Case.&.Grammatical.relations,2008

8/22/2019 Case.&.Grammatical.relations,2008

1/305


2/305

Case and Grammatical Relations


3/305

Volume 81

Case and Grammatical Relations. Studies in honor of Bernard Comrie

Edited by Greville G. Corbett and Michael Noonan

General Editor

Michael NoonanUniversity of Wisconsin-Milwaukee

Editorial BoardWallace ChafeSanta Barbara

Ronald W. LangackerSan Diego

Bernard ComrieLeipzig / Santa Barbara

Charles N. LiSanta Barbara

R.M.W. DixonMelbourne

Andrew PawleyCanberra

Matthew S. DryerBuffalo

Doris L. PayneOregon

John HaimanSt Paul

Frans PlankKonstanz

Jerrold M. SadockChicago

Bernd HeineKln

Assistant Editors

Spike GildeaUniversity of Oregon

Suzanne KemmerRice University

Paul J. HopperPittsburgh

Sandra A. TompsonSanta Barbara

Andrej A. KibrikMoscow

Dan I. SlobinBerkeley

Edith MoravcsikMilwaukee

A companion series to the journal Studies in Language. Volumes in this

series are functionally and typologically oriented, covering specific topics in

language by collecting together data from a wide variety of languages and

language typologies.

Typological Studies in Language (TSL)


4/305

Case and Grammatical RelationsStudies in honor of Bernard Comrie

Edited by

Greville G. CorbettUniversity of Surrey

Michael NoonanUniversity of Wisconsin-Milwaukee

John Benjamins Publishing Company

Amsterdam / Philadelphia


5/305

Library of Congress Cataloging-in-Publication Data

Case and grammatical relations : studies in honor of Bernard Comrie / edited by GrevilleG. Corbett, Michael Noonan.

p. cm. (ypological Studies in Language, - ; v. )

. Grammar, Comparative and general--Case. . Grammar, Comparative and general. I.

Comrie, Bernard, - II. Corbett, Greville G. III. Noonan, Michael (Michael P.)

P..C

--dc

(Hb; alk. paper)

John Benjamins B.V.

No part of this book may be reproduced in any form, by print, photoprint, microfilm, or anyother means, without written permission from the publisher.

John Benjamins Publishing Co. P.O. Box Amsterdam Te Netherlands

John Benjamins North America P.O. Box Philadelphia -

Te paper used in this publication meets the minimum requirements of

American National Standard for Information Sciences Permanence of

Paper for Printed Library Materials, z39.48-1984.

8TM


6/305

Table of contents

Preface vii

Determining morphosyntactic eature values: Te case o case 1

Greville G. Corbett

Does Hungarian have a case system? 35

Andrew Spencer

Case in Ingush syntax 57

Johanna Nichols

Cases, arguments, verbs in Abkhaz, Georgian and Mingrelian 75

George Hewitt

Te degenerate dative in Southern Norrbothnian 105

sten Dahl

Case compounding in the Bodic languages 127

Michael Noonan

Leipzig fourmille de typologues Genitive objects in comparison 149Martin Haspelmath & Susanne Michaelis

An asymmetry between VO and OV languages: Te ordering o obliques167

John A. Hawkins

On the scope o the reerential hierarchy in the typology o 191

grammatical relations

Balthasar Bickel

Does passivization require a subject category? 211

Marianne Mithun

Te definiteness o subjects and objects in Malagasy 241Edward L. Keenan

Without aspect 263

Maria Polinsky


7/305

Table of contents

Author index 283

Language index 285

Subject index 287


8/305

Preface

One o the pleasant difficulties o organizing a Festschrif or Bernard Comrie was

selecting a suitable topic around which to organize a coherent volume. For most

scholars, even very distinguished ones, the choice o topic or a Festschrif volume

is a simple one: choose the subject area in which the scholar has made his or her

mark. Since Bernard has made his mark in so many areas o linguistics, there was

no one obvious candidate, bur rather a host o possibilities. Afer some deliberation,

we chose case and grammatical relations, but there were many other candidates.Bernards work in the area o case and grammatical relations spans several decades

and retains its relevance. His most important works in this area include Keenan &

Comrie 1977, a seminal work which significantly strengthened the field o Relational

Grammar and encouraged typologists to investigate grammatical relations in particu-

lar constructions; Comrie (1986, 1991) and Comrie & Polinsky (1998) which deal with

the issue o how we determine how many cases a language has; and Comrie 1981,

one o the first comprehensive works on contemporary linguistic typology, in which

numerous issues dealing with case, grammatical relations, the relational hierarchy, andso on are discussed with elegance and clarity. His works dealing with issues o case

and grammatical relations in specific languages number in the dozens and represent a

broad sampling o the worlds languages.

Te papers in this volume can thereore be grouped into two broad though

overlapping classes: those dealing primarily with case and those dealing primarily

with grammatical relations. Corbett and Spencer use Comrie (1986) as a starting

point or their analyses o the case systems o Russian and Hungarian, respectively.

Corbett addresses the question o how many cases Russian has. Views on this issue

range rom the standard six o most grammars o Russian to as many as eleven. He

concludes that to answer this question, we have to recognize that case values vary

in their status in a language like Russian, with some being more canonical than

others. Spencertakes on a similar issue, asking whether Hungarian has any cases

at all. Analyses o the number o cases o Hungarian vary even more than those

o Russian, ranging rom 17 to 28. Spencer concludes that Hungarian has no case

system at all, employing rather a set o used postpositional portmanteaus.

Nichols and Hewitt examine the case systems o languages in the Caucasus.

Nicholsdescribes the case system o Ingush, concluding that the case system andagreement are consistently ergative, the ew exceptions exhibiting neutral [not

accusative] alignment, while the syntax o argument sharing is split ergative/


9/305

viii Preace

(loosely) accusative. Hewitt surveys some eatures o case and grammatical rela-

tions on the basis o examples rom languages belonging to the North West Cau-

casian and South Caucasian (or Kartvelian) amilies.

Dahldiscusses the retention o the dative case in some dialects o northern

Sweden. Tese dialects have retained the historical dative, which has been lost in

the standard varieties o all the mainland Scandinavian languages. Te situation

described by Dahl is interesting because the historical dative ollows a grammati-

calization path otherwise unattested: the dative plural develops into an indefinite

plural marker in premodified noun phrases.

Noonaninvestigates an issue in typology, namely the issue o case combining.

Afer presenting a typology o the kinds o case combining, he details the modes

o case combining ound in the Bodic languages o the ibeto-Burman amily. Tepaper by Haspelmath and Michaelissurveys genitive objects in a set o European

languages: German, English, Latin, French and Italian. Tey examine both geni-

tives marked by cases and genitives marked by prepositions and find a common

meaning among a diverse range o apparent unctions and orms.

Te second, related theme o grammatical relations is considered rom differ-

ent perspectives in the ollowing papers. Hawkinsexamines the order o obliques

in OV and VO languages, using the data collected or WALS (Haspelmath et al.

2005), o which Bernard Comrie is a coeditor. Bickel discusses the effects o thereerential hierarchy on the distribution o grammatical relations. In line with Ber-

nards own work on this, he finds weak statistical support or effects on splits in

case alignment, but he urthermore shows that there are no such effects on splits

in agreement systems and that the available typological evidence is indecisive

or other constructions. Mithunposes the question: Does passivization require

a subject category? She shows that languages without a robust subject category

generally lack passives, but there are exceptions, or example Pomoan languages

which show little evidence o a subject category yet still have a construction typi-

cally translated as a passive.

In linguistic theory, especially in its more ormal instantiations, attempts have

been made to link the distribution o case to aspect (or example, in the Minimal-

ist Program, tense or aspect is ofen viewed as a case assigner, and languages may

differ in allowing such case assignment or not). It is also well-known that aspect

or mood are closely linked to case alternations, or example, in Finnish or Slavic,

where the partitive is more ofen than not ound in the imperective, irrealis, or

under negation. Case, definiteness, and aspect orm another well-known cluster

o properties, and Bernards work has addressed the interaction among all three.Te consideration o tense, aspect and definiteness introduces a broader context in

which case and grammatical relations can be considered, as shown particularly by

two papers in this volume. Keenaninvestigates subjecthood, looking specifically


10/305

iPreace

at Malagasy rom the perspective o the generalization that subjects in West Aus-

tronesian languages are definite. He examines three types o quantified NPs which

could be interpreted as counterexamples to this generalization. Polinskydiscusses

an extreme case o language reanalysis when both case marking and correlated

aspectual distinctions disappear under the incomplete acquisition o Russian by

heritage speakers (people who learned Russian as children, but later abandoned it,

completely or partially, in avour o another dominant language). Crucially, while

the morphological realizations o case and aspect are absent rom heritage Rus-

sian, new mechanisms come into place which compensate or the absence o mor-

phology and express the conceptual contrasts that are seemingly universal.

Each paper was written especially or this volume, linking to research in which

Bernard has been involved, and taking it a step urther.We offer our sincere thanks to Lisa Mack, yko Dirksmeyer and Deborah

Mulvaney or their careul help in the preparation o the volume and to Edith

Moravcsik or numerous helpul comments. And Bernard is involved again here:

having received a pre-publication version on his birthday, he made some o his

trademark suggestions, which as authors we were all grateul to adopt.

References

Comrie, B. 1981[1989]. Language Universals and Linguistic ypology. Chicago IL: University oChicago Press.

Comrie, B. 1986. On delimiting cases. In Case in Slavic, R.D. Brecht & J.S. Levine (Eds), 86106.Columbus OH: Slavica.

Comrie, B. 1991. Form and unction in identiying cases. In Paradigms: Te Economy of Inflec-tion, F. Plank (Ed.), 4155. Berlin: Mouton de Gruyter.

Comrie, B. 2005. Alignment o case marking. In Te World Atlas of Language Structures,M. Haspelmath et al. (Eds), 398405. Oxord: OUP.

Comrie, B. & Polinsky, M. 1998. Te great Daghestanian case hoax. In Case, ypology, andGrammar, A. Siewierska & Jae Jung Song (Eds), 95114. Amsterdam: John Benjamins.

Haspelmath, M., Dryer, M., Gil, D. & Comrie, B. 2005. Te World Atlas of Language Structures(WALS). Oxord: OUP.

Keenan, E.L. & Comrie, B. 1977. Noun phrase accessibility and universal grammar. LinguisticInquiry8: 6399.


11/305


12/305

Determining morphosyntactic feature values

Te case o case1

Greville G. CorbettSurrey Morphology Groupa

Determining the number o cases (case values) in a given language may be a

challenging analytical task. In establishing the techniques, special attentionhas been devoted to Russian, since it exhibits a whole set o difficult problems.It has been claimed to have as ew as six case values or as many as eleven. Teevidence is considered again, taking the valuable work o the Set-theoreticalSchool, and moving on to a Canonical approach in which we construct alogical scheme against which to evaluate the different case values. We see clearlythat the case values differ dramatically in status, rom those at the centre o thesystem to those which are peripheral and in decline, yet maintaining a presencein the case system.

A long-running and still vital debate concerns the way in which we can deter-mine the number o cases (case values) in a given language. Tis matters bothor the description o particular languages, and even more or typology, given theimperative or the typologist to compare like with like. Within this debate specialattention has been devoted to Russian. And rightly so, since Russian exhibits awhole set o difficult analytic problems with respect to case. As a result it has beenclaimed to have as ew as six case values or as many as eleven.

. Tis is an issue to which Bernard Comrie has made important contributions (1986, 1991).One o his papers on the topic includes the line: the ideas are thereore put orward in thespirit o initiating a discussion (1991: 104); the current chapter takes orward that discussion

Te support o the ESRC under grant RES-051-27-0122 and o the AHRC under grant B/RG/AN4375/APN18306 is grateully acknowledged, as is the hospitality o the Department oLinguistics, Max Planck Institute or Evolutionary Anthropology, Leipzig, where I started workon this paper. I wish to thank all the ollowing or discussion and comments: Peter Arkadiev,Matthew Baerman, Noel Brackney, Dunstan Brown, Marina Chumakina, Bernard Comrie,Nicholas Evans, Anna Kibort, Maria Koptjevska-amm, Alexander Krasovitsky, Alison Long,Olga Mladenova, Edith Moravcsik, Irina Nikolaeva, Vladimir Plungian, Andrew Spencer andCarole iberius. Some o these comments were given at a meeting o Surrey Linguistic Circleon 7 June 2006. Special thanks or extended illuminating discussions are due to Michael Danieland Igor Meluk.


13/305

Greville G. Corbett

Tis contribution continues the debate, again giving Russian a central place.Our concern with case is partly with case as a eature (comparable to gender, num-ber and person), but mainly with the values o the eature (nominative, accusativeand so on).2What is novel about it is first the adoption o a canonical approach,in which we construct a logical scheme against which to evaluate the differentcase values (see 1 below), and second the act that the criteria we discuss areshown to be relevant to morphosyntactic eatures more generally, rather thenbeing restricted to case.

Te debate on case has a distinguished earlier history, including among othersHjelmslev (193537), Jakobson (1936, 1958),3de Groot (1939) and Kuryowicz(1949).4 It is not surprising that Russian has figured large, since its case sys-

tem combines many o the difficult issues. Uspenskij (1957) gives interestingdetail on earlier discussions, notably within the group which started work onmathematical models in linguistics in Moscow in 1956; responding to ques-tions posed by Andrej Kolmogorov.5Tis work in the Set-theoretical School ledto important contributions by Zaliznjak (1967, 1973), which are o continuing

value. One reason or this is Zaliznjaks modern approach to morphology (seeKrylov 2002: 705). Meluk (1986/2006) takes up the issues discussed by Zalizn-

jak critically and constructively. An extensive and sympathetic technical sur-

vey o this work is given by van Helden (1993).6

Tere is continuing discussion

. Te related issue as to when case should be recognized as a eature is considered in Spencer &Otoguro (2005) and Spencer (this volume). Features are ofen called categories, and their valuesare sometimes called properties, sometimes also grammemes, notably by Zaliznjak.

. Jakobsons approach is discussed and developed in many places, or instance, in Chvany (1986)and in Franks (1995: 4155). Unlike work in this tradition, we will not decompose Jakobsonseight case values into sub-eatures, except where there is specific evidence or structuring (as with

the second locative). Tere are two main grounds or this decision. First the analyses o this typehave proved problematic (see, or example, Baerman, Brown & Corbett 2005: 210). And second,as Gerald Gazdar has pointed out (personal communication), there are 6720 possible ways to de-scribe eight values using three binary eatures. In view o this, unless there are principled reasonsor postulating particular binary eatures rom the outset, it should not be taken as significant ithere is an analysis using binary eatures which is partially successul.

. For a critical account o some o the early attempts to describe the Russian system see Kilby(1977: 142); or much earlier accounts o case see Serbat (1981), and or more recent develop-ments see Anderson (2006) and Butt (2006). Te importance o the issue can be seen rom the act

that the general topic o case, taken broadly, has 6643 entries in Campes bibliography (1994).. See van Helden (1993: 138) or sources.

. van Helden (1993: 554557, 10621115) is particularly relevant to Zaliznjaks work; Meyer(1994) is a valuable review o van Helden (1993).


14/305

Determining morphosyntactic eature values

on determining the case values o Russian, see or instance Gladkij (1999)7andPlungian (2000: 161180); a useul survey rom a computational perspective isprovided by Koval (2004).

Each o the case values identified (such as nominative or genitive) may havedifferent unctions; as Zaliznjak (1973: 56n2) points out, this is recognized in thetraditional names: genetivus possessivus, genetivus partitivus, dativus commodi,dativus possessivus, and so on, where the first part o the name specifies the ormalcase, and the second part indicates its unction. While we shall concentrate oncase values, these various case unctions also deserve typological investigation, asproposed by Ferguson (1970), and continued recently through the use o semanticmaps (Haspelmath 2003).8

. Te canonical approach in typology

How are we to make progress in understanding a system like the Russian casesystem? We must examine the data careully using different approaches (includingwork with consultants and corpus studies) and work through the extensive litera-ture on the subject, some o which we noted above. Yet the data are so amiliar

to many linguists that it is hard to get urther. In trying to get a new view o thiscomplex phenomenon we may take a canonical approach. We extrapolate romwhat there is to what there might be. And within that scheme o theoretical pos-sibilities we can situate the real instances we have ound. An effect o this canonicalapproach is to separate out coincidental overlappings in the examples that exist;we may then start to ask which characteristics happento be the way they are andwhich haveto be the way they are. o take an easy example: in Russian the ormIvanu to Ivan can only be an instance o the dative case. Whatever the syntac-tic environment, Ivanuto Ivan must be a dative (and i another case value wererequired this orm would be ungrammatical). On the other hand Ivanacould fitinto syntactic slots where we find the accusative and those requiring the genitive.I every orm were like that, we would have no evidence or distinguishing genitiverom accusative; and continuing along that route to the logical end point, i wehad no means o distinguishing cases we would have no case system. We can say,

. Gladkijs dependency model (presented in Gladkij 1969, 1973a,b) is considered to be

the apex o case modelling (van Helden 1993: 849); see van Helden (1993: 849878) ordiscussion.

. Te issue o whether case values have meaning is taken up by Meluk (2004). For discussionboth o establishing case values and o their unctions see Arkadiev (2006: 1848).


15/305

Greville G. Corbett

thereore, that the situation represented by Ivanuto Ivan, where we have a uniquemapping rom orm to unction is canonical, that represented by Ivanais not. Wecan imagine a system in which every orm o every nominal were like Ivanu. Tatwould be ully canonical in terms o the orm-unction mapping. It does not mat-ter at this stage whether such a system exists: the point is that we can define it, wecan recognize it i we find it, and it gives us one measure o canonicity accordingto which we can calibrate the instances o case values in the system which we areexamining. We can consider the syntax o these case values in a similar way. Imag-ine we had a language whose case values were all determined by simple syntacticrules o the type: the direct object o a transitive verb stands in the accusative case.We could recognize and agree about such a system, and it would indeed be canoni-

cal. It is harder to classiy the numerous possible deviations rom such simple syn-tactic rules: we can agree readily that there is a deviation, but it is much harder toestablish and to agree whether we have a slightly different syntactic structure, or asemantic condition, and so on. Tus it can be helpul to have the logically possiblecanonical system in mind, whether or not we find examples o it.

Te canonical approach then requires clear definitions. We take these to theirlogical end points, in order to construct a theoretical space. Te convergence ocriteria fixes a canonical point rom which the phenomena actually ound can

be calibrated. Te instances which would qualiy as canonical according to ourdefinitions, the ones that are the indisputable instances, will almost certainlynot be requent. Tis is expected, and certainly not to be treated as a problem.Since the paper uses Russian data, it is appropriate to repeat the neat ormula-tion by Johanna Nichols (personal communication): Canonical constructions areall alike; each non-canonical construction is non-canonical in its own way.9Tiscanonical approach has been applied or both syntax and morphology (see Corbett(2007a) or reerences). Te canonical approach allows us to handle gradient phe-nomena in a principled way. For the current problem this has a nice consequence,in that we can locate instances o case values as more or less canonical, rather thanhaving an all or nothing requirement that a particular instance is a case valueor it is not. Tis is a particular advantage when considering diachronic change,since orms may gain or lose casehood gradually over time (Michael Daniel,personal communication).

. Compare: All happy amilies are alike; each unhappy amily is unhappy in its own way.(Lev olstoy,Anna Karenina). It is worth stressing, however, that there is no requirement or thecanonical point to be occupied, in act it typically is not. o continue the comparison, there isno need or us to point to the linguistic equivalent o a amily completely and perectly happyaccording to all criteria.


16/305


. Canonical features and their values

In attempting to establish what is canonical or case, we find that almost everycriterion holds equally well or other morphosyntactic eatures and their values.We shall concentrate on case, while making occasional comparisons with othereatures. We need to look first at how case values are identified, in order then toinvestigate how issues o canonicity relate to such case values (3). As a point oreerence, we give traditional paradigms o two types o Russian noun:

(1) Paradigmo twoRussiannouns

urnalmagazine

(inflectional class I)

komnataroom

(inflectional class II)

Singular Plural Singular Plural

Nominative urnal urnaly komnata komnaty Accusative urnal urnaly komnatu komnaty Genitive urnala urnalov komnaty komnatDative urnalu urnalam komnate komnatamInstrumental urnalom urnalami komnatoj komnatamiLocative urnale urnalax komnate komnatax

Tese are ully regular nouns: there are many thousands which inflect similarly.10However, these nouns represents only two inflectional classes; there are two othermajor classes and several smaller subclasses (Corbett & Fraser 1993).

Te cases proposed are relatively uncontroversial. But it is still worth asking, asKolmogorov and the members o the Set-theoretical School did, how we justiy suchan analysis. For instance, given that urnalhas the same orms, singular and plu-ral, or the case values nominative and accusative, how do we justiy claiming there

are two case values here? Te method is set out in Zaliznjak (1967/2002: 3642),but see also Goddard (1982) and Comrie (1986, 1991). We start rom the idea ocontexts. We find various syntactic contexts, such asja viu I see , u menjanet I havent got a and collate the orms which fit appropriately into thesecontexts (by tradition, the contexts are the rows, and the items examined nounsin this instance are arranged in columns, giving a table). Evidently urnalandkomnatuwould both fit in the ja viu context. However, there are other con-texts where komnatuwould not fit, but komnatawould be needed, and here urnal

. Animates like Ivan, mentioned earlier, inflect like urnal magazine but, being animate,have the accusative syncretic with the genitive.


17/305

Greville G. Corbett

would again fit. Hence we could claim that the two different instances o urnalhave different case values. I two contexts produce exactly the same results orevery noun we test, then we can discard one o the two contexts.

So ar this makes intuitive sense. However, the context must also be semanti-cally constrained. Tus the context ona piet she is writing would allow both

pismo a letter and karandaomwith a pencil. We do not want to suggest thesetwo nouns are in the same case, rather that the apparently single context is notadequate here.11In the terms o Comrie (1986: 91)12we also require identity ounction. And more generally, we may require alternative contexts to allow naturalreadings or different semantic classes o noun.

Te procedure works well while we constrain the contexts (consciously or

unconsciously). But suppose that like Zaliznjak we take the procedure seriouslyand include contexts likeja risuju svoju I am drawing my own ? Svojuonesown is eminine, and there is no possible orm o urnalmagazine that could fitinto this context, since it is masculine. Te next step in the procedure is to elimi-nate contexts like this one, which produce gaps in the table, provided that in allother respects (i.e., apart rom the gap) the context gives results which are identi-cal to those o another context. Tis has the neat effect o allowing us to abstractaway rom the eatures o number and gender. For a uller account see Zaliznjak

(1967/2002: 3642); the issues are well summarized in Blake (1994: 2930).

. Criteria for canonical features and values

Given this general approach to determining case eature values, let us now con-sider a canonical morphosyntactic eature and its values, and see how our casedata fit in. Tere are several criteria, which we group under more general prin-ciples. Tis part has two goals, thereore. o introduce the criteria or canonicity,

and then or each one to indicate briefly how the six main cases o Russian (as in(1)) measure up against it.

Principle I: Features and their values are clearly distinguished by formal means (andthe clearer the formal means by which a feature or value is distinguished,the more canonical that feature or value).

. Te notion o context is considered urther in Madojan (1989).

. Bl & Pettersson (1988) take issue with Comrie (1986), arguing against what they con-sider to be over-reliance on the analogical principle or determining the number o cases; theiranalysis, in my view, does not provide a convincing interace between syntax and morphology.As discussed in 2.2 below, syntax should be morphology-ree, and their analysis, I believe,does not guarantee this.


18/305


Formal means are clear to the extent that they allow a transparent and regu-lar mapping rom orm to unction. Te general point is straightorward: in thecanonical situation there is clear evidence or the eature and its values.

urning to Russian, it is evident that there is a morphosyntactic eature ocase. In most instances, accounting or the inflectional orm o the use o a noun,adjective or pronoun requires reerence to case. More interesting is to look at the

values o the eature and more interesting still to look at the contentious case val-ues. We thereore look quite briefly here at the main six values, going through thecriteria in turn, mainly in order to indicate that the six main case values vary con-siderably, and in some instances diverge substantially rom canonical. We shall notlinger over the detail here, in order to concentrate subsequently on the contentious

case values (in 4 and 5). We shall see that Principle I is largely observed, thoughwith great variation between the case values.

Principle I covers our more specific criteria:

Criterion 1: Canonical eatures and their values have a dedicated orm(are autonomous).

It is natural to assume that in order to postulate a eature, and its various values,

we should be able in each instance to point to an inflected orm and to show thatit can be explained only in terms o the particular eature and value. For instance,in (1) above, the orm urnalomrequires reerence to case and to instrumental: itcannot result rom any other specification. Tis is what Zaliznjak (1973: 6974)and Meluk (1986: 6670) treat as autonomous. Tere is a question, however,as to what is the standard o comparison. We might state the criterion in abso-lute terms, that is, we look or some marker with a unique unction. Tis wouldbe appropriate i the inflectional morphology in question was also canonical (seePrinciple III below). However, Zaliznjak and Meluk both treat autonomy relative

to a particular lexeme; i or a given lexeme there is a unique orm, then the eaturevalue is autonomous. It may seem obvious that we can argue or a eature only i itautonomous is in this sense. However, even that is not clear-cut; see Chumakina,Kibort & Corbett (2007) where it is argued that a person eature is required in thegrammar o Archi, even though there is no dedicated orm to support it.

Evaluation of Russian: some case values are clearly canonical in terms o criterion1:thus urnalom magazine can only be instrumental (singular). However, the ac-

cusative case ares poorly against this criterion. Only in the second inflectionalclass do we find a unique orm or the accusative: komnatu room is accusative(singular). For all other classes the accusative is expressed by orms syncretic withthe nominative or genitive. Zaliznjak (1973: 7475) talks o the accusative as beingclose to non-autonomous.


19/305

Greville G. Corbett

Criterion 2: Canonical eatures and their values are uniquely distinguishableacross other logically compatible eatures and their values.

In the canonical situation, we can distinguish a case and its values irrespectiveo other eatures and their values. In other words, we do not have to select particu-lar combinations: any o them will serve.13O course, in languages like Germanwe find numerous syncretisms, so that, or instance, to establish gender we look atorms which are singular. Allowing or the combination o eatures is a key part othe procedure or defining case; see Zaliznjak (1973: 59). Te non-canonicity pro-duced by various types o syncretism can produce difficult problems, as we shallsee. Indeed, Meyer (1994: 360) suggests that successul set-theoretic modelling o

a category in a given language requires ull knowledge o all the other categories.

Russian:while case is expressed together with number (and also with gender in ad-jectives and some pronouns) the different values are normally distinguishable in thedifferent combinations. However, the accusative is again ar rom canonical here.

Criterion 3: Canonical eatures and their values are distinguished consistentlyacross relevant word classes.

In a sense it is true to say that German and English both have case, and indeed

that German and English both have gender. However, these statements are some-what misleading. In German we have evidence or the case eature in articles,adjectives and pronouns (as well as limited evidence in nouns). For gender wehave evidence in articles, adjectives and pronouns. In English the evidence or caseand gender is restricted to pronouns.14Criterion 3 draws the distinction betweenthe two systems, German being largely canonical here and English clearly not.

Russian: the six main case values are canonical in this respect. Tey are distin-

guished consistently across nouns, adjectives and pronouns.Criterion 4: Canonical eatures and their values are distinguished consistently

across lexemes within relevant word classes.

. Zaliznjak (1973: 8486) and Meluk (1986: 61) talk o complete versus partial case values,where a complete case value embraces all nouns o a language throughout the whole paradigm,while a partial case value unctions or a subset o the nouns or or a subset o the paradigmsonly (Meluk 1986: 61). I separate out the notions o embracing all nouns and that o applyingthroughout the paradigm in criteria 4 and 2 respectively.

. For those who do not accept pronominal gender systems, as discussed in Corbett(1991: 169170), the gender example is not relevant here, but the argument rom case is clear.Numerous urther examples o case systems where the inventories o different word classesdiffer can be ound in Iggesen (2005).


20/305


In the canonical situation, given the morphosyntactic specification deter-mined by the syntax and the word class (part o speech) o the target,15no moreis required. Each member o the word class marks the eature and all its valuesconsistently. When more inormation isrequired, that is, when not all members othe word class behave consistently, deviations rom the canonical situation may beseen in two different ways:

1. In terms o level, the deviation may be in terms o thefeatureas a whole16oronly in terms o its values. Apart rom a ew indeclinables, any Russian adjectivemarks gender. In contrast, in Macedonian, while most adjectives mark gender andnumber, some mark number but do not mark gender (Friedman 1993: 266267).

Tat is, they do not mark thefeaturegender as a whole; it is not simply that theyail to distinguish particular values. Returning to case in Russian, we have alreadyseen in (1) an instance o how different nouns can each mark case, but mark itsdifferent valuesto differing degrees.2. In terms o range, the deviation may affect different numbers o lexemes. Tecanonical situation is that each lexeme marks the eature and its values. Devia-tions may involve larger or smaller subclasses. We may find inflectional classes,which distinguish the particular eature and its values to varying degrees (as inpoint 1). Tere may be subclasses at various levels, right down to individual lexicalexceptions. Te latter may be overdifferentiated (marking too many distinctionscompared with the other members o their subclass) or they may show additionalsyncretism and so show too ew distinctions. For such situations, approaches likethat o Network Morphology (as in Corbett & Fraser 1993 and Evans, Brown &Corbett 2002), which rely on deault inheritance, are particularly helpul, sincethey readily capture generalizations which apply to large classes and are overrid-den by specifications which are more and more restricted, right down to the idio-syncrasies o individual words. Moreover, this sort o deviation may be motivated

to varying degrees. We may find that a distinction is available or, say, all animates(highly motivated), or at the other extreme that the distinction affects a number olexical items each o which requires special indication in its lexical entry.

Te different types o non-canonicity o lexemes (such as marking too ew ortoo many distinctions) are covered by our third principle below, namely thatcanonical morphosyntactic eatures and their values are expressed by canoni-cal inflectional morphology. Tat principle is concerned with canonicity rom

. Note that we reer to the target since these criteria apply not just to case but more generallyto morphosyntactic eatures. For case, the target is the governed item.

. Tis criterion takes up lexical generality, as in Bybee (1985: 8486), and goes urther inclearly distinguishing its application to eatures and to their values.


21/305

Greville G. Corbett

the point o view o the lexeme. Here we take the perspective o the eature andits values, suggesting that a canonical eature will not be subject to the restric-tions we are considering. o clariy, we can return to those Macedonian adjectiveswhich, non-canonically, mark number but not gender. From the point o view othe lexeme (as in Principle III), it would be equally non-canonical to ail to distin-guish gender or to ail to distinguish number (since both eatures are available tothat word class). From the point o view o the eatures (Criterion 4), in respect othese Macedonian adjectives, number is more canonical than gender since there isa restriction on gender which number is exempt rom.

Te essential point o this criterion, then, is that in the canonical situationit is sufficient to have a syntactic rule (o the type: in Polish the preposition ku

towards requires the dative) and the word class o the target (e.g., noun). Anyrequirement or additional inormation about the particular lexeme(s) making upthe target is non-canonical.

Russian: the main case values are close to canonical in this respect, since they aredistinguished by almost all nouns, adjectives and pronouns. It is true that Russian hasquite a high proportion o indeclinable nouns. Tese ail to inflect, but or number asor case, so that case values are not singled out here. Tese indeclinables lack morpho-

logical case, but can occupy syntactic slots appropriate or the different case values.

Principle II: Te use of canonical morphosyntactic features and their values is deter-mined by simple syntactic rules.

Tis principle is vital or the interace between syntax and morphology. Animportant part o being simple is that syntax is morphology-ree (Zwicky1996: 301). Tat is, the rules o syntax do not have access to purely morphologi-cal eatures, such as inflectional class. Here we part company with Zaliznjak, who

rightly points to the issue o the complexity o the rules o government, but sug-gests that it is not significant whether we have simpler rules o government and alarger inventory o cases, or more complex rules and ewer cases (1973: 67). Spe-cifically, non-autonomous cases can be eliminated by a rule o government whichdistinguishes between subclasses o noun. We accept such a possibility when thesubclasses are semantically defined, but exclude the possibility o rules o govern-ment reerring to inflectional classes.

Russian: here we find all the main values o case are largely canonical, though againwith some surprising divergences.

Criterion 5: Te use o canonical morphosyntactic eatures and their values isobligatory.


22/305


Tis is a well-known and important criterion. It is highlighted by Jakobson; inhis discussion o Boas (1938: 132133) Jakobson produced the amous quote: Tusthe true difference between languages is not in what may or may not be expressedbut in what must or must not be conveyed by the speakers. (1959/1971: 492)17With regard to case, in the canonical situation case is required. In Russian anynoun must be in some case, and so i there is a nominal orm with no overt marker,this will still be interpreted as having a specific case value (the particular value willdepend on the paradigm, see the orms o the nominative/accusative singular oinflectional class I in (1), and the genitive plural o inflectional class II). Te barestem will not be interpreted as being outside the case system. Similarly, rom thepoint o view o the governor, in the canonical situation a governor requires a

particular case value (or instance, Russian k towards takes the dative, and thereis no more to be said).

However, there is a more subtle problem lurking here. Where there is somedegree o optionality, it is not straightorward, when trying to define case, to avoidadmitting additional case values such as the nominative/instrumental. Given acopula verb like bytbe, in the past or uture, a predicate nominal may be in thenominative or instrumental case. As Zaliznjak (1973: 62) says, no grammar oRussian posits a composite case here; to avoid this, he has a stipulation which

has the effect o eliminating cases that would arise rom choices between other-wise established cases.18Our criterion has the effect o making such choices non-canonical.19

Russian: it is certainly true that the use o the main six case values is obligatory.

. Note that Meluk (1960/1974) discusses this criterion and in (1974: 111) points out that he,

Meluk, wrote the article in 1958; beore seeing Jakobsons article. Percov (1996: 40, 2001: 71)traces the history o the notion back through Jakobson to Boas and beore him to Maspero(1934: 35). However, I think Jakobson is right to give primacy to Boas, since the idea can be alsoound in Boas (1911: 3543, especially 4043).

. And the problem caused by optionality becomes worse, since once we had a second loca-tive, we would also have to allow or a composite locative/second locative or those nouns whichhave an optional second locative.

. Michael Daniel points out (personal communication) that we might analyse such casechoices as reflecting slightly different diagnostic contexts (in the sense discussed in 2). Tisapproach would entail making context a gradient notion. Te contexts envisaged in 2 could bedistinguished by straightorward syntactic tests, and unproblematic speaker judgements, whilethe case choices like nominative versus instrumental in the nominal predicate cannot. Tus itseems appropriate to treat such choices differently, and to consider them non-canonical. Fordiscussion o some problems connected with the criterion see Maslova (1994).


23/305

Greville G. Corbett

Criterion 6: Canonical use o morphosyntactic eatures and their values does notadmit syntactic conditions.

Canonically there is a syntactic rule (mark the direct object with the accusa-tive) but no additionalsyntactic conditions.

Russian: there are instances o non-canonicity. Tus a possible condition mightinvolve word order; and indeed topicalization affects the use o accusative or geni-tive or the direct object negated verbs (imberlake 1975: 126).20

Criterion 7: Canonical use o morphosyntactic eatures and their values does notadmit semantic conditions.

I only semantics were involved we would have instead a morphosemantic ea-ture, as we find, or instance, in local cases notably in Daghestanian languages.Here we are concerned with eatures and values that are genuinely morphosyntac-tic in nature. Te sort o condition that would give non-canonicity would be casemarking or objects being conditioned by whether the object was definite or not.

Russian: the choice o case values or negated objects is an example here (imber-lake 1975: 125). Similarly the nominal predicate with copular and semi-copular

verbs may be in the nominative or instrumental. Te conditions on these choicesare numerous, and complex, and have changed considerably over the last two cen-turies

Criterion 8: Canonical use o morphosyntactic eatures and their values does notadmit lexical conditions rom the target (governee).

While Criterion 4 was concerned with whether marking was possible or not,here we assume that case is marked, and ask whether there can be urther lexicalconditions. Te Russian prepositionpo (whose meaning is too wide to give a gloss

or), allows surprising conditions, which have been changing over the recent past.Significantly or our purposes, different numerals stand in different cases whengoverned bypo(Comrie, Stone & Polinsky 1996: 153154 and reerences there). Inother usespo takes the dative, but with the meaning afer it may take the locative,

. Te conditions considered in criteria 69 could in principle be absolute or relative. I theywere always relative, allowing some degree o optionality, they could be treated as sub-instanceso criterion 5, which specifies that the canonical use o eatures and their values is obligatory.Te conditions noted in criteria 69 are treated separately since they can be absolute, and so donot necessarily come under criterion 5. Tey can apply to other morphosyntactic eatures too:thus the values o the number eature in agreement are also subject to word-order conditions(Corbett 2006: 179).


24/305


and according to Comrie (1991: 50) these collocations are largely lexicalized, anotherinstance o non-canonical case government. For more on the various governmentpossibilities oposee Iomdin (1991).

Tis criterion can be seen as relating to lexical semantics (while criterion 7relates to semantics not specific to the lexeme). One way o thinking o criterion 8is to say that in canonical use one can combine the lexical meaning o the lexemeand the grammatical meaning o the eature value in a compositional ashion. InRussian the instrumental (without preposition) can be used to indicate the timeo an event, but provided the governed noun denotes a part o the day, or a seasono the year. Tis constraint rom the lexical semantics o the noun is non-canonical.21

Russian: under this latter interpretation, the six main case values have many ullycanonical uses. However, there are also several instances o conditions, particu-larly or temporal expressions, which are non-canonical in this respect.

Criterion 9: Canonical use o morphosyntactic eatures and their values does notadmit additional lexical conditions rom the controller (governor).

Te idea is that in the canonical situation the controller has a single require-

ment (e.g., it governs the dative). Additional conditions rom this source arenot canonical.

Russian: the main six case values are largely canonical in this respect, but we findsome instances o non-canonicity. Tus verbs normally take their subject in thenominative, irrespective o polarity. Just a ew verbs, however, can have a genitivesubject, but only when negated. Depending on the analysis o these constructions,we might consider this as an additional condition rom the governor.

(2) pisem ne prixodi-l-o letter.pl.gen neg come-pst-n.sg noletterscame

Te verbs must be lexically specified, and they govern the genitive onlywhen negated.

. Examples which are not canonical in this regard are not restricted to case. While or manynouns the plural is in a ully straightorward relationship to the singular, there are some wherethe plural has rather specific meanings (see, or instance, Plungian (1997) on vremjatime andvremenatimes in Russian).


25/305

Greville G. Corbett

Criterion 10: Te use o canonical morphosyntactic eatures and their values issufficient (they are independent).

In canonical instances, eatures and their values can stand alone.

Russian: five o the main case values o Russian are canonical in this respect: eachcan stand alone to fill various syntactic slots. Tus the dative can signal the indirectobject. Only the locative is non-canonical in this respect, since it can occur onlytogether with a preposition.

Principle III: Canonical morphosyntactic features and their values are expressed bycanonical inflectional morphology.

Tis principle covers a whole set o criteria, which we should have in mind butneed not consider in detail (they are specified in Corbett (2007b)).22 In brie, acanonical system has a one-to-one correspondence between orm and unction.o discuss canonicity o lexemes, we rely on the preceding analysis, since it pre-supposes that the eatures and their values are established or the language underinvestigation. Given the inventory o the eatures and their values we can multiplythem together, to give the possible cells in a paradigm. Tus i we find six cases andtwo numbers in the nominal system, the paradigm o a noun should have twelve

cells. And then, to be canonical, a paradigm has to be consistent, according tothese requirements:

(3) Canonicalinflection

comparison acrosscellsof a lexeme

comparison acrosslexemes

composition/structure (o theinflected word)

same same

lexical material (shape o stem) same differentinflectional material (shape o

inflection)different same

outcome (shape o inflected word) different different

Tis schema has two levels o comparison. At the first level, we start rom theabstract paradigm (matrix) obtained by multiplying out the eatures and their

. Since we are dealing with Russian, we have assumed that we are indeed dealing with bonafide inflection. For some languages this would be a serious issue or establishing whether wewere dealing with a case eature or not.


26/305


values, and we examine any one lexeme fitted into this paradigm. Te middlecolumn involves comparing cell with cell, within a single lexeme. We considerthe composition and structure o the cells (lef column): i the first consists oa stem and a suffix, or this lexeme to have a canonical paradigm, every othercell must be the same in this regard. Finding a prefix or any different means oexponence would indicate non-canonicity. In terms o the lexical material in thecell, identity is required (there should be no change to the stem). On the otherhand, the inflectional material should be different in every cell. Te canonicaloutcome or such a lexeme, as shown in the last row, is that every cell in its para-digm will realize the morphosyntactic specification in a way distinct rom that oevery other cell.

Te second level o comparison (given in the right column) involves a com-parison o lexemes one with another. In a canonical system, the composition andstructure o each cell remains the same, when we compare across lexemes. Nat-urally we require that the lexical inormation be different or different lexemes.However, in the canonical situation, the inflectional material is identical. Tat is, iour first lexeme marks genitive plural in -a-t, so does every other.

Te overall outcome is that every cell o every lexeme is distinct. Inflectiono this canonical type would make perect sense in unctional terms. Every mor-

phosyntactic distinction is drawn unambiguously, or a small amount o phono-logical material. In real inflectional morphology we find great divergence rom thecanonical situation. Russian alone would be sufficient to show numerous diver-gences. Te point o the canonical scheme is to provide a clear measure againstwhich we can view such complex systems.

Tis schema makes inflectional classes non-canonical, but still allows orcumulation. Cumulation arises rom intersecting eatures and it is appropriatelycriterion 2, relating to eatures, which makes cumulation non-canonical in com-parison to agglutination (criterion 2 states that canonical eatures and their val-ues are uniquely distinguishable across other logically compatible eatures andtheir values).

Russian: i we apply this principle to Russian, we see that the expression o case isthrough non-canonical morphology in many instances. However, this is only a re-flection o the usional morphology o Russian. Case values are no less canonicallyrealized than the other eature values with which case interacts; case values are ianything more canonical in this respect.

o conclude this section we may say that the six main case values justiy thetraditional perception o them as ull cases. However, they are not equally canoni-cal. Te canonical view gives a new view on what or some will be very amiliardata, and this view highlights the disparities between the six main case values.


27/305

Greville G. Corbett

. Te more problematic case values in Russian

We now turn to the more problematic case values. While there are discussionsin the literature o individual difficult values, or instance, the instrumental inLatvian or the locative in Latin, it is truly remarkable that one language shouldhave a system with so many problematic case values. We present the essential dataor each case value in turn, and only then go on to the issue o canonicity (5).

. Vocative

Te vocative is o special interest, because it includes ossilized remnants o an

old vocative and the development o a new one. Tere are the old orms BoeOGod, Gospodi O Lord, which are distinct vocative orms, used or invocation.Ten the new orms are those like NataNatasha!, Pet Petya!, mam Mummy!and so on. Te new vocative is optional, the nominative always being an alterna-tive. Tese new vocative orms are reported rom the early 20th century, but theyappear to have become more requent in the second hal o the century (Com-rie, Stone & Polinsky 1996: 132, and reerences there). It is sometimes stated thatthe new vocative is available only or nouns in inflectional class II (see (1)), and

with the ending -abeing unstressed, and that it is ound only in the singular. How-ever, there are several acts which suggest a different account. First, rebjata, whichis hard to translate, but means something like lads, guys, is used only in theplural, having become split over time rom its original singular rebenokchild (seeChumakina, Hippisley & Corbett 2004 or an account o this development). Tishas the vocative rebjatlads!, though it is not rom inflectional class II, and it is plu-ral. Tis suggests a truncation, based on the nominative. Tere is also devatgirls!rom devata. A second remarkable point about the new vocative is that it defies aregular rule o Russian, that final consonants are not voiced. Tus gorodtown is

pronounced with final [t], the [d] being evident when there is an inflection so thatthe consonant is not final. Te new vocative can dey this constraint, in a gradientway (devoicing may be absent or partial); see Daniel and Spencer (orthcoming)or this point, citing Panov (1997: 108110), and Daniel (in press) or interestingdiscussion o vocatives in general. A third point, noted by Koval (2004: 5), ol-lowing Klobukov (1986: 21), is that or some speakers there is a vocative o wordsending in a -CCa cluster, where the vocative differs rom the bare stem. Tus someaccept the vocative Mik rom Mika(diminutive orm o Mixail); this vocative

differs rom the bare stem: when the bare stem is used as a genitive plural, a fleet-ing vowel is required givingMiek.All these acts point towards truncation o thenominative, as worked out in Yadroff (1996). (See also Floricic 2002: 160162 orurther reerences, and Meluk 2006: 503504.)


28/305


Te act that or some nouns the vocative has this otherwise impossible shape,without devoicing, means here there is particular autonomy o orm (as discussedby Koval 2004: 5).

. Second genitive

Te second genitive is a stark challenge or attempts to determine the number ocases o Russian (see Zaliznjak 1973; Worth 1984; and Comrie 1986).23Contrastthese orms o kiselkissel (a thickened ruit drink) and ajtea. Both are memberso inflectional class I (they vary somewhat rom the orms in (1), but in predictableways), and as expected both have the normal (first) genitive:

(4) vkus kiselj-a taste[sg.nom] kissel-sg.gen thetasteokissel

(5) vkus aj-a taste[sg.nom] tea-sg.gen thetasteo tea

We find a contrast, however, in certain partitive expressions:

(6) stakan kiselj-a glass[sg.nom] kissel-sg.gen aglassokissel

(7) stakan aj-u glass[sg.nom] tea-sg.gen2 aglasso tea

In contemporary Russian, in the active use o the speakers I have consulted, kiselkissel is an example o a regular noun, while ajtea is one o the subclass which

has a separate second genitive. Te number o nouns with this second genitive isrestricted and declining.24Tey are all members o the inflectional class I. O the

. his material on the second genitive is presented in Corbett (orthcoming), in thecontext o a discussion o higher order irregularity, where it is pointed out that theRussian second genitive shows the rather unexpected interaction o overdierentiationand syncretism.

. Ilola & Mustajoki (1989: 4141) reporting on Zaliznjak (1977), identiy 396. However, theorm has been in decline, continuing since the publication o Zalizjaks dictionary. Our examplehelps show this: kiselkissel is given by Zaliznjak as having a second genitive, but consultantsdo not offer this orm. Google gives over 200 examples o stakan kiseljaglass o kissel and noneo stakan kiselju; however, other web searches reveal occasional sporadic instances o kiseljuas asecond genitive. Tis noun has almost lost its second genitive, but not quite. For urther discus-


29/305

Greville G. Corbett

nouns which have a second genitive, or some the second genitive is normally usedin partitive expressions, or the others the second genitive is a possibility, but in com-petition with the ordinary genitive; or data on this see Panov (1968: 180), Graudina,Ickovi & Katlinskaja (1976: 121125), Comrie, Stone & Polinsky (1996: 124125),and especially Paus (1994). Tus the second genitive varies rom being normallyused, to being optional, to unusual according to the particular lexical item.

Te orm itsel is o interest too, as these partial paradigms show:

(8) Russianpartialsingularparadigms

nominative kisel aj

genitive kiselja ajagenitive 2 as genitive ajudative kiselju aju

Te extra orm o aj tea, the second genitive, is syncretic with the dative. Tereis no unique orm or it. We cannot consign the problem to syntax by claimingthat the orm used is the dative, since any modifiers marked or case are indeedgenitive. Tis is not obvious, since in the modern language at least or some speakers the inclusion o an agreeing modifier disavours the use o the second genitive;

instead the ordinary genitive is more likely:

(9) stakan krepk-ogo aj-a glass[sg.nom] strong-m.sg.gen tea(m)-sg.gen aglasso strongtea

Here the presence o the modifier krepk-ogo strong-m.sg.gen seems to makeit more likely, or some speakers at least, that the ordinary genitive ajawill beused.25However, in those instances where the noun stands in the less likely secondgenitive in an expression similar to (9) genitive agreement is still required. Tuskrepkogo aju strong tea is possible as a second genitive. Tus the distributionaltest shows that we have to distinguish the second genitive rom other orms. Just tobe sure, we check what happens i we put the attributive modifier in the dative:

sion o the second genitive see Wierzbicka (1983: 249252), Uspenskij (2004: 1126) and or arecent analysis o the second genitive in the Russian National Corpus see Brehmer (2006).

. A urther syntactic restriction is suggested by Irina Nikolaeva (personal communication),

namely that the head noun (stakanglass or similar) should be in a direct case. Speakers varyhere, but some have are unwilling, or less willing, to accept a second genitive when the head isin an oblique case. Initial corpus investigation supports this intuition.


30/305


(10) krepk-omu aj-u strong-m.sg.dat tea(m)-sg.dat strongtea

(10) can be used only in syntactic positions where a dative is required. It is not asecond genitive, and could not be used in (9). Te problem is thereore a morpho-logical one and not a syntactic issue: second genitives are not syntactic datives. osum up: the issues presented by the second genitive are that it has a unique distri-bution but no unique orm; it is available with a relatively small number o nouns,and with these its degree o optionality varies considerably; it may be subject to thesyntactic condition that modifiers intervening between the governor and the nounin the genitive disavour the second genitive.26

. Second locative

Russian has a locative case, ofen called the prepositional case because it occursonly together with a preposition (v in, na on, o about, concerning, or pri by,at, attached to, in the time o, and rarely with po, as noted in 3, discussion ocriterion 8). Examples would be v urnale in the magazine andv komnatein theroom. For many nouns (including those o inflectional class I, like urnal maga-zine) the locative has a distinct orm in the singular, and or all inflecting nounsthe locative is distinct in the plural.

For our purposes the second locative is o greater interest. It is restricted inseveral ways. It occurs only with v in and naon, primarily when used in theirbasic locative or temporal sense, and not with the other two prepositions whichtake the locative.27It is distinguished only in the singular, and or relatively smallgroups o nouns. However, these are o different inflectional classes, and within thefirst inflectional class there are nouns belonging to different subclasses which havethe second locative.28Second locatives are always stressed on the inflection: e.g., v

sadin the garden; similarly or nouns o inflectional class III. Tis is importantwhen considering whether there are unique orms. For inflectional class I, which

. While ollowing Jakobson (1958) in accepting a second genitive and a second locative, wedo not also have to decompose the cases, and to accept Jakobsons amous representation in theorm o a cube. We may agree with Worth (1984: 298) that the cube was an enticing mistake.

. See Torndahl (1974), Brown (2007) and reerences there, or the development o the

second locative, and see Plungian (2002) or recent discussion o its semantics.. Tus sad garden and les orestorm the nominative plural differently (sadybut lesa), butboth have the second locative (sadand les).


31/305

Greville G. Corbett

has most o the nouns with a second locative, the -uinflection is shared with thedative, but the orms are normally distinct, the second locative being stressed onthe inflection and the dative singular typically being stressed on the stem (or thosenouns with a second locative). In inflectional class III there are also nouns with thesecond locative, or example v krovin blood (covered in blood). Te inflection isthe same as or the dative singular and or the normal locative singular; these twohave stem stress, while the second locative is stressed on the inflection.

It is limited too in the number o nouns which have it:

(11) Nounswiththesecondlocative(Ilola&Mustajoki (1989: 42 43)29romZaliznjak (1977)30

inflectional class examplenouns with secondlocative available

o these, secondlocative optional

I na beregon the bank 128 33III v stepin the steppe 31 8

For some nouns (in locative contexts) the second locative is obligatory: it has tobe v sadin the garden (and not *v sade). For some nouns it is only optional. Tenumber o nouns involved is declining.

Te restrictions continue in that the second locative, like the second genitive,has special orms only or nouns. When case is marked on other word classes,notably adjectives, the orms are as or the normal locative:

(12) my by-l-i v star-om sad-u 1pl be-pst-pl in old-m.sg.loc garden-loc.2 wewereinanoldgarden

(13) my razgovariva-l-i o star-om sad-e 1pl converse-pst-pl about old-m.sg.loc garden-loc

weweretalkingaboutanoldgarden

Te adjective is identical, or the second locative in (12) and the normal locative in(13). Unlike the second genitive, there is no difficulty with including adjectives in

. Brown (2007) reports similar but slightly lower figures, and adds interesting inormation onrequency. Te orm is in decline and so published figures tend to be overestimates o its use.

. Zaliznjak also includes v zabyt in oblivion and v poluzabyt in semiconsciousness(Zaliznjak 1967/2002: 287), and does not mark these as optional. However, Plungian (2002)states that the distinction is largely lost or these two nouns, with one or other orm (in -eorin -) being generalized or all uses. Tis view is adopted in the 2003 revision o Zaliznjaksdictionary.


32/305


such phrases. And as Brown (2007) demonstrates, such modifiers are not restrictedto set phrases. Speakers are ully comortable with new expressions.

How then should we treat the second locative? Brown (2007) draws the anal-ogy with sub-genders, specifically the animate/inanimate sub-genders o Russian.Tese are distinct or a small proportion o the available paradigm cells. Te sameis certainly true o the second locative (and is equally true o the second genitive).Tis particular sort o non-canonicity has a satisying account. Browns solution isto treat the second locative as a structured case (1998: 198200). In his NetworkMorphology analysis, which relies heavily on the notion o deaults, the deaultrealization o the second locative will be as the normal locative (this deault isoverridden or the relative ew nouns with a distinct second locative). Inflectional

morphology in this model is specified according to ordered paths, where morespecific inormation is ordered afer more general. For instance the specificationo the ordinary locative singular may be given as:31

(14) singular locative

Tis ordering is justified by the act that some nouns have different stems orsingular and plural (and these are specified within the lexical entry), while nonehave a special stem or locative. Ten the specification o the second locative is an

extension o this path:(15) singular locative locative2

Te effect is that when the second locative is required by the preposition, i thenoun has a orm matching the specification (15) completely, this is the appropri-ate orm. I not, by deault, the closest matching path specifies the orm, and thatpath is as in (14). Tis latter will always occur with adjectives, which Brown coverselegantly with no extra machinery. Te analysis in effect says that all nouns have asecond locative, but very ew nouns reallydo.

wo important points need to be made here. First, Browns solution orthis particular type o non-canonicity is not just an elegant idea. It is worked outas a ull implementation within Network Morphology (one o the inerential-realizational approaches). Browns implementation gives the right outcomes andothers can test it to ensure that it is indeed a valid analysis (the ragment is providedin Brown 1998). And second, it develops an idea ound in Comrie (1991: 102), that

. Brown (1998, 2007) uses prepositional and locative rather than locative and locative2;the labels are not important here, and I have retained locative and locative2 or consistency inthe text.


33/305

Greville G. Corbett

o having a hierarchical eature analysis or the second cases,32one which capturesthe specific nature o these case values and does not involve invoking other artifi-cial eatures or other case values.

. Te adnumerative

Te loss o the dual has lef strange remnants scattered across Slavonic. Few arestranger than the adnumerative. Te normal situation in Russian is that thenumerals dvatwo, tri three, etyreour (as well as oba both,poltoraone and ahal,polhal in compounds, or the latter see Billings 1998), provided they arein the nominative case (or the accusative and are quantiying an inanimate) take

a noun in the genitive singular. For example, dvaurnala two magazines, whereurnal is in the genitive singular as in cenaurnala the price o the magazine.An extremely small number o nouns have a special orm in these circumstances.Tey are in inflectional class I, and the orm is distinguished by stress only. Tusdvaastwo hours, two oclock shows the special adnumerative, in contrast withokolosaabout an hour with the normal genitive singular.

Te restrictions, in addition to those alluded to already, are severe. Te nounswhich have the orm are: ashour, agstep, pace, arball, sphere, rjadrow, sledootprint. (Zaliznjak (1977) gives the orm as being optional with arball, sphere

and sled ootprint, while Meluk (1985: 431432) implies that it is optional ormore o them.) Te use o the orm is best maintained with as hour. Meluk(1985: 430437) offers interesting discussion, and suggests that with the nouns orwhich the adnumerative is not obligatory, an intervening modifier requires the useo the ordinary genitive (1985: 432):

(16) *dva irok-ix ag-(gaisrequired,accordingtoMeluk) two.nomwide-pl.gen step-adnumerative twowidesteps

Te eature specification o the modifier (genitive plural) is perhaps surprising;however, this orm can be used both with nouns with the adnumerative where amodifier occurs (Meluk (1985: 433) and or those nouns without an adnumera-tive, which would stand in the genitive singular. For details o the choice betweengenitive plural and nominative plural or the modifier in comparable phrases seeCorbett (1993).

Returning to the adnumerative, we should still ask i it is a case value. It appearsto be a rather odd combination o case and number. As Zaliznjak (1967/2002: 47)

. Brown discusses the locative and Comrie the genitive, but in both instances the accountgeneralizes to both second cases.


34/305


says, it is special as a case in being available only in the singular.33In its avour, ithas a unique orm. Furthermore, Russian numerals typically govern a case, whilethe number o the noun is limited (generally it must be plural but with lowernumerals singular), and so the adnumerative fits into this pattern. Te adnumera-tive seems to be on the extreme edge o what could be included as a case value.

. Te inclusive

Tis case was suggested by Zaliznjak (1967/2002: 50), as one possible analysis toaccount or expressions o the type idti v letiki to become a pilot. Tese expres-sions are particularly challenging or synchronic analysis. Tere is helpul initial

discussion in Zaliznjak (1967/2002: 5052), and careul consideration o variousalternatives in Meluk (1978/1985: 461489). Afer reviewing the evidence weshall conclude, with Meluk that there is no need to postulate an additional casehere, but that the considerable problems lie elsewhere.

Te essential problem with expressions like idti v letiki to become a pilot isthe orm o the noun letiki. Tis has the apparent orm o the nominative plural.However, the preposition v in(to) takes the accusative or the locative. Te accusa-tive plural o animate nouns like letik pilotis as the genitive, that is, letikov. Tisis a ully general syncretic pattern in Russian: all animate nouns in the plural have

accusative as genitive, while all inanimates in the plural have accusative as nomi-native. Tus the nominal orm in the expression above cannot be accounted or bythe normal rules, and Zaliznjak proposes an additional case as a result.

O course, it is tempting to say that we are dealing merely with a set phrase.But that simply will not do. Tere are some one hundred verbs with the appropri-ate semantics to take the place o idtigo in the first syntactic slot, and even somenouns, like kandidatcandidate (Meluk 1985: 461). Moreover, any animate nouncan in principle take the place o letikipilots in the third slot (though it is usu-

ally those denoting proessions or social groupings). What is constant is that theseexpressions always involve becoming something. Te two ormal requirements arethe preposition v, which normally means in(to) and that the noun ollowing thepreposition must be in the plural. Interestingly, this requirement is maintained

. Meluk (1985: 430431) argues that orms used with particular nouns (including supple-tion in two instances) in the plural, when quantified by higher numerals (those above our asin the examples discussed above), should also be considered instances o the adnumerative. Forinstance, dvenadcat elovektwelve people not *dvenadcat ljudejtwelve people. I do not findthe argument convincing, but nevertheless urge the reader to check Meluks reasoning inde-pendently. Te issue is difficult.


35/305

Greville G. Corbett

even when semantically it makes no apparent sense, as in this example (Meluk1985: 465):

(17) Derjagin pro-lez v sekretar-i Derjagin[nom] through-climb[pst.sg.m] into secretary-pl na-ego partbjuro our-n.sg.gen party.office34

Derjaginwormedhisway intobeingsecretary oourparty office.

Te point is that there is only one secretary, but still the plural is required here.Meluk (1985: 462) rejects Zaliznjaks additional case here, on the grounds

that it is being postulated on the basis o a single rather specific construction. Tat

is reasonable, provided an alternative analysis can be given. Tere are two options,either nouns like letiki pilots in such expressions are in the nominative or theyare in the accusative.

Te nominative option has the advantage o simplicity, in that the orm inthe expression matches the normal nominative orm. However, Meluk producesthree specific arguments against this option. o take just one, he points out thatidti v soldatyto become a soldier has a near synonym inidti v armijuto join thearmy and here armijuis indisputably in the accusative. Meluk also gives what hecalls a metapravilo metarule, namely that in Russian prepositions never govern

the nominative. (Tere are other potential instances o prepositions governing thenominative in Russian, Zaliznjak 1967/2002: 51n23, but Meluk 1985: 438452has alternative analyses or these too.) Apart rom the problematic instances inquestion, Russian prepositions govern all cases but the nominative. Meluk is keento maintain that generalization, though another linguist might look precisely ora preposition governing the nominative to complete the set. I agree with Melukthat this construction is not sufficient to override the generalization.

Tus Meluk takes the second option, and claims that letiki and similar

nouns are in the accusative case. Tis seems an odd claim at first, since the nor-mal accusative is letikov (as the genitive). Meluk argues that in this constructionthe preposition vrequires the governed noun phrase to be in the plural and to beinanimate. Since the inanimate accusative is as the nominative, this would give theright orm. Tis solution has the great advantage o pinning the unusual behaviourprecisely on the preposition vin this construction (and avoids not only postulatingan extra case, but also doubling up o lexical entries). It is a remarkable require-ment o a preposition, however.

. Tis noun is indeclinable. Note that not all Russian speakers share Meluks intuition con-cerning this example; some find it strange i naegoour is included, which highlights the actthat there is just one post.


36/305


Tere is a urther very interesting complication. In this construction the nounphrases are not ully inanimate. In attributive position, we find inanimate agree-ment orms (Meluk 1985: 466):

(18) popa-l v star-ie inener-y. (not*star-ix) get.pst[sg.m] into chie-pl engineer-pl *chie -pl.acc=gen (he)madeittochie engineer

Here the orm starie chiecan be treated as being in the accusative, identicalto the nominative, which is what we find with inanimates. Te orm *starix,with accusative identical to genitive, as or animates, is unacceptable. In relativepronoun position, however, we find animate agreement (Igor Meluk personal

communication):35

(19) pro-lez v sekretari partbjuro, through-climb[pst.sg.m] into secretary-pl party.office kakov-yx u nas preziraj-ut which-pl with 1pl.gen despise-3pl wormedhisway intobeingsecretary o theparty office,which(sort) we despise

Here the orm kakovyxis accusative as genitive, marking animacy; the accusative-

nominative (inanimate) *kakovyeis not accepted. Meluk gives other examples osuch splits in animacy; the restriction to the accusative case together with the actthat personal pronouns always show accusative-genitive syncretism means thatonly two distinct agreement positions are available or us to test (attributive andrelative pronoun). In this limited sense, all Meluks examples are in accord withthe Agreement Hierarchy (Corbett 2006: 206237) in that we find syntactically

justified orms in attributive position, contrasted with semantically justified ormso the relative pronoun.

As Meluk points out, the basic idea can be ound in Rothstein, who writes:

Te mechanism o the metonymy here involves changing o a single lexical eaturespecification rom [+animate] (or more likely rom [+human], which implies[+animate]) to [animate].[ootnote omitted] (1977: 98).

Meluk traces the idea back to Vinogradov (1947: 165), but the care-ul argumentation is all Meluks. Tere is later discussion in Bl (1988) andUspenskij (2004: 2738).

. Te example is modified rom that in Meluk (1985: 466) to avoid the complication oanimacy in the original example. Meluks judgement on the example remains the same.


37/305

Greville G. Corbett

o conclude this particular part o the discussion, we may say that these expres-sions are challenging and ascinating, but that ollowing Meluk (and disagreeingwith Zaliznjak) there is no need to postulate a separate case to account or them.

. A canonical view of the problematic case values of Russian

Since we have eliminated the potential inclusive case, this leaves our problematiccase values to consider in terms o canonicity. Tese are the vocative (presented in4.1 above), the second genitive (4.2), the second locative (4.3) and the adnu-merative (4.4). We thereore return to the criteria or canonicity, and assess these

our case values.

Principle I: Features and their values are clearly distinguished by formal means.

Te case values we are considering show largely non-canonical behaviourwith respect to the criteria which all under Principle I.

Criterion 1: Canonical eatures and their values have a unique orm.

Te value which is non-canonical here is the second genitive, whose orm is

always the same as that o the dative.Criterion 2: Canonical eatures and their values are uniquely distinguishable

across other eatures and their values.

Here all the values under discussion are non-canonical: each o them isrestricted to one number (almost exclusively the singular).36

Criterion 3: Canonical eatures and their values are distinguished consistentlyacross relevant word classes.

Again all our values are non-canonical here, since they are restricted to nouns,while case is a eature also o adjectives and pronouns in Russian.

Criterion 4: Canonical eatures and their values are distinguished consistentlyacross lexemes within relevant word classes.

.

wo provisos are in order here. First, recall that Meluk (1985: 430) claims that the adnu-merative is ound in the plural too. For this criterion, however, the effect would not change, sinceany given noun has an adnumerative only or the singular or or the plural. Second there are thevocatives rebjatlads! and devatgirls!, ormed rom original plurals; again the effect does notchange since there is no corresponding singular vocative or these items.


38/305


Here all our values are non-canonical.37We noted that each is dramaticallyrestricted in the nouns which can mark it. Tis is particularly true o the adnu-merative, available or only a handul o nouns.

Principle II: Te use of canonical morphosyntactic features and their values isdetermined by simple syntactic rules.

Once again, in different ways, the our values under discussion show non-canonicity or the criteria which all under this principle.

Criterion 5: Te use o canonical morphosyntactic eatures and their values isobligatory.

Te new vocative is optional (the nominative is available or this unction).Te remaining three values are optional or at least some o the nouns whichhave them.

Criterion 6: Canonical use o morphosyntactic eatures and their values does notadmit syntactic conditions.

As we noted in 4.2, the second genitive is disavoured i there is a modi-fier with the noun. We saw a similar issue with the adnumerative in 4.4. Tis is

another non-canonical aspect o these two case values.Criterion 7: Canonical use o morphosyntactic eatures and their values does not

admit semantic conditions.

Here the second locative is non-canonical. We have already noted that its useis restricted to two particular prepositions. Furthermore, they must be used in alocational sense. Tus the second locative na moston the bridge is normal, in thelocational sense.38Plungian (2002) argues urther that the second locative is usedparticularly (though not exclusively) or instances o strict localization. Tis is

just one reflection o treating syntactic uses o case values as canonical, as opposedto semantic uses.

. Isaenko (1962: 82) believes that neither the second genitive nor the second locative shouldbe recognized as cases, since too ew nouns have the orms; perhaps treating these case valuesrather as non-canonical is preerable to a black/white decision.

. With non-locational expressions the picture is complex: or some speakers, the normallocative is available or expressions likeja nastaivaju na mosteI insist on a bridge (as opposedto a ord or a tunnel, or instance), while others find this awkward or even unacceptable. Teimportant point is that the choice is na mosteor some speakers, or avoiding the orm or others;no speaker offered na mostuon a bridge using the second locative here.


39/305

Greville G. Corbett

Criterion 8: Canonical use o morphosyntactic eatures and their values does notadmit lexical conditions rom the target.

Tere is no problem here. Tere are issues as to which nouns have the casevalues in question, but we have not noted any urther lexical conditions. Te typeo condition discussed under criterion 8 in 3 involving different case values deter-mined by the particular governee is not ound with the problematic case values.

Criterion 9: Canonical use o morphosyntactic eatures and their values does notadmit additional conditions rom the controller (governor).

Te idea is that the controller should have a single, simple requirement (e.g.,

dative). Te adnumerative is non-canonical here, since the ew controllers thatgovern it have an additional condition: the controller must stand in the nomina-tive (or the accusative identical to the nominative as occurs when the governednoun is inanimate). Tis is seen in the ollowing contrast:

(20) dva as- two.nom hour-adnumerative twooclock

(21) k dvum as-am by two.dat hour-pl.dat by twooclock

In this respect the adnumerative is again non-canonical.39

Criterion 10: Te use o canonical morphosyntactic eatures and their values issufficient (they are independent).

Just as the first locative, the second locative is non-canonical in this respect.So too is the adnumerative, since it is available only together with one o the smalllist o quantifiers which governs it. Tus the adnumerative ashour cannot beused on its own (without a quantifier) to mean two-our hours. (For this reason,alternative terms like paucal are better avoided.)

Principle III: Canonical morphosyntactic features and their values are expressed bycanonical inflectional morphology.

. Te syntax o Russian quantified expressions is well known or its complexity. Othernumeral expressions also have matching in case values in the obliques; however, the case valuesinvolved all have numerous other uses outside quantified expression. Te adnumerative isnon-canonical in that it is available only according to the constraint imposed by quantifiedexpressions.


40/305


As we noted earlier, there are several deviations rom canonicity, but mosto these are not significantly different rom those ound with the main cases. Teone that stands out is the new vocative, ormed by truncation. Tis is highly non-canonical since this device is not used to mark case elsewhere in the paradigmsunder discussion.

(22) Summary o non-canonical characteristics o the problematic case values

criterion (and brief

description) vocative second genitive second locative adnumerative

1. dedicated orm

(autonomous)

X

2. distinguishable acrossother eatures and values

X X X X

3. distinguished consistentlyacross relevant word classes

X X X X

4. distinguished consistentlyacross lexemes

X X X X

5. obligatory X (X) (X) (X)6. no syntactic conditions X X7. no semantic conditions X

8. no lexical conditions romthe governee

9. no additional lexicalconditions rom thegovernor

X

10. use is sufficient (eatures/values are independent)

X X

In (22), X indicates non-canonical behaviour, and (X) somewhat non-canonicalbehaviour. From (22) it is clear that the problematic case values are all non-canonicalto a high degree. It must be recalled that the six main case values were not ullycanonical either, yet the difference between the main six and these our is strik-ing. O the our problematic case values, the least difficult is the vocative. Tis isthe case value which is on the rise, expanding its range, while all the others are indecline. Moreover, the vocative is typically a ringe case value, not well integratedinto the case system, and so the non-canonical properties o the Russia

Case.&.Grammatical.relations,2008

Documents

Transcript of Case.&.Grammatical.relations,2008