside VP, surfaces preverbally in (17a)).7 Third, V+v moves to T in theinfinitival clause. This operation is the key to deriving the Aux-V order here. Wetake it that this infinitive-movement is triggered by a selectional property of themain-clause verb VR. Specifically, we propose the relevant selectional property tobe the nature of the (defective) TP that VR selects (cf. Evers 1975; and Haege-man & van Riemsdijk 1986 on VPR and the verbs that trigger it, and Kayne 1991;

. It should, of course, be noted that object-raising (i.e. stranding rather than pied-piping)would also have been available as an alternative to VP-raising at this stage of the OE deriva-tion. We would therefore expect OE also to exhibit structures in which non-object VP-internalmaterial surfaces after the infinitival V, and leaking structures of this kind are in fact attested:

(i) <Meone

schal>shall

leouedear

sustrensisters

þeosethese

storienstories

tellentell

eftafterwards

outo-you

“One shall tell these stories to you afterwards, dear sisters”(Ancrene Riwle II.122.1552; Kroch & Taylor (2000)’s (32), p. 155)

Here both an adverbial (eft) and an indirect object pronoun (ou) surface to the right of theverb (tellen) which we assume to have raised to adjoin to infinitival T (cf. main text); the directobject (þeose storien) contrasts with the just-mentioned VP-material in surfacing to the left oftellen, indicating that it must have undergone movement. Following the loss of VP-raising (i.e.the pied-piping means of satisfying v’s EPP-requirements), only object raising would have beenavailable, with the result that structures of this type would be expected to have been even morefrequent than during the OE period where they still alternated with the VP-raising type illus-trated in (17a) in the main text. The consequences of the loss of obligatory object shift (cf. §2)are discussed in more detail in §4.1 and 4.2.

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.21 (1214-1240)


Roberts 1997b, on the connection between restructuring and infinitive-movementin Romance). We argue that this selectional property also accounts for the absenceof the infinitive marker to in the complements to these verbs: since the selectedT triggers infinitive-movement to itself, there is no position for the infinitival T-marker (cf. the often-noted absence of zu or te in such complements in Germanand Dutch respectively).8 The next step in the derivation of the VR structure in(17a) is then vP movement to the specifier of the selected T (this is another in-stance of “pied-piping” satisfying an EPP-feature); we return to the consequencesof employing the DP-raising/non-pied-piping mode of EPP-satisfaction in §4.2below, but note here only that we do not exclude it and that structures derivedvia this mode of EPP-satisfaction (i.e. “pseudo” modern English-like S-Aux-V-Ostructures) are indeed attested in both OE and ME (cf. B&R for further discus-sion). Returning to the derivation of (17a): following merger of VR and V-to-vraising, the vP is raised to the specifier of the matrix vP under the influence ofmatrix v’s EPP-feature, and the (matrix) external argument is merged. Follow-ing completion of the vP phase, the material in v’s complement (thus, crucially,the embedded V which has undergone raising to infinitival T) is sent to Spellout,whereafter matrix T is merged. T then Agrees with and attracts VR in accordancewith our assumptions about modals, and it also probes for D-features, which mustthen be raised to SpecTP in order to satisfy T’s D-related EPP-requirements. Atthis point, the same two options that apply in the monoclausal environments dis-cussed above are once again available: either the subject DP or the vP can undergoraising to SpecTP. VR structures like (17a) are derived via vP-raising (i.e. pied-piping), whereas VPR structures like (17b) are the result of DP-raising. On ouranalysis, therefore, the difference between VR and VPR structures lies solely inthe manner in which matrix T’s EPP-requirements are met. This would seem tobe a desirable outcome as modern counterparts of these structures – attested inmany non-standard German and Dutch varieties and also in spoken Afrikaans –are interpretively equivalent (Hans Broekhuis, Liliane Haegeman, p.c.), i.e. theyinstantiate precisely the kind of “true optionality” that our optional pied-pipinganalysis would predict (cf. further discussion in Biberauer & Richards 2006).

. It is worth noting that to was not as systematically found in infinitives in ME as it is in present-day English (cf. Los 1999, 2005). In particular, it consistently failed to appear in the complementsof modals, a characteristic which the modern-day English modals, with the well-known excep-tion of ought, retain and which has also frequently been said to have played an important rolein the reanalysis of the English modals (cf. Lightfoot 1979; Roberts 1985, 1993 and the discus-sion in §4). It is possible that ought in fact derives from some other source than verb-projectionraising, since its cognates in German and Dutch do not trigger verb-projection raising. We leavethis matter for further research.

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.22 (1240-1274)


The above discussion has therefore shown the optionality in respect of themanner in which T’s EPP-requirements are met is not restricted to monoclausalcontexts; exactly the same optionality appears to have been available in biclausalV(P)R structures. In connection specifically with VPR structures, it is furtherworth noting that the same optionality is also likely to have been available in thelower clause, i.e. like matrix T, infinitival T’s EPP-requirements could also be meteither via vP- or DP-raising in VPR structures, whereafter the contents of infiniti-val SpecTP undergo raising into the matrix clause as outlined above. As both vP-and DP-raising in the lower clause will ultimately deliver the requisite S-Aux-O-Vstring, it is clear that the precise manner in which infinitival T’s EPP-requirementsare met is opaque. Like the V2 structures discussed in §3.1 above, VPR structurestherefore constitute ambiguous input in respect of the manner in which matrix T’sEPP-requirements are satisfied. Both VR and VPR structures would thus have con-stituted points of ambiguity in the input available to the OE and ME child, pointsof ambiguity that, we believe, may well have played an important role in allowingchildren to continue postulating optionality-permitting grammars until approxi-mately the mid-15th century when the combined weight of evidence in favour of agrammar that permitted only one of the original options (i.e. DP-raising) becametoo great. We will consider this latter point in more detail in §4 below.

. Conclusion

In this section, we have seen that residual head-final structures and also vari-ous other structures that are usually regarded as anomalous in the ME contextcan be given a unified analysis as the output of a single grammar, one which li-censes both DP- and vP-raising to satisfy T’s EPP-requirements. Specifically inconnection with the two kinds of V-Aux order usually thought to coexist in ME –Styl-F and a residue of “genuine” V-Aux orders reflecting the OE system – we haveshown that a unified analysis in terms of vP-raising eliminates numerous problemsassociated with more traditional ‘grammars in competition’-based analyses. Addi-tionally, a vP-raising analysis facilitates an understanding of various coterminouschanges that are usually viewed as independent and unrelated, i.e. as accidentallycoterminous. We have also shown that various ME structures, including simpleverb-containing V2 and V(P)R structures, would have been ambiguous in respectof the manner in which T’s EPP-requirements are satisfied, with the result thatthe learner could interpret these structures either way, with other evidence in theinput therefore proving crucial in maintaining the optionality-licensing grammar.Next, we turn to some of the consequences of what we have proposed here.

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.23 (1274-1329)


. Some consequences

The analysis of V(P)R in ME put forward in the previous section has three interest-ing consequences. These concern the nature of the order subject–modal–negativeobject–infinitive in 15th-century English, the ultimate loss of vP-movement toSpecTP, and the reanalysis of modal verbs (a subclass of the class VR men-tioned in the previous section; see van Kemenade 1993). We now describe eachof these in turn.

. OV orders with modals in 15th-century English

In terms of our analysis, we can account for van der Wurff (1997, 1999) and Ing-ham’s (2002, 2003) observation that in 15th-century English, the only survivingOV orders involving non-pronominal objects are of the general type subject–modal–negative object–infinitive as illustrated in (19):

(19) þeithey

shuldshould

nono

meyhirmayor

hauehave

(Capgrave Chronicles 199.6; cited in Fischer et al. 2000:163)

As we mentioned in the previous section, modals are the principal exponents ofthe class VR. We would therefore expect “clause union” effects in the complementsof modals, with the result that objects and other infinitival vP-material can surfacebefore the matrix verb (VR). Note, however, that VP-raising to SpecvP had alreadybeen reanalysed as object-movement by the ME period (i.e. a variety of factors,discussed in more detail in Biberauer & Roberts 2006a, c had obscured the inputjustifying the postulation of the pied-piping option, with the result that the oper-ation of the Subset Principle had led to the stranding option having become thenorm).9 Thus the analysis discussed in the preceding section actually only predictsthe availability of the lower-clause object to the left of V during ME. Note further,though, that object-movement became restricted to negative or quantified DPs inlater ME (cf. van der Wurff and Ingham, op. cit.). B&R analyse this as a restriction

. Biberauer & Roberts (ibid.) identify the paucity in early ME of unambiguous “signposts” forthe pied-piping (i.e. rigidly head-final) grammar as the reason for the demise of VP-pied-piping.Thus, for example, the fact that particle verbs, were, as noted by Spasov (1966; cited in Kroch &Taylor 2000:146), vanishingly rare during the 12th and 13th centuries removed one of the clearindicators that the OE grammar at least had the option of requiring all VP-internal materialto raise into the vP-domain. Similarly, the loss of dative case and the compensatory encod-ing of indirect objects via PPs led to an increase in “leaking”/extraposition structures featuringa post-verbal indirect object, which further weakened the evidence for VP-piedpiping (whichwould have produced structures featuring the indirect object and other VP-internal material inpreverbal position).

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.24 (1329-1344)


on the nature of v’s object-movement-triggering D-feature: whereas D on v wasassociated with a movement-triggering EPP-feature during OE, a sub-feature of D(Neg) bore this feature during the later ME period (cf. also Biberauer & Roberts2006a). A structure like (18) was thus reanalysed in ME as (20):

(20) TP

vP2 T’

T vP2

vP1

DP-Subj v’

v’

v VP

TPINFVR

T’vP1

T vP1

V+v T

Already sent to Spellout: VP

(V) (DP-Obj)

where vP1 has the following internal structure (bracketed elements once again be-ing those that have undergone movement out of vP1 or, at the appropriate point,been sent to Spellout and material which is rendered inaccessible by the PIC isonce again given in outline font):

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.25 (1344-1411)


(20’) vP1

DPPRO

v’

v’

(V+v) (VP)

DP-Obj

In (20) we have object-movement to SpecvP, triggered by the specialised D-featureon v, and vP-movement to SpecTP in the lower clause. Exactly the kind of structureillustrated in (19) is thus predicted, given (a) the analysis of V(P)R put forward inthe previous section, (b) the idea that former remnant VP-movement was reanal-ysed as object-movement in Early ME, and (c) the idea that only negative objectscould undergo object-movement by Late ME.

A question that arises at this point is what would prevent the matrix v fromalso having the relevant type of D-feature. This would attract the object into themain clause, giving, other things being equal, the unattested order subject – object –auxiliary – verb. We propose that these orders are not found for an independentreason, namely the fact that in finite clauses V raises to T at this period (see §4.3and also Pollock 1989; Roberts 1985, 1993, 1999). Hence the finite verb (VR orauxiliary) will always precede a raised object. We therefore do not absolutely ex-clude this possibility, but conclude that, in general, it has no effect on surface wordorder and certainly does not give rise to subject – object – auxiliary – verb orders.

In fact, there is some evidence for object-raising into the higher clause fromthe rare examples where VR is also non-finite, giving rise to two infinitive clauses.One such example is given in Beukema & van der Wurff (2000):

(21) . . . soso

foulfoul

þatthat

þouthou

schaltshalt

nou3hhtnought

elleselse

mowemay(infin)

sesee

þerynnetherein

botbut

fylþehedefoulness

andand

wrecchednessewretchedness

“so filthy that you will not be able to see anything else in it but foulness andwretchedness”(The Fyve Wyttes, p. 14,1.28ff.; cited in Beukema & van der Wurff 2000:86)

This is exactly predicted by our analysis.A further point should be mentioned in this connection. It is well–known that

some modals were defective in argument structure in ME (see Warner 1983, 1993),and that the reanalysis of modals involved the loss of argument structure (Light-foot 1979; Roberts 1985, 1993; Roberts & Roussou 2003). It is therefore likelythat, by Late ME, at least some members of VR were associated with a defectivev, one which lacks the relevant D-feature by virtue of its lack of argument struc-

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.26 (1411-1450)


ture. There is some indication that v associated with modals could be defective inME, from examples with what appears to be a quirky dative subject:

(22) hwiwhy

mime

ouhought

andand

hwiwhy

meme

scalshall

iesuJesus

cristChrist

luuienlove

(Ancr. R. (EETS 1952) 6.23, Visser 1963–1973, §1712; cited in Roberts1985:38)

To the extent that verbs with a quirky dative subject argument resemble unac-cusatives (cf. Belletti & Rizzi 1988), this kind of example would feature a defectivev in the main clause, i.e. one which does not bear an EPP-feature and consequentlycannot attract a negative object to its specifier.

Finally, in order to account for van der Wurff ’s observations about the natureof relic OV-orders, we also need to exclude a monoclausal structure in which theobject raises, giving SOV order. This can clearly be excluded by appealing to V-movement to T (see §4.3 below) or C, to the extent that the language was still V2at this point, on which see Fischer et al. (2000:129ff.): even if negative object shifttakes place in this case, its effects will be masked by V-to-T(-to-C) movement.

Hence the analysis we propose correctly predicts that the last OV structures inME would have presented the surface ordering illustrated in (19) (where the objectis not a pronoun; pronominal object shift survived into Early Modern English, asshown in Roberts 1995).

. The loss of vP-movement

The second consequence of the analysis of V(P)R proposed above is that we cansee how the structure corresponding to (20) with the type of object that by the15th century fails to raise (i.e. a non-negative object) led to the loss of vP-raisingas a means of satisfying T’s EPP-requirement.

To see how this works, consider the variant of the structure in (20) (whichrepresents the structure of examples like (19)) without object-movement:

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.27 (1450-1490)


(23) TP

T’DP-Subj

(DP-Subj)

vP

T’vP1

T

V+v T

T

DPPRO

(V+ )v ( P )v 1

v’

v VP

VR TP


(V) DP-Obj

As in the case of the adverbial PP in (8), the VP indicated in outline here is mergedas the complement of the lower v, and thanks to the operation of the PIC (as inChomsky 2000:108), this material is sent to Spellout and therefore becomes in-accessible for further operations as soon as the lower vP is completed. Hence,movement of this vP to SpecTP has no effect on the surface position of the ob-ject, which remains final. What this means is that the choice between pied-pipingvP to the lower SpecTP and exclusively raising the subject to that position, whichwas operative throughout ME, has absolutely no effect on the surface order of el-ements, since the only overt material in vP which the PIC allows to be spelled outin its moved position is the subject.

Because of the PIC then, later ME acquirers had no clear evidence in VR-containing contexts to distinguish a derivation involving pied-piping of vP tosatisfy T’s EPP-requirement from one in which only the subject moves to satisfythat requirement: as soon as objects do not raise as readily as they used to in OEand earlier ME, with the only moving objects being negated ones which only movein clauses involving some form of “obscuring” V-movement (cf. §4.1), it becomesmuch harder to distinguish DP-raising from vP-raising in VR-structures (recallthat V does not come into play as an element contained in vP here as it has alreadybeen sent to Spellout in accordance with the PIC). It is of course possible that thepresence of vP-adverbials or other modifiers might disambiguate the two deriva-

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.28 (1490-1565)


tions (see §3.2 above), but in the vast majority of cases, the ambiguity would havebeen present. We take it that this situation led to the reanalysis of an embedded TPwith the structure in (24) (which corresponds to the embedded TP in (23)) as onewith the structure in (25):

(24) TPINF

vP T’

DPPRO

v’ T tvP

( )v V+v T


(V) DP-Obj

(25) TPINF

vP

T’DPPRO

v’

T

( )v

V+v T (DP)

VP

(V) DP-Obj

We propose that this is how the general option of pied-piping vP to SpecTP waslost in V(P)R infinitival contexts. As the structure in (24) shows, the fronted vP ininfinitival contexts may have contained no overt material at all: an empty subject(here indicated as PRO) and the trace/copy of v. Recall that VP has already beensent to Spellout, and hence is not realised in the moved position. Given the lackof evidence for vP-movement, the simpler option of DP-movement was preferred(assuming that language acquirers always take the simplest option consistent withthe trigger experience, where simplicity is taken to mean the smallest structureconsistent with the input – see Clark & Roberts 1993).10

. Note that the phrase consistent with the trigger experience is crucial here: as argued in de-tail in Biberauer & Richards (2006), economy considerations do not rule out the co-existence oftwo means of satisfying a given EPP-feature, i.e. they do not rule out the existence of optionality-

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.29 (1565-1575)


What we would now like to suggest is that V(P)R structures represent anothercontext in which changes in embedded-clause syntax in fact triggered a change inthe matrix-clause system, once again contra Lightfoot (1991). More specifically,we propose that the reanalysis indicated in (24) and (25) was not confined tothe TP complements of VR, but that it was actually extended to the matrix TPin VR-containing structures and also to matrix TPs more generally, i.e. also tothose in monoclausal contexts. What we would like to propose is that the factthat DP-raising was the only available option in contexts like (24)–(25) com-bined with the pre-existing availability of DP-raising as an option for satisfyingT’s EPP-requirements in matrix contexts strongly “biased” learners in favour oftaking the DP-raising option wherever they could, especially given the generalsimplicity preference just mentioned. Ultimately, this therefore led to the loss ofvP-movement (cf. B&R 24f.). As noted in §3.1, the consequences were that only aDP can satisfy T’s EPP-requirement, and hence expletive insertion became oblig-atory in contexts where no argumental DP was available for raising (see B&R:ibid. for further discussion). Furthermore, “Styl-F”, as analysed in §3.1, was lostas this possibility crucially depends on the availability of vP-fronting. Our analy-sis therefore predicts that Styl-F will not be available in languages which requireEnglish-style DP-movement in order to satisfy T’s EPP-requirement (i.e. languageswhich do not at least allow vP-raising as an option). This prediction would seemto be correct as Styl-F was also lost in the Mainland Scandinavian languages, all ofwhich underwent a subject-related change very similar to that which occurred inEnglish (cf. Biberauer 2004, 2006 for further discussion).

. The reanalysis of the modals in ENE

We now turn to the connection between the 15th-century reanalysis described inthe previous section and the well-known reanalysis of a subclass of the members of

permitting pied-piping grammars and there is no sense in which one can speak of an inherent“cost” associated with a grammar that permits options. Economy considerations do, however,come into play in the acquisition context, where a “bigger” grammar qualifies as “less economi-cal” in Subset terms and therefore needs to be robustly triggered (cf. Biberauer & Roberts 2006a,cfor more detailed discussion). There is therefore no contradiction involved in maintaining, onthe one hand, that a pied-piping grammar is “costless” – this is the case whenever it is suffi-ciently robustly triggered by the input – while asserting, on the other, that acquirers will not,for reasons of acquisitional economy, postulate a grammar of this kind whenever the input doesnot clearly support it. In the current context, we can thus say that maintaining the relevant kindof pied-piping grammars was “economical” in OE and earlier ME because it was empiricallymotivated, but that the same sort of system became “uneconomical” at the relevant points inEnglish’s history, likewise for input-related reasons.

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.30 (1575-1628)


VR, the modals, as auxiliaries in the early 16th century (see Lightfoot 1979; Roberts1985, 1993; Roberts & Roussou 2003).

Consider again the structure of a sequence containing a modal with an infini-tival complement after the reanalysis of (24) as (25). Following Roberts (1993:262)and Roberts & Roussou (2003:41–42), we take it that the loss of infinitival inflec-tion, which had taken place by 1500, removed the trigger for V-to-T movement inthe complement to VR (the assumption is therefore that the infinitival inflectionspecifically instantiated features that not only entered into an Agree relationshipwith T, but V also had to undergo movement under the influence of an associatedEPP-feature; cf. Biberauer & Roberts 2006b for further consideration of the corre-lation between overtly realised inflection and movement). In this way, the evidencefor the lower functional T-v system was removed from the trigger experience of ac-quirers. Hence (25) was reanalysed in the early 16th century as monoclausal, withthe modals merged in v or T and the lexical verb remaining in V – cf. (26):11

(26) TP

DP-Subj T’

TModal

vP

(Subj) v’

v VP

V Obj

As pointed out by Roberts (1985, 1993, 1999) and Warner (1997), this reanalysisin turn contributed to the conditions for the loss of (finite) V-to-T movement laterin ENE by creating a system in which T could always be realised by an auxiliary.The fact that do underwent the same reanalysis as the modals at about the sametime (see Roberts 1993:292ff.) is important in this connection since it meant thatany verb and any tense could be associated with an auxiliary. In other words, thetrigger for V-to-T raising was obscured by the development of a class of auxiliaries(Roberts 1999:293).

. This reanalysis cannot apply to ought, which has always been able to take a to-infinitive com-plement, but, as mentioned in Note 8, we consider it likely that this particular modal has adifferent diachronic source from the others.

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.31 (1628-1662)


Hence we see a clear case of a cascade of changes, each motivated by the onebefore, leading by late ENE to a radically different syntactic system from that ofEarly ME. We summarise the relevant changes as follows:

(27) a. Loss of VP-to-SpecvP pied-piping, and its replacement by object-move-ment (12th century; B&R);

b. Restriction of object-movement to negative objects (ca1400; van derWurff 1997, 1999 and Ingham 2002, 2003);

c. Loss of vP-to-SpecTP movement, and its replacement by subject-move-ment/expletive-insertion; and loss of Styl-F (15th century; B&R and§3.1 above);

d. Reanalysis of modals from VR to auxiliaries (ca1525–1550; Lightfoot1979; Roberts 1985; Warner 1997);

e. Loss of V-to-T movement (ca1575–1600; Kroch 1989; Roberts 1993;Warner 1997).

The Modern English system of do-support emerged in the 17th century and wascaused partly by other factors which space considerations do not permit us toconsider here (but see Biberauer & Roberts 2006b where we argue that this isconnected to the development of contracted negation in n’t; and see also Warner1997; Biberauer & Roberts 2006a offers a more detailed discussion of an elaboratedversion of the above cascade.).

. Conclusion

This paper proposes a new interpretation of the ME data concerning the loss ofsurface head-final orders, in line with the general proposals in B&R. It entails acompletely novel rethinking of the alleged instances of Styl-F in ME, which weconsider to be V-Aux orders of a standard, formerly productive type. We also putforward an analysis of Verb (Projection) Raising, which has a number of inter-esting consequences for Late ME syntax, notably concerning the last occurrencesof OV order, the rise of a canonical subject position and the concomitant loss ofStyl-F and the 16th-century reanalysis of earlier V(P)R triggers as auxiliaries. Fi-nally, we were able to propose the sequence of changes in (26). Here we observe aninteresting “domino effect” of syntactic changes, which may be of theoretical sig-nificance and certainly represents a fine-grained and interesting descriptive pictureof the development of a number of aspects of English syntax.

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.32 (1662-1769)


References

Ackema, P. & Neeleman, A. (2003). Context-sensitive spell-out. Natural Language and LinguisticTheory, 21, 681–735.

Alexiadou, A. & Anagnostopoulou, E. (1998). Parametrizing AGR: Word order, V-movementand EPP-checking. Natural Language and Linguistic Theory, 16, 491–539.

Baker, M., Johnson, K., & Roberts, I. (1989). Passive arguments raised. Linguistic Inquiry, 20,219–251.

Barbosa, P. (1995). Null Subjects. PhD dissertation, MIT.Belletti, A. & Rizzi, L. (1988). Psych verbs and θ-theory. Natural Language and Linguistic Theory,

6, 291–352.Berwick, R. (1985). The Acquisition of Syntactic Knowledge. Cambridge MA: The MIT Press.Beukema, F. & van der Wurff, W. (2000). Modals, objects and negation in late Middle English.

In S. Barbiers, F. Beukema, & W. van der Wurff (Eds.), Modality and its Interaction with theVerbal System (pp. 75–102). Amsterdam: John Benjamins.

Biberauer, T. (2003). Verb Second (V2) in Afrikaans: A minimalist investigation of word-ordervariation. PhD dissertation, University of Cambridge.

Biberauer, T. (2004). Reconsidering the EPP and Spec-TP in Germanic. In L. Astruc & M.Richards (Eds.), Cambridge Occasional Papers in Linguistics (COPiL), 1 (pp. 15–40).

Biberauer, T. (2006). Changes at the edge: Tracking the rise of a canonical subject position inGermanic. Paper presented at Edges in Syntax (Nicosia).

Biberauer, T. & D’Alessandro, R. (2006). Syntactic doubling and the encoding of Voice. Paperpresented at WCCFL21 (Seattle).

Biberauer, T. & Richards, M. (2005). Keeping movement motivated: Optionality without thetears. In F. Chalcraft & E. Sipetzis (Eds.), Cambridge Occasional Papers in Linguistics(COPiL), 2 (pp. 27–45).

Biberauer, T. & Richards, M. (2006). True optionality: When the grammar doesn’t mind. In C.Boeckx (Ed.), Minimalist Essays (pp. 35–67). Amsterdam: John Benjamins.

Biberauer, T. & Roberts, I. (2005). Changing EPP parameters in the history of English:Accounting for variation and change. English Language and Linguistics, 9(1), 5–46.

Biberauer, T. & Roberts, I. (2006a). Cascading parameter changes: Internally driven changein Middle and Early Modern English. In T. Eythórsson (Ed.), Grammatical Change andLinguistic Theory: The Rosendal Papers. Amsterdam: John Benjamins.

Biberauer, T. & Roberts, I. (2006b). Subjects, tense and verb movement in Germanic inRomance. In Proceedings of GLOW V in Asia.

Biberauer, T. & Roberts, I. (2006c). The return of the subset principle: A Germanic case study.Paper presented at CGSW21 (Santa Cruz).

Borer, H. (1986). I-subjects. Linguistic Inquiry, 17, 375–416.Breivik, L. (1990). Existential there: A synchronic and diachronic study. Oslo: Novus Press.Canale, M. (1978). Word Order Change in Old English: Base reanalysis in generative grammar.

PhD dissertation, McGill University.Chomsky, N. (1995). The Minimalist Program. Cambridge MA: The MIT Press.Chomsky, N. (2000). Minimalist inquiries: The framework. In R. Martin, D. Michaels, & J.

Uriagereka (Eds.), Step by Step. Essays on minimalist syntax in honor of Howard Lasnik (pp.89–156). Cambridge MA: The MIT Press.

Chomsky, N. (2001). Derivation by phase. In M. Kenstowicz (Ed.), Ken Hale: A life in language(pp. 1–52). Cambridge MA: The MIT Press.

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.33 (1769-1883)


Chomsky, N. (2004). Beyond explanatory adequacy. In A. Belletti (Ed.), The Cartography ofSyntactic Structures, Volume 3: Structures and beyond (pp. 104–131). Oxford: OUP.

Chomsky, N. (2005). On Phases. Ms., MIT.Cinque, G. (1999). Adverbs and Functional Heads: A cross-linguistic perspective. Oxford: OUP.Clark, R. & Roberts, I. (1993). A computational model of language learnability and language

change. Linguistic Inquiry, 24, 299–345.Evers, A. (1975). The Transformational Cycle in Dutch and German. PhD dissertation,

University of Utrecht.Falk, C. (1993). Non-referential Subjects in the History of Swedish. PhD dissertation, University

of Lund.Fischer, O., van Kemenade, A., Koopman, W., & van der Wurff, W. (2000). The Syntax of Early

English. Cambridge: CUP.Gray, D. (1985). The Oxford Book of Late Medieval Prose and Verse. Oxford: OUP.Haeberli, E. (1999a). Features, Categories and the Syntax of A-positions. Synchronic and

diachronic variation in the Germanic languages. PhD dissertation, University of Geneva.(Published as: Haeberli, E. (2002). Features, Categories and the Syntax of A-positions. Cross-linguistic variation in the Germanic languages. Dordrecht: Kluwer).

Haeberli, E. (1999b). On the word order ‘XP-Subject’ in the Germanic languages. Journal ofComparative Germanic Linguistics, 3, 1–36.

Haeberli, E. (2000). Adjuncts and the syntax of subjects in Old and Middle English. In S.Pintzuk, G. Tsoulas, & A. Warner (Eds.), Diachronic Syntax: Models and mechanisms (pp.109–131). Oxford: OUP.

Haeberli, E. & Pintzuk, S. (2004). Revisiting verb (projection) raising in Old English. Talkpresented at the 8th Diachronic Generative Syntax/DiGS Conference (Yale).

Haegeman, L. & van Riemsdijk, H. (1986). Verb projection raising, scope and the typology ofrules affecting verbs. Linguistic Inquiry, 7, 417–466.

Hinterhölzl, R. (1999). Restructuring Infinitives and the Theory of Complementation. PhDdissertation: University of Southern California. (Published as: Hinterhölzl, R. (2006).Scrambling, Remnant Movement, and Restructuring in West Germanic. Oxford: OUP.)

Holmberg, A. (2000). Scandinavian stylistic fronting: How any category can become anexpletive. Linguistic Inquiry, 31, 445–483.

Holmberg, A. & Platzack, C. (1995). The Role of Inflection in Scandinavian Syntax. New York NY:OUP.

Hornstein, N. (1999). Move! A minimalist theory of construal. Oxford: Blackwell.Ingham, R. (2001). The structure and function of expletive there in pre-modern English. Reading

Working Papers in Linguistics, 5, 213–249.Ingham, R. (2002). Negated subjects and objects in 15th century non-literary English. Language

Variation and Change, 14, 291–322.Ingham, R. (2003). The changing status of Middle English OV order: Evidence from two genres.

Sky Journal of Linguistics, 16, 75–92.Kayne, R. (1991). Romance clitics, verb movement, and PRO. Linguistic Inquiry, 22, 647–686.

(Republished as Kayne, R. (2000). Romance clitics, verb movement, and PRO. In R. Kayne(Ed.), Parameters and Universals (pp. 60–97). New York NY: OUP).

Kayne, R. (1994). The Antisymmetry of Syntax. Cambridge MA: The MIT Press.Kayne, R. (2000). Parameters and Universals. New York NY: OUP.van Kemenade, A. (1987). Syntactic Case and Morphological Case in the History of English.

Dordrecht: Foris.

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.34 (1883-1991)


van Kemenade, A. (1993). The history of the English modals: A reanalysis. Folia LinguisticaHistorica, 13(1), 143–166.

Kroch, A. (1989). Reflexes of grammar in patterns of language change. Language Variation andChange, 1, 199–244.

Kroch, A. & Taylor, A. (1994). Remarks on XV/VX alternation in Early Middle English. Ms.,University of Pennsylvania.

Kroch, A. & Taylor, A. (2000). Verb-object order in Middle English. In S. Pintzuk, G. Tsoulas, &A. Warner (Eds.), Diachronic syntax: Models and mechanisms (pp. 132–163). Oxford: OUP.

Lee-Schoenfeld, V. (2005). Beyond Coherence: The syntax of opacity in German. PhD disserta-tion, University of California at Santa Cruz.

Lightfoot, D. (1979). Principles of Diachronic Syntax. Cambridge: CUP.Lightfoot, D. (1991). How to set Parameters: Arguments from language change. Cambridge MA:

The MIT Press.Los, B. (1999). Infinitival Complementation in Old and Middle English. The Hague: HAG.Los, B. (2005). The Rise of the TO-infinitive. Oxford: OUP.Maling, J. (1990). Inversion in embedded clauses in Modern Icelandic. In J. Maling & A. Zaenen

(Eds.), Modern Icelandic Syntax (pp. 71–91). San Diego CA: Academic Press.Manzini, R. & Roussou, A. (2000). A minimalist theory of A-movement and control. Lingua,

110, 409–447.Marantz, A. (1997). No escape from syntax. University of Pennsylvania Working Papers in

Linguistics, 4(2), 201–225.Moro, A. (2000). Dynamic Antisymmetry: Movement as a symmetry-breaking phenomenon.

Cambridge MA: The MIT Press.Mossé, F. (1968). Manual of Middle English. Baltimore MD: Johns Hopkins Press.Müller, G. (2004). Verb-second as vP-first Journal of Comparative Germanic Linguistics, 7, 139–

274.Pintzuk, S. (1991). Phrase Structures in Competition: Variation and change in Old English

word order. PhD dissertation, University of Pennsylvania. (Published as Pintzuk, S. (1999).Phrase Structures in Competition: Variation and change in Old English word order. New York:Garland).

Platzack, C. (1988). The emergence of a word order difference in Scandinavian subordinateclauses. McGill Working Papers in Linguistics: Special Issue on Comparative Germanic Syntax,215–238.

Pollock, J-Y. (1989). Verb movement, UG and the structure of IP. Linguistic Inquiry, 12, 365–424.Richards, M. & Biberauer, T. (2004). Optionality, loss of inflection and the rise of expletives:

Why Faroese is a VO Afrikaans. Paper presented at the Workshop on Faroese (Reykjavik).Richards, M. & Biberauer, T. (2005). Explaining Expl. In M. den Dikken & C. Tortora (Eds.),

The Function of Function Words and Functional Categories (pp. 115–154). Amsterdam: JohnBenjamins.

Roberts, I. (1985). Agreement parameters and the development of the English modal auxiliaries.Natural Language and Linguistic Theory, 3, 21–58.

Roberts, I. (1993). Verbs and Diachronic Syntax. Dordrecht: Kluwer.Roberts, I. (1995). Object movement and verb movement in Early Modern English. In H.

Haider, S. Olsen, & S. Vikner (Eds.), Studies in Comparative Germanic Syntax (pp. 269–284).Dordrecht: Kluwer.

Roberts, I. (1997a). Directionality and word order change in the history of English. In A.van Kemenade & N. Vincent (Eds.), Parameters of Morphosyntactic Change (pp. 397–426).Cambridge: CUP.

JB[v.20020404] Prn:24/10/2006; 14:43 F: LA9709.tex / p.35 (1991-2061)


Roberts, I. (1997b). “Restructuring, Head Movement and Locality”. Linguistic Inquiry, 28, 423–460.

Roberts, I. (1999). Verb movement and markedness. In M. de Graff (Ed.), Language Changeand Language Creation: Creolization, diachrony and development (pp. 287–327). CambridgeMA: The MIT Press.

Roberts, I. & Roussou, A. (2003). Syntactic Change. A minimalist approach to grammaticalization.Cambridge: CUP.

Rutten, J. (1991). Infinitival Complements and Auxiliaries. PhD dissertation, University ofAmsterdam.

Spasov, D. (1966). English Phrasal Verbs. Sofia: Naouka Izkoustvo.Trips, C. (2002). From OV to VO in Early Middle English. Amsterdam: John Benjamins.Warner, A. (1983). Review of Lightfoot 1979. Journal of Linguistics, 19, 187–209.Warner, A. (1993). English Auxiliaries: Structure and history. Cambridge: CUP.Warner, A. (1997). The structure of parametric change and V-movement in the history of

English. In A. van Kemenade & N. Vincent (Eds.), Parameters of Morphosyntactic Change(pp. 380–393). Cambridge: CUP.

Williams, A. (2000). Null subjects in Middle English existentials. In S. Pintzuk, G. Tsoulas, & A.Warner (Eds.), Diachronic Syntax. Models and mechanisms (pp. 164–187). Oxford: OUP.

van der Wurff, W. (1997). Deriving object-verb order in late Middle English. Journal ofLinguistics, 33, 485–509.

van der Wurff, W. (1999). Objects and verbs in Modern Icelandic and fifteenth-century English:A word order parallel and its causes. Lingua, 109, 237–265.

Visser, F. (1963–1973). An Historical Syntax of the English Language. Leiden: Brill.Wurmbrand, S. (2001). Infinitives. Restructuring and clause structure. Amsterdam: John

Benjamins.Zwart, J-W. (1997). Morphosyntax of Verb Movement: A minimalist approach to the syntax of

Dutch. Dordrecht: Kluwer.Zwart, J-W. (2005). A comparative approach to syntactic change in the history of English. English

Language and Linguistics, 9, 157–179.

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.1 (55-151)

Syntactic sources of word-formation processes

Evidence from Old English and Old High German

Carola TripsUniversität Stuttgart

This paper surveys word-formation from a diachronic perspective and the ques-tion of whether word-formations are built by the same principles that governsyntax. It is assumed that word-formations like compounds and derivations his-torically start out as syntactic phrases and in the process of becoming morphol-ogical phrases lose structural syntactic properties like maximal projections andfunctional categories as well as semantic properties like e.g. referentiality. Thiswill be shown with diachronic data from German and English focusing on thephenomena of the development of suffixes like Modern English -hood, ModernGerman -heit, and the rise of genitive compounds. Based on these findings it willbe claimed that an analysis like Lieber (1992) or Ackema (1999) assuming thatmorphological operations are governed by syntactic principles is not borne outand that word-formation operations have to be attributed to an independentmodule of word-formation subject to its own governing principles. Nevertheless,the rise of genitive compounds shows that new syntactic structures can occuronce old syntactic structures have developed into morphological structures im-plying that there is interaction between syntax and morphology. Thus, looking atword-formation from a diachronic perspective provides new insights into thenature and place of morphology.

. Introduction

This paper1 investigates syntactic sources of word-formations assuming that word-formations like compounds and derivations are part of a morphological compo-nent in the grammar, and that they historically started out as syntactic phrasesbeing built in syntax. By doing so, the question of how syntactic phrases develop

. I would like to thank Jonny Butler, Klaus von Heusinger, Jürgen Pafel, Ingo Plag, AchimStein and two anonymous reviewers for helpful comments. All errors and oversights are myown.

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.2 (151-196)

Carola Trips

into words, i.e., compounds (and finally derivational affixes), will be addressedas well as the question of whether word-formation operations can be subsumedunder syntax (e.g. Baker 1988; Lieber 1992) or have to be attributed to an inde-pendent module of word-formation subject to its own restrictions (e.g. Lapointe1980, DiSciullo & Williams 1987).

According to the traditional transformational view, the ‘sentential source’ hy-pothesis of word-formation (Lees 1960:31; Marchand 1969:31), morphologicalcomposites (compounds, suffixal derivatives, prefixal combinations) are ‘reduced’sentences in the form of nouns, adjectives and verbs and as such can be explainedfrom full sentences. Thus, compounds like the ones in (1) can be explained fromthe sentences ‘(we) wash with the machine’ and ‘(he is) blind with regard tocolours’:

(1) a. washing machineb. color-blind

However, this does not explain the different structural properties of compoundsand sentences, as stated here. Marchand notes

A compound behaves like a simple noun and can therefore only be part of a sen-tence, but it cannot function as a sentence, except as response utterance. It is thusdifferent from the sentence itself which is an independent complete utterance. Asthe language has created two distinct, though basically related syntagmatic enti-ties, it will not be enough simply to state that they are related. If we say that acompound is reducible to a sentence we must explain the difference in terms ofthe particular structures of sentence and compound. (Marchand 1969:32)

In this approach to word-formation, the dependence of morphology on syntax isvery strong, i.e. the underlying structure of every element of word-formation isa sentence. Although this approach as such is outdated for a number of reasons,there has been an ongoing debate as to where word-formation takes place, and ifthere is such a thing as ‘word syntax’, meaning that words are built by the sameprinciples that govern syntax. To answer these questions, we have to understandthe difference between phrasal structures and morphological structures, as Marc-hand notes in his quotation. So, what are the differences between the examplesshown below2?

. I would like to stress here that there is, not surprisingly, a plethora of literature on syn-thetic compounds in the framework of word syntax but hardly any on root compounds. As itis assumed that synthetic compounds have argument structure they seem to be a suitable phe-nomenon to demonstrate that word syntax exists. And this is probably also why root compoundsare ignored. Since they have a number of properties different from synthetic compounds andcrucial for the question of whether word syntax can be assumed or not this paper focuses onthis type of compounds.

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.3 (196-267)


(2) a. [XP a green house]b. a [X◦ greenhouse]

(3) a. [XP der rote Wein]b. der [X◦ Rotwein]

In the literature, a number of differences have been proposed, some of which willbe briefly discussed here (for a comparison of the differences found in the litera-ture see Table 1). First, it has been claimed (a.o. Bauer 2002; Plag 2003) that thestress patterns of phrases and words differ: in (2a) we find nuclear stress (stresson the nucleus of the phrase, i.e., on the head) and in (2b) we find compoundstress (stress on the left-hand member of the compound). This also applies to theGerman examples in (3). Moreover, in a language like German, which exhibitsmorphological agreement marking on the noun and its modifiers (case, number,gender), we find differences between compounds and phrases, shown in (3b) and(3a): whereas the adjectival first element in the compound does not agree with thenominal head, the determiner and adjectival modifier in the phrase in (3a) do. Ifwe tried to modify the nominal head of the compound in (3b) we would get anungrammatical string as in (4b)

(4) a. [XP der rote gute Wein]b. *[X◦ der Rot gute wein]

If we compare this example to the grammatical (4a), we see that modification of aphrasal head is, of course, possible.

It has also been shown that compounds cannot contain a determiner (and, ingeneral, no functional elements):

(5) a. der Menschenhasser (‘the men-hater’)b. *der Die-Menschen-Hasser (‘the the-man-hater’)

Further, we find a difference in meaning: Syntactic phrases have a referential in-terpretation although they can have a non-referential, generic, interpretation in arestricted context. Compounds (of the type N+N and A+N), on the other hand,can only have a non-referential interpretation (see also Williams 1989b). This isillustrated with the examples below:

(6) a. Des Hippies Leben war kurz (ambiguous).b. Das Hippieleben wird heutzutage oft imitiert (non-referential).

(7) a. The hippie’s life was short (ambiguous).b. Nowadays, the hippie life is often being imitated (non-referential).

In both German and English, the examples under a. are ambiguous, they have areferential interpretation, the determiner and modifier of the head noun refer to aparticular entity. Due to the context, they can also have a non-referential interpre-

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.4 (267-348)

Carola Trips

tation. In the examples under b. where we have the compound Hippieleben ‘hippielife’ only a non-referential interpretation is possible.

With respect to the semantics of morphological composites Bauer (1988:102)notes that there is a further difference: compounds provide names for entities,properties or actions whereas a syntactic phrase provides a description. Thus, thecompound judo man provides a name for the person concerned, a sentence like‘He’s an expert in judo’ however provides a description.

The differences discussed in the literature (and to some extent here) are givenin Table 1:

Table 1. Properties of syntactic phrases vs. properties of compounds

Syntactic phrases Compounds

nuclear stress compound stress→ ‘nuclear stress rule’ → ‘compound stress rule’agreement no agreement→ roter Wein → Rotwein(red-AGR wine) (red wine)

can be modified cannot be modified→ des Spargels köstliche Spitzen → *die Spargel köstlichen Spitzen(the asparagus’s delicious tops) (the asparagus delicious tops)

nouns can be preceded by det. first noun cannot be preceded by det.→ der Menschenhasser → *der Die-Menschen-Hasser(the men hater) (the the-men-hater)

no idiosyncracies and no lexicalisation idiosyncracies and lexicalisationnot sensitive to internal structure of words sensitive to internal structure of wordsno productive/unproductive patterns of sentences productive patterns

The fact that compounds show idiosyncracies and lexicalisation3 phenomena, andthat they are sensitive to the internal structure of words, will be discussed in Sec-tion 3 and 4. With respect to the last property listed in Table 1, it should be notedthat in morphology productivity is defined as ‘the property of an affix to be usedto coin new complex words’ (Plag 2003:44) or, according to Schultink (1961:113)as:

. . . the possibility for language users to coin unintentionally an in principle un-limited number of new formations, by using the morphological procedure thatlies behind the form-meaning correspondence of some known words.

From these standard definitions it becomes clear that the notion of productivity insyntax is a completely different matter.

. Lexicalisation is defined here as the process whereby ‘a lexeme takes on a form which itcould not have if it had arisen by the application of productive rules’ (Bauer 2002:48). Althoughit is a diachronic process, it leaves traces in a synchronic grammar. This means that lexicalisedelements are stored as such in the lexicon.

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.5 (348-412)


So far we have seen that from a synchronic perspective there are differencesbetween syntactic and morphological phrases, affecting all levels of language. Inthe following, I will show that from a diachronic perspective these differences ariseduring the development of syntactic phrases into morphological complexes (in anumber of cases) and that they have to be analysed as a loss of structural andsemantic properties. I will present two phenomena from the history of Englishand German that support my assumption: 1) the development of suffixes and 2)the rise of genitive compounds.

In the next section I will start out describing the development of suffixes inEnglish and German. In Section 3 I will discuss approaches that claim that word-formation is governed by syntactic principles, like Lieber (1992) and Ackema(1999). I will show that these approaches run into serious problems taking intoconsideration the data from Section 2. In Section 4 I will then present anotherphenomenon, the rise of genitive compounds, which will lend further supportto the assumption that the once-syntactic phrases in older stages of English andGerman gain non-syntactic properties during their development into word-units,and that the ‘syntax of words’ clearly differs from general grammatical principles(e.g., phrase syntax). Further, I will discuss the consequences of this assumption –to have an independent morphological module in the grammar – and claim thatlooking at cases of desyntacticisation, i.e., taking a diachronic perspective into con-sideration, will shed light on the interaction between syntax and word-formation.Section 5 concludes.

. The development of suffixes in the history of English and German

The majority of native Germanic derivational suffixes developed from syntacticphrases via a stage where they are the second constituent of compounds losingstructural and syntactic properties on their way (Tschentscher 1958; Wiesner 1968;Marchand 1969). One such example is the development of the Modern English(ModE) suffix -hood from hâd which was still a free morpheme in Old English(OE). If we have a look at the entry of the Oxford English Dictionary (OED) wefind the following information on the suffix:

-hood, suffix(hâd) [ME. -hod (-hode):-OE. -hâd = OS. -hêd, OHG. -heit.] Orig. a distinct n.,meaning ‘person, personality, sex, condition, quality, rank’ (see hâd n.), whichbeing freely combined with nouns, as in OE. cild-hâd ‘child-condition’, mæð-hâd‘virgin state’, pâpan hâd ‘papal dignity’, ceased at length to be used as a separateword, and survived as a mere suffix, and is thus noteworthy as a late example ofthe process by which suffixes arose. (OED; version 3.0, 2002)

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.6 (412-476)

Carola Trips

As noted in the entry, in OE hâd was still a free morpheme and occurred as a nounhead preceded by a determiner:4,5

(8) . . . andand

cwædonsaid

anmodliceunanimously

þætthat

MartinusM.

wærewere

wyrðeworth

[þæsGen,Sg.the

hadesGen,Sg.],

office

andand

gesæligblessed

sacerdpriest

toto

swilcumsuch

bisceopdome.episcopate

(ÆLS_[Martin]:264.6135)

Further, we find cases of had modified by a numeral as in (9), an adjective as in(10), or an NP in the genitive as in (11) and (12). All these cases show agree-ment between the head noun and its modifer(s). 180 instances of the pattern“modifier(s)+ had” were found in the corpus.:

(9) &and

[seNom,Sg.the

ðriddaNom,Sg.third

hadNom,Sg.]

entity

onon

ðærethe

halganholy

ðrynnyssethreeness(Trinity)

isseis the

halgaHoly

FroforgastGhost

. . .

(ÆHom_12:1.1792)

(10) Þathen

forþonforthwith

sona(as) soon

swaas

hiretheir

gemæccaanxiety

wæswas

forðfered,departed

heoshe

wearpexpelled

hireher

framfrom

[þamDat,Sg.the

woruldlicanDat,Sg.secular

hadeDat,Sg.]

office(GDPref_and_4_[C]:14.279.19.4103)

(11) andand

seoshe

gefyldefulfilled

þysnethis

earfoðanhard

cwydespeech

ðurhthrough

[Acc,Sg. ðathe

annysseoneness

[CristesGen,Sg.(of) Christ

hadesGen,Sg.]];

person(ÆCHom_II,_28:224.113.4984)

(12) þætthat

hehe

þærthere

onfengereceived

[Dat,Sg. [ærcebiscopesGen,Sg.archbishop’s

hadeDat,Sg.]

office(Bede_3:21.248.11.2540)

. All the data presented here from the diachronic stages of English are extracted from thefollowing annotated diachronic corpora: The York-Toronto-Helsinki Parsed Corpus of Old En-glish Prose (Taylor et al. 2003), Penn-Helsinki Parsed Corpus of Middle English 2 (Kroch & Taylor2000) and The Penn-Helsinki Parsed Corpus of Early Modern English (Kroch et al. 2004). The Ger-man data are from the Bibliotheca Augustana (Harsch 1996) and from the MittelhochdeutscheBegriffsdatenbank (MHDBDB, Springeth et al. 2002).

. The following diachronic data from English and German are given only with glosses becausefor the points made a literal translation is not needed here.

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.7 (476-536)


The same can be shown for German where we find the free morpheme heit in OldHigh German (OHG) texts:

(13) ninichtnot

dhizdiesthis

siiseibe

chiuuissogewisssure

[dherNom,Sgderthe

anderNom,Sganderother

heitNom,Sg]

personperson

godesgottesgod

selboselbstself

druhtinherrlord

christ.christchrist

(Isidor; Paris, BN, Ms. lat. 2326: 8,21)

(14) ohohoh

ininin

[dhemDat,Sgdiesenthese

dhrimDat,Sgdreithree

heidimDat,Sg]

personenpersons

scalsollshall

manmanone

ziuuaarezuwahrindeed

einaeinenone

gotnissagottgod

beodan.verkündigenannounce

(Isidor; Paris, BN, Ms. lat. 2326: 21,5)

In (15) hâd also occurs with an adjectival modifier which is a further source forword-formations with -hood in Modern English. Interestingly, most of these ad-jectives are relational adjectives, they denote the office or rank of a person. Thus,e.g. both cyninges hade and cynelica had have the same meaning: ‘rank of a king’.

(15) Acbut

seðehe who

ærbefore

hæfdehad

himhimself

ana

clænepure

wif,wife

sehe

wærewere

gecorenchosen

toto

[þamDat,Sg.the

clænanDat,Sg.pure

hadeDat,Sg.],

office

swajust

swaas

sethe

apostolapostle

PaulusP.

onon

hishis

pistoleepistle

awrat.wrote

(ÆLet_1_[Wulfsige_Xa]:25.29)

Here, the author does not talk about the properties of the office that is pure butabout an office for pure persons: somebody was chosen for the pure office. Bothpatterns, the one with a nominal modifier in the genitive and the one with an ad-jectival modifier, show the beginning of a desyntacticisation process from syntacticphrase to morphological phrase.

Apart from the syntactic phrases with the free morpheme hâd as nominalhead we also find compounds with hâd as the second element (i.e., as head ofthe compound).6

. Trips (2006) shows that due to a semantic shift of hâd having the specific meaning ‘officeof priest’ in many religious contexts and losing this specific meaning compounding becomes anoption for the speaker to name persons holding an office in general.

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.8 (536-595)

Carola Trips

(16) Soðlicetruly

wæswas

gewordenbecome

þathat

ZachariasZ.

[hisGen,Sg.his

sacerdesGen,Sg.priest’s

hadesGen,Sg.]

office

breacused

onon

hyshis

gewrixlesvicarious’

endebyrdnesseorder

beforanbefore

Gode.God.

ÆfterAfter

gewunanthe habit

[ðæsGen,Sg.the

sacerdhadesGen,Sg.vicarious-order’s

hlotesGen,Sg.]

decision’s

hehe

eodecame

þætso that

hehe

hyshis

offrungasacrifice

sette;set

(Lk_[WSCp]:1.8.3569)

The example in (16) shows the formal and structural differences between hâd ashead of a syntactic phrase (sacerdes hades with agreement marking) and as head ofa compound (sacerdhades only with agreement marking on the head noun). Thefollowing examples also clearly show the absence of agreement marking on thefirst element which is taken to be a diagnostic for compounding.

(17) Þathen

hehe

noldenot-wanted

forfor

[hisSg.his

biscopbishop

hadeGen,Sg.]

office

hithey

awegaway

adrifan, . . .drive(GDPref_and_3_[C]:7.188.19.2363)

(18) &and

[hyreSg.her

mægðhadNom,Sg.]

virgin state

isis

ansund,sound,

. . .

. . .

(ÆHom_1:420.219)

(19) Þærethere

tidetime

þærthere

onon

CentKent

heoldheld

HonoriusHonorius

[þoneAcc,Sg.the

arcebysceophadAcc,Sg.],

archbishop office

. . .

(Bede_5:17.452.23.4545)

In Middle English (ME), suddenly a high number (82 cases of a total of 101 cases)of formations with adjectival bases appear7 which indicates that the free mor-pheme hâd has developed into a morphological element, a suffix building abstractnouns:

(20) þerthere

opwexeþgrows

alleall

guodes.good,

uayrhede.beauty,

richesse.richnesse,

worþssipe.honour,

blisse.bliss,

uirtue.virtue

(CMAYENBI,75.1436)

. This finding is from the Middle English period M2, according to the division of the periodsof the The Penn-Helsinki Parsed Corpus of Middle English 2, which is the time span between 1250and 1350.

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.9 (595-656)


The same can be shown for Old High German: here we find compounds where thefirst element does no longer show agreement marking with the head noun:

(21) meinsuuarteomeineideperjuries

entiundand

lukino,lügenlies

kyridonobegierdendesire

entiundand

[unrehteroGenunrechteunjust

fizusheitoGen],schlauheitslyness

(Altbair. Beicht; Orléans, StB, Hs 184, S. 328,3*)

(22) ininin

buachonbuchbook

manmanman

gimeintimeintmeans

[thiodie(se)the(this)

iroihretheir

chuanheiti].kühnheitboldness

Iajayes

ististis

iueuchyou

ininin

thesadiesenthese

zitizeitentimes

zizuto

[giuuonaheiti],gewohnheithabit

(Otfrieds Evangelienbuch I, 4.I)

In Middle High German (MHG) we find a high frequency of formations with-heit8 which indicates that this element has developed into a suffix building ab-stract nouns:

(23) ichI

wilwill

euchyou

diethe

warheittruth

sagen,tell

(MHDBDB, Der Bergmann, par. 56, line 261)

(24) dazthat

isthas

vonof

bosheitwickedness

nihtnot

geschen.happened

(MHDBDB, Die Eule und der Habicht, par. 168, line 54)

The development of -hood in English and -heit in German from free morpheme tobound morpheme9 presents a structural change: initially, the free morpheme hassyntactic properties and acts as a noun head in syntactic phrases. As soon as thiselement develops into a non-syntactic element it loses its syntactic properties: itcannot function as a noun head in syntactic phrases any longer which is formallymarked by the loss of agreement marking in the noun phrase. This developmentmust be seen as a desyntacticisation process during which a syntactic element de-velops into a word-unit with properties that cannot be attributed to syntax but tomorphology.

. In the MHDBDB which comprises 139 MHG texts I found 186 types and 9801 tokens.

. It is not clear to me why OE hâd and OHG heit were lost as free morphemes. Sometimes,these elements change phonologically and remain in the language as free morphemes like e.g.English doom vs. the suffix -dom. I assume that a.o. frequency plays a role here but at this pointI have to leave this matter for future research.

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.10 (656-700)

Carola Trips

. Are morphological structures the same as syntactic structures?

In the previous section it was shown that syntactic structures can develop intomorphological structures, and it was claimed that the latter ones have propertiesdifferent from the former ones. From a theoretical perspective, the diachronic ev-idence presented here strengthens the claim that morphological structures are notgoverned by the same principles that govern syntax. In the literature (mainly Baker1988 and Lieber 1992) however, it has been put forward that word-formation phe-nomena adhere to syntactic contraints and therefore should be characterised assyntactic phenomena (taking place in syntax). Lieber’s (1992) approach is mostradical in this respect. She claims

In order for phrasal categories to be the input to processes of derivation and com-pounding, at least some construction of words must be done in the syntax. Theconceptually simplest possible theory would then be one in which all morphologyis done as a part of a theory of syntax. (Lieber 1992:21)

Under Lieber’s assumption the construction of words must be done in syntax withgeneral principles of grammar like the principles of X-bar theory. From what wasshown with the data so far, this seems to be problematic for a number of reasonsthat will be discussed in this section. Even Lieber has to admit that

. . . no one has yet succeeded in deriving the properties of words and the propertiesof sentences from the same basic principles of grammar. (Lieber 1992:21)

In the following the properties of the X-bar schema will be discussed and it willbecome clear that principles of X-bar theory cannot account for word forma-tion phenomena which explains why an undertaking such as Lieber’s cannot besuccessful.

. Headedness, the X-bar schema and referentiality

Lieber’s (1992) syntactic approach to word-formation is a modification of the X-bar schema to adapt to morphological needs. However, to achieve this she hasto assume that 1) specifiers can appear within the X-bar level, 2) recursion is al-lowed within the X◦ level and 3) non-heads need not be maximal projections. Allthree modifications do not seem to have any independent syntactic justification.Although it has been claimed by Stowell (1981) that Japanese and German sup-port the assumption in 1) this claim has seriously been challenged recently. Withrespect to the claim in 2) it does not seem to be justified to propose this type ofrecursion for syntax, at least Lieber does not provide any evidence in this respect.The assumption in 3) leads to a generation of structures that are not attested at allfor syntactic units (for a thorough discussion see Borer (1998) and Sproat (1993)).

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.11 (700-762)


Another serious problem for Lieber’s approach is that she derives the orderingof elements in complex words from syntactic rules which imply relations betweencategories. However, as Meyer (1993:48) has pointed out

In morphology the concept ‘head’ is defined with regard to position in complexwords and not in terms of relations between categories. This is a crucial distinctionbetween the use of the head-concept in syntax and morphology.

The crucial distinction noted by Meyer is implicitly stated by Williams’ (1981)famous Right-hand head Rule (RHR) and Di Sciullo and Williams’ (1987) rule:

(25) Right-hand Head Rule Williams (1981:248):In morphology we define the head of a morphologically complex word to bethe righthand member of that word.

(26) DiSciullo & Williams (1987:26):the head (F) of a word is the rightmost element of the word marked forfeature F.

To illustrate this point let us compare a syntactic and a morphological structure:

(27) Syntactic structure: definition of head in terms of relations between categories

FP

Spec F’

F◦ XP

Spec X’

X◦ YP

(28) Morphological structure: definition of head in terms of linear position

X◦

Y◦ X◦

If we have a look again at some of the diachronic data presented in Section 2, wesee that it is the linear ordering of elements that is the crucial property here:

(29) Þa forþon sona swa hire gemæcca wæs forðfered, heo wearp hire fram[þamDat,Sg. woruldlicanDat,Sg. hadeDat,Sg.]

(30) and seo gefylde þysne earfoðan cwyde ðurh [Acc,Sg. ða annysse [CristesGen,Sg.hadesGen,Sg.]];

(31) Þære tide þær on Cent heold Honorius [þoneAcc.Sg.

Acc.Sg.arcebysceophad], . . .

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.12 (762-805)

Carola Trips

These examples show that the headedness of compounds is strictly directional be-cause they develop from noun phrases with prenominal modifiers where the headnoun is the rightmost element (modifier (A/N) + noun = compound of the typeN+N or A+N). As compounds develop from the linear ordering of modifiers andnominal heads, movement as a way of deriving this order is ruled out.

Further, if we try to apply the right-headedness of morphology to syntax wesee that we have two different notions here: if we consider a language of the typeSOVX then the presence of an element X to the right of the verb would not resultin interpreting this element as the right head of this structure. Rather we would as-sume movement to derive this configuration, and we would still say that the headof the clause is V (Borer 1998). This example illustrates the difference betweenthe two notions of right-headedness in morphology and syntax: in morphologyit is strictly linear, in syntax it is linear based on hierarchical structures (see alsoBooij 2005). This difference is also a problem for Lieber’s approach: for her therelation between syntactic and morphological headedness has to be consistentin the sense that if a language is syntactically left-headed, it is morphologicallyleft-headed, and if a language is syntactically right-headed it is morphologicallyright-headed. However, if we just have a look at the two Germanic languages in-vestigated here, we see that this is not borne out: English is left-headed syntacticallybut right-headed morphologically. Thus, with respect to synthetic compounds wefind right-headedness although the VP is left-headed. In German, on the otherhand, we find left-headed APs and NPs, and right-headed VPs and IPs syntac-tically, but right-headed morphological complexes (where prefixal heads are theexception). The differences between the syntactic and morphological headednessonly between these two languages shows that Lieber’s approach does not hold (seealso Sproat 1993).

Borer (1998) has noted that the different notions of head in syntax and mor-phology is also problematic for Kayne’s (1994) assumption that the only possibleword order at D-structure is [Specifier[Head Complement]]. Thus, he would anal-yse a language of the type SOV with movement of the object to the left out of itscanonical position to the right of the verb, and he would claim that the effects ofthis movement operation could be traced. So it is not the linear ordering, i.e. thelocation of the object (and the verb) that determines the head in a structure, butthis is precisely what is stated in Williams’ RHR. Therefore, this fact and the factthat in Kayne’s structures heads are always generated in the left periphery of X’makes it impossible to unify his notion of head with the RHR.10

. This applies also to word-formation theories that assume a Kaynian structure like e.g.Drijkoningen (1999) or Josefsson (1998).

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.13 (805-858)


As argued above, the right-headedness ensures that there is a strictly linear re-lation between the left-hand and the right-hand member of a compound where theleft-hand member semantically determines the right-hand member somehow, butthe nature of that relation is unpredictable which is illustrated with the examplesbelow:

(32) a. Fischfrau (‘fish woman’)b. Museumsbuch (‘museum book’11)c. stone wall

These compounds (in isolation) are all ambiguous, many relations can be intro-duced for interpreting them. Thus, Museumsbuch could at least have the meanings‘book located in a museum’, ‘book about a museum’, ‘book published by a mu-seum’, ‘book with a cover showing a museum’. This ambiguity results from thevariability of constituent meaning and the number of relations possible within thecompound (see Meyer 1993). This typical characteristic of compounds is anotheraspect different from syntax.

Another serious problem for Lieber’s approach pertains to her Licensing con-ditions and the way she uses specifiers and complements. With respect to the firstpoint she proposes the following conditions:

(33) Lieber’s Licensing conditions (English)

a. Heads are initial with respect to complements.b. Heads are final with respect to specifiers.c. Heads are final with respect to modifiers.

(Lieber 1992:49ff.)

Further, Lieber (1992:38f.) defines complements as internal arguments obligato-rily selected by a verb, specifiers as quantifiers, degree words, subjects, modals, andmodifiers as restrictive modifiers (that limit the reference of the modified item). Ifwe try to apply these conditions to types of word-formation we see that root com-pounds of the type N+N (‘towel rack’) or A+N (‘greenhouse’) adhere to Lieber’slicensing condition (33) c., illustrated here with the OE example (34):

(34) Þære tide þær on Cent heold Honorius [þone arcebysceophad], . . .

Here, the first element restricts the extension of the complex word and can as suchbe seen as a modifier. But suffixes are problematic: according to Lieber, suffixes areheads of their words providing the category and other morphosyntactic featuresof those words. Moreover, and crucially, suffixes select (categorically and seman-tically) their bases. The stems of derivatives on the other hand are specifiers for

. The -s is a linking morpheme typical of German N+N compounds.

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.14 (858-910)

Carola Trips

her because they do not transparently modify the whole derivative. Now the claimthat suffixes are heads that select bases and the claim that stems are specifiers leadsto a contradiction because in syntax only complements can be selected by heads(and clearly not specifiers). Consequently, her licensing conditions neither workfor syntax nor for morphology.

Ackema (1999) proposes an X-bar schema that is applied both to the struc-ture of XPs and X◦s without the modifications Lieber proposed in her theory. Hisschema is shown below:

(35) a. Xn → Xn-1, Yn

b. Xn-1 → Xn-2, Zn

c. where n = 0 or 2(Ackema 1999:4)

For syntax, the value of n is 2, for morphology it is 0. This results in structures like

(36) X2

Spec X1

X◦

Spec X-1

X-2 compl

compl

As in syntax, there is a structural distinction between specifiers and complements.Moreover, X◦ has a special status, because it functions as maximal projection inmorphology, and at the same time, it is the head of a maximal projection XP insyntax.

Although Ackema’s ‘syntax below zero’ model seems to be formally attractiveat first sight, there are a number of problems: first, he assumes that hierarchicalstructure and the relations between elements (difference between specifier, headand complement) are relevant in morphological structure. Second, the structurein (36) implies that specifiers and complements in syntax have the same propertiesas in morphology. Third, he has to assume subword structures, and, what is moreproblematic, two distinct heads with two distinct maximal projections within asingle maximal projection. Moreover, in examples like happy and unhappy thehead happy always has to be A-2, implying that the head has the following structure(Ackema 1999:6):

(37) [A0[A-1[A-2 happy]]]

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.15 (910-979)


It should be noted that although Ackema assumes that morphological operationsare governed by syntactic principles, he does not claim that they are built in syntax.

. Consequences for an analysis of the diachronic data: Loss of structure andloss of referentiality

The critical discussion of Lieber’s and Ackema’s approach in the previous sectionhas shown that for theories assuming the same principles for syntax and mor-phology serious problems arise, especially if we look at fundamental notions suchas headedness. This becomes even more evident if the diachronic data presentedin Section 2 are considered. What I will suggest here is that when free morphemesfunctioning as phrasal heads develop into bound morphemes functioning first asheads in morphological complexes, and finally as suffixes, there is a transitionfrom syntactic structure to morphological structure implying loss of functionalcategories and maximal projections. Thus, the syntactic phrase

(38) [DP [DP ærcebiscopes] [NP hade]]

has a structure like

(39) DP

DP D’

D◦ NP

N’

N◦ YP

On its way to becoming a morphological element, the phrase is stripped of struc-ture, more precisely functional structure. What remains is lexical structure on theword level (i.e. it lacks maximal projections):

(40) [N◦[N◦ arcebysceop] [N◦ had]]

(41) N◦

N◦ N◦

This structure can then be inserted as such at D-structure, under N, resulting in aphrasal syntactic structure NP.

Formal evidence for this development is the fact that agreement cannot oc-cur any longer (for an analysis of the so-called linking morphemes see Section4). Moreover, if we were to analyse morphological complexes syntactically, e.g.via head-to-head movement, we would have to explain why material that appearsin syntactic structure does not appear in word structure. This has been notedby Ackema & Neeleman (hence A & N, 2001; 2004). They claim that verbal id-

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.16 (979-1034)

Carola Trips

ioms (a. o.) like the Dutch heartenbreeker ‘heart breaker’ cannot be explained byincorporation:

(42) a. datthat

hijhe

Marie’sMary’s

hartheart

vaakoften

heefthas

gebroken.broken

(Ackema & Neeleman 2001:45)b. *hij

heisis

eena

[NP [N [V hartenhart

breek]breaker

er] [VP vaakoften

tV [NP Marie’sMary’s

enand

Sue’sSue’s

tN]]]

(Ackema & Neeleman 2001:48)

The example in (42a) a. shows the syntactic idiom iemands hart breken ‘to breaksomeone’s heart’, and the example in (42a) b. the synthetic compound derived byincorporation from the syntactic idiom. It should be possible to strand material ifthis element was derived syntactically. But, as the example in b. clearly shows, thisis not the case, the example is ungrammatical. Therefore, in word-formation ma-terial that is obligatorily present in syntax must be omitted, which is only possibleif the element is realised in morphology.12

The observed loss of referentiality in compounds (see also Section 4) followsfrom this assumption. Williams (1989b:286) claims that “. . . the notion of refer-ence is tied to the syntactic notion of maximal projection, and hence should playno role in morphology.” If we further assume, in line with Williams (1989a) andAckema (1999), that the argument structure of nouns contain a R(eferential)-rolethat has to be discharged by D, then a definite determiner is needed to make a ref-erential reading possible. But since the functional categories of syntactic structureare lost during the development described above, there is no way to discharge theR-role. These two assumptions explain why compounds like (40), and compoundsin general, can only have a non-referential interpretation.

There is another observation showing the structural and semantic differencebetween syntactic structure and morphological structure: It has been noted by A& N (2001) that in N+V compounds in English (and other languages) the nouncan never function as the internal argument of the verb.13 The following exampleslillustrate this:

(43) a. to window shop �= to shop windowsb. to hand-make �= to make handsc. to air-condition �= to condition air

. This observation is also a problem for the Distributed Morphology model proposed byMarantz (1993).

. This seems also to be true of A+N compounds in English, e.g. greenhouse.

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.17 (1034-1057)


These compounds all have an unpredictable interpretation, an idiomatic meaning.So the N+V compound hand-make in (43b) cannot mean ‘to make hands’ etc.,but, of course, it would be possible to express this meaning with a sentence like ‘hemakes twenty hands every day’ where we have a fully compositonal meaning, i.e.where hands functions as the internal argument of the verb make (this sentenceis, of course, only plausible in a specific context, like e.g. ‘he’ is a potter makinghands of clay). According to A & N this can be explained if it is assumed thatlisting is costly (lexical storage should be kept to a minimum). Since there is aregular process in syntax to express the relation between a verb and its internalargument, and since generation of syntactic structure is the unmarked case (intheir model), there is no way to express this relation in morphology. According toA & N syntax blocks morphology in this case. This implies that syntactic structuresand morphological structures have different properties leading to the activation ofeither syntax or morphology.

In this section, it has been shown that morphological structure differs fromsyntactic structure structurally and semantically, and that this difference is trace-able if the development of syntactic structures into morphological structures istaken into consideration. Therefore, it is assumed here that ‘word-syntax’ (inthe sense that word-formation is governed by syntactic principles) does not existand, moreover, that morphological complexes are built in an autonomous word-formation component.14 To strengthen this claim, further diachronic evidence willbe presented in the next section.

. Further evidence: The rise of genitive compounds in the historyof German and English

In this section, the rise of genitive compounds in German and English will be dis-cussed. It will be shown that with respect to this phenomenon there are parallelsbetween the two languages involving a reanalysis of syntactic structure as mor-phological structure. This reanalysis serves as further evidence for the claim thatword-formation is not governed by syntactic principles and that it is part of anautonomous morphological component of grammar.

. This does not imply per se, however, that there is no interaction between syntax andmorphology.

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.18 (1057-1136)

Carola Trips

. German

In OHG we find two types of compounds: 1) the type N+N where the first con-stituent is a nominal stem with a compositional vowel as in tag-a-sterro (‘morningstar’). In some cases this vowel is replaced by -e as in hove-e-stat (‘farmstead’)or completely deleted as in got-Ø-spel (‘gospel’); this is an Indoeuropean word-formation pattern which is very productive in OHG (see e.g. Paul [1880]1995);2) the type N+N where the first element bears genitive case and thus shows asyntactic relation between the first and the second constituent as in senefes korn(‘mustard seed’). According to Grimm (1826:599) this type of compound is ratherseldom in OHG. In fact, it is not clear whether we have a phrase here or a com-pound. This structural ambiguity is due to the fact that in OHG we predominantlyfind genitive attributes in prenominal position:

(44) senefes korn (‘mustard seed’)

a. [NP [NP senefes] korn]b. [NP senefes korn]

(Demske 2001:298)

The string senefes korn can be interpreted as a complex noun with a prenominalgenitive as in (44) a. or as a nominal compound as in (44) b. Only in the first casecan we speak of a syntactic phrase since referentiality is a property of maximalphrases.

In the literature, it has generally been assumed that genitive compounds de-velop from lexicalised phrase structures and that the rise of these compoundscorrelates with the postposing of genitive attributes in Early New High German(ENHG). Pavlov (1983) assumes that the rise of genitive compounds is due to thestructural ambiguity of nominal phrases. There are two patterns that show thisambiguity: 1) structures with a determiner where the determiner either refers tothe head noun or to the prenominal genitive attribute of a syntactic phrase, and 2)structures without a determiner:

(45) wegen der Kirchen Ceremonien (‘because of the churches’ ceremonies’)

a. [ [ der Kirchen] Ceremonien]b. [ der Kirchen Ceremonien]

(Demske 2001:300)

(46) fewres flammen (‘fire’s flames’)

a. [ [ fewres] flammen]b. [ fewres flammen]

(Demske 2001:301)

In both cases the first constituent (der Kirchen and fewres) can have either a ref-erential or a non-referential (generic), interpretation. Like the example in (44),

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.19 (1136-1174)


only the first structures can be interpreted as syntactic phrases since referential-ity is a property of maximal phrases. Pavlov (1983) appeals to the principle ofisomorphism which assumes a one-to-one relation between form and meaning.Thus, he sees this development as a means to avoid structural ambiguity. Demske(2001), however, claims that the rise of genitive compounds is not simply the re-sult of reanalysing structurally ambiguous patterns but the result of fundamentalchanges in the nominal phrase in the history of German. She shows that thesechanges affected the relation between the head noun and its prenominal attributeswhich led to the postposing of genitival complements.15 She points out that underPavlov’s assumption the sudden increase of genitive compounds in ENHG can-not be explained since the structural ambiguity existed already in OHG.16 For her,the ENHG examples shown above are lexicalised syntactic structures rather thancompounds that still existed in ENHG. Evidence for this claim is the fact that thesestructures clearly show a referential interpretation:

(47) ausserapart-from

dessenthat

wollenwant

siethey

(. . . ). . .

auchalso

[deßthis

vorigenprevious

Landtagsparliament’s

schluss]resolution(Demske 2001:302)

(48) waswhat

dannthen

[landesland’s

werung]currency

ist/is

[wowhere

ainerone

mittwith

demthe

seckelbag

ist]is

(Demske 2001:303)

In both examples the first elements refer to a particular entity (a particular parlia-ment in (47) and a particular land in (48)). In (47) this interpretation is forced bythe determiner ‘deß’ and the adjective ‘vorigen’.

Demske further claims that not until ENHG are these lexicalised structuresreanalysed as nominal compounds, evidence of which is the mentioned increase ofthis pattern that could otherwise not be explained. Thus, due to the fundamentalchanges in the nominal system, animate and inanimate particular nouns that inENHG could be used as prenominal genitives as shown in (49):

(49) a. des hertzogen von Burgundien diener (‘the duke of Burgundy’s servant’)

. Due to a number of changes in the nominal system in ENHG the relation between thearticle and the head noun is newly defined. This change has consequences for the possessivepronoun which is reanalysed as possessive article as well as the attributive genitive which is alsoreanalysed as part of the article system.

. Demske notes that the (rare) cases of N+N compounds in OHG are isolated cases that arethe result of analogical processes. These processes, however, could not motivate the establish-ment of a new word-formation pattern because the frequency of these cases is too low.

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.20 (1174-1223)

Carola Trips

b. der vinde vffsatz (‘the enemy’s essay’)c. der sachen vrteiler (‘the thing’s judge’)

(Demske 2001:215)

could not be used in this position any longer which resulted in postposing them.Demske notes that these nouns do not have a possessive relation with its headnoun (‘genitivus qualitatis’, ‘genitivus explicativus’, ‘genitivus definitivus’) and arethus not interpretable as possessive. On the other hand, those nouns that have apossessive relation with its head noun, like the ‘genitivus possessivus’, ‘the genitivussubjectivus’ and the ‘genitivus objectivus’ can also be expressed with a possessivepronoun and therefore they can occur in prenominal position. Thus, we find thefollowing distribution of attibutive genitives in Modern German (ModG):

(50) a. Pauls Ernennung (‘Paul’s nomination’)b. ein Mann der Vernunft (‘a man of reason’)c. die Möglichkeit der Entspannung (‘the possibility of relaxing’)d. das Laster der Trunksucht (‘the vice of alcoholism’)

(Demske 2001:247)

Since the prenominal genitive was not available any longer, the existing lexicalisedstructures were reanalysed as a morphological pattern, and in this way the riseof a ‘new’ word-formation pattern was possible. It should also be noted that theprenominal genitive with proper nouns that is still possible in ModG (see example(50a) is analysed by Demske as a lexically marked affix of possession and not as aninstantiation of genitive case. For her, it is a lexical rule for proper nouns whichhave a headlike status (i.e., they are not phrases).

According to Demske the reanalysis of syntactic lexicalised structures as mor-phological structures can be described as the loss of a word boundary but it alsoincludes a structural reinterpretation, i.e., the language learner reinterprets thegiven surface structure by attributing to it another base structure. Under the as-sumption that language change happens during language acquisition (e.g. vanKemenade 1987; Kroch 1989) the learner deduces a syntactic or morphologicalstructure from the input she gets that has to be compatible with her grammar. Inthe case at hand, the learner has evidence that the existing prenominal genitiveshave properties attributed to morphological complexes (e.g. non-referentiality,compound stress etc.) and therefore the learner reanalyses the lexical syntacticstructures as morphological structures. However, this can only happen if old andnew structures have a number of properties in common. Since syntactic and mor-

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.21 (1223-1280)


phological structures share a number of properties (binary structures, recursivestructures, endocentric structures) this seems to be a plausible assumption.17

Evidence for Demske’s claim that morphological structures exist in ENHG isthe reanalysis of the suffix marking genitive case: here, the reanalysis of the nounphrase as morphological phrase leads to the reanalysis of the inflectional endingas a linking morpheme (‘Fugenmorphem’), evidence of which is the fact that thisgenitive suffix also occurs in genitive compounds where the first element originallydoes not show it because it is not part of its paradigm (strong feminine nouns andweak nouns):

(51) a. von den Regierungs-Räthen (der Regierung Räthe) (‘government civil ser-vant’)

b. des Bawers-man (des Bauern-mann) (‘farmer man’)(Demske 2001:308)

In cases like (52) below we find the genitive singular form as first element althoughthe interpretation in terms of a transparent syntactic phrase would require theplural form here:

(52) weibß kleider (‘woman’s clothes’), mynchs orden (‘monk’s order’)(Demske 2001:310)

These examples can only be explained if it is assumed that their structures havebeen reanalysed as word structures implying that 1) inflectional affixes are nolonger governed by inflectional rules and 2) the first part of the phrase is reanalysedas part of the compound, and the compound as a whole gets case.

Moreover, the referential (possessive) interpretation becomes obsolete andonly a non-referential, generic, interpretation is possible:

(53) 1546: diewho

inin

Schafskleidernsheep clothes

zuto

euchyou

komencome

(Demske 2001:309)

Further support for this assumption is the rise of copulative compounds in ENHGas shown in (54). In this type of compound we find a coordination relation be-tween the two elements that cannot be lexicalised as a syntactic phrase.

(54) alle Manspersonen ‘all man’s persons’, Leibs Erben ‘body’s heirs’(Demske 2001:311)

. I would further claim that the learner has evidence that these structures are more marked(e.g. they are more restricted at least wrt their semantics and frequency of occurrence) asopposed to the syntactic structures with postnominal genitive.

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.22 (1280-1329)

Carola Trips

Demske points out that especially the rise of the linking morphemes presents evi-dence for a lexical analysis of genitive compounds that could not be explained witha syntactic analysis.

For the development of compounding as a morphological process Demskeproposes the following three stages: In OHG noun phrases with prenominal geni-tive attribute occur. They clearly have the status of syntactic phrases. In the courseof time some of these phrases are lexicalised; nevertheless they are syntacticallytransparent as they show a referential interpretation and allow modifiers of thefirst constituent. Finally, these lexicalised syntactic structures are reanalysed asmorphological structures that then display syntactic opacity.

The rise of genitive compounds (in German) is a further instance of desyntac-ticisation: once-syntactic structures are reanalysed as morphological structures. InGerman, due to the fact that the relation within the noun phrase changes, a re-analysis of the prenominal genitive as possessive phrase becomes possible whichthen leads to the postposing of genitive complements in ENHG implying the riseof new syntactic structures. In the next section we will see that in English we finda development resembling the one described for German.

. English

With respect to the development of genitive compounds we find a similar picturein English. In OE two types of compounds occur: 1) the Indoeuropean word-formation pattern of N+N compounds where the first constituent is a nominalstem with compositional vowel (Sauer 1992:81). However, in almost all cases thisvowel has been reduced to a linking morpheme -e as in hilderinc ‘battle man’; 2)the genitive compound where the first element bears genitive case as in cinnesmen‘kinsmen’. As in German, we have a situation where morphological structures co-exist with syntactic structures, and sometimes it is not clear whether we have aphrase or a compound. According to Sauer (1992:152) a clear case is

(55) a. cinnesmen ‘kinsmen’b. . . . wið [heora agenes cynnes mannum] ’with their own kin’s men’

As in German, this structural ambiguity is due to the fact that in OE we predomi-nantly find nominal genitives in prenominal position.

McLagan (2003) notes that factors for preplacement are animacy, high pro-totypicality of possessive relationship and one-word-hood. The second factor canalso be described by the contrast between a referential and a non-referential inter-pretation as mentioned above. Thus, we find ambiguous cases like (56)

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.23 (1329-1395)


(56) Eodewent

þathen

toto

anumone

drymensorcerer

[þewho

deoflesdevil’s

cræft]craft

cuðe,knew

(ÆLS_[Basil]:364.697)

where both a referential and a non-referential interpretation are possible. Bothreadings are given in (57).

(57) a. referential: ‘the craft of devil (itself)’b. generic: ‘craft typical of the devil’

In her study, McLagan (2003) observes that during the eleventh century preposedgenitive phrases become even more frequent (rising from 69.1% around 1000 to77.4% around 1100). She also cites a study from Yerkes (1982) which shows thatthe reviser of Gregory’s Dialogues (who worked some time between 950 and 1050)transposed many genitive phrases in the text from after the nominal head to beforethe head noun in such a way that the placement clearly differed from the originalLatin version. Allen (1998:20) observes that

. . . even in the OE period, post-head genitives were on the decline, and . . . thisinitial decline cannot be explained as a consequence of reduced inflection.

According to what was said above for German, there seem to be parallels betweenGerman and English. Demske (2001) notes, that the fundamental changes ob-served in the nominal system in the history of German can also be found in thenominal system in the history of English.18 These changes consequently led to therise of genitive compounds, which can be attributed to the reanalysis of syntac-tic structure (genitive phrase) as morphological structure (genitive compound).Under this assumption, the sudden rise of the pattern observed by McLagan canbe explained along the lines of the rise of the pattern in ENHG as will be shownbelow.

Although there does not seem to be a direct correlation between the changesin the nominal system and the loss of morphological distinctions between the OEand the ME period, the latter could have been one factor that promoted some ofthese changes. In her study, Allen (1998, see also Allen 1995) shows that in thetwelfth century morphological case marking was still well-preserved in the south-ern dialects but started to disappear in the northern dialects. In the thirteenthcentury, agreement inflection had become optional (depending on dialect). Allenfurther notes that at that time the correlation between agreement and postnom-inal genitive breaks down resulting in the fact that the postnominal genitives areconsistently restricted to partitive relations:

. As in German, in English the relation between the article and the head noun is newly de-fined. According to Demske, this change has the same consequences as in German: the possessivepronoun and the attributive genitive are both reanalysed as part of the article system.

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.24 (1395-1459)

Carola Trips

(58) Andand

hehe

wiðwith

þtthat

hefraised

upup

hetelestcruellest

alreallGen

wepne.weaponsgen

‘And with that, he raised up the cruellest of all weapons’(Allen 1998:19)

Allen assumes that these genitives are all selected lexically by their head implyingthat postnominal genitives were no longer structurally case marked. Moreover,she notes that genitives of the ‘objective’, ‘subjective’ or ‘possessive’ type were nolonger found in this position. As shown in 4.1, the same semantic restriction canbe found for the postnominal genitives in German which is evidence for changesin the nominal system (the relevant change here is that genitival attributes are re-analysed as part of the article system). The restriction of postnominal genitivesin EME also depends on semantic factors, which makes it plausible to assumethat this finding reflects the same change in the nominal system as proposed forGerman by Demske, independent of the loss of agreement marking. Since the pos-sessive relation could be expressed with a prenominal genitival noun and headnoun as well as with a possessive pronoun the latter could occur in prenominalposition. As this relation could no longer be expressed with postnominal geni-tives, it started to be expressed more frequently in prenominal position, and thisexplains the rise of prenominal genitives observed by McLagan. Strikingly, in astudy on prenominal genitives (nouns) in EME, I found that 451 out of 457 casesshow 1) a non-referential interpretation, 2) no article:

(59) Þattthat

lareteaching

þattthat

wasswas

allall

bisettfilled

Wiþþwith

[nahhtess þessterrnesse].night’s darkness

(CMORM,II,236.2482)

(60) II

chargecharge

þeyou

atat

mymy

blissyngblessing

kepekeep

þiyour

bodybody

kleneclean

atat

þethe

lestleast

frofrom

[womanys feleschep]woman’s fellowship

tyltil

þuyou

taketake

aa

wyfewife

aftyrafter

þethe

lawelaw

ofof

þethe

Chirche.church

(CMKEMPE,222.3577)

These findings show that the genitive -es ending was no longer analysed as aninflectional marker (non-referential interpretation of the prenominal genitives).Thus, it has to be assumed that these structures were reanalysed as a morpho-logical pattern and the rise of the word-formation pattern NGen+N was possible.Moreover, the gradual loss of inflectional endings led to syncretism of vocalic gen-itive endings in -e (-e for Sg., -a for Pl.)19 and consequently to an unclear statusof inflectional endings in general. Furthermore, the -s genitive form increased (as

. In ME, the -e ending was also a reflex of the syncretism of the OE nominative endings -a,-e, -o, -u.

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.25 (1459-1503)


well as the syntactic of-genitive20). Accordingly, we find cases where different formsof ‘prenominal genitives’ (nouns) coexist (in one text):

(61) a. heuene lyhte ‘heaven light’b. heueryche ‘heaven realm’c. heuen kynge ‘heaven king’

(Sauer 1992:89)

These forms all have the word heaven as first element. In a. we find an -e-ending onheaven, in c. we find the same context but no -e-ending. The form in b., heueryche,is a contracted form of heaven and ryche, it shows phonological changes and islexicalised. These examples illustrate that the genitive endings here are no longergoverned by transparent inflectional rules, the examples are all isolated elements,i.e., compounds.

In some texts, e.g., in the Ormulum, we find the coexistence of an N+N com-pound, an N-es+N compound (note the non-referential interpretation) and thesame noun hirde ‘shepherd’ with a postnominal of -phrase:

(62) Þa wass he [shepess hirde], . . .(CMORM,I,123.1067)

(63) Crist iss all se Daviþþ wass [Shephirde], king, & kemmpe.(CMORM,I,123.1063)

(64) & Crist iss [hirde off hise shep], . . .(CMORM,I,123.1068)

As noted in Section 4.1, Demske claims that the occurrence of linking morphemesare strong evidence for the reanalysis of the suffix marking genitive case. Thisseems also to be borne out for English: in ME a number of linking morphemesstart to arise:

(65) a. OE nihtegale, ME nihtingaleb. OE lof-song, ME lof tsongc. OE handgeweorc, ME handiwerc

(Sauer 1992:82)

Moreover, as in German, the genitive -es suffix now also occurs on the first elementin compounds although in OE this form is originally not part of that element’sparadigm:

. It has been claimed that the rise of the of -phrase is due to French influence (e.g. Lightfoot(1999), Allen (2002)). To my knowledge, however, there is no empirically well-founded studythat proves this assumption (see also Faiß (1992)).

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.26 (1503-1556)

Carola Trips

(66) a. OE burg-tun, ME burgess tunb. OE land-man, ME londes-men

(Sauer 1992:157)

Therefore, it can be assumed that at this stage in the history of English inflectionalaffixes are no longer governed by inflectional rules. The genitival modifier of thehead noun is reanalysed as part of the compound since the whole complex wordgets case.

Another parallel between German and English is that copulative compoundsalso arise as another productive pattern (not possible as syntactic structure) toform compounds:

(67) a. ME burgess tun (castle-town)b. ME leodes-folk (men-folk)c. ME shepess lamb (sheep-lamb)

(Sauer 1992:420)

From what was said above we can assume then that, as in German, the syntacticlexicalised structures (prenominal genitive) have been reanalysed as word struc-tures. This reanalysis leads to the rise of postnominal syntactic structures, thegenitive of -phrase. The development of new syntactic structures can be seen asa consequence of this reanalysis.

In Section 3 A & N’s model was briefly discussed. If it was applied to the rise ofgenitive compounds in German and English we would have to assume that in thecourse of time the syntactic pattern loses its properties (transparency, referential-ity, functional structure) and is replaced by morphological structure. In this case,morphology blocks syntax. This blocking then leads to new syntactic structures.

In English, another possibility arose: in ME the possessive marker was reanal-ysed as a clitic21 being part of syntax, rather than an inflectional marker, being partof morphology. Allen (2002) claims that a clitic analysis has to be assumed at leastfor some possessives by the late fourteenth century, when the first examples of thegroup genitive started to appear:

(68) þethe

kingking

ofof

FrauncesFrance’s

menmen

(Allen 2002:63)

. Allen has shown that the Saxon-Genitive (’s) in ModE is a clitic attached to the precedingelement (DP). The so-called ‘group genitive’ where the clitic is attached to a DP group is a lateME innovation (first to coordinate genitive phrases and appositive genitive constructions, andlater on to possessive NPs that did not end with a possessor).

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.27 (1556-1628)


This phenomenon is still found in ModE:

(69) a. an old man’s bookb. the man with the whistle’s book

(Demske 2001:245)

Although there are striking similarities between German and English with respectto the rise of genitive compounds, the following differences should be pointedout: in German, the changes described above happen in ENHG (1350–1650), inEnglish these changes happen much earlier, in Early Middle English (1150–1350).It was argued above that the loss of the postnominal genitive as well as the rise ofgenitive compounds seem to be the result of fundamental changes in the nominalsystem and not the result of the loss of morphological case marking. However, thefact that these changes happened much earlier in English could well be attributedto the latter change because it affected all elements (determiners, adjectives, pos-sessives) that were involved in the change of the nominal system and thereforepossibly catalysed this change. This would also explain why we find many N+Ncompounds with a possessive interpretation (like e.g. beehive etc.) but without alinking morpheme and just few of that type with this element:

(70) a. ME domes-dei, pentecostes-dei, cinnes-men, Cristes-tyde, steoress-mannb. EModE kinswoman, kinsmen, huntsman, bedeswomanc. ModE doomsday, kinsmen, Wednesday, Thursday, daisy

Moreover, the structural differences between the prenominal genitive in ModGand ModE could also be attributed to the loss of agreement in ME: in the for-mer language it is a possessive affix governed by lexical rules, whereas in thelatter language it is a clitic (non-head status) that attaches to phrases. Althoughthe correlation between the changes in the nominal system, the rise of genitivecompounds and the loss of morphological case marking need further in-depth in-vestigation, the observations made above indicate that there are parallels betweenGerman and English, and that there are desyntacticisation processes that shed newlight on the difference between syntactic and morphological structure.

Table 2 summarises the similarities and differences of this development inboth languages:

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.28 (1628-1704)

Carola Trips

Table 2. The rise of genitive compounds: comparison German and English

German English

prenominal genitive possible prenominal genitive possible→ ‘Vaters Buch’, ‘Erics Buch’ → ‘father’s book’, ‘Eric’s book’(nouns [+animate], personal reference) (nouns [+animate], personal reference)but never with generic reference but never with generic referencepostnominal genitive possible postnominal genitive not possible→ ‘die Ordnung des Landes’, → ‘*the law the lands’‘die Bücher der Kinder’ but ‘Saxon-Genitive’ (clitic ’s)→ both generic and referential → group genitive ‘Peter and John’s house’

independent genitive (‘at Fred’s’)postnominal genitive with von postnominal genitive with of→ ‘Das Buch von Vater’, → ‘the front of the house’,‘die Bücher von den Kindern’ ‘the books of the children’

. Conclusion

In this paper it was argued that there are syntactic sources of word-formation pro-cesses showing a transition from syntactic structures to morphological structuresand thus can be seen as instances of desyntacticisation. The phenomena describedin Section 2 and 4 have strengthened the assumption that morphological struc-tures are not the same as syntactic structures, especially by looking at the notion ofheadedness and structural hierarchy. These findings demonstrate that an analysislike Lieber (1992) or Ackema (1999) claiming that all word-formation phenom-ena are governed by syntactic constraints and hence are syntactic phenomena isnot borne out. Further, it was shown that morphological structures have semanticproperties that are different from syntactic structures (only generic interpretationpossible vs. ambiguity of generic and referential interpretation). The fact that newsyntactic structures occur (of -genitive) once the old syntactic structures (prenom-inal genitive) developed into morphological structures (genitive compounds) areevidence for the assumption that morphology can block syntax (in the sense ofAckema & Neeleman 2001). The phenomena and their properties described aboveshow that diachronic studies provide new insights into the interaction betweensyntax and morphology.

References

Ackema, P. (1999). Issues in Morphosyntax [Linguistik Aktuell/Linguistics Today 26]. Amster-dam: John Benjamins.

Ackema, P. & Neeleman, A. (2001). Competition between syntax and morphology. In G.Legendre et al. (Eds.), Optimality-Theoretic Syntax (pp. 29–60). Oxford: OUP.

Ackema, P. & Neeleman, A. (2004). Beyond Morphology. Oxford: OUP.Allen, C. L. (1995). Case Marking and Reanalysis: Grammatical relations from Old to Early Mod-

ern English. Oxford: Clarendon Press.

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.29 (1704-1881)


Allen, C. L. (1998). Loss of the post-head genitive in English. In ALS98 papers in proceedings,1–26. http://www.als.asn.au/: last viewed 12/03/06.

Allen, C. L. (2002). Case and Middle English genitive noun phrases. In D. Lightfoot (Ed.), Syn-tactic Effects of Morphological Change (pp. 57–87). Oxford: OUP.

Baker, M. (1988). Incorporation: A theory of grammatical function changing. Chicago IL: Univer-sity of Chicago Press.

Bauer, L. (1988). Introducing Linguistic Morphology. Edinburgh: EUP.Bauer, L. (2002). English Word-Formation (9th edition). Cambridge: CUP.Booij, G. (2005). The Grammar of Words. Oxford: OUP.Borer, H. (1998). Morphology and syntax. In A. Spencer & A. M. Zwicky (Eds.), The Handbook

of Morphology (pp. 151–191). Oxford: Blackwell.Demske, U. (2001). Merkmale und Relationen. Berlin: Walter de Gruyter.DiSciullo, A.-M. & Williams, E. (1987). On the Definition of Word. (2nd edition). Cambridge

MA: The MIT Press.Drijkoningen, F. (1999). Antisymmetry and the lefthand in morphology. Catalan Working Pa-

pers in Linguistics (7), 71–87.Faiß, K. (1992). English Historical Morphology and Word-formation: Loss versus enrichment. Trier:

Wissenschaftlicher Verlag Trier.Grimm, J. (1826). Deutsche Grammatik. Hildesheim: Olms (Nachdruck).Harsch, U. (Ed.). (1996). Bibliotheca Augustana. Augsburg: Fachhochschule Augsburg.Josefsson, G. (1998). Minimal Words in a Minimal Syntax [Linguistik Aktuell/ Linguistics Today

19]. Amsterdam: John Benjamins.Kayne, R. (1994). The Antisymmetry of Syntax [Linguistic Inquiry Monograph 25]. Cambridge

MA: The MIT Press.van Kemenade, A. (1987). Syntactic Case and Morphological Case in the History of English. Dor-

drecht: Foris.Kroch, A. (1989). Reflexes of grammar in patterns of language change. Language Variation and

Change, 1, 199–244.Kroch, A., Santorini, B., & Delfs, L. (Eds.). (2004). The Penn-Helsinki Parsed Corpus of Early

Modern English (PPCEME). Philadelphia PA: University of Pennsylvania.Kroch, A. & Taylor, A. (Eds.). (2000). The Penn-Helsinki Parsed Corpus of Middle English, Second

Edition (PPCME2). Philadelphia PA: University of Pennsylvania.Lapointe, S. G. (1980). A Theory of Grammatical Agreement. PhD dissertation, University of

Massachusetts, Amherst.Lees, R. B. (1960). The Grammar of English Nominalizations. The Hague: Mouton.Lieber, R. (1992). Deconstructing Morphology. Chicago IL: The University of Chicago Press.Lightfoot, D. (1999). The Development of Language: Acquisition, change and evolution. Oxford:

Blackwell.Marantz, M. H. A. (1993). Distributed morphology and the pieces of inflection. In K. Hale & S.

J. Keyser (Eds.), The View from Building 20 (pp. 3–76). Cambridge MA: The MIT Press.Marchand, H. (1969). The Categories and Types of Present-Day English Word- Formation (2nd

edition). München.McLagan, H. (2003). The Syntax of Genitive Constructions in Old English: Placement of genitive

phrases in AElfric’s second series of Catholic Homilies. MA thesis, School of Language Studies,Australian National University.

Meyer, R. (1993). Compound Comprehension in Isolation and Context. Tübingen: Max Niemeyer.Paul, H. ([1880]1995). Prinzipien der Sprachgeschichte. Halle: Niemeyer.

JB[v.20020404] Prn:24/10/2006; 14:58 F: LA9710.tex / p.30 (1881-1988)

Carola Trips

Pavlov, V. (1983). Zur Ausbildung der Norm der deutschen Literatursprache im Bereich derWortbildung (1470–1730). Von der Wortgruppe zur substantivischen Zusammensetzung. Berlin:Akademie-Verlag.

Plag, I. (2003). Word-Formation in English. Cambridge: CUP.Sauer, H. (1992). Nominalkomposita im Frühmittelenglischen. Tübingen: Max Niemeyer.Schultink, H. (1961). Produktiviteit als morfologisch fenomeen. Forum der Letteren (2), 110–

125.Springeth, M., Schmidt, K., & Pütz, H. (Eds.). (2002). Mittelhochdeutsche Begriffsdatenbank

(MHDBDB). Salzburg: Universität Salzburg/ Paris: Lodron.Sproat, R. (1993). Morphological non-separation revisited: A review of Lieber’s Deconstructing

Morphology. In G. Booij & J. van Marle (Eds.), Yearbook of Morphology 1992 (pp. 235–258).Dordrecht: Kluwer.

Stowell, T. (1981). Origins of Phrase Structure. PhD dissertation, MIT.Taylor, A., Warner, A., Pintzuk, S., & Beths, F. (Eds.). (2003). The York-Toronto-Helsinki Parsed

Corpus of Old English Prose (YCOE). Toronto: University of York.Trips, C. (2006). Lexical Semantics and diachronic morphology: The development of the deriva-

tional suffixes -hood, -dom and -ship in the history of English. Ms, University of Stuttgart.Tschentscher, C. (1958). Geschichte der germanischen Bildungssilbe TUM. PhD dissertation,

Universität Stuttgart.Wiesner, J. (1968). Das Wort heit im Umkreis althochdeutscher personaÜbersetzungen. Ein

Beitrag zur Lehngut-Theorie. Beiträge zur Geschichte der deutschen Sprache und Literatur, 90,3–67.

Williams, E. (1981). On the notions ‘lexically related’ and ‘head of a word’. Linguistic Inquiry,12, 245–274.

Williams, E. (1989a). The anaphoric nature of 0-roles. Linguistic Inquiry, 20(3), 425–456.Williams, E. (1989b). Maximal projections in words and phrases. In M. R. Baltin & A. S. Kroch

(Eds.), Alternative Conceptions of Phrase Structure (pp. 280–293). Chicago IL:The Universityof Chicago Press.

Yerkes, D. (1982). Syntax and Style in Old English: A comparison of the two versions of Waer-ferth’s translation of Gregory’s Dialogues. Binghampton NY: Center for Medieval and EarlyRenaissance Studies.

JB[v.20020404] Prn:26/10/2006; 9:27 F: LA97IND.tex / p.1 (47-144)

Index

Aaccusative case –, –,

, , , , ,

Afrikaans , , , , ,

–, , –, –,

, , ,

Alemannic , , , , ,

,

Anglo-Saxon

argument(s) –, , , ,

, –, , , –, ,

–, , , , ,

, , , , , ,

, , , , , ,

, ,

auxiliary , , –, ,

,

see also auxiliary selectionauxiliary selection , ,

, , , , –,

, –

BBavarian

be , , –

Burzio’s Generalization ,

, –

see also Sibling Correlation

Ccanonical subject position

case see accusative casesee dative casesee nominative casesee predicative casesee structural case

compounds , , –,

, , , , –,

–, –

concord –, , –,

, , ,

counterfactual , , ,

–, –

DDanish , , , , , ,

–, , , , ,

dative case , –, , ,

,

desyntacticisation , ,

, , ,

development of suffixes ,

dialect(s) , , , , , ,

, , , –, ,

,

DP , –, , , ,

, , , –,

–, –, ,

Dutch , –, , , , ,

, , , , , , ,

, , , , , ,

, , , , ,

–, –, , ,

, , , –, ,

–, –, , ,

EEarly Modern English

(EModE, ENE) , ,

, , –, –,

, –, , , ,

, –, ,

Early New High German

ellipsis , , –,

English –, , , ,

–, –, , –, ,

, , –, , , ,

, , –, , ,

, , –, , ,

, , , , ,

–, , , –,

, , , , , ,

, , , , , ,

, , , , , ,

–, –, , ,

, , –

EPP feature , –

FFaroese , , , , , ,

, , ,

feature checking –,

feature movement ,

feature(s) –, , , ,

, , , , –,

, –, , ,

, –, ,

–, ,

see also EPP-featuresee also feature checkingsee also feature movementsee also number

neutralisationFrisian , , –, , , ,

, –

GGerman –, –, , ,

, , , , , , , ,

, , , , ,

–, –, , ,

, , , –, ,

, , , –,

–, , , –,

, , –, , ,

, , , , –,

, , , , –,

–

Gothic

Hhead-movement , , ,

, , , ,


Index

headedness , , ,

history of English , ,

, , ,

Holmberg’s generalization ,

–, , , , , ,

–

IIcelandic –, –, , ,

–, –, , , –,

, , , , –, ,

, , , –,

inherent accusatives ,

intensional reading ,

LLow-Saxon

MMainland Scandinavian ,

, ,

mereology , ,

Middle Dutch , –,

, , –, ,

Middle English (ME) ,

–, , , , ,

, , , , , ,

–, –, , ,

, , , –

Middle High German (MHG),

NNom/Acc alternation

see nominative/accusativealternation

nominal phrase(s) –,

–, –, ,

number neutralisation–

nominative/accusativealternation –, , ,

, , , , ,

nominative case , , –,

–

Norwegian , , , , ,

, , , , , , –,

, , , , , ,

, , ,

null elements

number agreement ,

Oobject shift –, , ,

–, ,

Old English (OE) , , ,

, –, , , ,

, –, , , ,

, , , , ,

–

Old High German (OHG,OldHGerman) , ,

, , , , ,

Old-Frisian

Old-Kentish ,

Old-Norse

Old Swedish

optionality , , , ,

, , ,

see also true optionality

Ppartial functions –

passive –, , –, ,

, –, , , , ,

, ,

pied-piping , , –,

, –, –, ,

, –,

predicate , , , , ,

–, , , , , ,

, , , , , ,

–, , ,

complex ∼ –, , ,

, , ,

nominal ∼ –,

–, , –,

, –

∼ accusative ,

∼ nominative , , ,

, ,

predication –, , ,

, , ,

predication theory , , ,

–,

predicative case –, ,

see also predicate accusativesee also predicate

nominativepronominal determiners

–, , , , ,

, , , , –,

, ,

pronoun(s) , , , , ,

, , , , –,

, –, , ,

, , , ,

resumptive ∼ –,

, ,

pseudo-coordination ,

R

reconstruction , , ,

–

referential reading ,

referentiality , , ,

–, , , , ,

relational case –, , ,

–, –,

relativization –,

–, , , , ,

, , , –

remnant-movement , ,

–

resultative , , –,

, ,

resumption , ,

resumptive pronoun –,

, , , ,

rise of genitive compounds, , –, , ,

–

S

Scandinavian , , , , ,

, , –,

see also MainlandScandinavian

shape conservation –,

–, , , , , , ,

–

Sibling Correlation , , ,

–, ,

structural case –, , ,

, , ,

see also relational case

Stylistic Fronting (Styl-F), , –, , ,

,

Swedish –, , , , ,

, , –, , , –,

–, , , , , ,

,


Index

Ttrue optionality ,

UUpper Franconian

VV2 , , –, –, ,

–, , , , ,

, –, , , ,

,

verb movement –, , ,

, , , , , , , ,

, ,

Verb Projection Raising (VPR), , , –,

Verb Raising (VR) , ,

, , , –

verb-second see V2

WWest-Flemish

word-order change , ,

,

Y

Yiddish , , , , , ,

,

Z

Zurich German (ZG)–, , , –,

, , –, –

Linguistik Aktuell/Linguistics TodayA complete list of titles in this series can be found on the publishers’ website, www.benjamins.com

101 Karimi, Simin, Vida Samiian and Wendy WilKinS (eds.):PhrasalandClausalArchitecture.Syntacticderivationandinterpretation.InhonorofJosephE.Emonds.vi,411pp.+index.Expected December 2006

100 SchWabe, Kerstin and Susanne WinKler (eds.):OnInformationStructure,MeaningandForm.Generalizationsacrosslanguages.x,557pp.+index.Expected December 2006

99 martínez-Gil, Fernando and Sonia colina (eds.):Optimality-TheoreticStudiesinSpanishPhonology.vi,548pp.+index.Expected December 2006

98 PireS, acrisio:TheMinimalistSyntaxofDefectiveDomains.Gerundsandinfinitives.2006.xiv,188pp.97 hartmann, Jutta m. and lászló molnárFi (eds.):ComparativeStudiesinGermanicSyntax.From

AfrikaanstoZurichGerman.2006.vi,331pp.96 lynGFelt, benjamin and torgrim SolStad (eds.):DemotingtheAgent.Passive,middleandother

voicephenomena.viii,326pp.+index.Expected December 200695 VoGeleer, Svetlana and liliane taSmoWSKi (eds.):Non-definitenessandPlurality.2006.vi,358pp.94 arche, maría J.:IndividualsinTime.Tense,aspectandtheindividual/stagedistinction.2006.xiv,281pp.93 ProGoVac, ljiljana, Kate PaeSani, eugenia caSielleS and ellen barton (eds.):TheSyntaxof

Nonsententials.Multidisciplinaryperspectives.2006.x,372pp.92 boecKx, cedric (ed.):AgreementSystems.2006.ix,346pp.91 boecKx, cedric (ed.):MinimalistEssays.2006.xvi,399pp.90 dalmi, Gréte:TheRoleofAgreementinNon-FinitePredication.2005.xvi,222pp.89 Velde, John r. te:DerivingCoordinateSymmetries.Aphase-basedapproachintegratingSelect,Merge,

CopyandMatch.2006.x,385pp.88 mohr, Sabine:ClausalArchitectureandSubjectPositions.ImpersonalconstructionsintheGermanic

languages.2005.viii,207pp.87 Julien, marit:NominalPhrasesfromaScandinavianPerspective.2005.xvi,348pp.86 coSta, João and maria cristina FiGueiredo SilVa (eds.):StudiesonAgreement.2006.vi,285pp.85 miKKelSen, line:CopularClauses.Specification,predicationandequation.2005.viii,210pp.84 PaFel, Jürgen:QuantifierScopeinGerman.2006.xvi,312pp.83 SchWeiKert, Walter:TheOrderofPrepositionalPhrasesintheStructureoftheClause.2005.

xii,338pp.82 Quinn, heidi:TheDistributionofPronounCaseFormsinEnglish.2005.xii,409pp.81 FuSS, eric:TheRiseofAgreement.Aformalapproachtothesyntaxandgrammaticalizationofverbal

inflection.2005.xii,336pp.80 burKhardt, Petra:TheSyntax–DiscourseInterface.Representingandinterpretingdependency.2005.

xii,259pp.79 Schmid, tanja:InfinitivalSyntax.InfinitivusProParticipioasarepairstrategy.2005.xiv,251pp.78 diKKen, marcel den and christina m. tortora (eds.):TheFunctionofFunctionWordsand

FunctionalCategories.2005.vii,292pp.77 ÖztürK, balkız:Case,ReferentialityandPhraseStructure.2005.x,268pp.76 StaVrou, melita and arhonto terzi (eds.):AdvancesinGreekGenerativeSyntax.InhonorofDimitra

Theophanopoulou-Kontou.2005.viii,366pp.75 di Sciullo, anna maria (ed.):UGandExternalSystems.Language,brainandcomputation.2005.

xviii,398pp.74 heGGie, lorie and Francisco ordóñez (eds.):CliticandAffixCombinations.Theoreticalperspectives.

2005.viii,390pp.73 carnie, andrew, heidi harley and Sheila ann dooley (eds.):VerbFirst.Onthesyntaxofverb-

initiallanguages.2005.xiv,434pp.72 FuSS, eric and carola triPS (eds.):DiachronicCluestoSynchronicGrammar.2004.viii,228pp.71 Gelderen, elly van:GrammaticalizationasEconomy.2004.xvi,320pp.70 auStin, Jennifer r., Stefan enGelberG and Gisa rauh (eds.):Adverbials.Theinterplaybetween

meaning,context,andsyntacticstructure.2004.x,346pp.69 KiSS, Katalin É. and henk van riemSdiJK (eds.):VerbClusters.AstudyofHungarian,Germanand

Dutch.2004.vi,514pp.

68 breul, carsten:FocusStructureinGenerativeGrammar.Anintegratedsyntactic,semanticandintonationalapproach.2004.x,432pp.

67 mišeSKa tomić, olga (ed.):BalkanSyntaxandSemantics.2004.xvi,499pp.66 Grohmann, Kleanthes K.:ProlificDomains.OntheAnti-Localityofmovementdependencies.2003.

xvi,372pp.65 manninen, Satu helena:SmallPhraseLayers.AstudyofFinnishMannerAdverbials.2003.xii,275pp.64 boecKx, cedric and Kleanthes K. Grohmann (eds.):MultipleWh-Fronting.2003.x,292pp.63 boecKx, cedric:IslandsandChains.Resumptionasstranding.2003.xii,224pp.62 carnie, andrew, heidi harley and maryann Willie (eds.):FormalApproachestoFunctionin

Grammar.InhonorofEloiseJelinek.2003.xii,378pp.61 SchWabe, Kerstin and Susanne WinKler (eds.):TheInterfaces.Derivingandinterpretingomitted

structures.2003.vi,403pp.60 triPS, carola:FromOVtoVOinEarlyMiddleEnglish.2002.xiv,359pp.59 dehÉ, nicole:ParticleVerbsinEnglish.Syntax,informationstructureandintonation.2002.xii,305pp.58 di Sciullo, anna maria (ed.):AsymmetryinGrammar.Volume2:Morphology,phonology,acquisition.

2003.vi,309pp.57 di Sciullo, anna maria (ed.):AsymmetryinGrammar.Volume1:Syntaxandsemantics.2003.

vi,405pp.56 coene, martine and yves d’hulSt (eds.):FromNPtoDP.Volume2:Theexpressionofpossessionin

nounphrases.2003.x,295pp.55 coene, martine and yves d’hulSt (eds.):FromNPtoDP.Volume1:Thesyntaxandsemanticsofnoun

phrases.2003.vi,362pp.54 baPtiSta, marlyse:TheSyntaxofCapeVerdeanCreole.TheSotaventovarieties.2003.

xxii,294pp.(incl.CD-rom).53 zWart, c. Jan-Wouter and Werner abraham (eds.):StudiesinComparativeGermanicSyntax.

Proceedingsfromthe15thWorkshoponComparativeGermanicSyntax(Groningen,May26–27,2000).2002.xiv,407pp.

52 Simon, horst J. and heike WieSe (eds.):Pronouns–GrammarandRepresentation.2002.xii,294pp.51 Gerlach, birgit:CliticsbetweenSyntaxandLexicon.2002.xii,282pp.50 Steinbach, markus:MiddleVoice.Acomparativestudyinthesyntax-semanticsinterfaceofGerman.

2002.xii,340pp.49 alexiadou, artemis (ed.):TheoreticalApproachestoUniversals.2002.viii,319pp.48 alexiadou, artemis, elena anaGnoStoPoulou, Sjef barbierS and hans-martin Gärtner

(eds.):DimensionsofMovement.Fromfeaturestoremnants.2002.vi,345pp.47 barbierS, Sjef, Frits beuKema and Wim van der WurFF (eds.):ModalityanditsInteractionwiththe

VerbalSystem.2002.x,290pp.46 PanaGiotidiS, Phoevos:Pronouns,CliticsandEmptyNouns.‘Pronominality’andlicensinginsyntax.

2002.x,214pp.45 abraham, Werner and c. Jan-Wouter zWart (eds.):IssuesinFormalGerman(ic)Typology.2002.

xviii,336pp.44 taylan, eser erguvanlı (ed.):TheVerbinTurkish.2002.xviii,267pp.43 FeatherSton, Sam:EmptyCategoriesinSentenceProcessing.2001.xvi,279pp.42 alexiadou, artemis:FunctionalStructureinNominals.Nominalizationandergativity.2001.x,233pp.41 zeller, Jochen:ParticleVerbsandLocalDomains.2001.xii,325pp.40 hoeKSema, Jack, hotze rullmann, Víctor Sánchez-Valencia and ton van der Wouden

(eds.):PerspectivesonNegationandPolarityItems.2001.xii,368pp.39 Gelderen, elly van:AHistoryofEnglishReflexivePronouns.Person,Self,andInterpretability.2000.

xiv,279pp.38 meinunGer, andré:SyntacticAspectsofTopicandComment.2000.xii,247pp.37 lutz, uli, Gereon müller and arnim von StechoW (eds.):Wh-ScopeMarking.2000.vi,483pp.36 Gerlach, birgit and Janet GriJzenhout (eds.):CliticsinPhonology,MorphologyandSyntax.2001.

xii,441pp.35 hróarSdóttir, Thorbjörg:WordOrderChangeinIcelandic.FromOVtoVO.2001.xiv,385pp.34 reuland, eric (ed.):ArgumentsandCase.ExplainingBurzio’sGeneralization.2000.xii,255pp.

33 PuSKáS, Genoveva:WordOrderinHungarian.ThesyntaxofĀ-positions.2000.xvi,398pp.32 alexiadou, artemis, Paul laW, andré meinunGer and chris Wilder (eds.):TheSyntaxof

RelativeClauses.2000.vi,397pp.31 SVenoniuS, Peter (ed.):TheDerivationofVOandOV.2000.vi,372pp.30 beuKema, Frits and marcel den diKKen (eds.):CliticPhenomenainEuropeanLanguages.2000.

x,324pp.29 miyamoto, tadao:TheLightVerbConstructioninJapanese.Theroleoftheverbalnoun.2000.

xiv,232pp.28 hermanS, ben and marc van ooStendorP (eds.):TheDerivationalResidueinPhonological

OptimalityTheory.2000.viii,322pp.27 růžičKa, rudolf:ControlinGrammarandPragmatics.Across-linguisticstudy.1999.x,206pp.26 acKema, Peter:IssuesinMorphosyntax.1999.viii,310pp.25 FelSer, claudia:VerbalComplementClauses.Aminimaliststudyofdirectperceptionconstructions.

1999.xiv,278pp.24 rebuSchi, Georges and laurice tuller (eds.):TheGrammarofFocus.1999.vi,366pp.23 GiannaKidou, anastasia:PolaritySensitivityas(Non)VeridicalDependency.1998.xvi,282pp.22 alexiadou, artemis and chris Wilder (eds.):Possessors,PredicatesandMovementinthe

DeterminerPhrase.1998.vi,388pp.21 Klein, henny:AdverbsofDegreeinDutchandRelatedLanguages.1998.x,232pp.20 laenzlinGer, christopher:ComparativeStudiesinWordOrderVariation.Adverbs,pronouns,and

clausestructureinRomanceandGermanic.1998.x,371pp.19 JoSeFSSon, Gunlög:MinimalWordsinaMinimalSyntax.WordformationinSwedish.1998.ix,199pp.18 alexiadou, artemis:AdverbPlacement.Acasestudyinantisymmetricsyntax.1997.x,256pp.17 beermann, dorothee a., david leblanc and henk van riemSdiJK (eds.):RightwardMovement.

1997.vi,410pp.16 liu, Feng-hsi:ScopeandSpecificity.1997.viii,187pp.15 rohrbacher, bernhard Wolfgang:Morphology-DrivenSyntax.AtheoryofVtoIraisingandpro-

drop.1999.viii,296pp.14 anaGnoStoPoulou, elena, henk van riemSdiJK and Frans zWartS (eds.):MaterialsonLeft

Dislocation.1997.viii,349pp.13 alexiadou, artemis and t. alan hall (eds.):StudiesonUniversalGrammarandTypological

Variation.1997.viii,252pp.12 abraham, Werner, Samuel david ePStein, höskuldur thráinSSon and c. Jan-Wouter zWart

(eds.):MinimalIdeas.Syntacticstudiesintheminimalistframework.1996.xii,364pp.11 lutz, uli and Jürgen PaFel (eds.):OnExtractionandExtrapositioninGerman.1996.xii,315pp.10 cinQue, Guglielmo and Giuliana GiuSti (eds.):AdvancesinRoumanianLinguistics.1995.xi,172pp.9 Gelderen, elly van:TheRiseofFunctionalCategories.1993.x,224pp.8 FanSeloW, Gisbert (ed.):TheParametrizationofUniversalGrammar.1993.xvii,232pp.7 ÅFarlí, tor a.:TheSyntaxofNorwegianPassiveConstructions.1992.xii,177pp.6 bhatt, christa, elisabeth lÖbel and claudia maria Schmidt (eds.):SyntacticPhraseStructure

PhenomenainNounPhrasesandSentences.1989.ix,187pp.5 GreWendorF, Günther and Wolfgang SterneFeld (eds.):ScramblingandBarriers.1990.vi,442pp.4 abraham, Werner and Sjaak de meiJ (eds.):Topic,FocusandConfigurationality.Papersfromthe6th

GroningenGrammarTalks,Groningen,1984.1986.v,349pp.3 abraham, Werner (ed.):OntheFormalSyntaxoftheWestgermania.Papersfromthe3rdGroningen

GrammarTalks(3eGroningerGrammatikgespräche),Groningen,January1981.1983.vi,242pp.2 ehlich, Konrad and Jürgen rehbein:Augenkommunikation.MethodenreflexionundBeispielanalyse.

1982.viii,150pp.Withmanyphotographicills.1 KlaPPenbach, ruth (1911–1977):StudienzurModernenDeutschenLexikographie.Auswahlaus

denLexikographischenArbeitenvonRuthKlappenbach,erweitertumdreiBeiträgevonHeleneMalige-Klappenbach.(WritteninGerman).1980.xxiii,313pp.

Comparative Studies in Germanic Syntax (Linguistik Aktuell 97)

Documents

Transcript of Comparative Studies in Germanic Syntax (Linguistik Aktuell 97)