Rivista di Grammatica Generativalear.unive.it/jspui/bitstream/11707/420/1/2008compl.pdf · Rivista...

Rivista di Grammatica Generativa

33 – 2008

I

SELECTED PROCEEDINGS OF THE XXXIV

INCONTRO DI GRAMMATICA GENERATIVA

INTRODUCTION

Paola Benincà, Federico Damonte, Nicoletta Penello

In this special issue of the Rivista di Grammatica Generativa are collected some of the papers presented at the XXXIV Incontro di Grammatica Generativa, held at the Department of Linguistics, Communication and Performing Arts of the University of Padua, on 21 - 23 February 2008. The annual Incontro started back in 1975 as an informal gathering of linguists working within Generative Grammar (as reflected in the original name Incontro Informale di Grammatica Generativa); over the years it has become one of the most important conferences of formal linguistics in Europe, with a low rate of accepted abstracts, a high rate of speakers coming from foreign institutions, and several invited speakers who are established authorities in their own fields of research. The conference has also grown bigger and more articulated, with poster sessions and parallel sessions now being common. To publish the proceedings of such a conference is a considerable editorial effort, which inevitably requires some kind of selection.

In our case, editorial constraints implied that we could either publish most of the papers presented at the conference with a very low limit on the number of pages, or allow the authors more space but then make a severe selection on the papers that would form the proceedings. We followed the last route, and tried to turn this limit into an opportunity to publish a more coherent set of contributions. Since a selection based on the quality of the paper would have been rather difficult and arbitrary, given the average high quality of the papers presented at the Conference, we took the decision to choose the contributions by topic, and consider only syntactic works. For the same reason we also decided to have the papers reviewed, and to ask the reviewers to provide the authors with comments about specific points that could be improved, rather than with a general evaluation of the paper. The fact that our own expertise lies mainly in syntax also contributed to the decision to only consider articles on syntax. Our overall goal was to produce a volume similar to a selected and edited set of papers on a single, albeit general, topic. We are pleased to report

II

that our original reviewing process, while inevitably delaying publication, was appreciated by some of the authors, who remarked on how useful the comments had been in improving their papers. With the benefit of hindsight, we can also say that a higher number of pages was indispensable to allow the authors to make their cases convincingly.

As for the topics covered by the authors, despite the inevitable variety of subjects and frameworks that can be found in any volume of proceedings, some interesting patterns can be observed. The Cartographic Programme (on which see the introduction of Cinque 2002 and Rizzi 2004) has proved its usefulness in providing a very clear set of theoretical tools that can be used at different levels of generality: in our collection, this programme is represented more directly by the wide-ranging contributions by Adriana Belletti and Guglielmo Cinque. Belletti’s contribution illustrates the role that the “low periphery” (or VP periphery, the set of discourse-linked functional projections located in the lower part of IP) plays in apparently unrelated syntactic constructions, in this case cleft sentences. Cinque’s paper deals with internally and externally headed relative clauses further refining his proposal (Cinque 2003/8) of a unique underlying structure for all types of relative clauses, in which there are two nouns, one inside and one outside the relative clause. It is therefore crucial that both nouns match in their indefinite character. The paper by Federica Cognola builds on Belletti’s low periphery hypothesis and proposes that phenomena belonging to this area are involved in the alternation between OV and VO syntax in the Mòcheno dialect, a German variety spoken in northern Italy.

Also noteworthy is the fact that all the other contributions in this volume that assume the cartographic approach revolve around the same general problem, namely how does the internal structure of a complex modifier reflect on its external syntactic properties. This is the theoretical problem underlying Silvio Cruschina and Eva-Maria Remberger’s paper on some evidentiality markers in Romance, which derive from grammaticalised clausal structure that have been re-analysed as adverbial modifiers, spelling out evidentiality and occupying a precise position, the authors claim, in a specifier within the IP layer. Likewise, Jacopo Garzonio’s contribution deals with a case illustrating an “incomplete” Jespersen’s cycle of negation in Florentine, namely the grammaticalisation of a negative adverb as a postverbal element reinforcing the preverbal negation. Again, the analysis involves the relationship between the internal structure of negative modifiers and the syntactic feature they spell out in the functional structure. Finally, Andrea Cattaneo’s paper on the expression of deontic modality through periphrastic constructions in some Romance dialects tackles the same general problem, in that it is concerned with the derivation of the cross-linguistically different meanings which can be associated with a single modal verb from the same underlying structure containing a pseudo-relative clause.

III

Regarding the remaining papers, we observe that, while they take many notions and formal tools from the Minimalist Programme, they share with the Cartographic programme the assumptions that the functional structure of language is crucially based on features carrying specific meanings, and that the inventory of features is the same in all languages and can be discovered only through a comparative approach. This seems to go against the stress, within the Minimalist programme, on uninterpretable features as the main trigger for syntactic derivations, and on the differences in the distribution of these features across categories as the main reason for language variation. In short, while these papers do not deal with cartographic fixed hierarchies of functional projections, they share the basic assumption that specific meanings are represented as interpretable features that project maximal projections.

The idea that syntactic derivations are driven by the same interpretable features in different languages can be clearly seen in the explicitly comparative papers by Lena Baunaz and Genoveva Puskás on wh-movement in French and Hungarian, and by Mélanie Jouitteau and Milan Rezac on possessive predication in different Breton dialects. In the former work, the authors propose that covert wh-movement has the property of “feature stripping”, i.e. the wh-item can get rid of a feature in order to move to a higher position. The authors then apply this hypothesis to wh in-situ in French, and to partial wh-movement in Hungarian. Jouitteau an Rezac consider the different ways attested in Breton dialects to express possessive predication to be on a “path” from a periphrastic “mihi est” construction to a fully lexicalised transitive verb “have”. The paper assumes that these different forms correspond to different underlying configurations, which are all connected by simple changes and reanalyses. The same basic assumption is also visible, we think, in those contributions which deal with a single language, namely Lena Baunaz’ paper on floating quantifiers and N-words in French, Daniela Isac and Alison Kirk’s work on the split DP hypothesis in Ancient Greek, and Michèle Vincent and Andrew Radford’s article on French transitive participle agreement. Baunaz’ paper crucially assumes that universal quantifiers have a universally fixed set of properties and tests N-words such as personne in French for these properties, to conclude that they are syntactically and semantically similar to universal quantifiers. Daniela Isac and Alison Kirk’s paper on discontinuous DPs in Ancient Greek follows standard minimalist assumptions in proposing that syntactic projections, in this case the DP, can iterate, and uses phase theory to motivate DP internal movement of phrases; interestingly, though, their analysis shares the cartographic assumption of a layered functional structure within DP, with each layer spelling out different semantic features. They also assume that these DP projections have their own CP layer, in a parallel way to what has been proposed for the VP periphery by Belletti. Finally, Vincent and Radford’s paper, while concerned with a central mechanism of the latest version of the minimalist programme, namely percolation of features from a

IV

phase head down to the selected head, crucially use comparative evidence from complementiser agreement phenomena to sustain their view that percolation of uninterpretable features does not occur in participle agreement in French, and that the uninterpretable features in questions remain on the head throughout the course of the derivation.

All these papers, it seems to us, share a basic assumption, namely that the syntactic structure of language, even if not described in cartographic terms, must be the same for all languages, and variation must be derived from properties of the structure itself. The idea that basically all languages are alike has not been universally accepted within Generative Grammar, and in certain respects even within the cartographic framework. It seems to us very significant that all the papers we are presenting in some way share this assumption, which appears to be taken seriously by different authors working on different topics within different frameworks.

References

Belletti, A. (2004) Aspects of the low IP area, in Rizzi (ed.) Cinque, G. (1999) Adverbs and Functional Heads. Oxford and New York: Oxford University

Press. Cinque, G. (2002), The Structure of DP and IP: The Cartography of Syntactic Structures, Vol.

1, Oxford and New York: Oxford University Press. Cinque ([2003] 2008) “The prenominal origin of relative clauses”, Paper presented at the

NYU Workshop on Antisymmetry and Remnant Movement, Oct. 31- Nov.1 2003; at the LSA Summer School at Harvard and MIT, 2005; at the Colloque de Syntaxe et

Sémantique à Paris , 4-6 October 2007, and at the XXXIV Incontro di Grammatica

Generativa at the University of Padua, February 23 2008. Rizzi, L. (2004) ed., The Structure of CP and IP: The Cartography of Syntactic Structures,

Vol. 2, Oxford and New York: Oxford University Press.


33 – 2008

3

MORE ON THE INDEFINITE CHARACTER OF THE

HEAD OF RESTRICTIVE RELATIVES

Guglielmo Cinque

Introduction

The literature on relative clauses makes occasional reference to the fact that the Head internal to a (restrictive) relative clause is indefinite.

For example, Browning (1987, 129-131) observes (also see Bianchi 1999,43) that the trace within a restrictive relative clause is interpreted as indefinite, appearing in contexts that exhibit an indefiniteness restriction: The men that there

were in the garden vs. *There were the men in the garden (cf. There were (some/

many/three) men in the garden). Similarly, Kayne (1994, chapter 9, 124), suggests that an indefinite determiner

should not necessarily be taken to occupy the same position as the definite determiner that takes scope over the Head and the relative clause (and that marks the uniqueness or maximality of the intersection of the set of things denoted by the Head and the set of things denoted by the relative clause). Rather it could be taken to head “some “smaller” category, perhaps a QP” (Kayne 1994,p.167, fn.15; also see Kayne 2006,§7).1

In what follows I will discuss three sets of facts that seem to provide further support for the conclusion that the Head internal to a (restrictive) relative clause is indeed inherently indefinite.

A. The first comes from the existence of languages that bear the indefinite character of the Head of restrictive relative clauses on their sleeves, so to speak. Kusaiean (Austronesian) has headed postnominal relative clauses, which (as in most other Austronesian languages) precede demonstratives: N… RC Dem. An interesting feature of Kusaiean is the regular co-occurrence of a lower indefinite

1 For the idea that the determiner takes scope over the Head and the relative clause see Stockwell, Schachter and Partee (1973), which also contains one of the earliest proposals that the Head internal to the relative clause should be taken to be (specific-)indefinite.

Guglielmo Cinque

4

article with the demonstrative. The indefinite article follows the Head and precedes the relative clause and the demonstrative. See:

(1) [mwet se [elthal uniyah] ah] pa Sohn (Kusaiean - Sohn 1973,114f) [person a [they killed] Dem] TOP John ‘The person whom they killed was John’

This may be taken to suggest that the Head of the relative clause is an “indefinite DP” embedded in the larger (definite) DP.2

B. The second piece of evidence for the indefinite character of the Head of restrictive relatives comes from a number of constructions in Italian (and other languages) which contain a DP that can only be indefinite except when it heads a restrictive relative clause; in which case it is allowed to be definite. See for example (2) to (6):3

(2) a. Ho una/*la/*0 fame terribile4 I.have a/the/0 hunger terrible ‘I am terribly hungry’ b. La fame terribile che ho.. The hunger terrible that I.have..

(3) a. Ha un/*il/*0 bel viso She.has a/the/0 beautiful face

2 The word order in (1) arguably derives from a Merge order [Dem [RC [indef.det. [NP]]]], via successive roll-up movements (cf. Cinque 2005, in preparation). Mooré (Gur) also allows the co-occurrence of an indefinite determiner (closer to the N) with the definite determiner. See (i), from Tellier (1989, 308): (i) m karma [sebr ninga wa] saame 1sg read [book a (certain) the] yesterday ‘I read the certain book (that we already talked about) yesterday’ All of this should not be taken to mean that indefinite determiners are necessarily lower than definite ones. Lakhota, to be discussed below, offers evidence for the presence of a lower indefinite determiner and a higher one, which possibly occupies the same position as definite determiners (see (14)a). For comparative evidence that one should assume more than one position for (definite) determiners within DP, see Kayne (2004). 3 These cases are reminiscent of the contrast between (i)a and b, due to Noam Chomsky, mentioned in Browning (1987,129): (i) a. *John had the question for the teacher b. The question that John had for the teacher 4 On the determinerless nature of the non modified counterpart of (2)a (Ho fame, J’ai faim ‘I am hungry (lit. I have hunger)’), see Kayne (2005,41f), and Săvescu (2008a,b). As apparent from (2)a, when modified, fame ‘hunger’ obligatorily takes an indefinite determiner.

More on the indefinite character of the head of restrictive relatives

5

b. Il bel viso che ha.. The beautiful face that she has..

(4) a. Hanno preso una/*la/*0 posizione diversa They.have taken a/the/0 position different ‘They have taken a different position’ b. La posizione diversa che hanno preso.. the position different that they.have taken.. ‘The different position that they have taken..’

(5) a. Ha preso un/*il/*0 granchio (in the idiomatic reading ‘(S)he made a mistake’) (S)he.has caught a/the/0 crab b. Il granchio che ha preso.. the crab that (S)he.has caught.. ‘The mistake that (s)he made..’

(6) a Pensava di essere un/*il/*0 genio incompreso He.thought he was a/the/0 genius undiscovered ‘He thought he was an undiscovered genius’ b. Non era il genio incompreso che pensava di essere not he.was the genius undiscovered that he thought to be ‘He wasn’t the undiscovered genius that he thought he was’

It is tempting to take all of the b. cases of (2) to (6) as evidence for the presence in front of the Head of a (lower) unpronounced indefinite determiner in Italian, as shown in (7):5

(7) a. la [UNA fame terribile] che ho.. the a hunger terrible that I.have.. b. Il [UN bel viso] che ha.. the a beautiful face that she.has.. c. La [UNA posizione diversa] che hanno preso.. the a position different that they.took.. d. Il [UN granchio] che ho preso.. the a crab that I.have caught.. (the mistake that I made..) e. Il [UN genio incompreso] che pensava di essere.. the a genius undiscovered that he.thought to be.. ‘The undiscovered genius that he thought he was..’

5 We abstract away here from the question whether the Head is internal to the relative clause and raises in front of the relative clause, or is external, matched by an identical (indefinite) internal Head within the relative clause. See below for discussion.

Guglielmo Cinque

6

If we assume that, nothing special needs to be said concerning the exceptional determiner that occurs with the NPs in (2)b-(6)b.

Other interesting evidence for the presence of a null indefinite determiner within the Head of a restrictive relative comes from two special interpretative properties of indefinite DPs, not shared by their definite counterparts.

The first involves a specific interpretation of adjectives like sconosciuto ‘unknown’ observed in Abusch and Rooth (1997). They note that if the DP in which such adjectives occur is indefinite the adjectives, in addition to their meaning roughly paraphrasable as ‘little known, insignificant’, can also be interpreted in an ‘epistemic’ sense roughly paraphrasable as “that it is not known where it is”. See for example the ambiguity of (8)a, which contrasts with the non ambiguity of (8)b (if the latter sentence is at all good):6

(8) a. Vive in un villaggio sconosciuto del Sud della Francia 1. ‘he lives in a village of the South of France and it is not known which one it is’ 2. ‘he lives in some insignificant/little known village of the South of France’ b. Vive nel villaggio sconosciuto del sud della Francia 1. *‘he lives in the village of the South of France and it is not known where it is’ 2. ‘he lives in the insignificant/little known village of the South of France’

Now, consider the example (9), where the adjective sconosciuto appears in a definite DP containing a relative clause:

(9) Nel villaggio sconosciuto del sud della Francia in cui vive.. In the village unknown of the South of France in which he lives.. ‘In the unknown village of the South of France in which he lives..’

Here, differently from (8)b, the ‘epistemic’ interpretation is again available, which makes it plausible to assume the hidden presence of an indefinite determiner within the Head, as shown in (10):

(10) Nel [UN villaggio sconosciuto del sud della Francia] in cui vive.. In the [a village unknown of the South of France] in which he lives.. ‘In the unknown village of the South of France in which he lives..’

A second property typical of indefinite DPs in Italian (and, more generally, Romance) is the fact that a postnominal adjective is compatible with both a specific and a non-specific interpretation of the DP (while a prenominal one forces the specific reading). This was originally observed by Bosque (1993, 1996, 2001) for Spanish and is discussed by Picallo (1994), and Cinque (forthcoming) for Catalan 6 Better Vive in quel villaggio sconosciuto nel sud della Francia ‘He lives in that unknown village of the South of France’, which only has the non-epistemic sense. On the fact that the epistemic sense is lost in Italian if the adjective appears prenominally see the discussion in Cinque (forthcoming, chapter 2).


7

and Italian, respectively. Definite DPs cannot have the non-specific interpretation. Consider, for example, (11)

(11) a. So che un attore famoso interverrà alla festa I.know that an actor famous will.come to.the party ‘I know that a famous actor will come to the party’ b. So che l’attore famoso interverrà alla festa I.know that the actor famous will.come to.the party ‘I know that the famous actor will come to the party’

While (11)a is ambiguous between a reading in which the speaker has in mind a specific famous actor (the specific reading) and one in which he does not know the identity of the famous actor who will come to the party (the non-specific reading), (11)b cannot have the non-specific reading.

Again, it is interesting to observe that the non-specific reading (in addition to the specific one) becomes available in a definite DP if this contains a restrictive relative clause. See (12):

(12) L’attore famoso che interverrà alla festa sicuramente avrà lo smoking The actor famous that will.come to the party will surely wear a tuxedo ‘The famous actor that will come to the party will surely wear a tuxedo’

This interpretive effect can once more be understood if we take the Head of the relative clause to be indefinite:7

(13) L’ [UN attore famoso] che interverrà alla festa sicuramente avrà lo smoking The [an actor famous] that will.come to the party will surely wear a tuxedo ‘The famous actor that will come to the party will surely wear a tuxedo’

C. The third piece of evidence for the indefinite character of the Head of restrictive relatives comes from a reinterpretation, in the light of Cinque (2003/8, in preparation), of the indefinite restriction of Lakhota internally headed relative clauses discussed in Williamson (1987).

As Williamson shows, the Head of Lakhota (restrictive) relative clauses is internal to the relative clause and displays an indefiniteness restriction. Like the English existential there-construction, it can only contain ‘weak determiners’ (in the sense of Milsark 1974), i.e. indefinite articles like ‘a’, weak quantifiers like ‘some’, ‘many’, and ‘few’, and cardinal numerals. The presence of any ‘strong determiners’

7 If a silent indefinite determiner is present in definite DPs containing a restrictive relative, recourse to a special accommodation mechanism (as in Heim 1982) to account for the fact that such definite DPs can be bound by a quantifier binding a pronominal inside them (Every

man saw the dog that barked at him), while definite DPs ordinarily cannot (p. 245ff), may prove unnecessary. Interestingly, Heim explicitly says (p. 247) that the above sentence receives the same truth conditions as Every man saw a dog that barked at him.

Guglielmo Cinque

8

(like definite articles, demonstratives, quantifiers such as ‘all’, ‘every’, ‘most’, etc.) renders the sentence ungrammatical (see p.175f). See for example the contrast between (14)a and b:

(14) a. [[ Mary [owįža wą] kağe] ki/cha/k’ų] he ophewatų (Williamson 1987, 171) M. quilt a make the/a/the aforementioned Dem I-buy ‘I bought the/a quilt that Mary made’ b. *[[ Mary [owįža ki] kağe] ki] he ophewatų (Williamson 1987, 171) M. quilt the make the Dem I-buy

I would like to suggest that this indefinite restriction on the internal Head of internally headed relative clauses in Lakhota is to be expected under the unified analysis of relative clauses that I proposed in (2003/8) (see Cinque in preparation for more detailed discussion), and constitutes further evidence for the indefinite character of the Head of (restrictive) relative clauses in general.8

The core of Cinque’s (2003/8) proposal is that a single structure underlies all types of relative clauses (externally headed postnominal, externally headed prenominal, internally headed, headless, and correlative), in both the raising and the

matching derivations; a structure in which the relative clause is merged prenominally, with every difference among the distinct relative clause types due to different derivational options.9 The prenominal merger of relative clauses is arguably a consequence of a more general property of UG. In Cinque (2002, 2005, 2008a), on the basis of a general left-right asymmetry of natural languages, I suggested that all elements found to the right of a lexical head (N(P),V(P),etc.) are not merged there, but come to be there as a consequence of the lexical head raising above them, merged in a lefthand specifier position. If so, also relative clauses, a sort of “syntactic adjectives” in Benveniste’s (1966, 222) terms, are like adjectives merged prenominally in one of the NP’s functional projections. In Cinque (2003/8, in preparation), it is suggested on the basis of cross-linguistic evidence that the Merge position of (finite) restrictive relative clauses is above the Numeral, the Adjectives, and the NP (in fact above all of Milsark’s ‘weak determiners’, (one type of) indefinite articles included), and below Universal Quantifiers, Demonstratives and definite articles (more generally, all of Milsark’s ‘strong determiners’), roughly

8 Also Peterson (1974) took the obligatorily indefinite Head of the internally headed relative clauses of Mooré (Gur) to be evidence for the indefinite character of the Head of restrictive relatives, in an analysis that has some points of contact with the one I propose below. 9 The prenominal Merge of relative clauses renders a “matching” derivation possible in Kayne’s (1994) Antisymmetry theory. This may be a welcome result if both “raising” and “matching” derivations turn out to be necessary, as argued in Åfarli (1994), Sauerland (1998, 1999, 2003), Aoun and Li (2003), Szczegielniak (2005), Salzmann (2006), and Cinque (in preparation), among others.


9

as in (15), which I take to be a fragment of the universal structure of nominal phrases:10

(15) [DemP D° [RC X° [NumP Y° [AP… Z° [NP]]]]]

As more clearly apparent from (16), which is to be thought of as built bottom up (with Merge and Move interspersed), this unified structure has both an external Head, and a Head internal to the relative clause, which are exact matches of each other. Given that the external Head (the chunk of the extended projection of NP modified by the relative clause) is, as noted, ‘indefinite’, the Head internal to the relative clause must also be ‘indefinite’. This will be at the basis, as I suggest below, of the indefinite restriction holding of Lakhota’s internally headed relative clauses.11

10 I will ignore nonrestrictive relative clauses here. For discussion see Cinque (2008b), and fn25there for evidence that they are merged above demonstratives (and universal quantifiers). 11 For simplicity, I am taking that to be a complementizer inserted under a C head, but see Kayne (forthcoming) (and Sportiche 2008, Koopman and Sportiche forthcoming on que/qui in French) for arguments that they are (weak) relative pronouns/phrases, which would require merger into the specifier position of an additional C head.

Guglielmo Cinque

10

The “raising” and “matching” derivations can be seen as two different derivational options open to this structure; in the “raising” one, it is the Head internal to the relative clause that ends up being the overt Head; in the “matching” one, it is the external Head that ends up being the overt Head.

In other words, if only the Head internal to the relative clause raises (say to Spec, C2) causing the c-commanded external Head not to be pronounced, we have the “raising” derivation, in which reconstruction and island effects are detectable as the overt Head is in a chain with the relative clause internal position (see (17)):


11

If on the other hand the external Head raises (say, to Spec,C1) above the position to where the Head internal to the relative clause has raised (Spec,C2), causing the latter not to be pronounced, we have the “matching” derivation, in which reconstruction effects are not detectable as the surviving, overt, Head is not in a chain with the relative clause internal position (see (18)).12

12 Island effects are still detectable if, as we have assumed in the text, the Head internal to the relative clause raises, but here much variation exists. See below, and especially Cinque (in preparation) for illustration. Bulgarian offers interesting evidence that (if it moves) the Head internal to the relative clause indeed raises to a position lower than that to which the external Head raises. When the “raising” derivation is not forced, the overt Head (which is the external Head) can be separated from the complementizer/relative pronoun by topic or focus phrases, suggesting that it raises to a high position within the CP field. However, when the “raising” derivation is forced, the overt Head (which is the internal one) cannot be similarly separated from the complementizer/relative pronoun, suggesting that it raises to a position lower than that occupied by topic and focus phrases. See Krapova (2008) for discussion.

Guglielmo Cinque

12

Under a copy theory of movement (Chomsky 1995, 202ff), and deletion as non pronunciation in PF of full structures in the computation, a number of problems arise. For one, in sentences like (19) we would expect a principle C violation when the Head internal to the relative clause is “reconstructed” in its position of Merge, contrary to fact.

(19) [The pictures of Marsdeni [which pictures of Marsdeni hei displays which

pictures of Marsdeni prominently] pictures of Marsdeni ] are generally the attractive ones (cf. Safir 1998)

For another, in the idiom chunk case of “raising” derivations (cf. (20)), we would expect ungrammaticality (or at least marginality) due to the second, external, occurrence of the idiom chunk not being able to pair with the rest of the idiom:

(20) [The headwayi that [he made headwayi] headway] was satisfactory

For the first problem, we refer to Sauerland’s (1999, 2003) solution in terms of the notion of “vehicle change”. For the second (and for other problems), we refer to Cinque (in preparation). There the idiom case is tentatively treated in terms of a


13

silent Head: AMOUNT, TYPE, etc. (in the case at hand: [The AMOUNT of

headway that [he made AMOUNT of headway] AMOUNT] was satisfactory). After sketching how the externally headed postnominal type of restrictive

relatives is derived from (16) in both the “raising” and the “matching” derivations, let us briefly review how the other types of restrictive relatives can be derived, under the two derivations, from the same, unique, structure.

Externally Headed Prenominal RCs: Raising (cf. (21)):

dP2 is attracted to Spec,C2, from where it controls the deletion of dP1; after which the remnant raises to Spec,C1.

13 Reconstruction effects are expected as the overt Head is the ‘internal’one (linked to

the trace). And so is sensitivity to islands, due to the movement of the ‘internal’ Head.

13 I abstract here from the further possibility of dP1 raising above dP2 and controlling its deletion before the raising of the IP remnant. Within Chinese, and in other languages with prenominal relatives, some variation exists also with respect to the position of C(P)1 and C(P)2, which either are or can be merged above the position of definite articles, demonstratives, and universal quantifiers. For more detailed discussion of both cases, see Cinque (in preparation). Also the problem raised by the violation of proper binding caused by movement of the remnant needs to be addressed.

Guglielmo Cinque

14

This case seems to be instantiated by Chinese, which displays both relativization of idiom chunks (hence reconstruction) and island sensitivity (Aoun and Li 2003, 177), and Modern Tamil, where, according to Annamalai and Steever (1998, 123) and Vasu (1994, section 2.2), prenominal relative clauses are sensitive to islands.14

Matching (cf. (22)):

dP1 directly controls the deletion of dP2 backward. No reconstruction effects are expected, as the overt Head is the ‘external’ one (the ‘internal’ Head not having moved). Nor is sensitivity to islands, as no movement of the internal Head is involved.

This case may be instantiated by (among other languages) Tsez (Northeast Caucasian), which apparently shows no island sensitivity (Comrie and Polinsky 1999).

Internally headed RCs (which often alternate with prenominal RCs – Cole 1987): 14 Also see the Inverse Case Attraction option of the Malayalam relative clause in (i),which Abraham (1978,64) takes as evidence for the movement of the Head (cf. the analysis of Inverse Case Attraction in terms of “raising” in Cinque 2007, 99-101): (i) [saar innale sakaariccillee oru vidyaarthi-ye ] avan innu vannilla teacher yesterday scold.PST.TAG one student-ACC he today come.PST.NEG

‘The student whom the teacher scolded yesterday did not come today’


15

If internally headed relative clauses always displayed the indefiniteness restriction of Lakhota, as Williamson (1987) originally conjectured (p. 169) (also see Culy 1990), and necessarily showed the other properties characterizing Lakhota (possibility of stacking and absence of island sensitivity – Williamson 1987, 173 and 177), internally headed relative clauses could be taken to involve just a “matching” derivation; one which in fact is the converse of the “matching” derivation of externally headed prenominal relatives. In this case, it is the internal Head that controls the deletion of the external Head forward (rather than viceversa). Compare (23) with (22):15

I take this to be correct, even if for just one type of internally headed relative clauses. Those of Lakhota, as well as those of Diegueño (Yuman – Gorbet 1976),

15 Note that in the “matching” derivation of internally headed relative clauses ((23)), as well as in the “matching” derivation of externally headed prenominal relatives ((22)), neither Head c-commands the other from its in situ position, so that deletion of one by the other is not forced. As with VP deletion, which can take place either backward or forward in the same language, one should expect deletion here to freely apply either backward or forward, with the consequence that the language may give the impression of having two separate strategies of relative clause formation (external prenominal and internal) (cf. Cole’s observation that often externally headed prenominal relatives alternate with internally headed relatives within the same language). However, things are more complex (see Cinque in preparation for discussion).

Guglielmo Cinque

16

and Mojave (Hokan – Munro 1976), which show the same cluster of properties (indefiniteness restriction, the possibility of stacking and the absence of island sensitivity).16

In the recent literature (Basilico 1996, Grosu 2000, Grosu and Landman 1998), another type of internally headed relative clauses is recognized, which displays no indefinite restriction, and also the impossibility of stacking and the presence of island sensitivity (the opposite properties of the first type).

This type is apparently found in Japanese, Korean, Quechua, Navajo, and Haida, among other languages.

Given especially its property of island sensitivity, it is tempting to see this second type as involving movement (differently from the first type); more specifically to involve the “raising” derivation in (24), where the internal Head, dP2, is attracted to Spec,C2, from where it controls the deletion of dP1, the external Head. After that a phrase of the Remnant must be taken to raise to Spec,C1, higher than the strong determiners.17 In this case, reconstruction effects are expected, as the overt Head is the ‘internal’ one, linked to the trace, as is sensitivity to islands, due to the movement of the internal Head.

16 The head internal relative clauses of the Gur languages Mooré (Peterson 1974, Tellier 1989) and Buli (Hiraiwa 2005, section 5.3.2) also show the indefiniteness restriction. However, the fact that the internally headed relatives of Buli show sensitivity to islands, and those of Mooré license parasitic gaps (Tellier 1989) suggests that the internal Head does move, though not as high as to cross over the strong determiners (which is what the “left-headed” variant of the same construction in (i) in Buli apparently does): [ná:-m’Ʊ [àtì núrú-wá swà] lá] (Hiraiwa 2005,198) cow-the COMP man-the own Dem ‘the cow that the man owns’ 17 In the “left-headed” internally headed relative clauses of the Gur languages discussed by Hiraiwa (2005) there is no additional raising of a phrase of the Remnant.


17

If dP1 also raises above dP2 and controls its deletion before a phrase of the remnant in turn raises above the strong determiners, the expected properties will be partly different. See Cinque (in preparation). Once the different types of internally headed relative clauses are teased apart, the indefiniteness restriction of the first type, where nothing moves, can be taken to provide independent evidence for the indefinite nature of the Head of restrictive relative clauses, as the internal Head and the external one, the traditional Head of relative clauses, are an exact match of each other under the analysis sketched above.18

18 Lakhota lacks (internally headed) nonrestrictive relative clauses (see (i)), fromWilliamson 1987,175): (i) *[[(Miye) makuže] ki/cha] wichawota ki ekta mnį kte Iemph I-sick the/a feast the to I-go FUT ‘I, who am sick, will go to the feast’ Given the merger of nonrestrictives above demonstratives (and universal quantifiers) (see fn. 8 above), one could however expect there to be languages with internally headed nonrestrictive relatives. One such language appears to be Haida. See (ii), from Enrico (2003, 570): (ii) [tuut-ee-raa qung-ee 7ij-aa-n]-raaga ’la 7waa-gaa-n [box-DF-in moon-DF be-EVID-PST-]for 3PERS do-EVID-PST ‘He did it for the moon, which was in the box’

Guglielmo Cinque

18

We will be even more sketchy on the remaining two types, referring to Cinque (in preparation) for more detailed discussion.19

For headless, or free, relative clauses, which arguably involve movement of just the internal Head, we take there to be a silent external Head (of a restricted class: THING, AMOUNT, PLACE, TIME, PERSON, MANNER) as shown in (25):

(25) a. (I don’t like) [[ what THING you said] (SUCH) THING] b. (He weighs) [[ what AMOUNT you weigh] (SUCH) AMOUNT] c. (Here is) [[ where PLACE they slept] THERE PLACE] d. (I was there) [[ when TIME he said that] THEN TIME] e. (She hates [[ whoever PERSON does that ] (SUCH) PERSON]

In certain languages the “dummy” external Head (thing, place, time, person, etc.) is necessarily overt (‘thing (that) you said’ = ‘what you said’; ..). See Rapanui (Austronesian) – Du Feu 1996,47; Obolo (Niger-Congo) – Faraclas 1984,45; Abun (Papuan) – Berry and Berry 1999,146ff.

In Lakhota, it is instead in situ, within the relative clause, and optional. See (26):

(26) [ Mary (taku) kağe] ki] ophewatų M. (something) make the I-buy ‘I bought what Mary made’

As to (Relative-)Correlative clauses, they are, strictly speaking, not a separate type, as they involve one or the other of the core types of relative clauses as one component. Abstracting away from the multiple headed adjunct correlative construction, simple correlatives can be analysed as the ‘left dislocation’ of a full DP containing (depending on the language) either an externally headed postnominal relative20, or an externally headed prenominal one21, or an internally headed one22, or a free relative (as in many Hindi correlatives), resumed in the matrix clause by a DP preceded by a demonstrative or by a demonstrative or pronominal alone. See Cinque (in preparation) for more detailed discussion.

19 The possible further type of ‘adjoined relatives’ (Hale 1976) can be analyzed as a case of (obligatory?) relative clause ‘extraposition’. See Keenan and Comrie (1977,80f) and Cinque (in preparation). 20 As in (some of the correlatives of) Marathi. See Cinque (in preparation) for discussion. 21 As in the Malayalam case discussed in fn.14 above, which also shows that it may be too strong to claim, as Downing (1978,400) did, that “[i]f a language has correlative relative constructions, it does not have prenominal ad-relative clauses.” 22 As in Wappo (a language isolate of California). See Li and Thompson (1978) and Thompson, Park and Li (2006, 6.2).


19

For the facts reviewed here to be construed as evidence for the indefinite character of the Head of restrictive relative clauses, some questions and some apparent counterexamples should also be addressed.23

A potential difficulty could come from the so-called definite conjugation of Hungarian, which marks (in main clauses) the definite character of the object. However, it seems that the trace of the relativized internal Head is indeed marked as indefinite, despite the apparent definite character of the external Head (and of the relative pronoun), thus confirming the evidence given above. See (27), from MacWhinney and Pléh (1988,100):24

(27) A kutya kergeti a macskat, amelyet nez az eger. The dog(NOM) chase-3S-DEF the cat-ACC, which-ACC watch(3SINDEF) the mouse(NOM). ‘The dog chases the cat whom the mouse watches’

But things may be more complex and need to be looked into more carefully. For example, Bianchi (1999,82f) claims that the relative determiner ami (from the interrogative mi ‘who’) triggers the indefinite conjugation, while the relative determiner amely (from mely ‘which’) triggers the definite one (though this appears contradicted by (27) above). Should the which-type relative pronouns indeed turn out to sometime trigger the definite conjugation, it could be that they exploit the higher nonrestrictive Merge position, like the formal il quale restrictives of Italian in Cinque’s (1982, 2008b) analysis, which also show a definite relative pronoun.

Another possible difficulty for the indefinite character of the Head of restrictive relative clauses is the existence of restrictive relative clauses with definite resumptive pronouns, like the Palestinian Arabic case in (28), from Shlonsky (1992, 445)25, or the Bulgarian example (29)26:

(28) l-bint ?illi šufti-*(ha) the-girl that (you.fem.) saw-her ‘the girl that you saw’

23 I thank Gaby Hermon for raising these questions and for helpful discussion. 24 As MacWhinney and Pléh (1988,100) put it, “[i]f the head noun plays the role of the object in the relative clause the verb is obligatorily indefinite even though the head of a relative is usually thought of as semantically definite (Kuno, 1986) and this definiteness is even marked morphologically on the relative pronoun. If the head noun is the subject of the relative clause the conjugation varies according to the definiteness of the object in the relative clause”. 25 Resumptive pronouns are also obligatorily required in Lebanese Arabic (Aoun and Li 2003, sect. 4.3). 26 Examples (29) and (30)b were suggested by Iliyana Krapova.

Guglielmo Cinque

20

(29) Poznavam edin colleg ot našta katedra deto to ku-što (go) uvolnixa I know a colleague from our institute that just (him) they.fired ‘I know a colleague from our institute that they just fired’

One possibility to reconcile these facts with the evidence reviewed above for the indefinite character of the Head of restrictive relatives would be to say that ‘definite’ pronouns can also stand for the smaller (indefinite) dP constituent postulated above.

Although more work is needed for an understanding of the phenomenon, a suggestive piece of evidence for this idea seems to come from Clitic Left Dislocation. If the left dislocated nominal and the ‘resumptive’ clitic start out as a ‘big DP’ constituent, which is subsequently split apart27, we have to conclude that a definite clitic can double (‘resume’) an indefinite nominal (in Italian even a non specific one). See (30)a from Italian and (30)b from Bulgarian:

(30) a. Una ragazza prima o poi la troverà a girl sooner or later her he.will.find ‘Sooner or later he will find a girl’ b. Edin colleg ot našta katedra to ku-što go uvolnixa a colleague from our institute just him they.fired ‘A colleague from our institute they just fired him’

References

Abraham, P.T. 1978. Relative Clause in Malayalam. Annamalainagar: Department of Linguistics, Annamalai University (Publication no.62).

Abusch, Dorit and Mats Rooth. 1997. Epistemic NP Modifiers. In A. Lawson, ed., Proceedings from Semantics and Linguistic Theory (SALT) VII, 1-18. Ithaca: CLC Publications (Cornell University).

Åfarli, Tor. 1994. A promotion analysis of restrictive relative clauses. The Linguistic Review 11.81-100.

Annamalai, E. and Sanford B. Steever. 1998. Modern Tamil. In S.B. Steever, ed., The

Dravidian Languages. 100-128. London: Routledge. Aoun, Joseph and Y. Audrey Li. 2003. Essays on the Representational and Derivational

Nature of Grammar. The Diversity of Wh-Constructions. Cambridge, Mass.: MIT Press.

Basilico, David. 1996. Head Position and Internally Headed Relative Clauses. Language 72. 498-532.

27 As proposed by Kayne (1972, sect. 3) for Complex Inversion in French and Uriagereka (1995,81) for Spanish. A movement derivation of Clitic Left Dislocation (cf. Cinque 1977) is compatible with the observations presented in Cinque (1990, chapter 2) if successive cyclic movement is for independent reasons unavailable.


21

Benveniste, Émile. 1966. La phrase relative, problème de syntaxe générale. In É.Benveniste Problèmes de linguistique générale, I. 208-222.Paris: Gallimard.

Berry, Keith and Christine Berry. 1999. A Description of Abun: a West Papuan language of

Irian Jaya. Research School of Pacific and Asian Studies. The Australian National University, Canberra.

Bianchi, Valentina. 1999. Consequences of Antisymmetry: Headed Relative Clauses. Berlin: Mouton de Gruyter.

Bosque, Ignacio. 1993. “Degree Quantification and Modal Operators in Spanish”, paper presented at the 1993 Going Romance meeting, Utrecht.

Bosque, Ignacio. 1996. On Specificity and adjective position. In J. Gutiérrez-Rexach and L. Silva-Villar, eds., Perspectives on Spanish Linguistics. Vol.1, 1-13. Department of Linguistics, UCLA.

Bosque, Ignacio. 2001. Adjective position and the interpretation of indefinites. In J. Gutiérrez-Rexach and L. Silva-Villar, eds., Current Issues in Spanish Syntax and

Semantics, 17-63. Berlin: Mouton de Gruyter. Browning, Marguerite. 1987. Null Operator Constructions. Ph.D. Dissertation. MIT. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. Cinque, Guglielmo. 1977. The Movement Nature of Left Dislocation. Linguistic Inquiry 8.397-

412. Cinque, Guglielmo. 1990. Types of A'-Dependencies. Cambridge, Mass.: MIT Press. Cinque, Guglielmo. 2002. Complement and Adverbial PPs: Implications for Clause Structure.

Paper presented at the 25th Annual GLOW Colloquium, Amsterdam, April 9-11 2002 (later published in Cinque 2006, pp. 145-166).

Cinque, Guglielmo. 2003/8. The prenominal origin of relative clauses. Paper presented at the NYU Workshop on Antisymmetry and Remnant Movement, Oct. 31- Nov.1 2003, at the LSA Summer School at Harvard and MIT in 2005, at the Colloque de Syntaxe et Sémantique à Paris 2007, 4-6 October 2007, and at the XXXIV Incontro di Grammatica Generativa at the University of Padua on February 23 2008.

(Handout and abstract available online at: http://dspace-unive.cilea.it/handle/10278/210 and http://dspace-unive.cilea.it/handle/10278/211) Cinque, Guglielmo. 2005. Deriving Greenberg’s Universal 20 and Its Exceptions. Linguistic

Inquiry. 36. 315-332. Cinque, Guglielmo. 2006. Restructuring and Functional Heads. The Cartography of Syntactic

Structures, vol.4. Oxford, Oxford University Press. Cinque, Guglielmo. 2007. A note on linguistic theory and typology. Linguistic Typology 11.

93-106. Cinque, Guglielmo. 2008a. The Fundamental Left-Right Asymmetry of Natural Languages.

In S. Scalise, E. Magni, A. Bisetto, eds., Universals of Language Today. Berlin: Springer.

(available online at: http://dspace-unive.cilea.it/handle/10278/214) Cinque, Guglielmo. 2008b. Two Types of Nonrestrictive Relative Clauses. In O. Bonami and

P. Cabredo Hofherr (eds.) Empirical Issues in Syntax and Semantics 7. Paris.

Guglielmo Cinque

22

(available online at: http://dspace-unive.cilea.it/handle/10278/386) Cinque, Guglielmo. Forthcoming. The Syntax of Adjectives. A Comparative Study. Cinque, Guglielmo. In preparation. The Prenominal Origin of Relative Clauses. Cole, Peter. 1987. The Structure of Internally Headed Relative Clauses. Natural Language

and Linguistic Theory 5. 277-302. Comrie, Bernard and Maria Polinsky. 1999. Form and Function in Syntax: Relative clauses in

Tsez. In M. Darnell, E. Moravcsik, F. Newmeyer, M. Noonan, K. Wheatley, eds., Functionalism and Formalism in Linguistics II: Case Studies. 77-92. Amsterdam: Benjamins.

Culy, Christopher. 1990. The Syntax and Semantics of Internally Headed Relative Clauses. Ph.D. Dissertation, Stanford University.

Downing, Bruce T. 1978. Some Universals of Relative Clause Structure. In J.H. Greenberg, ed., Universals of Human Language. Vol.4. Syntax. 375-418. Stanford: Stanford University Press.

Du Feu, Veronica. 1999. Rapanui. London: Routledge. Enrico, John. 2003. Haida Syntax. Vol. I. Lincoln: University of Nebraska Press. Faraclas, Nicholas G. 1984. A Grammar of Obolo. Bloomington: Indiana University

Linguistics Club. Gorbet, Larry. 1976. A Grammar of Diegueño Nominals. New York: Garland. Grosu, Alexander. 2000. The semantic diversity of internally-headed relative clauses. In C.

Schaner-Wolles et al., eds., Naturally! Linguistic studies in honour of Wolfgang

Ulrich Dressler presented on the occasion of his 60th birthday. 143-152. Torino: Rosenberg & Sellier.

Grosu, Alexander and Fred Landman. 1998. Strange relatives of the third kind. Natural

Language Semantics 6. 125-170. Hale, Ken. 1976. The adjoined relative clause in Australia. In R. M. W. Dixon (ed.)

Grammatical Categories in Australian Languages. 78-105. Canberra: AIAS. Heim, Irene. 1982. The Semantics of Definite and Indefinite Noun Phrases. Ph.D.

Dissertation, University of Massachusetts at Amherst. Hiraiwa, Ken. 2005. Dimensions of Symmetry in Syntax: Agreement and Clausal

Architecture. Ph.D. Dissertation, MIT. Kayne, Richard S. 1972. Subject Inversion in French Interrogatives. In J. Casagrande and B.

Saciuk, eds., Generative Studies in Romance Languages. 70-126. Rowley, Mass.: Newbury House.

Kayne, Richard S. 1994. The Antisymmetry of Syntax. Cambridge, Mass.: MIT Press. Kayne, Richard S. 2004. Some Preliminary Comparative Remarks on French and Italian

Definite Articles. Ms., New York University (available online at: http://www.nyu.edu/gsas/dept/lingu/people/faculty/kayne/papers/Kayne1004DefiniteArticles.

pdf)


23

Kayne, Richard S. 2005. Some Notes on Comparative Syntax, with Special Reference to English and French. In G. Cinque and R. Kayne, eds., The Oxford Handbook of

Comparative Syntax. 3-69. New York: Oxford University Press. Kayne, Richard S. 2006. Expletives, Datives, and the Tension between Morphology and

Syntax. Ms., NYU (available online at: http://ling.auf.net/lingBuzz/000286). Kayne, Richard S. Forthcoming. Why Isn’t This a Complementizer. In P. Svenonius et al.

(eds.) Functional Structure from Top to Toe. (available online at: http://ling.auf.net/lingBuzz/000726).

Keenan, L. Edward and Bernard Comrie. 1977. Noun Phrase Accessibility and Universal Grammar. Linguistic Inquiry 8. 63-99.

Koopman, Hilda and Dominique Sportiche. Forthcoming. The que/qui Alternation: New Analytical Directions. In P. Svenonius et al. (eds.) Functional Structure from Top to

Toe. (available online at: http://ling.auf.net/lingBuzz/000638). Krapova, Iliyana. 2008. A Unified Analysis of the Bulgarian Relative and Factive

Complementizer deto. Ms. Università Ca’ Foscari, Venezia. Kuno, Susumu. 1986. Functional syntax. Chicago: Chicago University Press. Li, Charles N. and Sandra A. Thompson. 1978. Relativization strategies in Wappo. BLS

4.106-113. MacWhinney, Brian and Csaba Pléh. 1988. The processing of restrictive relative clauses in

Hungarian. Cognition 29. 95-141. Milsark, Gary. 1974. Existential Sentences in English. Ph.D. Dissertation, MIT. Munro, Pamela. 1976. Mojave Syntax. New York: Garland. Peterson, Thomas H. 1974. On Definite Restrictive Relatives in Mooré. Journal of West

African Languages IX. 71-78. Picallo, M. Carme. 1994. A Mark of Specificity in Indefinite Nominals. Catalan Working

Papers in Linguistics 4.1. 143-167. Safir, Ken. 1998. Reconstruction and bound anaphora: Copy theory without deletion at LF.

Ms., Rutgers University. Salzmann, Martin. 2006. Reconstruction in German restrictive relative clauses. Linguistics in

the Netherlands 2006. 186-198. Sauerland, Uli. 1998. The Meaning of Chains. Ph.D. Dissertation, MIT. Sauerland, Uli. 1999. Two Structures for English Restrictive Relative Clauses. In M. Saito et

al. (eds.) Proceedings of the Nanzan Glow. 351-366. Nagoya: Nanzan University. Sauerland, Uli. 2003. Unpronounced heads in relative clauses. In K. Schwabe and S. Winkler,

eds., The Interfaces. Deriving and interpreting omitted structures. 205-226. Amsterdam: Benjamins.

Săvescu, Oana (2008a). ‘Hungry Experiencers’. In E. Elfner and M. Wakov, eds., Proceedings of the 37th Annual Meeting of the North Eastern Linguistic Society. Amherst, Mass.: GLSA Publications.

Săvescu, Oana (2008b) ‘Hungry Experiencers-An Argument for Silence’. Submitted. Shlonsky, Ur. 1992. Resumptive pronouns as a last resort. Linguistic Inquiry 23. 438-468.

Guglielmo Cinque

24

Sohn, Ho-Min (1973) “Relative Clause Formation in Micronesian Languages”. Working

Papers in Linguistics. University of Hawaii, 5/8. 93-124 (also in Oceanic Linguistics 12. 353-392 (1973).

Sportiche, Dominique. 2008. Inward Bound: splitting the wh-paradigm and French relative qui. Ms., UCLA (available online at: http://ling.auf.net/lingBuzz/000623).

Stockwell, Robert, Paul Schachter and Barbara Partee. 1968/1973. The Major Syntactic

Structures of English. New York: Holt, Rinehart and Winston. Szczegielniak, Adam. 2005. Relativization that you did… MIT Occasional Papers in

Linguistics 24 (Updated version of a Harvard Ph.D. Dissertation under the title Relativization and Ellipsis).

Tellier, Christine. 1989. Head-Internal Relatives and Parasitic Gaps in Mooré. In I. Haïk and L. Tuller, eds., Current Approaches to African Linguistics (vol. 6). 298-318. Dordrecht: Foris.

Thompson, Sandra A., Joseph Sung-Yul Park, Charles N. Li. 2006. A Reference Grammar of Wappo. Berkeley: University of California Press (available online at: http://repositories.cdlib.org/ucpress).

Uriagereka, Juan. 1995. Aspects of the Syntax of Clitic Placement in Western Romance. Linguistic Inquiry 26. 79-123.

Vasu, R. 1994. On the nature of wh-trace in Tamil relative clauses. International Journal of

Dravidian Linguistics, 23:1. 44-64. Williamson, Janis S. (1987) “An Indefinite Restriction for Relative Clauses in Lakhota”, in

E.J. Reuland and A.G.B. ter Meulen (eds.) The Representation of (In)definiteness. Cambridge: MIT Press, pp. 168-190.


33 – 2008

25

FLOATING QUANTIFIERS: FRENCH UNIVERSAL

QUANTIFIERS AND N-WORDS*

Lena Baunaz

Canonical quantification describes the configuration of canonical quanti-ficational sentences, (1a), while non-canonical quantification occurs when Q and its restriction are not adjacent anymore, (1b)1. Floating Quantifiers (FQs) are typical instances of non-canonical quantification: they involve movement of the noun restriction past the lexical Q (Kayne 1975, Sportiche 1988,)2. FQ structures depend on the association of a Q and an associate DP: toutes [DP les filles] and chacune [DP

des filles] 3. Tous les N ‘all the N’ and chacun des N ‘each of the N’ can overtly

reflect (1a) and (1b): (2a)/(3a) exemplify (1a); and (2b)/(3b) illustrate (1b). FQs and non-FQs are truth conditionally equivalent (Doetjes 1997, Bobaljik 2003): toutes

* I am extremely indebt to Genoveva Puskás, Tabea Ihsane and Christopher Laenzlinger, for their constant availability and patience in discussing data and judgments. Special thanks go to Genoveva Puskás, Ur Shlonsky, Marcel den Dikken and Eric Mathieu for their insightful comments on earlier version(s) of this article. I would also like to thank Gabi Soare. Finally, I thank the audience of the XXXIV° Incontro di Grammatica Generativa (February 21 – 23, 2008) for their precious comments and suggestions. I am also extremely thankful to Hamida Demirdache, Orin Percus, Anamaria Falaus and the audience of the Séminaire de Recherche (Université de Nantes, May 14, 2008), where I presented a preliminary version of this paper. All mistakes are of course mine. 1 ‘Quantifier’ refers to items that contribute to some quantity/amount meaning. In (1), it doesn’t say anything about its syntactic structures (or category), which is developed below. 2 FQs must be distinguished from Split-DP structures, which are characterized by movement of Q stranding the restriction in-situ, (ib) (Obenauer 1994, a.o): FQs and their non-FQ counterparts do not yield different presuppositions, whereas Split and unsplit do (i) a. Combien de toiles as-tu peint(e)s How many paintings have you painted.M.PL/F.PL b. Combien as-tu peint(*es) de toiles? How many have you paint. M.PL/*F.PL of. Paintings Both: ‘How many paintings did you paint?’ 3 Chaque ‘every’ and tout ‘all.sg’ are D°, selecting NPs, and don’t float. See Baunaz 2008.

Lena Baunaz

26

and chacune quantify over the entities les filles ‘the girls’. FQs and non-FQs syntactically agree with the associate DP: in (2), the DP les filles is plural/ feminine, just like toutes that modifies it. In (3), only gender agreement is displayed. FQ structures also overtly show up with (some) N-words, (4b).

(1) a. Canonical Quantification: Q– restriction – scope b. Non-Canonical Quantification: restriction– Q – scope

(2) a. Juliette a vu toutes les filles

J. has seen all.fem.pl the.fem.pl girls b. Juliette les a toutes vues J them has all.fem.pl seen

(3) a. Juliette a vu chacune des filles

J. has read each.fem.sg of.the girls.fem.pl b. Juliette les a vues chacune

J. them has seen each.fem.sg

(4) a. Juliette a chanté aucune des chansons

J. has sang none of.the songs b. Juliette en a chanté aucune

J. cl. has sang none ‘Juliette sang none of them’

FQ is clause-bound: a strict locality restriction rules the dependency of FQ and the DP it modifies: (i) FQ is not licensed in finite embedded clauses, (5a), (5c) and (ii) the modified DP must c-command FQ, (5b):

(5) a. * Les filles pensent que Maya a toutes vu les garçons the girls think that M. has all.F.PL seen the.M.PL boys.M.PL b. * Le manager des filles a toutes vu les garçons the manager of the girls has all. F.PL cured the.M.PL boys.M.PL c. *Il en pensent que Maya a vu aucune

he them think that M. has seen none

I first argue that argument N-words resemble semantically and syntactically to ∀Qs4. Then I propose that both types of Qs have similar internal structures.

4 One difference between tous les N and aucun des N is the fact that the first one can float from both the subject and the object position, while the latter can only float from the object position. I don’t discuss this in this paper, and leave it for further research.

Floating Quantifiers: French Universal Quantifiers and N-words

27

1. Universal Quantifiers: Internal Structures

I adopt the idea that tous are not heads, but modifiers (Fitzpatrick 2006). Two structures are available: either no selection between Q and its restriction is involved (contra Giusti 1997) (6a), or there is (6b):

(6) a. Adjoined structure: [DP tous [DP les [NP garçons]]] ‘all the boys’ b. Specifier structure: [D P tous (of) [D° the [NP garçons]]] ‘all the boys’

Selection is an argument in favour of (6a): no predicate selects for Qs alone, yet some verbs select for DPs. An argument against (6b) is (7): if demonstratives are maximal projections in [Spec, DP] (Giusti 1997), then no room in Spec is available for tous in (8b) (vs. (8a)).

(7) tous ces garçons ‘all these boys’

(8) a. Adjoined structure: [DP tous [DP ces [D° [NP garçons]]]] b. Specifier structure: [DP ces [D° [NP garçons]]]

I adopt the structure in (6a) (Fitzpatrick 2006)5: tous is adjoined to DP. Let’s now turn to the floating property of ∀Qs.

Two major approaches address the issue of FQs: the movement (syntactic) approach considers that the lack of semantic differences between FQ and non-FQ reflects a lack of syntactic differences (Sportiche 1988, a.o). FQ is a pure adnominal Q (9a). The adjunction (semantic) approach argues for the adverbial status of FQ (Dowty and Brodie 1984, a.o), (9b).

(9) a. [Les infirmières]i ont [VP[toutes/chacune ti] fait des points de sutures] b. Les infirmières ont [VP toutes/chacune [VP fait des points de sutures]] The nurses have all/each made stitches

The syntactic approach doesn’t account for the unavailability of postverbal FQs with passives / unaccusatives (vs. intransitives), nor does it treat object and subject related Qs equivalently (see Kayne 1975). The semantic approach doesn’t account for the Det-like behaviour of FQ.

I adopt an approach that accounts for both the adverbial positioning of these elements and their agreement behaviour: Doetjes 1997:202 claims that FQ is a genuine quantified noun phrase. iT is dependent on the presence of an associate DP with whom it agrees: she argues that FQ is an adnominal Q, selecting pro (of type e) as its restriction (and domain of quantification), (10a). The locality effects observed with FQ, as well as the rise of agreement is accounted for by pro, which syntactically and semantically mediates the relationship between the FQ and its

5 I refer the reader to Fitzpatrick 2006 and Baunaz 2008 for more developed arguments in favor of the adjunction analysis.

Lena Baunaz

28

associate DP. FQ binds the trace of the moved DP and syntactic agreement arises6. Fitzpatrick 2006:66 claims that ‘this pronominal element contributes a coreference–like relationship to the nominal associate’, in a A position. He proposes that FQ is semantically (and syntactically) equivalent to [all of them]. Pro is incorporated into the semantics of Dowty and Brodie 1984 for adverbial FQs. This semantics is similar to that of its non-floated counterpart: both are fully-fledged ∀Qs.

Because its distribution is that of adverbs7, the complex [FQQ + pro] is generated in an adjoined position from where it c-commands ec, a member of its chain, the A-trace of the DP over which it (indirectly) quantifies, (10b).

(10) a. FQ: [QP tous [DP pro]] b. [XP FQi [XP … eci…]] (Doetjes 1997:202, (5))

FQ and non-FQ have similar structures: Q adjoin to the DP hosting pro in FQ (its restriction); and to the overt DP restriction in non-FQs. In both cases, agreement comes from binding with the restriction (and the associate DP)8.

The aim of section 2 is to explore the internal structure of argument N-words (personne ‘nobody’, rien ‘nothing’, aucun des N ‘none of the N’).

2. N-words

Two issues arise as to the nature of N-words: (i) their negative status; (ii) their quantificational status. Based on Giannakidou 1998 (and subsq.) for Greek, I show that French N-words are inherently negative (Mathieu 2002 (a.o)). N-words can be translated by two truth conditionally equivalent formulae: either ¬ scopes over ∃, (15a); or ∀ scopes over ¬, (11b).

6 –un in chacun lexicalizes pro: chac- distributes over its restriction –un (Fitzpatrick 2006). 7 FQ is not of category ‘adverbs’ (Doetjes 1997). FQ must be distinguished from Q time adverbs which range over times and take scope over an event argument in the VP and from adverbial beaucoup ‘a lot’, which can appear clause medially, quantifying over an event argument and occupying VP-adjoined positions (Obenauer 1994). Doetjes treats adverbial beaucoup and FQ tous as occupying adverbial positions, yet states that whereas beaucoup can functions as an adverb, floated tous cannot: adnominal Qs do not theta-select their host, rather they select the category of their host, i.e., DPs and tous quantifies over individuals. 8 Adopting the adjoined structure for both constructions is advantageous, and the strict correlation between semantics and syntax is kept. Another advantage is that object and subject FQs are treated in a uniform way: what counts is the direct relation between FQ and the DP-trace (Doetjes 1997). It also accounts for the unavailability with unaccusatives / passives, and their availability with intransitives (iff the verb is followed by some material; FQ adjoins to a postverbal XP). I refer to Fitzpatrick 2006 for objections and his solutions.


29

(11) a. ¬∃x [P(x) ∧ Q(x)] (Existential negation) b. ∀x [P(x) →¬Q(x)] (Universal negation)

According to (11), N-words combine ¬ plus either ∃Q to its right or ∀Q to its left. Languages vary as to which logical translations they allow: Greek N-words are ∀Qs (Giannakidou 1998 (a.o)); Italian N-words combines ∀Qs with sentential negation (Zanuttini 1991); and French N-words are construed existentially (11a) (Déprez 1997, Mathieu 2002).

I focus on constructions involving the overt negative operator (Op) pas + an adjacent Q (unemphasized un N ‘a N’ and unemphasized tous les N ‘all the N’). In section 2.1, I discuss the status of pas un as well as pas tous and show that although pas modifies an adjacent un N, it takes sentential scope, building an anti-additive Op, while pas in pas tous les N only takes phrasal (local) scope (vs. sentential scope), building a monotone decreasing Op only in its VP argument. I show that pas

un lexicalizes (11a) and that N-words lexicalize the inverse (scope) pattern of the sequence pas tous les N, (11b), which never gives rise to universal negation. In section 2.4, I argue for the ∀ status of French N-words, using semantic and syntactic diagnostics, indicating contrasts between ∃Qs and ∀Qs, elaborated by Zanuttini 1991 and Giannakidou 1998. Further evidence in favour of this treatment is the fact that they may overtly display FQ structures. I show that N-words can (c)-overtly realize FQ, resulting in a structure à la Doetjes (section 2.5).

2.1. Monotonicity : pas un N vs pas tous les N

2.1.1. pas un N ‘not a single N’

pas un N ‘not a single N’ involves the (unemphasized) Det un ‘a’. Un can have a numeral reading. pas + un can also be idiomatic. Besides a numeral and an idiomatic reading, un Ns is potentially ambiguous between an indefinite and a specific reading (see Ihsane 2006, a.o). I claim that un N in pas un N is an indefinite that is locally licensed by the negative Op pas.

First the numeral reading is available if un is focalised, receiving heavy stress. pas un is not contrastive in (12): it doesn’t mean that not one single student bought

flowers, but two (or more). Yet pas can hardly negate a numeral, unless focalizer même ‘even’ is inserted: in (13), ten boys is focalised, shading doubts on the numeral status of un in pas un N, (12):

(12) Pas un étudiant a amené de roses not a student has brought flowers

(13) *(même) pas dix garçons ont bu du coca Even not ten boys have drank of.the coke ‘not even ten boys drank coke’

Lena Baunaz

30

Second, pas un N in (12) is not a minimizer: when co-occurring with pas, object un N can be treated as a Nominal minimizer. Yet ‘there is a certain degree of idiomaticity, both in the choice of the N [(14a vs. 14b)] (…) and in the choice of the V [(15a) vs. (15b)]’ (Tovena et al. 2004:414). If not c-commanded by negation, they appear with a non-idiomatic (specific) meaning (16b). Pas un in (13) is semantically and syntactically different from (14)-(15): it is in the subject position and is not idiomatic.

(14) a. Il n’a pas dit un mot

he did not say a word b. Il n’a pas bu une tasse

he did not drink a cup

(15) a. ne pas prendre une ride

to remain up–to–date b. ne pas avoir une ride not to have a wrinkle (Tovena and al. 2004:415)

(16) a. Il n’a pas dit un mot, (*c'est-à-dire ‘merci’) b. un mot n’a pas été dit, c'est-à-dire ‘merci’ (a word) ne has not been said (a word), i.e., ‘thank’

Non-idiomatic un N can appear post-verbally too, yet it must be split from pas by the past participle in (17a) and it must out scope negation. For negation to out scope ∃Qs, de N must show up (17b):

(17) a. Mika n’a pas vu (*pas) un cheval (∃ > ¬ only) M. ne has (not) seen (not) some horse b. Mika n’a pas vu de cheval M. ne has not seen of. horse ‘Mika didn’t see any horse’

If the two un N are similar, we expect un N in subject position to be able to take wide scope over pas. Yet it doesn’t and de N doesn’t show up in (13).

Pas un N is monotone decreasing in both its NP and VP (18) and anti-additive (19)9 (Corblin et al. 2004). Ladusaw 1992 argues that monotone decreasing contexts as well as c-command license NPIs10. The anti-additive pas un étudiant takes scope over the sentence, reversing its polarity, (20):

9 A function f is monotonically decreasing iff it reverses the subset relation. An anti-additive function is a subset of monotone decreasing functions: the union of two arguments corresponds to the disjunctive expression or in natural languages. 10 See also Zwarts 1995 and Giannakidou 1998 (subsq), for pure semantic analyses.


31

(18) a. Pas un être humain a amené de roses → pas un type a amené de roses Not a human being has brought roses → not a guy has brought roses b. Pas un type a amené de roses → Pas un type a amené de roses rouges not a guy has brought roses → not a guy ne has brought roses red

(19) Pas un type fume ou boit ↔ Pas un type fume et pas un type boit Not a guy smokes or drinks ↔ Not a guy smokes and not a guy drinks

(20) Pas un étudiant n’a bu quoi que ce soit Not a student ne has drunk anything ‘Not a student drank anything’

Semantically, then, pas un N is interpreted as existential negation (11a), i.e, pas

lexicalises ¬ and un N overtly realises ∃. Un N is licensed by pas and builds an anti-additive Op which in turn licenses NPIs in object position.

2.1.2. pas tous les N

Non-FQs and FQ structures have similar truth-conditions: the same if true of pas modifying tous: if the negative marker pas co-occurs with tous, neither (21a), nor (21b/c) express universal negation: pas negates the participant in subject position, not the event denoted by the VP: the unmarked reading is neg over ∀, it is not all the

children that have eaten the chocolate11. The scope of pas is not sentential (vs. pas

in pas un N), and the constituent does not sit in NegP. I claim that pas in (21) is like FQ, it operates locally: its domain is tous. (21d) shows that pas cannot float off the ∀Q, i.e., it must be parasitic on ∀. I argue that the combination of pas plus tous yield a complex Q. Whenever pas and tous co-occur, negation is interpreted as a modifier, just like the negative prefix in- in in-constant. Two consequences arise: (i) full ∀Qs reconstruct, parasitic pas marking their basic position12; (ii) that pas is syntactically similar to FQ.

(21) a. Tous les enfants n’ont pas mangé le chocolat ([¬ > ∀] ; *[∀ > ¬]) b. Les enfants n’ont pas tous mangé le chocolat ([¬ > ∀] ; *[∀ > ¬])

11 Sentences like (21a) are productive in spoken French and speakers give uncontroversial judgements: ‘When negation is present, it must take scope over the ∀Q and its restriction, but not over the whole sentence’ (http://www.lexilogos.com/, my translation). Intonation is neutral on both items. A reviewer pointed out to me that (21a) is ambigous in Italian. Yet, it seems that once intonation is controlled for, Italien behave like French (see Baunaz 2008, Andrea Cattaneo (p.c)). 12 If full ∀Qs must reconstruct, then they are interpreted in their base (A)-position: the argument of the verb is the associate-DP, over which the ∀Q adjoins. FQ cannot reconstruct, i.e., they must occupy a position distinct from full ∀Qs, i.e., an ‘adverbial’ adjoined position. The argument of the verb, namely, the DP-associate, is generated a A-position.

Lena Baunaz

32

c. Pas tous les enfants n’ont mangé le chocolat ([¬ > ∀] ; *[∀ > ¬]) d. * Pas les enfants n’ont tous mangé le chocolat (not) (all) the children ne have (not) (all) eaten the chocolate ‘Not all the children have eaten the chocolate’

The relation between pas and FQ or non-FQ tous is insecable and no Q can be inserted in between. In (22), the ‘not every’ meaning is not disrupted by the scope of neither the specific, or the indefinite reading of une femme:

(22) a. Tous les enfants n’ont pas pensé que Jean aimait une femme All the children ne have not thought that J. liked a woman b. Les enfants n’ont pas tous pensé que Jean aimait une femme The children ne have not all thought that J. liked a girl c. Pas tous les enfants ont pensé que Jean aimait une femme Not all the children have thought that J. liked a girl ‘Not all the children have thought that Jean liked a girl’

The sequence pas-FQ is not interrupted by neither a preceding nor a following VP adverb (23a vs. 23b,c)13. Nor can the universal adverb presque ‘almost’ be inserted in between: presque can modify the VP in (24) meaning that the children

are eating, but have not yet finished, or [pas tous] in both subject and object positions, but not tous alone (25):

(23) Ils ont (?péniblement) pas (*péniblement) tous péniblement mangé They have (painstakingly) not (painstakingly) all (painstakingly) eaten

(24) Les enfants ont presque mangé = (VP modification) the children have almost eaten

(25) a. Les enfants ont (presque) pas (*presque) tous mangé the children have (almost) not (almost) all eaten b. (Presque) pas (*presque) tous les enfants ont mangé Almost not (almost) all the children have eaten

The fact that medial and initial pas do not operate over the VP, but take scope over the ∀Q only, suggests that they form a syntactic unit. When pas is clause medial, its association with a non-FQ is covert in that ∀Q moves to the [Spec, TP], while pas indicates the base position of ∀Q. I conclude that pas is to tous what tous

is to the associate DP: a modifier, and as such it adjoins to it. Pas tous lexicalizes a complex Q.

13 The case of pas souvent tous… ‘not often all…’ is discussed in Baunaz 2008.


33

Pas ‘not’ is a downward entailing, while tous ‘all’ is decreasing in its NP, but increasing in its VP argument (see Beyssade 2006 :6, (28)). The pair in (26) shows that tous, licenses NPIs only in its restrictor.

(26) a. Tous les étudiants qui avaient quoi que ce soit à dire ont participé All the students that had anything to say have participated b. *Tous les étudiants avaient quoi que ce soit à dire. All the students have anything to say

If pas modifies tous, monotonicity gets reversed in all its arguments: pas tous is decreasing in its VP (27) and increasing in its NP (28) (it cannot license the NPI in its restrictor, (29a)). In (29b/c), pas is not strong enough to take sentential scope, and license NPIs. I conclude that the association of pas plus tous yields a complex anti-additive Q (31) (That is discontinuous in (30b)) and that pas operates over ∀Q, only:

(27) a. pas tous les gars sont arrivés → pas tous les gars sont arrivés tôt Not all the guys arrived → not all the guys arrived early b. Pas tous les gars sont arrivés tôt -/-> pas tous les gars sont arrivés Not all the guys arrived early -/-> not all the guys arrived

(28) Pas tous les gars de Genève sont arrivés →Pas tous les gars sont arrivés Not all the guys from Geneva arrived → not all the guys arrived

(29) a. * Pas tous les étudiants qui avaient quoi que ce soit à dire ont participé Not all the students that had anything to say have participated b. * (Pas) tous les étudiants n’ont vu quoi que ce soit c. * Tous les étudiants ont pas vu quoi que ce soit (not) all the students have (not) seen anything

(30) a. Pas tous les garçons dansaient ou chantaient → Pas tous les garçons dansaient et pas tous les garçons chantaient Not all the boys danced or sang → not all the boys danced and not all the boys sang b. Pas tous les garçons dansaient et pas tous les garçons chantaient --/-> Pas tous les garçons dansaient ou chantaient not all the boys danced and not all the boys sang --/-> not all the boys danced or sang

I have shown that phrasal pas is either parasitic on an adjacent Q (tous), or a licenser (for un, de N, see fn. 13). If c-command is a formal licensing condition on NPIs, then pas un N and pas tous les N involve distinct internal structures (section 2.3), which influence their syntax at the clausal level.

Lena Baunaz

34

2.3. Internal Structures

In section 2.2, I argued that pas occupies different positions in pas un and in pas

tous: (i) pas locally licenses un N, building an anti-additive Op which in turns reverses the polarity of the clause, while (ii) it only modifies tous.

Pas un is a complex Q, where pas negates un. I claim that un is an indefinite, that needs binding for interpretation. Because un is not bound at the VP level, or by any other c-commanding Op, pas is inserted, licensing it. If NPI licensing needs semantic ‘roofing’ and syntactic licensing, then pas must occupy a position from where it c-commands quoi que ce soit

14. The scope of pas un N is sentential. Yet the scope of pas alone is local, and builds an anti-additive Det. Pas is not parasitic on un, but a licenser. As such, it doesn’t adjoin to un N. Because pas un N forms a complex Q, I propose that pas occupies the Q position within the left-periphery of the DP, as in (31), where FP turn the NP into an argument (Ihsane 2006). The whole cluster is negative, and movement to [Spec, NegP] compulsory.

(31) [QP pas [Q ° [+neg] [FP [F° un [NumP [Num t …. [NP t garçon]]]]]]]

Because the sequences pas tous / tous… pas never trigger sentential negation, there is no NegP, and the neg-criterion needs not be satisfied. Pas is a modifier and doesn’t c-command the object NPI, and NPI licensing fails.

(32) a. *[CP[TP[[[[DPpas [Dpi tous les N]]]]][T°ont [VP ti [V° lu] [DP quoi que ce soit]]]] b. *… [VP [[[[QP pas [QPi tous [DP pro]]]]]] [VP eci [V° lu] [DP quoi que ce soit]]] c. * [CP[TP[[[[DPi tous les N]]]] [T° ont] [VP [[[[DP pas ti]]]] [V°lu] [DP quoi que ce soit]]]]

The behaviour of pas vis-à-vis tous in both FQ and non-FQ constructions is similar: in the unmarked cases, pas scopes only over the ∀Q.

What about N-words? Either they are also translated as existential negation (11a), i.e., pas un N and N-words would be synonyms, or they involve universal negation. Because French N-words behave like ∀Qs I claim that (11b) is the correct representation of French N-words.

2.4. The Universal Negative status of N-words

N-words are traditionally taken to express sentential negation, involving movement to [Spec, NegP] (Zanuttini 1991, Haegeman 1995, a.o). If nothing morphological indicates that they are negative in French, there are yet semantic and syntactic reasons to claim that French N-words are negative.

14 Pas un only occurs in subject positions, un N being the counterpart of de N ‘of.N’ in object position. Is pas in pas .. de N moved to NegP, or is it based generated there ? My analysis of pas un N advocates for the former idea (vs. Doetjes 1997; Ihsane 2006). I leave this open.


35

One of the finest analyses of N-words is Giannakidou’s 1998 on Greek. She argues that Greek N-words are not negative, but ∀Qs scoping over sentential negation. In Greek, N-words are emphatics. She shows that emphatics should be distinguished from nonemphatics which have the same lexical realisation.

(33) I Maria *(dhen) enekrine kanena/ KANENA sxedhio (Greek) the Maria not approved.3sg n. plan ‘Maria didn’t approve any plan’ (kanena-version) ‘Maria approved no plan’ (KANENA-version)

The two items have distinct prosodic and licensing mechanisms. Emphatics, but not nonemphatics (i) bear stress; (ii) cannot appear in non-negative environments (they have to appear in the same clause as the negative marker dhen); (iii) they can appear in subject position above a negative marker and (iv) can be modified by almost; (v) they can appear in fragment answers.

Giannakidou argues that emphatics are dependent (semantically non-negative) items: first, the (semantically) negative marker is always required, i.e., Greek is a strict NC language15; second, in elliptical contexts (fragment answers), the elided material must contain a negation that associates with the N-words when ellipsis is resolved. Hence, emphatics cannot be negative. She claims that they are ∀Qs taking scope over negation, while nonemphatics are ∃ taking narrow scope over negation.

This analysis is inadequate for French: the presence of negation in the elliptical part is impossible, since pas is excluded from the NC-system, (34a). If the full structure is spelled out, only (34b) is grammatical, since the elided material must be identical to its antecedent (Merchant 2001, Watanabe 2004). (34) suggests that personne is intrinsically negative: if the elliptical part contains a negative item, the sentence is ungrammatical itl is interpreted. DN is unavailable in (35) since pas occurs in the elided material, i.e cannot be emphasized (see fn. 15):

(34) a. *Je veux marier Anne et/ou [je ne veux pas marier] personne (d’autre) I want marry A. and/or [I ne want not marry ] nobody (else) b. Je veux marier Anne et/ou [je (ne) veux marier] personne (d’autre) I want marry A. and/or [I want marry ] nobody (else) ‘ I want to marry either anne or I don’t want to marry anybody (else)’ 15 When two N-words co-occur, yielding a positive proposition, Double Negation (DN) arises. A language is a Negative Concord language when the co-occurrence of two or more N-words yields a negative reading. The negative sentential marker must be present in strict NC languages (Greek), whereas it may be absent in non-strict (Italian). Within the non-strict class, languages are not uniform either: French pas is incompatible with NC (de Swart and Sag 2002). When multiple N-words co-occur in the same clause, the unmarked reading is NC, but depending on the intonation, DN can show up (see Corblin and Tovena 2003, Baunaz 2008 a.o.)). Yet DN is extremely difficult with pas, unless it triggers emphatic stress.

Lena Baunaz

36

(35) * Je ne veux pas marier Anne et/ou je ne veux pas marier] personne I ne want not marry A. and/or [I ne want not marry ] nobody

French N-words must be distinguished from Greek on at least three grounds: i) they do not bear stress; ii) French is not a strict NC language: pas ‘not’ cannot co-occur with N-words (de Swart and Sag 2002)16; iii) they are semantically negative since they can appear in fragment answers (36a); they cannot have a non-negative meaning in non-veridical contexts (36b); they are anti-additive (36c) and as such they can license NPIs (36d):

(36) a. A: Whom / Which girl did you see? (Fragment answers) B: Personne/ aucune / *qui que ce soit Nobody / No one / anybody b. Personne /aucun des hommes n’a appelé? (*non-negative / negative) Did nobody telephone? c. personne chantait ou dansait ↔ personne chantait et personne dansait nobody sang or danced ↔ nobody sang and nobody danced d. Personne / aucune des filles n’a vu qui que ce soit (NPIs licensing) Nobody / none of the girls ne has seen anybody

Because N-words fulfil the tests proposed in the literature stating the negative status of a constituent, Mathieu 2002, Déprez 1997, (a.o) conclude that they are semantically negative. What about their quantificational status?

N-words can logically be interpreted either as ∃Qs or as ∀Qs. Conversely, I argue they are composed of a ∀Q combined with negation.

Déprez 1997 argues that French N-words are equivalent to the numeral zero N. As such, they display a strong (universal negation) and a weak reading (existential negation) (cf. (11)). On the weak reading they stay in-situ, under the scope of an Op (negation, conditional etc). Under the strong reading, the indefinite introduces a variable interpreted within the restriction of negation. Strong Dets are specific indefinites undergoing QR: (37) is ungrammatical as a result of weak island violation. Ne being a scope marker indicates where personne is interpreted (i.e., post-QR), quand being an intervener, blocks QR.

(37) *Tu ne te demandes quand voir personne. (French) you NE yourself ask when to see N-word ‘You do not wonder when to see anyone.’ (Déprez 1997:57)

16 This propriety is crucial to distinguish French from Greek: Greek is strictly NC, French is non-strict. Greek emphatics occurs in the same environments as French’s but the elliptical construction test gives different results, which are related to the status of the negative marker in the two languages. Another point of differentiation is the fact that two N-words may create a DN reading in French, if one N-word is emphasized. Only NC shows up in Greek.


37

(i) zero personne is not compatible with clitic ne (38), (ii) zero personne is acceptable in stylistic inversion, but personne is not, (39). Then French N-words are not numerals (Haegeman 1996).

(38) a. Je n’ai vu personne/(*zéro personnes) I NE have seen no one / zero people ‘I did not see anyone’ b. Je n’ai pas vu Marie I NE have NOT seen Marie (Haegeman (1996: 6, (13)))

(39) a. Qu’a donné Jacques à trois personnes/à zero personnes de son groupe ? what has given Jacques to three people/ zero people of his group b. *?? Que (n’) a donné Jacques à personne de son groupe? what (NE) has given Jacques to no one of his group

Just like French ∀Qs (vs. ∃Qs), (i) the scope of N-words is clause-bound; (ii) they can float; (iii) they cannot be interpreted as predicate nominals; (iv) they can appear in exception-clauses (see Corblin et al 2004).

(40) a. * Je n’ai dit que personne/aucun des garçons allait venir17 I ne have said that no one / none of the boys would come b. Une fille a dit que tous les garçons étaient sympas (∃ > ∀) ; *(∀ > ∃) a girl has said that all the boys were nice c. Tous les garçons ont dit qu’une fille était sympa (∃ > ∀) ; (∀ > ∃) All the boys have said that a girl was nice

(41) a. Je n’en ai soulevé aucun

I ne cl. have lifted none ‘I didn’t lift any’ b. Un clown les a soulevés chacun. A clown them has lifted each (Puskás 2002 : 108(8))

(42) a. * Juliette est personne / aucune des filles Juliette is nobody / none of the girls b. *Juliette, Patsy, Edina et Louise sont toutes les amies à moi Juliette, Patsy, Edina and Louise are all the friends of mine c. Juliette est une amie à moi Juliette is a friend of mine

(43) a. Personne/aucun des mecs n’a parlé à personne, sauf Marie à son frère Nobody / None of.the guys ne has talked to anyone, except Marie to her brother b. Tous les mecs ont parlé à toutes les filles, sauf Marie à son frère

17 Note that there is no subject-object asymmetry with respect to LF-raising (vs. Kayne 1981)

Lena Baunaz

38

all the guys have talked to all the girls, except Marie to her brother c. * Une fille a parlé à un mec, sauf Marie à son frère a girl has talked to a guy, except Marie to his brother

Yet, N-words resemble tous les N (vs. chacun des N): (v) tous les N, unlike chacun des N can enter ClLD constructions, (44a) vs. (44b). In certain registers of French, personne can be dislocated (44c) (vs. ∃Qs (44e):

(44) a. Tous les étudiants, je les ai aidés all the students, I them have helped b. * Chacun des étudiants, je l'ai aidé each of the students, I him have helped c. Personne, il fiche rien, à Toulon (Zribi-Hertz 1994) Nobody he does nothing in Toulon d.? Aucun des garçons, je les ai aidés none of the boys, i them have helped e.* Une étudiante, je l’ai aidée a student, I her have helped

(vi) Italian and Greek ∀Qs (vs. ∃Qs) can be modified by the degree adverb

almost (Zanuttini 1991, Giannakidou 1998): in (45a) French N-words pattern like tous les N (45b) (vs. chacun des N (45c) or existential un N (45d)). (vii) N-words, like tous les N can occur as subject of both distributive and collective predicates (46) (vs. chacun des N is obligatory distributive, (46b)).

(45) a. Presque personne / aucune des filles n’a mangé des sushis almost nobody / none of the girls ne has eaten of.the sushis b. Presque toutes les filles ont mangé des sushis almost all the girls have eaten sushis c. ?? J’ai vu presque chacune des filles I have seen almost each of the girls d. * Presqu’ une fille a mangé des sushis almost a girl has eaten of.suchis

(46) a. Tous les enfants se sont rassemblés dans le parc pour manifester all the children ref. are gathered in the park to demonstrate b.* Chacune des filles s’est rassemblée dans le parc pour manifester Each of the girls refl. are gathered in the park to demonstrate c. Personne/aucune des filles ne s’est rassemblé dans le parc Nobody/none of the girls ne refl is gathered in the park d. Personne / aucune des filles n’a entouré le château. Nobody / none of the girls ne has surrounded the castle


39

The tests (i)-(iv) show that N-words and ∀Qs are similar. The tests (v)-(vii) suggest that they behave like tous les N (vs. chacun des N). Because French N-words are i) intrinsically negative, ii) behave like ∀Qs, I conclude that their internal composition is that the order of Ops is ∀ > ¬ (being interpreted as for all x, it is not

the case that x, i.e, reversing the order pas tous exhibits). In the next section, I investigate their internal structure. The parallelism between ∀Qs and N-words is taken seriously, and I show that N-words can (c)overtly realize derivations of the FQ type.

2.5. The structure of N-words: N-words are (covert) FQs

The presence of an overt restriction with ∀Qs allows FQ structures. Aucun takes des N as its associate DP and it can float, (47). (47) is reminiscent of FQ structures involving chacun (48): just like object FQ, en-cliticisation of the moved argument is obligatory, (47b).

(47) a. Je n’ai peint aucune des toiles I ne have painted none of.the paintings ‘I painted none of the paintings’ b. Je n’en ai (*aucune) peinte *(aucune) I ne cl. have (none) painted.fem. (none) ‘I painted none of them’

(48) a. Un clown a soulevé chacun des pianos. a clown has lifted each of.the pianos ‘A clown lifted each of the pianos.’ b. Un clown les a (*chacun) soulevés *(chacun) A clown them has (each) lifted (each) (Puskás 2002: 108(8))

A natural analysis is to assimilate aucun des N to the FQ structures in (48). I claim that aucun des N is essentially composed of an adnominal negative ∀Q and associate DP (des N), (49a). Its internal structure involves FQ, intrinsically negative (∀ plus ¬ lexicalize, resulting in the lexical item aucun

18, where -un realizes pro, the restriction in (49b)), plus a associate DP, des N). En is an effect of genitive under negation and shows up after cliticisation. Strictly speaking, en-cliticisation in (47b) is like les-cliticisation in (48b). I propose that en witnesses Agree of the negative FQ and its associate DP which gets negative through agree with the FQ, much like 18 Why is there no inner island effect in (49)? The answer has to do with the availability of weak Islands (WI) extraction: phrases involving existential presupposition can escape WIs (Starke 2001 for French). En-cliticization can trigger participle agreement in transitive constructions (not all informants like it in (47b), yet). (47b) means that from the set of paintings, I painted none of them, i.e., the set of paintings (the restriction)–to which en refers- is known, and as such carries some presupposition , i.e., no inner island effect arises.

Lena Baunaz

40

agreement arises, (49c). I claim that aucun lexicalizes the ∀Q plus the negative Op, via-Agree, agreement shows up.

(49) a. aucune des toiles : [DP [FQaucunei] [DPdes toilesi]] b. aucun: [QP ∀ [qP ¬ [DP un]]] c. en … aucune : enij … [XP [FQ ¬]j … [DP tj ]]

Not all argument N-words appear with associate DPs, though. I argue that rien ‘nothing’ and personne ‘nobody’ are composed of a FQ and a silent associate DP. In complex tense constructions the bare N-word rien cannot occur post-verbally (unless it is emphasized). Its position in (50) is highly reminiscent of the position of tous in (51):

(50) a. Je (n’) ai *(rien) vu (*rien) I ne have (nothing) seen (nothing) ‘I saw nothing’

(51) Je les ai *(tous) vus (*tous) I cl. have (all) seen (all) ‘I have seen them all’

(50) is easily accounted for, if assimilated to (51). Recall that the presence of FQ is licensed only if FQ can bind an ec. I claim that rien realises the adjoined negative FQ that binds an abstract DP, the ec THING (as in no-thing) in object position (52a), with which it agrees: when rien is in subject position, its structure is similar to full ∀Qs, (52b).

(52) a. rien : [VP [QP ∀ [QP¬] [DP pro]]i … [VP [DP THINGi ]]] b. rien: [DP [QPi ∀ [QP ¬]] [DPi THING]]

The bare N-word personne has a slightly different distribution: it cannot float in standard French (53a). Yet in some varieties of (Swiss) French, personne can appear either in clause medial position, resembling rien in (50), or in argument position (53b):

(53) a. J’ai (*personne) vu *(personne) (standard French) b. J’ai (personne) vu (personne) (varieties of Swiss French) I have (nobody) seen (nobody) ‘ I didn’t see anybody’

Two structures are possible: (i) a negative Full ∀Q (as in (53a)), and (ii) a negative FQ (as in (53b)). Because the complex Q personne is the negative version of tous les N, I propose that it involves a null BODY as its DP-associate (as in no-

body). Either BODY is bound by a negative FQ, witnessing a FQ structure (54b), or the negative Q is adjoined over it, as in full ∀Q structure (54c). Since ‘¬’ is phrasal, i.e., it takes scope over the restriction (pro ̧ the overt restriction) which gets bound by ∀. Default agreement results (3rd.p.sg). So when the FQ construction is covert, the ec involved doesn’t move. What counts is that the associate DP be non-overt.


41

(54) a. personne: [XP [FQ¬]i [XP … [DP BODYi ]]] (FQ, (53b)) b. personne: [XP [QP ∀ [¬][pro]]i [XP …[DP BODYi ]]] (FQ, (53b)) c. [DP [QPi ∀ [QP ¬]] [DPi BODY]] (full DP structure (53a))

N-words are anti-additive and license NPIs in object position: local negation is embedded within the Q structure and doesn’t c-command the NPI. Yet, the complex Q is negative: both ¬ and ∀ are lexicalized, forming a negative ∀Q which itself c-commands the NPI, licensing it. N-words are intrinsically negative, and subsequently move to NegP. The semantics of N-words follow from their structures: the negative import is provided by ¬, which negates and c-commands its restrictor (pro or an overt DP). N-words and complex Qs formed by the overt combination of pas ‘not’ and tous ‘all’ differ in that the former lexicalises both ∀ and ¬ as one Q (aucun; personne; rien), whereas the latter lexicalise two Qs. In pas tous, pas does not c-command the NPI and is not negative.

3. Conclusion

Taking into account syntax and semantics, I have elaborated a new and detailed characterization of ∀Qs and N-words in French. I have shown that ∀Qs are best analysed as adjoined to a maximal projection. Witnessing that N-words successfully pass the tests for negativity discussed in the literature, I argued for their negative status. As for the Q status of N-words, I have shown that they involve ∀ that adjoins to an XP. I have also argued that both ∀Qs and N-words are (potential) FQ structures and that they are clause-bound as a result of their internal structure, their (overt or covert) associate DP undergoing A-movement, a clause-bound movement.

Bibliography

Baunaz 2008. Split-DP and Floating Quantifiers: a Syntactic Approach to French

Quantification. University of Geneva. Beyssade, C. 2006. “la polarité”, séminaire transversal UPMF Grenoble II. Bobaljik, D. 2003. “Floating quantifiers: Handle with care”. In Cheng and Sybesma (eds), The

Second Glot International State-of-The-Article Book, 107-148. Berlin: Mouton de Gruyter.

Corblin, F. et al. 2004. “Negative Concord”, in Corblin and de Swart (eds), Handbook of

French Semantics, CSLI, 427-461. Déprez, V. 1997. “Two Types of Negative Concord”, Probus 9: 103-143. Doetjes, J. 1997. Quantifiers and selection. On the distribution of quantifying expressions in

French, Dutch and English. Doctoral dissertation, HIL, Leiden University. The Hague: HAG.

Dowty, D. and B. Brodie. 1984. “The Semantics of Floated Quantifiers in a Transformational Grammar”. Proceedings of WCCF 3,75-90.

Lena Baunaz

42

Fitzpatrick, J.M 2006. Syntactic and semantic routes to floating quantification. Doctoral dissertation, MIT, Cambridge, Mass.

Giannakidou, A. 1998. Polarity Sensitivity as (Non)veridical Dependency. John Benjamins. Giusti, G. 1997. “The categorial status of determiners”. In Haegeman, L. (ed), The New

Comparative Syntax, 95-123. London & NY: Longman. Ihsane, T. 2008. The Layered DP: Form and meaning of French indefinites.Linguistik Aktuel:

John Benjamins. Haegeman, L., 1995. The Syntax of Negation. Cambridge: CUP. Haegeman, L., 1996. “The Syntax of N-words and the NEG-Criterion”, in GenGenP 4.1: 1-

14. Kayne, R. 1975. French Syntax. MIT Press, Cambridge, MA. Ladusaw, W., 1992. “Expressing negation”, in Barker, C. & D. Dowty (eds). SALT II

Proceedings. Columbus, Ohio: The Ohio State University. Mathieu, E. 2002. The Syntax of Non-Canonical Quantification: A Comparative Study.

Doctoral Dissertation, UCL. Merchant, J, 2001. The syntax of silence: Sluicing, islands, and the theory of ellipsis. Oxford

University Press: Oxford. Obenauer, H. 1994. Aspects de la Syntaxe A-Barre, Thèse de Doctorat d’Etat, Université de

Paris VIII. Puskás, G. 2002. “Floating Quantifiers: what they can tell us about the syntax and semantics

of Quantifiers”, in GG@G 3: 105-128. de Swart, Henriette & Ivan Sag 2002. “Negation and negative concord in Romance”,

Linguistics and Philosophy 25: 373-417. Sportiche, D.1988. “A Theory of Floating Quantifiers and its Corrolaries for Constituent

Structure”, LI. Tovena L et al. 2004. “Polarity Sensitive Items”. In F. Corblin and H. de Swart (eds.),

Handbook of French Semantics, 403-411. CSLI. Watanabe, A. 2004. “The Genesis of Negative Concord: Syntax and Morphology of Negative

Doubling”, in LI 35.4: 559-612. Zanuttini, R. 1991. Syntactic Properties of Sentential Negation: a Comparative Study of

Romance Languages, PhD. Dissertation, UPenn. Zribi-Hertz, A., 1994. “La syntaxe des clitiques nominatifs en français standard et en français

avancé”. Travaux de Linguistique et de Philologie: 131-147. Strasbourg-Nancy: Klincksieck.

Zwarts, F., 1998. “Three types of Polarity”, in Plurality and Quantification. F. Hamm and E. Hinrichs (eds), 177-238.


33 – 2008

43

FEATURE STRIPPING AND WH-MOVEMENT

IN FRENCH AND HUNGARIAN∗

Lena Baunaz, Genoveva Puskás

1. Introduction

One of the main characteristics of wh-movement is that it is traditionally split into overt and covert movement. Whereas overt movement involves moving a whole wh-phrase, or at least a wh-word, covert movement may involve either phrasal movement of wh-chunks of variable sizes or feature movement (Pesetsky 2000). However, all of these movements are considered as uniform. In this paper, we argue in favor of a ‘dynamic’ approach to wh-movement. More precisely, we propose that covert wh-movement can transform in the course of the derivation. We label this phenomenon feature stripping, in that in the course of the derivation, a wh-phrase gets rid of a feature in order to move higher. Essentially, we claim that, in French, a wh-phrase in situ moves to reach its scope position covertly. In the course of this derivation, it undergoes feature stripping. We extend this analysis to Hungarian partial wh-movement.

2. Feature Stripping

French wh-questions are well-known to show a strikingly great number of variants, as attested in the paradigm below:1

(1) a. Qui as-tu vu? who have-you seen

∗ We thank the audiences of the 34th IGG and of the 1stWH-orkshop in Geneva. We are grateful for the many useful, challenging and encouraging questions and comments we received from Andrea Cattaneo, Marcel den Dikken, Mélanie Jouitteau, Eric Mathieu, Jean-Yves Pollock, Luigi Rizzi, Ur Shlonsky, Gabi Soare and Michal Starke. 1 Except for (1a), which is by now rather formal, all the other variants, which occur in spoken French, sound more natural with the contracted form t’as rather than the full form tu as (‘you have’). Note that (1a) displays genuine wh-movement with I-to-C movement comparable to standard English wh-movement.


44

b. Qui t’as vu? who you-have seen c. Qui est-ce que t’as vu? who is-that that you-have seen d. Qui c’est que t’as vu? who that-is that you-have seen e. C’est qui que t’as vu? that-is who that you-have seen f. T’as vu qui? you-have seen who ‘Who did you see?’

Of these six forms, only the last one involves no overt movement at all of the wh-phrase. The literature on French wh-questions has discussed at length the differences between these forms, essentially from the point of view of the differences in meaning that they support (see Obenauer 1994, Chang 1997, Boeckx 1999, Cheng & Rooryck 2000, Mathieu 2004 a.o.). Wh-in situ has been associated with some version or other of D-linking à la Pesetsky (1987). However, as discussed in Starke (2001) and shown at length in Baunaz (2008), French wh-in situ is ambiguous. Baunaz claims that wh-in situ occur in thee different versions, associated with three intonational profiles, and which exhibit three different syntactic behaviors. She claims that only presuppositional types of wh-in situ are interpreted as having wide-scope, even when occurring in an island. These are the wh-in situ that we are concentrating on in this paper.

2.1. The in-situ strategy

In addition to the well-known root phenomenon (2a), the in situ strategy is also widely attested in embedded contexts (2b):

(2) a. Jean a insulté qui? Jean has insulted who ‘Who did Jean insult?’ b. Tu crois qu’on les trouve où ? you think that one them finds where ‘Where do you think one finds them?’

As discussed in Starke (2001), Baunaz (2008), the wh-phrases in (2) have wide-scope if they are interpreted as having some presuppositional import.2

2 In fact, both Starke and Baunaz argue in favor of a three way distinction, including a non-presuppositional version and two types of presuppositional versions for the wh-phrase. However, the more subtle distinction between the two types of presupposition is irrelevant

Feature stripping and wh-movement in French and Hungarian

45

Presuppositionality has, among other properties, that of enabling a wh-phrase to escape weak islands. This is attested in overt wh-movement:

(3) a. ??/*Qui n’as-tu pas vu? who ne have-you not seen ‘Who didn’t you see?’ b. Quel film n’as-tu pas vu ? the-which film ne have-you not seen ‘Which film didn’t you see?’

Whereas qui is preferably interpreted as an indefinite, quel film (‘which film’) presupposes the existence of a set of films. Wh-movement over negation is fine in this case. The same is valid for in situ wh-phrases:

(4) a. Elle a pas vu quel film? she has not seen which film ‘Which film has she not seen?’ b. *Elle a pas vu qui diable? she has not seen who devil ‘Who the hell has she not seen?’

Here again, quel film (‘which film’) carries a presupposition of existence, and is fine in the context of a negative sentence (4a). The expression qui diable (‘who the hell’) is aggressively non-D-linked (see Pesetsky 1987, den Dikken &Giannakidou 2002, Obenauer 1994, Poletto&Pollock 2004). As illustrated in (4b), it is ungrammatical in a negative context. This leads Baunaz (2008) to conclude that in-situ wh-phrases undergo covert movement. In addition, as shown in (4), only presuppositional wh-phrases can escape neg-islands. We adopt Baunaz (2008) and assume that a presuppositional wh-phrase carries a feature [+presupp].

2.2. Stripping

The above discussion leads to a strong prediction, namely that a wh-phrase with a feature [+presupp] will escape weak islands, wherever they occur. However, the contrast in (5) below invalidates the prediction:

(5) a. Tu crois [qu’elle a pas vu quel film]? you think that she has not seen which film ‘Which film do you think she has not seen?’ b. *Tu crois pas [qu’elle a vu quel film]?

here, and we will simply use the term presuppositional for those wh-phrases which attain wide-scope and can escape islands.


46

you think not that she has seen which film (= which film don’t you think has seen)

The contrast between (5a, b) is unexpected, given that quel film (‘which film’) is [+presupp]. On the other hand, the ungrammaticality of (5b) reveals some sensitivity to negative islands, a characteristic of (some types of) wh-movement. We are facing here a curious case of incompatibility: on the one hand, a presuppositional wh-phrase can escape weak islands, as indeed illustrated in (5a); but on the other hand, the same wh-phrase is blocked by a higher neg-island, a rather unexpected fact.3

The same asymmetry appears with overt wh-movement:

(6) a. C’est quoi que tu n’as pas acheté ? that is what that you ne have not bought ‘What is it that you didn’t buy?’ [Mathieu 2002: 61] b. *Ce n’ est pas quoi que tu as acheté? that is ne is not what that you have bought (=What is it not that you have bought?)

In (6), quoi (‘what’) moves overtly. In (6a), it can escape the neg-island because it is presuppositional.4 In (6b), however, the presence of pas in the matrix clause leads to ungrammaticality. Note that the overt landing site of the wh-phrase is below the negative marker, so that the latter is not expected to intervene. However, another wh-strategy reveals that the same asymmetry holds throughout:

(7) a. Qui c’est [que t’as pas vu ]? who that-is that you-have not seen ‘Who is it that you didn’t see?’ b. *Qui c’est pas [que t’as vu ]?

3 A reviewer suggests that the ungrammaticality of (5b) might be simply due to “neg-raising”, with pas raised to the matrix clause but interpreted as an embedded negation. However, as the core of our analysis lies in the properties of the (wh)-movement types, whether the negative bare operator syntactically occurs in the matrix clause as the result of base-generation or movement is contingent to the problem. 4 We assume that quoi in wh-clefts is base-generated in the embedded clause and not as an in situ relative (see Baunaz 2008). Crucially, we consider c’est and quoi as two separate constituents, a fact which enables quoi to move covertly beyond c’est. The independence of c’est and quoi can be indirectly shown by the following: (i) c’est pas Mike qu’elle a embrassé it is not Mike that-she has kissed ‘It is not Mike that she kissed’ Example (i) shows that c’est and Mike are two different constituents which can be separated by an adverbial element, assuming that wh- and focus-clefts have an identical structure.


47

who that-is not that you-have seen (=who is it not that you saw?)

In the ‘qui c’est’ strategy, we observe an overt movement of qui to the edge of the matrix clause, most likely to the scope position of the wh-phrase.5 As illustrated in (7a), the movement is not sensitive to an embedded neg-island. The presence of the negative marker in the matrix clause, on the other hand, blocks the movement of the wh-phrase to its scope position. Putting the three pairs of data in (5), (6) and (7) together, we reach the conclusion that what blocks overt movement in (7b) is the same phenomenon which blocks access to a scope position via partial covert movement in (6b) and covert wh-movement in (5b). In all of these cases, movement out of an embedded island is fine, while crossing an intervening negation in the matrix clause leads to unacceptability. Assuming that movement is cyclic, we can reformulate the problem in the following way: the first step is insensitive to weak islands, as is expected for wh-phrases which carry a [+presupp] feature. The second step, however, exhibits the sensitivity to weak islands characteristic of [-presupp] wh-phrases.

So the data in (5,6,7) suggest that the second step of the cyclic process is different from the first one. We propose that the blocking effect of negation in the second step is due to the fact that in the course of the derivation, the wh-phrase loses its presuppositional properties: it strips off its [+presupp] feature and moves on as a ‘naked’ Q.

What we identify as “feature stripping” is the extraction of a bare wh-operator out of a phrase which bears a [+presupp] feature. Therefore, we claim that there may be two types of movement involved in the wh in-situ construction:

(i) a “heavy” covert XP-movement , which moves the wh-phrase to an edge position in French, as a presuppositional element:

(8) a. [ [wh-phrase]i…. ti]

(ii) a “stripped” Q-movement; further movement is then peeled off its presuppositional layer and can only carry on as bare Q movement. The idea is that presuppositional movement is restricted to (relatively) local movement:

(8) b. wh-Qj………. [ [wh-phrase tj]i…. ti] 5 Note that ‘qui c’est’ and ‘qui est-ce que’ are two different constructions. While the former is necessarily presuppositional, the latter is not. This is illustrated by the contrast with aggressively non-D-linked qui diable:

(i) a. Qui diable est-ce que t’as vu ? who devil is-this that you-have seen ‘Who the hell did you see? ‘ *qui diable c’est que t’as vu ? who devil it-is that you-have seen


48

The wh-phrase is able to escape from a neg-island, if the latter occurs in the portion of the derivation where the phrase still carries its [presupp] features; if negation occurs in a portion of the clause in which the wh-phrase has already been stripped off its feature, the movement will be blocked, as is expected of a bare Q-movement:6

(9) [CP wh-Qj …*NEG … [CP [wh-phrase tj ]i …NEG… ..ti ]

2.3. Further application

Let us now turn to another case where the in-situ strategy differs strongly from overt wh-movement. Wh-movement is also known to be sensitive to Strong Islands:

(10) a. Quii crois-tu [que Jean aime ti ]? who think-you that Jean loves ‘Who do you think that Jean loves?’ b. * Quii pleures-tu [parce que Jean aime ti ]? who cry-you because Jean loves (=who do you cry because Jean loves?)

On the other hand, a wh-phrase in-situ can be interpreted with matrix scope despite the fact that it occurs in a Strong Island (see Starke 2001):

(11) Tu crois [qu’elle a dit ça [ pour inciter Pierrot à séduire qui ]]? you think that-she has said that to incite Pierrot to seduce whom? (=whoi do you think she said that to incite Pierrot to seduce ti ?) [Starke 2001]

However, the wh-phrase in-situ can only be interpreted as presuppositional, as attested by the following contrast:

(12) a. Elle a ri [ parce que Jean s’est moqué de qui ?] she has laughed because Jean has made fun of who (=who did she laugh because Jean made fun of?)

6 Note that French exhibits some version of overt "stripping", not necessarily linked to presupposition: (i) a. Tu penses qu’elle a acheté combien de livres? you think that -she has bought how-many of books ‘How many books do you think that she bought?" Combien tu penses qu’elle a acheté de livres? How-many you think that-she has bought of books However, stripping already applies to the lower clause, leaving the restriction in-situ and moving a bare Q up to the matrix CP. This is attested by (ii): (ii) *Combien tu penses qu’elle a pas acheté de livres? How-many you think that-she has not bought of books.


49

b. *Elle a ri [parce que Jean s’est moqué de qui diable?] she has laughed because Jean has made fun of who devil

Recall that in situ wh-phrases have a non-presuppositional version (see fn 2). We expect (12b) to be grammatical, on a par with (12a). However, the aggressively non-D-linked qui diable (‘who the hell’) wh-phrase is not licensed within a strong island, showing that the wide scope reading is only accessible to a [+presupp] wh-phrase. In other words, the interpretation of (12a) presupposes the existence of the individual known to the speaker, or at least of a set from which the individual is selected.7

We claim that the wide scope interpretation obtains via covert movement, even in the case of strong islands. Indeed, the presence of weak islands reveals the same kind of asymmetry as the one that was discussed in section 2.2 above. Whereas the wide scope reading of a [+presupp] wh-phrase is not blocked by negation in the adjunct clause (13a), it is not legitimate in matrix negative contexts (13b):

(13) a. Elle a ri [parce que Jean n’a pas aimé quel film ? ] she has laughed because Jean neg has not liked which film =For which film is it the case that she laughed because Jean didn’t like the film? b. *elle a pas ri [parce que Jean a aimé quel film]? she has neg laughed because Jean has liked which actor

If quel film (‘which film’) is [+presupp], (13b) is unexpected. But at the same time, this shows that there is some movement, since we have taken sensitivity to weak islands to be a diagnosis for wh-movement. Clearly, it seems unreasonable to claim that there is no movement in the adjunct clause, but that there is movement to the matrix scope position. On the other hand, if there is covert movement of a presuppositional wh-phrase, there is no explanation as to why it is insensitive to negation within the adjunct clause but not outside it.8

We propose that (12a) and (13) involve two movements: a first step involves the +presuppositional wh-phrase:

(14) a. [ parce que [wh-phrase]i …NEG…..ti ]

The second step is an instance of feature stripping: the wh-phrase gets rid of its [+presupp] feature. It moves further as a bare Q. As such, it is sensitive to neg-islands:

(14) b. *Wh-Qj…..NEG….[ parce que [[wh-phrase tj]i …NEG…..ti ]

Note that we argue for an analysis in which the wh-in situ undergoes covert movement throughout to reach its scope position. This might be challenged by an alternative analysis. One might want to claim that the movement of the [+presupp]

7 On these differences in the interpretation of presuppositional wh-phrases, see Baunaz 2008. 8 See Starke 2001 for a similar observation on strong island extraction.


50

wh-phrase is restricted to the adjunct clause. In other words, the wh-phrase moves covertly to each the edge of the adjunct clause, where it has scope over the clause. Then, the whole clause moves up covertly to the matrix scope position. This would then give the illusion that the wh-phrase has moved to the matrix scope position. Such an analysis has been proposed for Hungarian (see section 3 below). However, on the basis of examples such as (12b), we reject this approach. Indeed, if qui

diable, an aggressively non-D-linked wh-phrase occurs in an adjunct clause, it can never have wide scope. Given the clausal movement analysis, it could move to the scope position of the adjunct clause as a non-presuppositional element (as it does overtly, see 15 below). Then, the clause would move as a whole to the matrix scope position, giving the illusion that the non-D-linked wh-phrase has wide scope. But this is not the case: the in-situ version cannot have wide scope, contrary to moved versions:

(15) Qui diable as-tu vu? who devil have-you seen ‘Who the hell did you see?’

2.4. Why to strip?

The question we investigate in this section is the motivation for the feature stripping we propose. Using the same tool, namely negative islands, a diagnosis for bare Q movement, we observe that not all environments trigger identical restrictions on covert movement. Consider the following:

(16) a. Avec quel artiste t’as dit [que Marie avait déjeuné t ]? with which artist you-have said that Marie had had-lunch ‘With which artist did you say that Mary had lunch?’ b. ??Avec quel artiste t’as pas dit [ que Marie avait déjeuné] ? with which artist you-have not said that Marie had had –lunch (=with which artist did you not say that Mary had lunch?)

(17) a. Avec quel artiste il a admis [ que Marie avait souvent travaillé t ]? with which artist he has admitted that Marie had often worked ‘With which artist did he admit that Marie had often worked?’ b. Avec quel artiste il a pas admis [que Marie avait with which artist he has not admitted that Marie had souvent travaillé t ]? often worked ‘With which artist did he not admit that Marie had often worked?’

The contrast in (16, 17) comes as rather unexpected, given our previous discussion. Indeed, we had observed that when a [+presupp] wh-phrase, such as avec quel artiste (‘with which artist’) occurs in an embedded clause, it is insensitive


51

to the presence of negation within the embedded clause, but is blocked by matrix negation (see 5 above). And this is what (16) confirms. But (17), which embeds the same kind of clause, allows the wh-phrase to have matrix scope even in the presence of matrix negation. We propose that the crucial difference is linked to the nature of the matrix predicate. Indeed, as already proposed by Cattell (1978), complements of factive predicates are presupposed, while complements of non-factives are asserted. As also observed in Horváth (1997), verbs vary as to the kind of clausal complement they select. She claims that some verbs select a ‘D-linked’ complement while others select an ‘open-ended, non D-linked’ clause. We observe, for French, a similar distinction, while giving the phenomenon a slightly different analysis. A verb like croire (‘think’) does not presuppose the content of the embedded clause.9 The use of a verb like croire expresses that the subject of the matrix clause has no guarantee that what is expressed in the embedded clause is an actual state of affairs in the world. In contrast, a verb like admettre (‘recognize, admit’) presupposes the content of the embedded clause. The following contrast can illustrate this:

(18) a. Tu crois que [Jean a parlé avec quel acteur?] you think that Jean has spoken with which actor ‘Which actor do you think that Jean spoke with?’ b. Le plus connu: Robert de Niro! the most famous: R.de.N. c. Avec aucun, il est bien trop timide! With none, he is far too shy

(19) a. T’ admets que [Jean a parlé avec quel acteur?] you-admit that Jean has spoken with which actor ‘Which actor do you admit that Jean spoke with?’ b. Le plus connu: Robert de Niro! c. #Avec aucun, il est bien trop timide!

Whereas (18c), which denies the content of the embedded clause, is consistent with a question like (18a), such an answer is unacceptable with (19a). We conclude that the two predicates come with different expectations as to their embedded clauses. Croire (‘think’) does not presuppose anything about the embedded clause, whereas admettre (‘admit’) presupposes that the proposition described in the embedded clause actually holds, that is, corresponds to some state of affairs in the world. This is what we call ‘presuppose’ in the context of an embedded clause.

To summarize, we observe that in contexts where the matrix predicate selects a presuppositional embedded clause, matrix negation does not block covert movement

9 This is true at least for one version of croire which means thinking that P, but not being sure that P is true.


52

of a wh-in situ. Context where the predicate selects a non-presuppositional clause will be islands to covert wh-movement. Other examples in (20) through (23) further illustrate the point:

(20) a. Tu crois que [Jean a pas invité quelle célébrité]? you think that Jean has not invited which celebrity ‘Which celebrity do you think that Jean did not invite?’ b. T’admets que [Jean a pas invité quelle célébrité]? you-recognize that Jean has not invited which celebrity ‘Which celebrity do you recognize that Jean did not invite?’

(21) a. *Tu crois pas [que Jean a invité quelle célébrité ] ? you think not that Jean has invited which celebrity (=which celebrity don’t you think that Jean invited?) b. T’admets pas [que Jean a invité quelle célébrité] ? you-recognize not that Jean has invited which celebrity ‘Which celebrity don’t you recognize that Jean invited?’

(22) a. Tu penses [qu’elle avait pas vu quel film ]? you think that she had not seen which film ‘Which film do you think that she hadn’t seen?’ b. Il a avoué [qu’elle avait pas vu quel film]? he has admitted that she had not seen which film ‘Which film did he admit that she hadn’t seen?’

(23) a. *Tu penses pas [qu’elle avait vu quel film]? you think not that she had seen which film (= which film don’t you think that she had seen?) b. ?Il a pas avoué [qu’elle avait vu quel film ]? he has not admitted that she had seen which film ‘Which film did he not admit that she had seen?’

All the examples in (20-23) contain a [+presupp] wh-phrase. As such, the wh-phrase raises past the embedded negation (recall that we have argued that [+presupp] wh-phrases are insensitive to the presence of negation). While (21a) and (23a) exhibit typical cases of feature stripping, covert movement being blocked by matrix negation, (21b) and (23b) do not. The wh-phrase reaches the matrix scope position despite the presence of a neg-island. We conclude that wh-phrases undergo feature stripping when they occur in a clause which is non-presuppositional. Note that it is not a phenomenon that might be attributed solely to the nature of the predicates. A non-presuppositional wh-phrase cannot escape matrix neg-islands, even with a verb like admettre:


53

(24) *Avec qui diable il a pas admis [que Jean a déjeuné ]? with who devil he has not recognized that Jean has had-lunch (=with who the hell did he not recognize that Jean had lunch?)

Recall that extracting from an adjunct island across negation is also impossible (see 13b). We attribute this behavior precisely to the fact that adjunct islands are, by definition, non-presuppositional. Therefore, the split is not between selected versus non-selected (adjunct) clauses, but rather between presuppositional versus non-presuppositional clauses.

We claim that the absence of a presuppositional component in the embedded clause will trigger feature stripping. The mechanism we argue for is the following.

As discussed above, the type of predicate will determine the nature of the embedded clause. Some verbs, like admettre (‘admit’) avouer (‘admit’) etc, will license a +presuppositional clause. We propose that the feature is encoded on C0, in the form of an interpretable feature P which is unvalued. We adopt Pesetsky and Torrego’s (2004) approach (henceforth P&T) and assume that valuation and interpretability are two independent concepts. We also adopt the notation they propose, namely that unvalued features are signaled by bare square brackets. C will then have iP [ ]. C also carries an uninterpretable Q feature which is unvalued, uQ [ ], as proposed in P&T for cyclic wh-movement. The wh-phrase in situ bears an uninterpretable Q feature with the value +interrog, as well as an interpretable feature P with the value +presupp.10 The unvalued Q and P features on C0 receive a value by Agree. However, in this phase, the uninterpretable Q features have not been deleted. The wh-phrase moves to the next phase, which contains the C head of the matrix clause. The latter bears a feature iQ [ ]. By Agree with the wh-phrase, the feature is assigned a value (see P&T for a discussion of successive Agree relations). The uninterpretable Q feature of the wh-phrase is finally checked off against the interpretable Q feature of matrix C. As for the P feature on the wh-phrase, it values the unvalued P feature on C. But the feature being interpretable, no feature checking actually takes place, and the wh-phrase retains its full P feature specification. It is as a full [+presupp] wh-phrase that it enters the higher phase, and is thus able to ignore a neg-island. For a visual aid to the process, we propose a schematic representation à

la P&T:

(25) a. _____________________ __________________Agree

↓ ↓

C0……admettre…. C0 …………… wh-phrase

iQ [ ] uQ [ ] uQ +interrog ⇒ 10 On the legitimacy of assigning an interpretable presuppositional feature to wh-phrases, see Baunaz 2008.


54

iP [ ] iP +presupp

b. _____________________ __________________

↓ ↓ ↓

C……admettre…. C …………… wh-phrase

iQ [1 ] uQ [1 ] uQ +interrog [ 1]

iP [2] iP +presupp [2]

Let us now turn to the feature stripping case. As in the previous case, the wh-phrase bears an uninterpretable Q features valued for +interrog., as well as an interpretable P feature valued for +presupp. The intermediate C0 head also bears an uninterpretable Q feature which is unvalued. However, because the matrix predicate is of the type croire, C bears an uninterpretable P feature, which is unvalued. Unvalued features will act as probes and Agree values the Q and P features. The Agree operation will be represented as follows:

(26) a. _____________________ __________________Agree

↓ ↓

C……croire…. C …………… wh-phrase

iQ [ ] uQ [ ] uQ +interrog ⇒

uP [ ] iP +presupp

b. _____________________ __________________

↓ ↓ ↓

C……admettre…. C …………… wh-phrase

iQ [1 ] uQ [1 ] uQ +interrog [ 1]

uP [2] iP +presupp [2]

The difference between the two cases illustrated in (25) and (26) lies in the nature of the P feature. In (26), the P feature on C0 is an uninterpretable feature. The wh-phrase moves cyclically up to the matrix C. In the process, it checks off the uninterpretable P feature of the intermediate C. We assume that feature checking makes features inactive.11 The wh-phrase thus checks off the +presupp feature and

11 This can be illustrated with agreement features. Once the features have been checked, they are not active anymore and seem invisible for further operations: (i) ??Quels artistes tu as dit qui t vont venir ?


55

moves on – it still has an uninterpretable Q feature to check – as a bare, non-presuppositional Q.

The last question which remains to be discussed is that of the stripping itself. A simple assumption could have been that since a feature is checked, the wh-phrase does not have to – or even is not allowed to – be active in a higher phase. Actually, as is already suggested in the discussion, we view the wh-phrase as a bundle of features, which are hierarchically organized. In a cartographic approach, we propose the following simplified structure for a presuppositional wh-phrase:

(27) PresuppP

3

[+presupp] QP

3

[+Qwh] DP

4

When the structure reaches the intermediate C0 and checks the P feature, the top layer strips off the structure and leaves the lower layers "bare", as a QP. This QP moves up to the matrix C, to check off its uninterpretable Q feature. See also Starke 2005 who independently proposed a similar mechanism for case-feature peeling.

3. Hungarian partial wh-movement

The feature stripping mechanism discussed above turns out to have another application, in a language which does not have wh-in situ. As Hungarian cannot leave a wh-phrase in-situ, the most straightforward strategy is overt movement which forces the wh-phrase to appear in its scope position. However, the language also has a partial movement strategy, whose constraints are similar to those of French wh-in situ, in that it appears, among others, as a strategy to escape strong islands. We propose that the same feature stripping mechanism is at play in the Hungarian partial wh-movement, and that the second part of the ‘partial movement’ is actually covert bare Q-movement.

Which artists you have-2sg said that-agr will come ‘Which artists did you say will come?’ (ii) *Quels artistes ont dit qui t vont venir? Which artists have-3pl said that-agr will come See also Rizzi & Shlonsky 2007.


56

3.1. Partial wh-movement

As mentioned above, Hungarian wh-phrases move overtly to their scope position:

(28) a. Kinek telefonált Mari ? who-dat telephone-pas-3s Mari-nom ‘Who did Mari call?’ b. Kinek gondoltad [hogy telefonált Mari t ] who-dat think-pas-2s that telephone-pas-3s Mari-nom ‘Who did you think that Mari called up?’

The language also has a strategy in which the wh-phrase moves up to an intermediate position, the embedded scope position, while an expletive wh-element appears in the matrix scope position:

(29) Mit gondolsz, [ hogy kineki telefonált Mari ti ]? WH think-pres-2s that who-dat telephone-pas-3s Mari-nom ‘Who do you think that Mari called up?’

Partial wh-movement is optionally available in many contexts. We come back to the relevant cases below. What is of particular interest, though, is that partial wh-movement typically enables to by-pass strong island restrictions. While full wh-movement is sensitive to strong islands, the partial wh-movement construction is not:

(30) a. *Kiveli vagy dühös [mert találkoztál ti] ? who-with are-you angry because you-met b. Miért vagy dühös [ mert kiveli találkoztál ti] ? Why-expl are-you angry because with-whom you-met [Horváth 1997]

Horváth (1997) analyses the partial wh-strategy in the following way: the contentful wh-phrase moves to the spec of the embedded CP;12 there is gets rid of its wh-feature, which percolates up to the CP. The CP thus “inherits” the [+wh] specification and becomes a +wh clause. The matrix clause contains an expletive element with a [+wh] morphological form which moves to the focus position, where it formally satisfies the wh-criterion and indicates the scope of the wh-element. Expletive replacement moves the whole [+wh] CP to the matrix scope position, giving the illusion that the embedded wh-phrase has matrix scope. This is what

12 Note that standard analyses of wh-movement in Hungarian claim that it targets a position FocP below CP (see a.o. Brody 1990, Puskás 1992). On the other hand, Lipták (2001) proposes a two step procedure, with overt movement to FocP followed by covert movement to specCP.


57

explains the apparent Strong Island violation in (30b): the contentful wh-phrase does not leave the island; it is the island as a whole which moves to the matrix scope position.

3.2. Constraints on the partial wh-strategy

We claim that the partial wh-strategy is more constrained than it seems. Consider the following:

(31) a. Mit hitt Mari, [ hogy kivel nem találkoztál?] WH think-pas-3s Mari-nom that who-dat neg meet-pas-3s (=who does Mari think that you did not meet?) b. Miért sírtál [ mert kivel nem találkoztál?] WH cry-pas-2s because who-instr neg met-pas-2s (=who did you cry because you didn’t meet?)

Recall that only presuppositional wh-phrases can escape a neg-island. Given that both (31a) and (31b) are fine, it seems that the wh-phrase occurring in islands in a partial wh-strategy is presuppositional.13 This is confirmed by the following:

(32) a. *Mi a fenét nem vettél? What the devil-acc neg buy-pas-2s (=what the hell didn’t you buy) b. *Miért sírtál [ mert ki a fenéveli találkoztál ti] ? Wh cry-pas-2s because who the devil-instr meet-pas-2s (=who the hell are you angry because you met?)

As expected, the aggressively non-D-linked expression ki/mi a fene (‘who/what the hell’) cannot escape a neg-island (32a). But, as illustrated in (32b), it is not licensed in a partial wh-construction either. So we conclude that the partial wh-construction with an apparent Strong Island violation involves a presuppositional wh-phrase, very much like the French wh-in situ strategy.

The second parallelism we can draw relates to the presence of a negative marker in the matrix clause. Horváth notes the following contrasts:

13 (31b) might not be straightforwardly acceptable to speakers of Hungarian. As in French, the bare wh-phrase is ambiguous between the non-presuppositional and the presuppositional readings, and it requires a special intonation to be acceptable as a presuppositional element. However, the acceptability of (i), an unambiguously presuppositional wh-phrase, confirms our claims: (i) miért sírtál [ mert melyik szinésznővel nem találkoztál?] WH cry-pas-2s because which actress-instr neg met-pas-2s (=which actress did you cry because you didn’t meet?)


58

(33) a. *Mit nem gondolsz [hogy kivel beszélt Mari ?] WH neg think-pres-3s that who-instr speak-pas-3s Mari-nom b. Mit nem ismert be János, [hogy hányszor WH neg admit-pas-3s part János that how-may-times hamisította az aláírásodat? forge-pas-3s the signature-2poss-acc (=how many times did Janos not admit that he forged your signature?)

Horváth concludes that the main predicate plays a role in the behavior of the partial wh-construction. Verbs like beismer (‘confess’, ‘admit’), as well as elárul (‘reveal’), tagad (‘deny’), észrevesz (‘notice’), megenged (‘permit’) select what she calls a “D-linked” complement, while verbs like gondol (‘think’) but also mond (‘say’), hall (‘hear’), érez (‘feel’) select a non-D-linked complement. We have translated this into a distinction between presuppositional and non-presuppositional complements. So, as in French, only presuppositional complements can extract over matrix negation.

We also observe that matrix negation blocks the partial wh-strategy from adjunct islands:14

(34) *Miért nem sírtál [mert kiveli találkoztál ti]?

WH neg cry-pas-2s because who-instr meet-pas-2s (= Who did you not cry because you met?)

14 While the negative counterpart of Horváth’s example (30b above) is also ungrammatical, as illustrated in (i), a reviewer observes that a positive adjective as in (ii) does not give rise to these intervention effects, questioning thus the parallel with French: (i) *Miért nem vagy dühös [mert kivel találkoztál] ? WH neg be-2s angry because who-instr meet-pas-2s ‘Why aren’t you angry because you met whom’ (ii) Miért nem vagy vidám [mert kivel találkoztál] ? WH neg be-2s happy because who-instr meet-pas-2s ‘Why aren’t you happy because you met whom’ However, we observe that the same applies to French (iii) and (iv) (iii) *Tu n’est pas fâché [parce que tu as rencontré quelle actrice?] You are not angry because you met which actress (iv) Tu n’es pas heureux [parce que tu as rencontré quelle actrice ?] You are not happy because you met which actress We think that the examples with BE+ predicative adjective are special, as they involve some kind of ‘internal’ or constituent negation (see e.g. Giannakidou 1998 for a discussion of these). The negation is not sentential, as it cannot deny the fact of ‘being’, but rather amounts to asserting a state of negative happiness (i.e. unhappiness). The details of such an analysis as well as why exactly there is a contrast between (i) and (ii) and between (iii) and (iv) remain to be explained, but are beyond the scope of this paper.


59

Given the striking parallelism between the French in-situ strategy and the Hungarian partial wh-movement, we would like to propose, contra Horváth (1997), that it is not the whole CP which moves covertly. Such an analysis could not account for the ungrammaticality of (32b). Rather, the same mechanism of feature stripping applies. The difference, on the other hand, resides in the fact that Hungarian has no in situ wh-construction. Some element must raise overtly to the scope position. The presuppositional feature is able to pied-pipe the whole wh-phrase to the embedded scope positions. But once it has stripped only feature-movement can take place. This is by essence covert movement. But as the matrix scope position needs to be filled overtly by expletive material, the expletive wh-element appears in the matrix clause. The covert features will raise to the expletive, accounting for the wide scope reading.

4. Conclusion

In this paper, we have shown that French wh-movement is able to circumvent island restrictions due to the fact that it is not a uniform movement. We have proposed a mechanism which raises covertly wh-phrases in two different steps, a full phrasal movement, followed by a stripped movement.

We have shown that a stripping analysis as discussed for French can be extended to other wh-phenomena, such as partial wh-movement in Hungarian.

However, overt wh-movement cannot strip. Feature stripping seems to be a property of covert wh-movement, at least in languages like French and Hungarian. Why this is so remains to be investigated.

References

Baunaz, Lena 2008. Split-DP and Floating Quantifiers: a syntactic approach to French Quantification. Doctoral dissertation, Université de Genève.

Boeckx, Cédric 1999. “Decomposing French Questions”. In J. Alexander, N.R. Han & M. Minnick Fox (eds) University of Pennsylvania Working Papers in Linguistics 6.1, 69-80.

Brody, Michael 1990. “Some remarks on the Focus field in Hungarian”. UCL Working

Papers in Linguistics 2, 201-225. Cattell, Ray 1978 “On the source of interrogative adverbs.” Language 54, 61-77. Chang, L. 1997. Wh-in situ in French. MA thesis, U. of British Colombia. Cheng, Lisa and Johan Rooryck 2000. “Licensing wh-in-situ”. Syntax 3.1: 1-19. Den Dikken, Marcel and Anastasia Giannakidou 2002. “From hell to polarity: ‘Aggressively

non-D-linked’ wh-phrases as polarity items”. Linguistic Inquiry 33(1): 31-61. Giannakidou, Anastasia 1998. Polarity Sensitivity as (Non) Veridical Dependency.

Amsterdam: John Benjamins.


60

Horváth, Julia 1997. “The status of ‘wh-expletives’ and the partial wh-movement construction of Hungarian”. Natural Language and Linguistic Theory 15: 509-772.

Lipták, Aniko 2001. On the Syntax of wh-items in Hungarian. PhD dissertation, Universiteit Leiden.

Mathieu, Eric 2004. “The Mapping of Form and its Interpretation: the Case of Optional WH-Movement in French”. Lingua114: 1090-1132.

Obenauer, Hans 1994. Aspects de la Syntaxe A-Barre, Thèse de Doctorat d’Etat, Université de Paris VIII.

Pesetsky, David 1987. Wh-in-situ: Movement and unselective binding. In E. Reuland and A. ter Meulen (eds) The Representation of (In)definiteness, 98-129. Cambridge, Mass.: MIT Press.

Pesetsky, David 2000. Phrasal Movement and its Kin. Boston, Mass: MIT Press. Pesetsky, David and Esther Torrego 2004. “The Syntax of Valuation and the Interpretability

of Features”. In S. Karimi, V. Samiian and W. K. Wilkins (eds) Phrasal and Clausal

Architecture: Syntactic derivation and interpretation. In honor of Joseph E. Emonds, 262-294. Amsterdam: John Benjamins.

Poletto, Cecilia and Jean-Yves Pollock 2004. “On wh-clitics and wh-doubling in French and some North Eastern Italian dialects”. Probus 16 (2): 241-272.

Puskás, Genoveva 1992. “The Wh-criterion in Hungarian”, Rivista di Grammatica

Generativa 17: 141-186. Rizzi, Luigi and Ur Shlonsky 2007. “Strategies of Subject Extraction”. In H.M Gärtener, and

Uli Sauerland (eds) Interfaces + Recursion =Language? Chomsky’s Minimalism and

the View from Syntax-Semantics, 115-160. Berlin: Mouton de Gruyter. Starke, Michal 2001. Move Dissolves into Merge: A Theory of Locality. Doctoral

dissertation, Université de Genève. Starke, Michal 2005. Nanosyntax. Class lectures. Spring 2005, CASTL, University of

Tromsø.


33 – 2008

61

gò da mangià (‘i have to eat’) VS. ho da mangiare (‘i have to eat’)*

Andrea Cattaneo

0. Introduction

In this paper, I will focus on a construction whose template is fairly common throughout Romance and non-Romance languages: the construction Subj have to

Vinfinitive. The interesting characteristic of this construction is that–despite the form is exactly identical in North Italian Dialects (NIDs), Italian, French, Spanish, English, Dutch, and many other languages–the interpretation corresponding to such structure varies sensibly cross-linguistically and even within the very same language (in that multiple interpretations can be associated with such structure).

Here, I will only discuss the phenomenon w.r.t. to Bellinzonese1 and Italian,2 and I will leave for a future paper the extension of my analysis to the other languages which also have the same construction.

* I am extremely indebt to Paola Benincà, Cecilia Poletto, Richard Kayne, Stephanie Harves, Nicoletta Penello, Jacopo Garzonio, and Diego Pescarini for discussing the data with me and for their very insightful comments. A special thank you goes to Paola Benincà, Cecilia Poletto, Nicoletta Penello, Diego Pescarini, and Jacopo Garzonio for their constant availability and (extreme) patience in discussing data and judgments. Finally, I need to thank Judy Bernstein, the audience of the Seminario di Ricerca (February 14, 2008) of the University of Padua and the audience of the XXXIV° Incontro di Grammatica Generativa (February 21 – 23, 2008) for their precious comments and suggestions. I am also extremely thankful to Genoveva Puskás, Christopher Laenzlinger, Jacques Moeschler and the audience of the Séminaire de Recherche (Université de Genève, April 22, 2008), where I presented a preliminary version of this paper. All mistakes are of course mine. 1 It is a Lombard Alpine dialect spoken in Canton Ticino, Switzerland, which I am a native speaker of. In particular, my variety is the combination of the dialect spoken in S.Antonino–where I was born and grew up–and the dialect of Monte Carasso spoken by my mother. I refer to my variety as Bellinzonese for simplicity. In reality, my idiolect differs in some respects from the actual Bellinzonese variety spoken in Bellinzona, the capital of Ticino. 2 Please note that the word order reported here is the one that holds in standard and substandard Italian. As for the judgments, they are usually expressed with respect to the subvariety of Italian spoken in Ticino. Some of the readings reported here as ungrammatical

Andrea Cattaneo

62

1.1 The state of the art

Bellinzonese and Italian both have in common the structure in (1) and (2).

(1) Gò da mangià (Bellinzonese) have.1sg by to.eat ‘I have to eat’

(2) Ho da mangiare (Italian) have.1sg by to.eat ‘I have to eat’

A first observation that can be made is that, in (2), the infinitival IP is selected by da and not by the prepositional complementizer di (cf. Kayne (1980, 1999; 2000)), which usually introduces infinitival clauses. Da is the complementizer that introduces Italian infinitival relatives (3).

(3) a. Voglio qualcosa da mangiare (Italian) want.1sg something from to.eat ‘I want something to eat’ b. *Voglio qualcosa di mangiare

As for the nature of this da, I will not take the investigation further in this paper. It is however not unconceivable to think that da is the combination of the prepositional complementizer di and a preposition a, whose nature is still obscure.3, 4 are in fact well-formed in other varieties of Northern and Central Italy, including Florentine. These cases will be properly signaled. 3 den Dikken (1996) claims that à in (ia) is the same dative preposition as in (ib), and that Italian da could be di + adat (den Dikken (1996): 13 (fn16)). (i) a. Jean cherche un livre (facile) à lire (French) John looks-for a book easy to eat ‘John is looking for a book to eat’ b. Un livre à Jean a book to John ‘One of John’s book’ (den Dikken (1996): 9 (3a & 4a)) Also, Walloon clearly instantiate a complex preposition that corresponds to the combination of de and the dative à (ii). (ii) c’è d-à mîne (Walloon) it is of-to mine (Remacle (1952): 342) 4 Meyer-Lübke (1968) points out that da etymologically derives from the combination of de (‘from’) and the ablative a(b). It could thus be the case that a is here the marker of ablative. It is not clear though why the ablative would surface in and only in non-finite relatives. Are non-finite relatives similar to passive sentences? I have no answer at present.

Gò da mangià (‘i have to eat’) VS. Ho da mangiare (‘i have to eat’)

63

In the case of Bellinzonese, da covers both the function of Italian declarative complementizer di and the function of Italian infinitival relativizer da (4).

(4) a. G ho bisögn da na a ca’ (Bellinzonese) ‘I need to go home’ b. Pensi da mia mangià think.1sg by not to.eat ‘I am thinking of not eating’ c. Vöri quaicòss da mangià want.1sg something from to.eat ‘I want something to eat’

The a of declarative da looks like the a that is found on all person clitics–ma, ta, ga, va as opposed to standard Italian mi, ti, gli, ci, vi. As Paola Benincà points out to me, Italian pronominal vowel i is not an etymological vowel, rather a support vowel (see Cardinaletti (2008)). I maintain that Bellinzonese pronominal vowel a is also a default support vowel and that declarative da has this form because d–the complementizer–combines with the default vowel. Now, if it is true that Italian da is in fact d + a, it would seem plausible to postulate that Bellinzonese da is: either d + a null preposition a (the same as in Italian) that blocks the realization of the default vowel, or we have d + default vowel + silent preposition A (thanks to Richie Kayne (p.c.)).

This having been said, let us now come back to the interpretation of (1) and (2). In the next section, we will see that despite the identical form, (1) and (2) have very interesting interpretive differences. In Bellinzonese, (1) has three main modal readings: a Deontic Reading (DR), a Necessity Reading (NR), and a Possibility Reading (PR). As for (2), Ticino Italian only allows one modal reading – a PR. In other varieties of Italian, (2) can have a DR or a PR.

1.1.1 Bellinzonese gò da and its readings

In (5) and (6), I illustrate the 3 modal readings that can be associated to (1). The reading in (5a) corresponds to what I call a Necessity Reading, whereas in (5b) we have a Deontic Reading.

(5) a. Gò da mangià, altrimenti manchi via (Bellinzonese) ‘I need to eat, otherwise I will faint’ b. Al dutur u m’à dii che gò da mangià püsee ‘The doctor told me that I must eat more’

In a different context, (1) also supports a Possibility Reading, as illustrated in (6).

Andrea Cattaneo

64

(6) I bought so much food that now: Gò da mangià (per 3 setiman) (Bellinzonese) have.1sg of to.eat for 3 weeks ‘I have something/stuff to eat/that I can eat (for 3 weeks)’

1.1.2 Italian ho da and its readings

When it comes to (2), the interpretation varies depending on the variety of Italian one takes into consideration. In Ticino Italian, (2) can only have a Possibility Reading as the grammaticality contrast between (7) and (8) suggests.

(7) a. * Ho da mangiare, altrimenti svengo (Tic. Italian) ‘I have a need to eat, or I will faint’ b. * Il dottore mi ha detto che ho da mangiare di più ‘The doctor told me that I must eat more’

(8) I bought so much food that now: Ho da mangiare (per 3 settimane) ((Tic.) Italian) have.1sg of to.eat for 3 weeks ‘I have something to eat/that I can eat (for 3 weeks)’

In order to encode a DR or a NR, Ticino Italian resorts to the structures in (9).

(9) a. Ho bisogno di mangiare (Tic. Italian) ‘I have a need to eat’ b. Devo mangiare ‘I have an obligation to eat something’

Now, in some varieties of Italian spoken in Northern and Central Italy (including Florence), besides the PR in (10), (2) can also have a DR (11b). Crucially, in these varieties, (2) cannot have a NR (11a).5

5 Crucially, Ticino Italian also differs from other Italian varieties with respect to (i). As an epistemic modal, (ia) is out in all varieties. (ib) and (ic) are out in Ticino Italian. (i) a. * Bussano alla porta, ha da essere il postino ((Reg.) Italian) knock.1pl at.the door has by to.be the mailman ‘Somebody is knocking at the door, it must be the mailman’ b. % Per viaggiare all’estero, hai da avere il passaporto for to.travel abroad have by to.have the passport ‘To travel abroad one needs to have a passport’ c. % Per andare in Africa, hai da essere vaccinato for to.go in Africa have by to.be vaccinated ‘To go to Africa you have to be vaccinated’


65

(10) I bought so much food that now: Ho da mangiare (per 3 settimane) (Italian) have.1sg of to.eat for 3 weeks ‘I have something to eat/that I can eat (for 3 weeks)’

(11) a. * Ho da mangiare, altrimenti svengo (Italian) ‘I need to eat or I will faint’ b. Ho da mangiare altrimenti la mamma si arrabbia ‘I must eat otherwise my mom will get mad’

2. The problem

The data that I have illustrated so far raise two major questions. On the one hand, one wonders why the structures in (1) and (2), which look identical in form and distribution, have different readings in Bellinzonese and in different regional varieties of Italian, see Table I.

Table I Deontic Reading Necessity Reading Possibility Reading

gò da mangià

Bellinzonese

+

+

+

- - +

ho da mangiare

Ticino Italian Other varieties

+ - +

The other side of the problem is represented by the possibility of the same structure to be associated with different interpretations. How is this possible?

Here, I will try to supply a structural explanation for both problems.

It is interesting to notice that the equivalent of any sentence in (i) is well-formed in Bellinzonese, and other Italian dialects (Northern as well as Southern). In particular, if (iia) only has a PDMR in Italian (whether it be Ticino or Florence Italian), the latter can have the full range of modal readings in Bellinzonese and other NIDs.

(ii) a. Ho da dormire ((Reg.) Italian) ‘I have of what/where to sleep’ *‘I have to sleep’ *‘I need to sleep’ b. Gò da durmi (Bellinzonese) ‘I have of what/where to sleep’ ‘I have to sleep’ ‘I need to sleep’

Andrea Cattaneo

66

3. The proposal in a nutshell

I follow Kayne’s (1993) and I maintain that the verb have results from the incorporation of an oblique preposition into be. Also, I espouse Kayne (2005) and I maintain that for an element to be null, the latter has to move to a phase edge.

I argue that NR, DR, and PR, all derive from an underlying relative structure: an infinitival relative that contains a null light noun encoding modality and the argument of the infinitival verb, whether it be overt or not. Depending on what the relative head is, we get the different interpretations. In (12), we have the derivation for the NR of gò da mangià (‘I have to eat’). I propose that a null BISÒGN (‘need (n.)’) heads the relative and from there it noun-strips into have.6

(12) [gòp BISÒGN] [DPpossessor tp [ [ da [ ... mangià [QUAICOS]]]]]

In (13), we have the derivation for the DR of gò da mangià (‘I have to eat’). (13) only differs from (12) in the nature of the null noun that heads the relative.

(13) [gòp ÒBLIG] [DPpossessor tp [ [ da [ ... mangià [QUAICOS]]]]]

In (14), I give the derivation for the PR of gò da mangià (‘I have to eat’). In this case, the head of the relative is the null QUAICOSS (‘something’): the argument of the non-finite verb. The noun encoding modality stays in the embedded infinitival.

(14) gòp [DPpossessor tp [(la) [QUAICOS da [ PUSIBILITÀ ... [mangià ]]]]]

It looks like in Ticino Italian, the null light noun encoding modality cannot head the relative, hence the impossibility of having the equivalent of (12) and (13). For those varieties of Italian that admit a DR of ho da mangiare, the derivation is like in (13).

Finally, I argue that modals are not primitives. Hence, (15) is not derived from a relative clause but rather via incorporation of a nominal root into have as in (16). Other modals are derived in a similar way, although I will not discuss them here.

6 The motivation behind this noun-stripping is related to the analysis I have of constructions like (i). Here, I will not discuss constructions like (i). For the present purpose, the step of noun incorporation in (12) is not crucial. (i) Gò bisògn da mangià (Bellinzonese) have.1sg need by to.eat ‘I have a need to eat’


67

(15) Devi stcudià (Bellinzonese) must.1sg to.eat ‘I have to eat something’

(16) a. ESSERE [ Pj [DOV- [ andare a casa]]]] b. [IP Devoj+z [ tj [ tz [ andare a casa]]]]

4. Towards an analysis for gò da/ho da constructions

4.1 Bellinzonese gò da

If we insert an overt object in a structure like (1), here in (17), we realize that the latter can be the argument of have or of the non-finite verb.

(17) Gò da mangià (Bellinzonese) have.1sg by to.eat ‘I have to eat’

If the overt object–la turta (‘the cake’)–is the object of the non-finite verb (18), the sentence can only have a NR (18a) or a DR (18b).7

(18) a. Gò (*la turta) da mangià (la turta), (Bellinzonese) have.1sg the cake of to.eat the cake sennò manchi via otherwise miss away ‘I need to eat the cake, or I will faint’ b. Al dutur u m’ à di che the doctor SClIIIsg Cldat has said that gò (*la turta) da mangià (la turta) have.1sg the cake of to.eat the cake ‘The doctor told me I must eat the cake’

If the overt object is the argument of the finite verb (19), the interpretation can be a Focal Possibility Reading (FPR) (19a) or a Focal Deontic Reading (FDR) (19b)–

7 This is confirmed by the contrast in (i). Thanks to Diego Pescarini for suggesting this test. Judgments in (i) are mine. In (i), a PDMR is not available. (i) a. ?/?? Ghevi da mangià la turta, però l’ho perdùda (Bellinzonese) ‘I had to eat the cake, but I lost it’ b. Ghevi da mangià la turta, però l’ho mia legiùda ‘I had to study a poem, but I didn’t read it’

Andrea Cattaneo

68

where I use a different non-finite verb so that the relevant reading is more natural–but not a NR or even a Focal Necessity Reading, supposing that one exists. 8,9

(19) a. A: I wasn’t able to take you grocery shopping yesterday! B: Don’t you worry! Gò la turta da mangià (*la turta) (Bellinzonese) have.1sg the cake of to.eat the cake ‘I have the cake to eat’ / ‘I have the cake I can eat’ b. A: I know you have to read an article every week B: True, but this time Go al sò libru da lecc (*al sò libru) have.1sg the his book of to.read the his book ‘I have his book to read’ / ‘I have his book that I must read’

Interestingly, in (19), the DP (la turta/al sò libru) must be focalized (not contrastively, but just stressed), hence the fact that an intonational break is required between the object and da mangià/da lecc.

The situation in (19) is different if the overt object is a bare noun like quaicòs (‘something’) or ròba (‘stuff’). Under normal circumstances, these nouns do not bare stress in a context like (19), although they could easily be contrastively focused. What ends up being stressed, in the case of (19), is the finite verb. Here, the behavior of quaicòs and ròba seems to pattern with the interpretive contrast in (20).

8 Notice that, although (19b) can have a DR as (i) further confirms, this DR reading is slightly different from the DR in (18b). For one thing, the equivalent of (19b) is not licit in a context like (18b), as (18b) shows. The acceptability of (18b) with la turta that looks like the complement of have does not improve even with a focal intonation on the object. (i) a. Ghevi un libru da lecc, però l’ho perdù (Bellinzonese) ‘I had a book to read, but I lost it’ b. Ghevi un libru da lecc, però l’ho mia legiù ‘I had a book to read, but I didn’t eat it’ 9Interestingly, if the DP corresponds to the indirect object of the embedded verb, like in (i), we see that in the non-finite relative in (ib) necessarily has a resumptive clitic on the infinitive. I take this to indicate that (ib) and (19) are relative clauses. (i) a. Gò una dòna cun chi parlà (Bellinzonese) have.1sg a woman with who to.speak ‘There is a woman whom I can/haveto talk to’ b. Gò una dòna da parla-*(g) inseam have.1sg a woman by to.speak-Cldat together ‘There is a woman whom I can/have to talk to’


69

(20) a. Gò al libru/*al LIBRU da lecc e (Bellinzonese) have.1sg the book the BOOK by to.read and al Gianni chèl da culurà the John that by to.color ‘I have the book to read and John has the one to paint’ b. Gò al LIBRU/*al libru da lecc e have.1sg the BOOK the book by to.read and mia al giurnal not the newspaper ‘I have the book to read and not the newspaper’

In (20a), the DP cannot be focused and have seems to be lexical. The sentences is in fact equivalent to I am in possession of the book to read and John of the one to paint. Importantly, in (20a), al libru da lecc only has a DR, a PDMR is not possible. In (20b), the DP must be focused and here have is not the possessive one. In fact, the sentence is equivalent to I have to read the BOOK, not the newspaper and not *I am in possession of the BOOK to read and not the newspaper. In (20b) al LIBRU da lecc has either a DR or a PR. The difference between the status of have in (20) is confirmed by the following contrast10:

(21) a. *Gh è al libru da lecc e (Bellinzonese) there is the book by to.read and al Gianni chèl da culurà the John that by to.color ‘I have the book to read and John has the one to paint’ b. Gh è al LIBRU da lecc e there is the BOOK by to.read and mia al giurnal not the newspaper ‘I have the book to read and not the newspaper’

Now, let’s briefly go back to (6)–repeated here in (22). In (22), it is impossible to have an FDR. Moreover, even if a DP were realized, focus remains unavailable.

10 Notice that there is at least one North Italian Dialect where lexical have requires the presence of the locative, but the latter is incompatible with deontic have to constructions (data on Sonogno collected via questionnaire by Nicoletta Penello and Andrea Cattaneo). This suggests that (contrary to Kayne (1993) and Bhatt (1997)) lexical have and deontic have are different.

Andrea Cattaneo

70

(22) I bought so much food that now: Gò da mangià (per 3 setiman) (Bellinzonese) have.1sg of to.eat for 3 weeks ‘I have something/stuff to eat/that I can eat (for 3 weeks)’

The reading in (22) is necessarily a PR and cannot be a DR. Quaicos (‘something’) cannot be overt in (22); only ròba (‘stuff’) can. Still, ròba (‘stuff’) cannot bear stress in this case and crucially the reading here is equivalent to a DR.

4.2 Italian ho da constructions

Let us now see what happens if an overt object is inserted in a structure like (2)–repeated here in (23).

(23) Ho da mangiare (Italian) have.1sg by to.eat ‘I have to eat’

In Ticino Italian, an overt object can only be the argument of have and not of the non-finite verb. If la torta (‘the cake’) is inserted in (23), the latter cannot be the argument of the non-finite verb (24).

(24) *Ho da mangiare la torta (Tic. Italian) have.1sg of to.eat the cake ‘I need to/have to eat the cake’

If the object is on the contrary the argument of have (25), the sentence is well-formed and can have, like in Bellinzonese, a PR or a DR.11

11 It is interesting to notice that PR of (25) supports the insertion of ci (i). Ci is excluded from this context if (25) has a DR (ii) (thanks to Diego Pescarini for pointing this out to me). (i) A: I am worried you’ll get bored in the train B: Don’t worry, (C’) Ho una tesi da studiare ‘I have a dissertation to study’ (ii) A: I hear you have to read an article every week for your supervisor B: True, but this week (*C’) Ho una tesi da studiare ‘I have a dissertation to study’ Moreover, notice that the fact that the DR available in (25) has a slightly different flavor than the one in (24) (in those varieties that admit it) is also confirmed by the possible range of arguments that (24) and (25) allow. Consider (iii) (thanks to Nicoletta Penello and Diego Pescarini for pointing this out to me). (iii) a. ?/?? Ho da studiare due storie/un nuovo libro


71

(25) a. A: I wasn’t able to take you grocery shopping B: Don’t you worry! Ho la torta da mangiare (Tic. Italian) have.1sg the cake of to.eat ‘I have the cake to eat’ / ‘I have the cake I can eat’ b. A: I know you have to read an article every week B: True, but this time Ho il suo libru da leggere (Tic. Italian) have.1sg the his book of to.read ‘I have his book to read’ / ‘I have his book that I must read’

Like in Bellinzonese, the two DPs in (25) must be focalized. If qualcosa (‘something’) and roba (‘stuff’) in (25) are overt, they do not bear stress (even though they can if contrastively focused). The contrast between il libro (‘the book’) focalized or not in (26) holds in Italian, not just in Ticino Italian.

(26) a. Ho il libro/*il LIBRO da leggere e (Italian) have.1sg the book the BOOK by to.read and Gianni quello da colorare John that by to.color ‘I have the book to read and John has the one to paint’ b. Ho il LIBRO/* il libro da leggere e have.1sg the BOOK the book by to.read and non il giornale not the newspaper ‘I have the book to read and not the newspaper’

In (26a), the DP cannot be focused and have seems to be lexical. The sentences is in fact equivalent to I am in possession of the book to read and John of the one to paint. Un libro da leggere only has a DR. In (26b), the DP must be focused and here have is not the possessive one. In fact, the sentence is equivalent to I have to read the BOOK, not the newspaper and not *I am in possession of the BOOK to read and not the newspaper. Il LIBRO da leggere has both a PR and a DR. In (26a), have is lexical, while in (26b) have is part o a deontic construct. The difference between the status of have in (26) is confirmed by the following contrast:

‘I have to study two stories/a new book’ b. Ho due storie/un nuovo libro da studiare ‘I have two stories/a new book to study’

Andrea Cattaneo

72

(27) a. *C’ è il libro da leggere e (Bellinzonese) there is the book by to.read and Gianni quello da colorare John that by to.color ‘I have the book to read and John has the one to paint’ b. C’ è il LIBRO/* il libro da leggere e have.1sg is the BOOK the book by to.read and non il giornale not the newspaper ‘I have the book to read and not the newspaper’

In the case of (10)–repeated here in (28)–the reading is only a PR and not a DR. If overt, qualcosa (‘something’) is not available but roba (‘stuff’) is in (28). If roba is overt, the reading we obtain is a DR.

(28) I bought so much food that now: Ho da mangiare (per 3 settimane) (Italian) have.1sg of to.eat for 3 weeks ‘I have something to eat/that I can eat (for 3 weeks)’

Notice that in those varieties of Italian that allow (23), only a DR, and not a PR, is available for (23).

4.3 Gò da/Ho da constructions and concealed modals

As Cecilia Poletto points out to me (p.c.), the DR of (19) and (25) can be paraphrased with a finite relative as in (29) and (30).

(29) a. Gò la turta che devi mangià (Bellinzonese) have.1sg the cake that must.1sg to.eat ‘I have the cake that I have to eat’ b. Gò la turta che pòdi mangià have.1sg the cake that can.1sg to.eat ‘I have the cake that I can eat’ c. Gò la turta che gò bisògn da mangià have.1sg the cake that have.1sg need of to.eat ‘I have the cake that I need to eat’

(30) a. Ho la torta che devo mangiare (Italian) have.1sg the cake that must.1sg to.eat ‘I have the cake that I have to eat’


73

b. Ho la torta che posso mangiare have.1sg the cake that can.1sg to.eat ‘I have the cake that I can eat’ c. Ho la torta che ho bisogno di mangiare have.1sg the cake that have.1sg need of to.eat ‘I have the cake that I need to eat’

Now, (29) and (30) do not admit (31) and (32), respectively, where the embedded clause has an infinitival modal.

(31) ??/*Gò la turta da duve/ pude/ (Bellinzonese) have.1sg the cake by to.must to.can veg bisògn da mangià to.have need of to.eat ‘I have the cake that I have to/can/need to eat’

(32) ?? Ho la torta da dover/ poter/ (Italian) have.1sg the cake by to.must to.can aver bisogno di mangiare to.have need of to.eat ‘I have the poem that I have to/can/need to eat’

Interestingly, (18) cannot be paraphrased with a finite or infinitival embedded relative clause, as (33) and (34) show.

(33) Gò che devi/pòdi/gò bisògn da studià una puesia (Bellinzonese) ‘The matter is that I must/can/need to study a poem’ * ‘I need to/have to study a poem’

(34) Gò da duve/pude/*veg bisògn da studià 'na puesia (Bellinzonese) * ‘I need to/have to study a poem’ # ‘I think I have to/must study a poem’

(33) and (34) are well-formed but it is impossible to have a deontic reading both of the matrix verb and of the embedded modal.12 In particular the only reading

12 (34) becomes perfectly acceptable if an overt bisògn or òblig is realized as the internal argument of matrix have (i). The same holds for (35b), the latter becomes acceptable even in Ticino Italian (ii). (i) Go bisògn/l’ òblig da duve studià una puesia have.1sg need/ the obligation by to.must to.eat a poem ‘I need to/have to eat a poem’

Andrea Cattaneo

74

available in (34) is the one where the matrix verb is interpreted epistemically (I think that I will have to study a poem). In Ticino Italian and other varieties of Itlian (35) neither the counterpart of (29) nor of (30) are licit.

(35) a. ?? Ho da dover studiare una poesia (Italian) b. * Ho da dover studiare una poesia (Tic. Italian) have.1sg by must.1sg to.eat a poem ‘I have a poem that I have to eat’

5. Gò da/Ho da constructions and Relative clauses

Based on the fact that: i) in Bellinzonese we can have a resumptive pronoun in the non-finite clause; ii) (29) and (30) show that we can have an explicit relative, iii) the different behavior of have, I propose that gò da/ho da constructions derive from an underlying infinitival relative clause (36).13

(36) VÈS/ESSERE [SC DPpossessor P [DP [CP da [IP DPmodal ... [VP DPobject]]]]]

Let us recapitulate the crucial aspects of the configurations we discussed so far:

o If the DPobject remains in its base position (whether overt or silent), we obtain: i) a NR, or ii) a DR. No PR is possible.

o If the DPobject raises to the head of the relative we have 2 possible configurations:

� If DPobject is silent, DPmodal can only correspond to something similar to

PUSIBILITÀ/POSSIBILITÀ (‘possibility’) and we have a FPR � If DPobject is overt, DPmodal can only correspond to something similar to

OBLIG/OBBLIGO (‘obligation’) and we have a FDR � No FNR is here available

o If the DPobject is focalized we can have: i) a FPR, or ii) a DR. No FNR is

possible

(ii) Ho bisogno/l’obbligo di dover studiare una poesia have.1sg need the obligation of to.must to.eat a poem ‘I have a need/an obligation to eat a poem’ The readings in this case are indeed both deontic, however the matrix modality can only be understood as an order/advice coming from someone else. These cases are not relevant for the discussion. 13 The idea of a null modal is present in Bhatt (1997) already. My proposal differs radically from his because the idea of the relative clause is not present in Bhatt (1997). His proposal captures the English data but cannot capture Bellinzonese’s and Italian’s data.


75

I argue that NR and DR derive from the structure in (37) (I give the derivation for Bellinzonese, but (37) holds for Italian as well) where I also illustrate a tentative and approximate derivation.

(37) a. [gòp BISÒGN] [DP tp [ [ da [ ... [mangià la turta/QUAICOS]]]]] b. [gòp OBLIG] [DP tp [ [ da [ ... [mangià la turta/QUAICOS]]]]] In (37a) we have the derivation for NR, while in (37b) we have the one for a DR.

In (37), a null (light) DP encoding modality (BISÒGN (‘need’), ÒBLIG (‘obligation’)) which is the “spell-out” of a modal node in the relative clause. After raising to the head of the relative clause, the modal DP noun-strips into have creating modal have. As of now, it remains unclear to me why in (37) it is not possible to have a PR, i.e. PUSIBILITÀ/POSSIBILITÀ cannot be the head of the relative.

In (38), I give the tentative derivation for PR and an DR where the DP is not focalized. In (38a), I illustrate the derivation for a PR and in (38b) one for an DR.

(38) a. gòp [DP tp [(la) [QUAICOSS/*quaicoss da [PUSIBILITÀ ... [mangià ]]]]] b. gòp [DP tp [(la) [quaicos/*QUAICOS da [ OBLIG ... [mangià ]]]]]

The internal argument of the embedded verb raises to the head of the relative clause. Depending on the embedded modal DP we obtain a FPR (38a) or a FDR (38b). I am not sure why here a FNR is not available and I am not able to explain why an FPR is incompatible with and over DPobject and why a silent DPobject is illicit with an FDR.

In (39), we have the derivation for a FPR and an FDR. I maintain that these constructions are basically derived from (37) above and that the overt DPobject is not in reality the head of the relative clause. Rather the DPobject is the focalized object of the embedded verb that reaches a left periphery (very likely to low VP one).

(39) a. [gòp PUSIBILITÀi]x [FOC° la TURTA] ... tx[DP tp [[ ti da [ ti ... [mangià ]]]]]

b. [gòp OBLIGi]x [FOC° la TURTA] ... tx[DP tp [ [ ti da [ ti ... [mangià ]]]]]

Andrea Cattaneo

76

In (39), we have the null DP that encodes modality raises to the head of the relative clause (like in (37)). The internal argument of the embedded verb is then raised to a Focus position: if the focus position is within the periphery of the relative clause and the modal DP noun-strips into have; if the focus position is the one in the low periphery (Belletti (2004), Poletto (2006)) the derivation follows as above with the extra step of the unit gò PUSIBILITÀ/OBLIG that raises past the focalized element. I do not know how to explain the impossibility of having a FNR and the fact that here PUSIBILITÀ/POSSIBILITÀ can be the head of the relative (contrary to (38)). I am not sure of how to account for the impossibility of having a realized modal in the non-finite relative.

6. Conclusions and future research

The data analyzed show that null elements do not behave in the same way cross-linguistically. The licensing requirements for null categories are not the same in all languages and it seems that even within the same language different null nouns obey different requirements. Not only, but my analysis and data show that null nouns (or categories, more in general), do not always have the exact overt counterpart.

My analysis has some important implications for the derivation of modals: they are not primitives (see also Kayne (2007)).

Finally, my analysis supplies evidence in favor of the idea that Relative clauses are always pronominal (Kayne (1994), Cinque (2007)).

I showed that gò bisògn da/ho bisogno di type constructions have a different underlying structure from gò da/ho da type constructions. gò bisògn da/ho bisogno di configurations derive from a small clause where the bisògn/bisogno element can be the subject or the predicate of the small clause. If bisògn/bisogno is the subject of the small clause: i) the predicate can only be a non-finite IP which is selected by a null SOMETHING, ii) the complementizer must be realized in order to assign Kase to SOMETHING. If bisògn/bisogno is the predicate of the small clause: i) only a DP can be the subject of the small clause, ii) the complementizer is absent, iii) not all languages have this configuration and within the same language not all nouns can enter this configuration. gò da/ho da constructions are underlyingly post-nominal relative clauses. If the DPobject remains in its base position (whether overt or silent), we obtain: i) a NR, or ii) a DR. It is not clear why a PR is not possible. If the DPobject raises to the head of the relative we have 2 possible configurations: i) If DPobject is silent, DPmodal can only correspond to something similar to PUSIBILITÀ / POSSIBILITÀ (‘possibility’) and we have a FPR; ii) If DPobject is overt, DPmodal can only correspond to something similar to OBLIG/OBBLIGO (‘obligation’) and we have a FDR (it is not clear why no FNR is here available). If the DPobject is focalized we can have: i) a FPR, or ii) a DR. No FNR is possible, but not clear what the reason is.


77

Null categories do not all behave the same. They undergo different constraints within the same language and cross-linguistically. Null categories do not have the exact overt counterpart.

Duvé/dovere (‘to must’) is not a primitive. As or future developments, what needs to be investigated is: i) the nature of relative da; ii) the exact functioning of the relatives at stake needs to be made more clear; iii) if it is true that to must is not a primitive, it needs be understood how it differs/relate from other modals and what this tells us about clitic climbing.

7. Bibliography

Belletti, A. (2004) “Aspects of the low IP area”, in L. Rizzi ed. (2004), 16-51. Benincà and Poletto (1994) “Bisogna and Its Companions: The Verbs of Necessity”. In Paths

Towards Universal Grammar: Studies in Honor of Richard Kayne. Ed. by Cinque et alii, 35-57 . Washington: Georgetown University Press.

Bhatt, R. (1997) “Obligation and possession”. In Heidi Harley (ed.), Papers from the UPenn/MIT

roundtable on argument structure and aspect. MIT Working Papers in Linguistics, vol. 32. Cardinaletti, A. (2008) “On different types of clitic clusters”. In C. De Cat and K. Demuth (eds.)

The Bantu-Romance Connection, Amsterdam, John Benjamins. Cardinaletti, A. and L. Repetti (2004) “Clitics in Northern Italian Dialects: Phonology, Syntax and

Microvariation”, University of Venice Working Papers in Linguistics, 14, 7-106. Cinque (2007) “A Note on Verb/Object Order and Head/Relative Clause Order”. Ms., Ca’

Foscari, Venice. Dikken, M., den. (1996) “How external is the external argument?”. WECOL. Kayne, R. (1993) “Toward a Modular Theory of Auxiliary Selection”, Studia Linguistica, 47:3-31

(reprinted in Kayne (2000)). Kayne, R. (2000) “Prepositional Complementizers as Attractors”. In Kayne (eds.) Parameters and

Universals, Oxford University Press, 283-313. Kayne, R. (2002) “On Some Prepositions That Look DP-internal: English of and French de”,

Catalan Journal of Linguistics, 1, 71-115. Kayne, R. (2005) “On Parameters and on Principles of Pronounciation”, in H. Broekhuis,

N.Corver, R.Huybregts, U.Kleinhenz and K.Koster eds. Organizing Grammar. Linguistic

Studies in Honor of Henk van Riemsdijk, Mouton De Gruyter, Berlin. Kayne, R. (2007) “Some English (and Romance) Auxiliaries”, XVII Colloquium on Generative

Grammar, Girona, June 13, 2007. Handout. Meyer-Lübke, W. (1968/4 ), Romanisches Etymologisches Wörterbuch. Heidelberg (1st edition

Heidelberg 1911). Poletto, C. (2006) “Parallel Phases: a study on the high and low left periphery of Old Italian,” in

M. Frascarelli (ed.), Phases of Interpretation, Berlin: Mouton De Gruyter, 261-293. Remacle, L. (1952) Syntaxe du parler Wallon de la Gleize. Tome I : Noms e articles – Adjectifs et

pronoms. Paris : Les Belles Lettres.


33 – 2008

79

OV/VO SYNTAX

IN MÒCHENO MAIN DECLARATIVE CLAUSES

Federica Cognola

1 Introduction

This work1 is concerned with OV/VO alternations in main declarative clauses of Mòcheno, a Tyrolean dialect spoken in the Fersina valley in Eastern Trentino (North of Italy)2. This language has lived in a long situation of isolation from German varieties and in contact with Romance dialects, developing a series of innovations especially at the syntactic level not to be found in the other Tyrolean dialects.

I will try to make sense of OV/VO word orders in Mòcheno claiming on the one hand that the pattern of variation can be accounted for language-internally, refuting therefore the idea of the presence of two grammars (Romance and German) or of the influence of the Romance varieties on the German one (which is not able to make predictions). Starting from the descriptive generalisation that OV/VO syntax depends on what shows up in the high periphery, I will adduce other evidence in favour of the claim that the variation pattern can be accounted for by hypothesising that i) OV/VO word orders are the result of the interaction of low and high periphery and ii) the two peripheries are linked through movement.

2 V2 and the high left periphery

Before introducing the main topic of this article, it is worth saying few words on the syntactic characteristics of Mòcheno. This dialect is a V2 language of the old Romance type (Cognola 2007); this means basically that it displays a fine-grained

1 I would like to thank Cecilia Poletto for having gone through so much complicated data with me and to Paola Benincà for useful comments on several versions of this paper. Thanks to my mum and to an anonymous reviewer for turning my English into real English. All shortcomings are mine. 2 The data presented here refer to the most conservative variety of Mòcheno, namely the one spoken in the village of Palù del Fersina. I would like to thank my good informant Leo Toller for his patience and concern in answering my questions accurately.

Federica Cognola

80

high periphery (Rizzi 1997) whose structure is the one identified by Benincà (2001), Benincà/Poletto (2004) and Benincà (2006) and given in (1):

(1) [ Force [ TS [ scene setting [ wh-rel [ DS [ LI [ I Focus V2 [ II Focus V2 [ wh- interr V2 [ Fin]]]]]]]]]]

The V2 constraint can be satisfied only by XPs showing up in the focus field. In the examples in (2) it can be seen that Mòcheno is actually a V2 language

since if an XP shows up in a Spec position of the focus field, verb-subject inversion is obligatory: (2a) is a case of contrastive focalization of the direct object (from here on DO) and (2c) is a case of a low adverb moved to the high periphery.

(2) a. A PUACH UND NET A PENNA hot der Mario kaft gester a book and not a pen has the Mario bought yesterday b. *A PUACH UND NET A PENNA der Mario hot kaft gester ‘Mario bought a book and not a pen’ c. Ollbe hot der Nane kaft de oier afn morkt always has the John bought the eggs on-the market d. *Ollbe der Nane hot kaft de oier afn morkt ‘John has always bought the eggs at the market’

In the following examples I give some cases of V3 (3a) and V4 (3b), which are ruled out in standard German but possible in Mòcheno with base-generated XPs.

(3) a. Gester za tschaina hone de pasta gezen yesterday for dinner have-CL PRON the pasta eaten ‘Yesterday for dinner I had pasta’ b. Der Mario, gester petn Nane i hon gem a puach en sell tepp the Mario gester yesterday with-the John I have given a book to that crook ‘Mario, yesterday with John I gave a book to that crook’

To sum up, we can say that Mòcheno can be analysed as a V2-language of the old Romance type: this means basically that verb-subject inversion is triggered only by XPs with operator-like properties moved to a Spec position of the Focus field. Cases of V3 and V4 are admitted with XPs base-generated in the Topic and Frame fields.

In what follows I will introduce the pattern of variation concerning OV and VO syntax.

3 OV/VO alternations

As far as OV/VO word orders in main declarative clauses are concerned, this language allows for both OV (4a) and VO (4b) apparently with no difference in meaning, according to the judgements of my informants.

OV/VO syntax in Mòcheno main declarative clauses

81

(4) a. Gester hone s/a puach kaft yesterday hav-CL PRON the/a book bought b. Gester hone kaft s/a puach ‘I bought the/a book yesterday’

The two sentences in (4) do not display the same syntactic properties though; as shown in (5a), in fact, only with OV syntax is it possible to focalise the DO whereas OV is ruled out if the focalised DO follows the past participle as in (5b).

(5) a. Gester hone A PUACH UND NET A PENNA kaft yesterday have-CL PRON a book and not a pen bought b. *Gester hone kaft A PUACH UND NET A PENNA ‘I bought a book and not a pen yesterday’

In this work I reject the hypothesis that the pattern shown by the previous examples can be made sense of by allowing the possibility of having both head-final and head-initial projections, and I assume, following the antisymmetric theory (Kayne 1994), the base order of all projections to be head initial.

Also the hypothesis of morphological-case-driven movement appears to be weak both on general grounds3 and on account of the fact that Mòcheno has no case morphology on DPs, as can be seen in (6a,b).

(6) a. Der Mario hot der Nane pakemmt the Mario-NOM/ACC has the John-NOM/ACC met ‘Mario met John’ b. Der Mario hot a puach en Nane gem the Mario has a book to-the John given ‘Mario gave John a book’

In what follows I will try to make sense of Mòcheno OV/VO word orders making use of the hypothesis of the presence of a VP periphery (from here on low left periphery) of the clause (Belletti (2001, 2004), Jayaseelan (2001), and Poletto (2006)) showing that OV syntax in Mòcheno involves the use of the low left periphery and is possible as long as i) the Spec of low FocusP has not been saturated by an XP generated below the periphery moving to a Spec of the high Focus field; ii) the past participle remains in a low position; iii) the XP triggering verb-subject

3 This is actually not a very strong argument since in old English the verb kept moving to TP even when verb morphology was lost (Roberts 1993). With regard to the relation between word order and morphology, Chiara Polo (2005) has convincingly showed that OV word orders in Latin do not depend on morphology, since DOs in VO constructions always had accusative morphology; according to her VO has rather to be interpreted into relation to information structure.

Federica Cognola

82

inversion has been moved to the high Focus field from a position above the low left periphery.

The derivation is to be seen in the structure below in (7):

(7) [FocusP [SpecFocus XP ] [Focus0 auxiliary ] [AgrSP [SpecAgrS subject ] [Agr0 auxiliary] [TopicP [SpecTopicP YP ] [Topic0] [FocusP [SpecFocus ZP] [Focus0 ] [V P [SpecV ] [V 0 past

participle] [DP ] ] ] ] ] ]

VO syntax is obligatory iff: i) Spec low FocusP has been saturated by an XP generated below the periphery moving to a Spec of the high Focus field; ii) this stepwise movement (Rizzi 2004b) forces the past participle to rise to the edge of the phase, that is above the low periphery leaving TopicP below4.

The derivation is given in (8):

(8) [FocusP [SpecFocus XP ] [Focus0 auxiliary ] [AgrSP [SpecAgrS subject ] [Agr0 auxiliary] [AspP

[SpecAsp ] [Asp0 past participle] [TopicP [SpecTopicP YP ] [Topic0 past participle ] [FocusP

[SpecFocus XP ] [Focus0 past participle ] [V P [SpecV ] [V 0 past participle] [DP ] ] ] ] ] ]

In what follows I will adduce evidence in favour of the analysis just sketched for Mòcheno OV/VO syntax showing first that in Mòcheno OV syntax is really to be derived via movement of XPs to Spec positions of the low left periphery and that when Spec low FocusP has been saturated OV syntax is not possible anymore.

4 OV syntax involves the use of the low left periphery

If we take into consideration low adverbs when looking at sentences displaying both OV and VO word orders, we immediately see that with VO syntax the DO has to show up after all low adverbs (9a,c) whereas in OV syntax DO and low adverbs are mixed up (9b,d).

(9) a. Gester hot der Luca schua gonz galezen (*gonz) s puach (*gonz) yesterday has the Luca already completely read the book b. Gester hot der Luca schua s puach gonz galezen ‘Luca read the whole book yesterday’ c. Gester hot der papa schua schia puzt (*schia) s houz (*schia) yesterday has the dad already well cleaned the house d. Gester hot der papa schua s houz schia puzt ‘Yesterday dad cleaned the house well’

The pattern found with VO syntax5:

4 In this work I will refer to the area above the low periphery as ‘edge of the lower phase’ in which the head of an AspP dedicated to the past participle in VO syntax is to be found. 5 In this work, for reasons of space, I will not try to make sense of how VO syntax is technically to be derived.


83

(10) LOW ADVERBS - PPs - PAST PART – DO

The pattern found with OV syntax:

(11) LOW ADVERB - DO - LOW ADVERBS - PPs - PAST PART

Following Cinque (1999), who claims that adverbs occupy fixed Spec positions in the IP layer, patterns (10) and (11) are enough proof in order to say that DO always leaves VP with OV syntax. Below we will also see that the position to which DOs move in OV syntax is to be found above the area where non-argumental PPs show up that, following Schweikert (2004) and Cinque (2006), I assume to be immediately above the VP layer.

Roughly speaking (see Cognola 2008 for more on this), the area to which DOs move with OV syntax is to be found below the low adverb schua, ‘already’ and above schia, ‘well’; non-argumental PPs are to be found below low adverbs.

(12) a. Gester hot der Luca schua s puach gonz galezen yesterday has the Luca already the book completely read ‘Yesterday Luca read the whole book’ b. Gester hot de mama schua s houz schia puzt yesterday has the mum already the house well cleaned ‘Yesterday mum cleaned the house well’

ALREADY - DP – WELL *DP - ALREADY – WELL

It is important to notice that the DO with OV syntax cannot in any case precede the adverb ’already’ (13a,b), whereas – as we will see later on – ’well’ and ’gonz’ can precede it giving rise to marked orders. Anyway, in unmarked sentences (that is if the low adverb is not focalised itself, which is in principle not ruled out) low adverbs have to follow DOs (presumably showing up in a Spec position of the low left periphery) (13c,d) and their relative order is gonz-schia (13d,e)6.

(13) a. Gester hone schua a puach kaft yesterday have-CL PRON already a book bought b. *Gester hone a puach schua en de boteig kaft ‘I already bought a book yesterday in the shop’ c. Gester hot de mama schua s houz gonz putzt yesterday has the mum already the house completely cleaned d. Gester hot de mama schua s houz schia putzt

6 Sentences with ‘gonz’ preceding ‘schia’ are somehow marked (but in any case grammatical), since it would be more natural to use the QP ‘s gonze houz’. Anyway, the important point is that the relative order of the low adverbs is ‘gonz-schia’ and not the other way round.

Federica Cognola

84

yesterday has the mum already the house well cleaned e. *Gester hot de mama schua s houz schia gonz putzt ‘Yesterday mum cleaned the whole house well’

If we take into consideration also a non-argumental PP, its unmarked position with OV syntax is below the low adverb ‘schia’ and above the past participle, as shown in (14).

(14) De mama hot schua s hauz schia petn stapsauger putz the mum has already the house well with the hoover cleaned ‘Mum has already cleaned well the house with the hoover’

The structure we can derive from the examples seen so far is the one given in (15).

(15) ALREADY - TOPIC - FOCUS - COMPLETELY - WELL - PPs – PASTPART

This structure is strikingly similar to the one identified for standard German by Hinterhölzl (2006) who claims that the positions in which DOs show up are AgrOPs. My assumption, on the contrary, is that at least in Mòcheno the area between ’already’ and ’well’ contains the low left periphery.

In the following section I will bring evidence in favour of this last claim of mine basing myself mainly on the test of contrasting.

4.1 More on the low left periphery

The hypothesis that the area between ’already’ and ’well’ in which DOs show up in Mòcheno OV syntax is really the low left periphery seems to find confirmation in the following examples in which the DO can be contrasted only if showing up above low manner adverbs7.

(16) a. *Gester hot der papa schia petn staupsauger S HAUS UND NET DER AUTO puzt yesterday has the dad well with the hoover the house and not the car cleaned b. Gester hot der papa S HAUS UND NET DER AUTO schia petn staupsauger putzt ‘Dad properly hoovered the house and not the car yesterday’

7 It would not be possible to contrast the DO also if it showed up in VO syntax. My claim for the VO cases is that no XP following the past participle can be contrasted since it is either a Topic or is in its base position.


85

In the following structure (17) I give the derivation of (16b) with the DO moving from Spec PredP (where it is presumably to be found in (16a)) to Spec low FocusP8.

(17) [AspP [SpecAsp schua ] [Asp0] [TopicP [SpecTopicP ] [Topic0] [FocusP [SpecFocus s houz ] [Focus0] [AspP [SpecAsp schia ] [Asp0] [PredP [SpecPred s houz ] [Pred0 ] [V P [SpecV ] [V 0 putzt ] [DP houz ] ] ] ] ] ] ]

Also a non-argumental PP could in principle be contrasted (that is moved to Spec low FocusP), since we saw in (14) that in Mòcheno its base position seems to be below the periphery. As shown in the examples below this prediction is borne out since the PP can be contrasted only if showing up before the low adverb ’well’ (18b), whereas this is ruled out if it follows (18a).

(18) a. *Gester hot der papa schua s hauz schia PETN STAUPSAUGER UND NET PETN STROZ putzt yesterday has the dad already the house well with the hoover and not with the cloth cleaned b. Gester hot der papa schua s hauz PETN STAUPSAUGER UND NET PETN STROZ schia putzt ‘Yesterday dad cleaned well the house with the hoover and not with the cloth’

In (19) I give the structure of (18b) with the PP moving from its base position to Spec FocusP; the DO is in Spec TopicP.

(19) [AspP [SpecAsp schua ] [Asp0 ] [TopicP [SpecTopicP s houz ] [Topic0 ] [FocusP [SpecFocus petn

staupsauger ] [Focus0 ] [AspP [SpecAsp schia ] [Asp0 ] [PP [SpecP petn staupsauger ] [P0 ][PredP [SpecPred s houz ] [Pred0 ][V P [SpecV ] [V 0 putzt ] [DP houz ] ] ] ] ] ] ]

One last point concerns the possibility of contrasting low adverbs that in principle should not be ruled out, since at least ‘gonz’ and ‘schia’ show up below the Topic and Focus projections. As shown in (20) in Mòcheno it is possible to contrast a low adverb both with the order DO-low adverb (20a) and low adverb-DO (20b); I take this to mean that a low adverb can show up both in Spec low FocusP and in Spec low TopicP9.

8 Following Hinterhölzl (2006), I claim that also in Mòcheno there is evidence in favour of the presence of a projection called PredP below low manner adverbs and immediately above VP. In my analysis this projection represents the first step of the derivation of DOs; only if moving to a Spec position of the low periphery can they become real arguments. Evidence in favour of the presence of predicate DOs come for example from idiom chunks, which are the only constructions showing always OV syntax. Only real arguments can enter the pattern of OV/VO alternations shown so far. 9 The same pattern is valid also for the adverb ‘gonz’.

Federica Cognola

86

(20) a. Gester hone s houz SCHIA UND NET PRESAPOC/*UND NET DER AUTO petn staupsauger putzt yesterday have-CL PRON the house well and not superficially cleaned ‘Yesterday I hoovered the house properly and not superficially’ b. Gester hone schia S HAUZ UND NET DER AUTO/*UND NET PRESAPOC petn staupsauger putzt ‘Yesterday I properly hoovered the house and not the car’

In (21) I give the structure of (20a): ‘schia’ is in Spec FocusP and DO in Spec TopicP.

(21) [AspP [SpecAsp schua ] [Asp0] [TopicP [SpecTopicP s hauz ] [Topic0] [FocusP [SpecFocus schia ] [Focus0] [AspP [SpecAsp schia] [Asp0] [PP [SpecP petn staupsauger ] [P0] [PredP [SpecPred hauz] [Pred0] [V P [SpecV ] [V 0 putz] [DP hauz] ] ] ] ] ] ] ]

In (22) I give the structure of (20b): ‘schia’ is in Spec TopicP and DO in Spec FocusP.

(22) [AspP [SpecAsp schua ] [Asp0] [TopicP [SpecTopicP schia ] [Topic0] [FocusP [SpecFocus s hauz ] [Focus0] [AspP [SpecAsp schia] [Asp0] [PP [SpecP petn staupsauger ] [P0] [PredP [SpecPred

hauz] [Pred0] [V P [SpecV ] [V 0 putz] [DP hauz] ] ] ] ] ] ] ]

The data presented so far have shown that OV syntax in Mòcheno can be accounted for assuming that DOs move either to the Spec of low Topic or low FocusP, whereas the past participle remains in a very low position.

The low phase of the clause has in Mòcheno the following structure:

(23) ALREADY - TOPIC - FOCUS - COMPLETELY - WELL - PPs – PREDP- PAST PART

After having shown that OV syntax in Mòcheno actually involves the use of the low left periphery and having sketched its structure and position in the clause, in the following section I will show that the possibility of having VO syntax in Mòcheno is tightly linked to the structure given in (23) and in no sense optional.

5 VO patterns

I begin this section with a first descriptive generalisation:

(24) The possibility of using the low periphery and having therefore OV syntax is linked to the type of XP showing up in the high periphery

In what follows I will try to illustrate this claim taking into consideration i) sentences beginning with a focalised XP and ii) sentences beginning with the adverb gester, ’yesterday’.


87

5.1 Sentences beginning with a focalised XP

Sentences beginning with a focalised XP base-generated below the low periphery are obligatory VO. As I show in (25a,b), it is not possible to have OV word order if the DO is contrastively focalised in the high focus field; this is also valid for the case of contrastive focalization of a non-argumental PP, (25c,d)10.

(25) a. A MOTZ PLIAMBLER UND NET A PFLONZ hone kaft afn morkt/ gester a bunch flowers and not a plant have-CL PRON bought on-the market/ yesterday b. *A MOTZ PLIAMBLER UND NET A PFLONZ hone afn morkt/gester kaft ‘I bought a bunch of flowers and not a plant yesterday at the market’ c. PETN STROZ UND NET PETN STAUPSAUGER hone puzt s hauz with the cloth and not with the hoover have-CL PRON cleaned the house d. *PETN STROZ UND NET PETN STAUPSAUGER hone s hauz putzt ‘I cleaned the house with the cloth and did not hoover it’

The pattern found in the previous sentences follows straight from my account of OV/VO word orders: i) XPs targeting a Spec position in the high Focus field and generated below the low periphery have to move first to the corresponding position of the low periphery; ii) the activation of low FocusP by an XP on its way to the high periphery causes movement of the past participle to the edge of the phase; iii) with VO syntax also Spec low TopicP is in principle available and can be occupied by an XP following the past participle.

In (26) I give the structure of (25a) with the PP showing up in its base position, but recall that it could also appear in SpecTopicP.

(26) [FocusP [SpecFocus A motz pliamber ] [Focus0 hon-e ] [AgrSP [SpecAgrS] [Agr0 hon ] [AspP [SpecAsp ] [Asp0 kaft ] [TopicP [SpecTopicP ] [Topic0 ][FocusP [SpecFocus motzpliamber] [Focus0 ] [AspP [SpecAsp ] [Asp0 ] [PP[SpecP afn morkt ] [P0 ] [V P [SpecV ] [V 0 kaft ] [DP motz pliamber ]] ] ] ] ] ] ] ]

10 I will not take into consideration here the case of contrastive focalization of the indirect object (IO) in the double object construction, that represents apparently a counter example to my claims, since OV syntax in this case is obligatory. My claim is that this especial pattern can be made sense of assuming that IO cannot move on its own to Spec low FocusP, as it is clearly hint at by the data of mixed syntax (IO cannot cross DO).

Federica Cognola

88

5.2 Main interrogative clauses

Evidence in favour of the analysis proposed above comes from wh- main interrogative clauses, that show obligatory VO with any type of wh element except for the one corresponding to ‘why’. In (27) I give some examples of this11.

(27) a. Pet bem hos-o kaft a puach? with whom have-CL PRON you bought a book b. *Pet bem hos-o a puach kaft? ‘With whom did you buy a book?’ c. Bos hos-o kaft en de boteig? what has-CL PRON bought in the shop d. *Bos hos-o en de boteig kaft? ‘What did you buy in the shop?’ e. Ber hot kaft a puach? who has bought a book f. *Ber hot a puach kaft? ‘Who bought a book?’

As was the case with contrastively focalised XPs, the derivation involves: i) movement of the wh-element first to Spec low FocusP as first step of the derivation; ii) this intermediate step forces the past participle to rise to the edge of the phase; iii) an XP following past participle in VO syntax is either in Spec TopicP or in its base position.

In (28) I give the structure of (27c)

(28) [FocusP [SpecFocus bos ] [Focus0 hos-o ] [AgrSP [SpecAgrS ] [Agr0 hos ][AspP [SpecAsp ] [Asp0 kaft ] [TopicP [SpecTopicP ] [Topic0 kaft] [FocusP[SpecFocus bos ] [Focus0 kaft] [AspP [SpecAsp ] [Asp0 kaft] [PP [SpecP en deboteig [P0 kaft ] [PredP [SpecPredP [Pred0 ] [V P [SpecV ] [V 0 kaft ] [DP bos ] ] ] ] ] ] ] ] ] ]

Further arguments in favour of this analysis come from why- and yes/no main interrogative clauses.

As already said, the only wh- main interrogative clauses optionally showing OV syntax are the ones introduced by ‘why’. This follows straightforwardly from Rizzi’s (2001) claim that ‘why’ is base generated in the CP layer in the head position of a projection called InterrogativeP in whose Spec ‘if’ is to be found. ‘Why’ does not move through low FocusP. Notice that with ‘why’ the NP subject shows up after the wh-element (and is presumably to be found in a Topic position) and does not need to be dislocated to the right as is the case of (27). 11 Intentionally I give only sentences with pronominal subjects which always show subject-verb inversion. In the case we has a DP the pattern would be less straightforward, since both subject verb inversion and NP right dislocation are admitted. I will not go into this.


89

(29) a. Babai der Nane hot er kaft a puach? why the John has-CL PRON bought a book b. Babai der Nane hot er a puach kaft? ‘Why did John buy a book?’

In (30) I give the structure of why-main interrogative clauses with the past participle showing either above VP or at the edge of the lower phase12.

(30) [InterrP [SpecInterr [InterrO babai [TopicP [SpecTopicP der Nane ] [Topic0 ][FocusP [SpecFocus [Focus0 hot-der ] [AgrSP [SpecAgrS der Nane] [Agr0 hot] [AspP [SpecAsp ] [Asp0 KAFT ] [TopicP [SpecTopicP ] [Topic0 ] [FocusP [SpecFocus a puach] [Focus0 ] [AspP [SpecAsp ] [Asp0 ] [PP [SpecP [P0 [PredP [SpecPredP a puach ] [Pred0 ] [V P [SpecV ] [V 0 KAFT ] [DP a

puach] ] ] ] ] ] ] ] ] ]

With yes/no main interrogative clauses no such pattern is to be found, since no wh- element, that is no low Focus is activated by an XP on its way to the CP layer; therefore both OV and VO are possible.

(31) a. Hoso kaft a puach? have-CL PRON bought a book b. Hoso a puach kaft? ‘Did you buy a book?’

Notice that OV syntax is obligatory with special main interrogative clauses (both yes/no and wh).

(32) a. Benn hot der Mario a bain za tschaina trog?!! when has the Mario a wine to dinner brought ‘When has Mario ever come to dinner with a bottle of wine?!!’ b. Ber hot a puach en de Maria gem?!! who has a book to the Maria given ‘Who has ever bought Mary a book?!’

This last point seems to hint at the fact that the wh-element showing up in special interrogative clauses do not share the same properties as wh-elements of real interrogatives which block Spec low FocusP, as already noted in the literature (among others Obenhauer 1994, 2004).

To summarise we can say that in a construction in which the high focus is surely occupied either by a focalised argumental XP or by an interrogative wh-element, VO syntax is obligatory, that is, the DO can never be interposed. OV syntax is possible as long as the low periphery is available, that is iff i) no focalised XP is

12 VO syntax is derived also in this case through past participle movement to the edge of the lower phase, see the possibility of contrasting DO in the following example: Babai der Nane hot der kaft A PUACH UND NET A PENNA?

Federica Cognola

90

extracted from below the low periphery, ii) the wh-element is base generated as is the case of ‘why’, iii) the low periphery is not involved, as with yes/no main questions.

In the last section I will come back to the case I began with, that is the one of sentences beginning with ‘gester’, trying to make sense on the basis of the proposed analysis why they allow both OV and VO syntax.

5.3 Sentences beginning with ‘gester’

As shown in (5) repeated here as (33), sentences beginning with the adverb gester, ’yesterday’, allow for both OV and VO syntax; notice that the adverb triggers verb-subject inversion in both cases13.

(33) a. Gester hone A PUACH UND NET A PENNA kaft yesterday have-CL PRON a book and not a pen bought b. Gester hone kaft a puach c. *Gester hone kaft A PUACH UND NET A PENNA ‘I bought a book and not a pen yesterday’

The possibility of having both OV/VO syntax with a sentence beginning with ‘yesterday’ is somehow unexpected after what we have seen so far, but it can be accounted for by claiming, following Benincà/Poletto (2004), that this adverb can be generated in at least three positions in the structure.

As shown in (34), ‘gester’ can be generated in a lower position above the VP layer where also adverbial PPs and bare NPs are to be found according to Schweikert (2004) and Cinque (1999, 2006); in TP; in CP in the topic frame and functions as a scene-setter.

(34) [ScenesettingP [SpecScenesetting GESTER ][SceneSetting0 ] [FocusP [SpecFocus [Focus0 auxiliary ] [AgrSP [SpecAgrS subject ] [Agr0 auxiliary ][TP [SpecT GESTER ] [T0 ] [AspP [SpecAsp ] [Asp0 ] [TopicP [SpecTopicP] [Topic0 ] [FocusP [SpecFocus ] [Focus0 ] [AspP [SpecAsp ] [Asp0 ] [DP

[SpecD GESTER ] [D0 ] [PredP [SpecPredP ] [Pred0 ] [V P [SpecV ] [V 0 past participle ] [DP ] ] ] ] ] ] ] ] ] ] ] ]

VO syntax is possible only if the ‘gester’ showing up in CP and triggering V2 is the one generated above VP in the area where adverbial PPs and bare DPs (Schweikert 2004, Cinque 2006) are to be found. The adverb moved from that position has to pass through Spec low FocusP on its way to the CP layer, forcing the past participle to rise to the edge of the lower phase. The impossibility of contrasting a DO with VO syntax in sentences beginning with this ‘gester’ depends on the fact that below the past participle only a Topic projection is available, since Spec

13 ‘Gester’ can theoretically also not trigger inversion; in that case I claim, following Benincà/ Poletto (2004), that it is used as a scene setter and occupies a Spec position in CP.


91

lowFocusP has been saturated by the XP moving to CP and *high focus-low focus. Of course one could also think that the DO remains in Spec PredP.

Cases of VO syntax with sentences beginning with ‘gester’ have the same derivation as (25c) with a focalised adverbial PP.

In (35) I give the structure of (33b).

(35) [FocusP [SpecFocus Gester ] [Focus0 hon-e ] [AgrSP [SpecAgrS ] [Agr0 hon ][AspP [SpecAsp ] [Asp0 kaft ] [TopicP [SpecTopicP a puach ] [Topic0 kaft][FocusP [SpecFocus gester ] [Focus0

kaft] [AspP [SpecAsp ] [Asp0 kaft] [DP [SpecD gester] [D0 kaft] [PredP [SpecPred a puach] [Pred0 kaft] [V P [SpecV ] [V 0 kaft ] [DP a puach ] ] ] ] ] ] ] ] ] ] ]

OV syntax with sentences beginning with ‘gester’ is possible only if the adverb is the one generated in T or higher in CP. I assume that it moves as well to a Spec position of the Focus field since it triggers inversion, but this position has to be different from the one occupied by an XP extracted from the area below the low periphery. The fact that no Spec position was saturated in the derivation, allows for the DO to move to one of the Specs of the low periphery whereas the past participle remains lower down in the structure. Notice that elements generated above the periphery can trigger inversion but they cannot target the same position as XPs moved from below the periphery, otherwise it would not be possible to contrast the DO in the lower phase, given the incompatibility between high focus and lower focus. They are probably moved to a Spec position for themes to be found in the focus field, whereas foci extracted from below are moved to a real Focus projection.

In (36) I give the strucure of (33a) with focalization of the DO.

(36) [FocusP [SpecFocus Gester ] [Focus0 hon-e ] [AgrSP [SpecAgrS ] [AgrS0 hon ] [TP [SpecT gester ] [T0 ] [AspP [SpecAsp ] [Asp0 kaft ] [TopicP [SpecTopicP[Topic0 kaft] [FocusP [SpecFocus a puach ] [Focus0 kaft] [AspP [SpecAsp ] [Asp0 kaft][PredP [SpecPred a puach ] [Pred0 kaft] [VP [SpecV] [V 0 kaft ] [DP a puach ] ] ] ] ] ] ] ] ] ] ]

6 Conclusions

In this work I tried to propose an account for OV/VO word order alternations in a Tyrolean dialect spoken in a Romance area claiming that the variation pattern can be easily made sense of starting from the hypothesis of the presence of a low and a high left periphery connected through movement.

OV syntax always involves the use of the low left periphery, which is not to be found immediately above VP as generally assumed in the literature, but below the low adverb ‘already’ and above ‘well’. OV syntax is possible if the XP triggering subject-verb inversion is generated in a position above the low periphery. With OV syntax the past participle does not move or moves very little.

VO syntax is possible only if Spec low FocusP is saturated by an XP moving to a Spec of high FocusP. The stepwise movement of an XP on its way to the CP layer

Federica Cognola

92

forces the past participle to rise to the edge of the lower phase: what follows can be either in Spec low TopicP or in its base position and therefore cannot be contrasted.

References

Belletti, Adriana (2001): “Inversion as focalization”, in: A.Hulk and J.Y.Pollock (eds.), Inversion in Romance and the theory of Universal Grammar. Oxford University Press.

– (2004): “Aspects of the low IP Area”, in: L. Rizzi (ed.), The Structure of CP and IP. The

Cartography of Syntactic Structures, Volume 2. Oxford University Press, New York/Oxford.

– /Ur, Shlonsky (1995): “The order of verbal complements: A comparative study”, in: Natural

Language and Linguistic Theory. 13, 3. Benincà, Paola (2001): “The Position of Topic and Focus in the left Periphery”, in: G. Cinque

and G. Salvi (eds.), Current Studies in Italian Syntax: essays offered to Lorenzo Renzi. Elsevier, Amsterdam, pagg. 41-64.

– (2006): “On the Functional Structure of the Left Periphery of Medieval Romance”, in: R. Zanuttini, H. Campos, E. Herburger e P. Portner (eds.), Negation, Tense and Clausal

Architecture: Cross-linguistics Investigation. Georgetown University Press, Washington.

– /Cecilia, Poletto (2004): “Topic, Focus and V2: defining the CP sublayers”, in: L. Rizzi (ed.), The Structure of CP and IP. The Cartography of Syntactic Structures, Volume 2. Oxford University Press, New York.

Cinque, Guglielmo (1999): Adverbs and Functional Heads. Oxford University Press, New York.

– (2006): “Complements and Adverbial PPs: Implication for Clause Structure”, in: G. Cinque, Restructuring and Functional Structure. The Cartography of Syntactic Structure,

Volume 4. Oxford University Press, New York. pages 145-166. Cognola, Federica (2007): “Ordini delle parole e movimento wh- in mòcheno”, paper

presented at the XIIIth Giornata di dialettologia, University of Padua. – (2008): “Ordini OV/VO nelle frasi dichiarative del Mòcheno”, talk delivered at the

University of Padua, 7th January. Diesing, Molly (1997): “Yiddish VP Order and the Typology of Object Movement in

Germanic”, in: Natural Language and Linguistic Theory 15. Hinterhölzl, Roland (2006): Scrambling, Remnant Movement and Restructuring in West

Germanic. Oxford University Press, New York. Jayaseelan, Karattuparambil (2001): “IP-internal Topic and Focus Phrases”, in: Studia

Linguistica 55.1. Kayne, Richard S. (1994): The antisymmetry of syntax. The MIT Press, Cambridge, Mass. Obenauer, Hans Georg (1994): Aspects de la syntaxe A-barre. Effects d’intervention et

mouvements des quantifieurs. Thèse d’Etat, Université de Paris VII.


93

– (2004): “Non standard wh-questions and alternative checkers in Pagotto”, in: H. Lohnstein and S. Trissler (eds): The left periphery of Germanic Languages. Mouton de Gruyter, Berlin.

Poletto, Cecilia (2002): “The left periphery of V2-Rhaetoromance dialects: a new perspective on V2 and V3”, in: S. Barbiers, L. Cornips e S. van der Kleij (eds.): Syntactic

Microvariation. Meertens Institute, Amsterdam. – (2006): “Old Italian Scrambling: the low left periphery of the clause”, in: M. Frascarelli

(ed.), Proceedings of the 31st Meeting of Generative Grammar, Mouton de Gruyter. Polo, Chiara (2004): Word Order Between Morphology and Syntax. RGG Monography.

Unipress, Padova. Rizzi, Luigi (1997): “The fine Structure of the Left Periphery”, in: L.Haegeman (ed.),

Elements of Grammar. Kluwer, Dordrecht. – (2001): “The Position of Inter(rogative)P in the Left Periphery of the Clause”, in: G. Cinque

and G. Salvi (eds.), Current Studies in Italian Syntax: essays offered to Lorenzo Renzi. Elsevier, Amsterdam.

– (2004a): “Locality and Left Periphery”, in: A. Belletti (ed.), Structures and Beyond: The

Cartography of Syntactic Structure, Volume 3. Oxford University Press, New York. – (2004b): “On the Form of Chains: Criterial Positions and ECP Effects”, available on the

internet under: wwww.ciscl.unisi.it. Roberts, Ian (1993): Verbs and diachronic Syntax: a comparative history of English and

French. Dordrecht/Kluwer Academic Publishers. Rowley, Anthony (2002): Liacht as de sproch: grammatica della lingua mòchena.

Pubblicazioni dell’Istituto mòcheno di cultura, Palù del Fersina. Schweikert, Walter (2004): “The Order of Prepositional Phrases in the Structure of the

clause”. PhD dissertation, University of Venice.


33 – 2008

95

HEARSAY AND REPORTED SPEECH:

EVIDENTIALITY IN ROMANCE

Silvio Cruschina, Eva-Maria Remberger

«DICICA. Dice che. Non “si dice che”, ma uno solo, innominato, “dice che”. È l’incipit di ogni aneddotica malignità, di ogni racconto sulle disgrazie altrui. Il

“dicica” alleggerisce la responsabilità del narratore, come nel “si dice” italiano, ma al tempo stesso rende più segreta, più esclusiva, più preziosa e godibile la notizia.

Non lo sanno tutti. Era uno solo a saperla. E ora siamo in tre». (L. Sciascia, Occhio di Capra, 1984: 53)

1. Introduction

The grammatical category of evidentiality is crosslinguistically realized by different strategies. The use of forms derived from verba dicendi constitutes one possible device. For instance, evidentiality can be marked by functional elements like dizque in some American Spanish and Portuguese varieties (cf. Aikhenvald 2004). In this paper, we show that this strategy is not restricted to these languages, but is also widespread in European Romance varieties such as Galician (Gal.), Romanian (Ro.), Sardinian (Srd.), and Sicilian (Sic.). In all varieties these functional elements derive from the verb ‘say’ + complementiser (SAYC), with a meaning similar to English ‘apparently’ or ‘allegedly’:

(1) Sp. Esto dizque va a ser pantano. (Kany 1944:172) this SAYC go.PRES.3S to be.INF swamp ‘This is going to be swamp, they say.’

(2) Sic. Dicica ci avivanu finutu i grana. SAYC to-him.CL have.IMPF.3P finish.PP the money ‘Apparently they had run out of money.’

(3) Srd. Custas columbas nachi s’abbaidana e an cominzadu a faeddare. (AdS) these pigeons SAYC REFL look.PRES.3P and have.PRES.3P start.PP to talk.INF

‘These pigeons look at each other and started talking, they say.’


96

(4) Ro. Cică banul n-aduce fericirea. (L. Ardelean) SAYC the money not bring.PRES.3S the happiness ‘Money doesn’t give you happiness, they say.’

(5) Gal Disque a filla da Antonia marchou á Coruña vivir co mozo. SAYC the daughter of-the A. go.PAST.3S to C. live.INF with-the boyfriend. ‘Apparently Antonia’s daughter went to C. to live with her boyfriend.’

In reference to Colombian Spanish, Travis (2006:1276-1277) shows that the SAYC element, even if “not fully grammaticalized”,1 has undergone changes typical of grammaticalisation processes (decategorialisation, inseparability, phonological reduction, semantic generalisation). If we apply the tests given in Travis and several others to the data of the varieties under investigation here, the grammatical status of the SAYC element becomes clear and incontrovertible. Nevertheless, we also must assume different stages in the grammaticalisation process in order to account for the grammatical and functional differences found across the languages in question, in which the roles of the SAYC elements vary from a reported speech marker to an evidential marker proper. Following a cartographic approach to syntactic structures, we show that these markers occupy the functional projection dedicated to evidentiality, namely, the Evidential Mood projection, as identified by Cinque (1999). Moreover, our analysis will provide evidence for a clear-cut distinction between epistemicity and evidentiality, a distinction that is also represented in Cinque’s work in terms of two distinct and independent designated functional projections.

1.1 Types of evidentiality

In recent years, the phenomenon of linguistic evidentiality has aroused increasing interest among researchers in the field of semantics and pragmatics (cf. Dendale et al. 2001, Aikhenvald 2004). The definition of evidentiality given by Aikhenvald is “a linguistic category whose primary meaning is source of information” (2004:3). The name of this relatively new linguistic category comes from the “kind of evidence a person has for making factual claims” (Anderson 1982:273), which is expressed by grammatical means.

There has been a long debate on the relationship between evidentiality and modality, especially epistemic modality. Whereas recent approaches, such as that of Aikhenvald (2004) given above, strictly distinguish between the two and therefore recognize evidentiality as a linguistic category proper, classical works on modality, e.g. Palmer (2001), consider evidential marking to be simply a particular instance of

1 We use the term grammaticalisation in a broad sense, as the general process consisting of those phenomena involving change and reanalysis of linguistic forms.

Hearsay and reported speech: evidentiality in Romance

97

epistemic modality. Epistemic modality, i.e. the “degree of certainty” (or “kind of CAUSAL source”, cf. Anderson 2001:311) and evidentiality as defined above may indeed overlap. Thus, Dendale et al. (2001:342) list a tripartite use of the term evidentiality: Evidentiality in the broad sense (e.g. Palmer 2001) where evidentiality is part of the system of epistemic modality (or vice versa), since both characterize the speaker’s “attitudes about the epistemic status of information” (Dendale et al. 2001:340) towards the proposition encoded in an utterance; Evidentiality in a

narrow sense (e.g. Anderson 1982, Willett 1988, Aikhenvald 2004) where the focus lies on the expression of the “information or sources of knowledge behind assertions” (Dendale et al. 2001:340); if evidential marking results in an interpretation that reveals the speakers’ (e.g. positive or negative) attitudes towards the evidentially-marked proposition, this is purely a consequence of pragmatics and has nothing to do with grammatical epistemicity; Overlapping of evidentiality and

epistemicity: this view is adopted particularly for the subfield of inferential evidentiality, which might be easily interpreted as a type of epistemic modality (cf. van der Auwera & Plungian 1998:86 following Dendale et al. 2001:242, Plungian 2001:354). Although we admit that evidentiality and other grammatical phenomena are interacting, in this paper we assume the first position and treat evidentiality as a grammatical category standing on its own.

Given evidentiality as a grammatical category, several subsystems have been proposed. The most elaborate can be probably found in Aikhenvald (2004). Yet, since we concentrate on a subfield of evidentiality, i.e. hearsay and reported speech, for our aims the system delineated by Willett (1988) will suffice. Willett distinguishes between three types of evidentiality, namely direct, indirect and inferential evidentiality (Willett 1988:57). Direct evidentiality is given when the speaker has (and explicitly marks) first-hand sensory evidence for the assertion he makes. Indirect evidentiality covers second- and third-hand evidence and is equivalent to reportative evidence. Inferential evidence is given whenever the speaker marks his assertion as having been deduced. This paper, of course, is concerned with indirect, i.e. reportative evidence. For reportative evidentiality, there is a further distinction, to which we will refer throughout this paper: (6) Types of reportative evidence (cf. Willett 1988:96)

a. second-hand evidence (hearsay): “the speaker claims to have heard of the situation described from someone who was a direct witness”, ‘he says’

b. third-hand evidence (hearsay): the speaker claims to have heard about the situation described, but not from a direct witness”, ‘I heard’, ‘I hear tell’, ‘it is said’, ‘they say’ (impersonal)

c. folklore: the speaker claims that the situation described is part of established oral history (fairy tales, mythology, oral literature, proverbs and sayings)


98

All types of reportative evidence are found in the evidentials appearing in the Romance data at issue in this paper, as will be shown below.2

1.2 Evidentiality in Romance

Romance languages lack a dedicated paradigm or system of grammatical evidentials. This does not mean that there are no ways to express evidentiality in Romance. The Romance languages express evidentiality through various modal and temporal forms, following a pattern that is typologically characteristic of the languages of Western Europe, where the development of evidentials from modal morphemes is a typical feature (cf. da Haan 2005). Evidentiality as an actual grammatical phenomenon within the Romance verbal system has been recognized and argued for in recent work (cf. Squartini 2001, 2004, 2005, Calaresu 2004).

The conditional, the future, the imperfect, and periphrases with modal verbs (e.g. dovere/devoir/deber + infinitive), are traditionally assumed to take on evidentiality meanings in specific contexts. Although it is undoubtedly true that these verb forms convey evidentiality meanings, their primary role is not to mark the source of information. Evidentiality is just one of the modal and aspectual meanings they can express. Following Aikhenvald (2004), we call these forms “evidential strategies”, as opposed to evidentiality and evidentials proper. The conditional, for instance, can be used with a quotative meaning in French:

(7) Fr. Il y aurait de nombreuses victimes. (Dendale and Tasmowski 2001:345) EXPL there have.COND.3S of numerous victims ‘It is said/it seems that there are many victims’

However, “[t]his does not mean that this conditional has ‘become’ an evidential” (Aikhenvald 2004:105). The conditional as well as the other Romance evidential strategies resorting to tense and modal specifications of the verbal system typically convey epistemic values, in particular uncertainty and non-responsibility for the information expressed. The interpretation associated with the non-firsthand source of information, therefore, must be considered a “semantic extension” of these forms which frequently overlaps and coexists with a range of modal notions including probability, possibility, factuality, subjectivity, and responsibility. Indeed the

2 There is a further confusion concerning a more specialised meaning for “reported” vs. “quotative” evidentiality. Aikhenvald (2004:177ff) e.g. states that “if a language has two reported type evidentials, the most common distinction is that between reported (stating what someone else has said without specifying the exact authorship) and quotative (introducing the exact author of the quoted report).” Thus, “reported” in this sense just means third-hand or folklore, i.e. (6)b and (6)c above, whereas “quotative” is synonym of second-hand evidence, i.e. (6)a.


99

boundaries between evidentiality and epistemic modality in these strategies are often blurred and dependent on the context.

In Romance, in addition to these strategies realized through the verbal system, various lexical strategies, such as adverbs (Fr. apparemment, visiblement), ‘parenthetic’ expressions (Fr. il paraît, il semble), and verbs of saying (Fr. ils disent

que), are used to express evidentiality or evidential-like meanings. These seem to be universal strategies. The impersonal use of verbs of saying, i.e. a matrix clause containing the verb of saying and followed by a complement clause introduced by the complementiser (Fr. on dit que, It. si dice che, Sp. se dice que), is also a very common and widespread means of marking third-hand reported speech.

As pointed out in Willett (1988:79) and Aikhenvald (2004:271ff), the development of evidential strategies into evidential markers is not rare, and the grammaticalisation of verbs of saying + complementiser represent a typical case, as confirmed by our data. The phenomenon has been observed in reference to some varieties of Spanish, particularly Old Spanish (Kany 1994)3 and American Spanish (cf. Travis 2006). We show that it is more wide-spread than it has been assumed so far, since it is found in several modern European Romance varieties, such as Galician, Romanian, Sardinian, and Sicilian. SAYC results from the reanalysis of a biclausal construction comprising a verb of saying with a complement clause. This construction is reanalysed as monoclausal and the matrix verb together with its complementiser is reinterpreted as a single and functional unit, that is a grammatical marker stripped of its lexical features (cf. Aikhenvald 2004:272). The primary meaning of the SAYC element develops into one directly connected to the marking of the source of information. The SAYC elements therefore represent fully-fledged grammatical evidentials, although the path towards grammaticalisation seems not to be complete in all the varieties under investigation here.

2. Tests of Grammaticalisation

In the Romance varieties at issue here, the evidential marker SAYC stems from a verbal form of the verb ‘to say’ plus the morphophonologically incorporated complementiser ‘that’. The original verbal form is normally the third person

3 Indeed, in the database CORDE (cf. Real Accademia Española, CORDE), we find several examples of dizque, especially in texts from the end of the fifteenth century, but with the first example dating back to 1293, cf. (i): (i) E tal postura dizque auien que luego que nascie y el ninno que luegol ponien en quitaçión. (Gran Conquista de Ultramar, anonymous) and such position SAYC have.PAST.3P that after that be-born.PAST.3S there the child that then-it.CL put.PAST.3P in dismission ‘And they were said to have such a position that, after the child was born, they were dismissed.’


100

singular, as in Sicilian, Spanish and Romanian. However, in Galician, it is an impersonal version of the third person singular, namely that constructed with the originally reflexive pronoun se. In Sardinian, there is diatopic variation as far as SAYC is concerned and the corresponding verbal form might stem either from a third person singular or from a third person plural. In any case, as has also been observed by Travis (2007:1276), in all varieties at issue the SAYC element is at an advanced, even if perhaps not completed, stage of a grammaticalisation process. That means that on all systematic levels we find the typical phenomena commonly used as diagnostics (cf. e.g. Heine 1993), i.e. phonological erosion (cf. 2.1), morphological decategorisation (cf. 2.2.), change in syntactic distribution and properties (cf. 2.3), as well as desemanticisation (cf. 2.4). In the Romance varieties under discussion, these phenomena manifest themselves to different degrees (cf. the table in (36)).

2.1 Phonology

In all varieties under examination here the SAYC elements show phonological reduction (phonological erosion) or, as in Sicilian and Sardinian, some other phonological peculiarities. In Colombian Spanish, the form dizque was originally dice que, i.e. ‘he/she says that’; the final vowel of the verbal form was apocopated and the complementiser fused with the verbal form to result in the SAYC-element. Following Travis (2007:1276) and Kany (1944) dizque can be further reduced to ’izque or also ihke and ike (e.g. in Venezuela, cf. Kany 1944:169). In Sicilian, the high vowels /i/ are different in quality in dicica, where they are [+ATR], from the full verb use of the same verb, where they are [-ATR].4 This evidence for dicica as a single unit comes from the dialect of Mussomeli (CL):

(8) a. dicica [ˈdi-ʧi-ka] b. dici ca [ˈdɪ-ʧɪ ˌka]

Furthermore, the complementiser ca does not bear a secondary stress in the simplex grammaticalised form, as it typically does when part of the complex form.

We find several forms of SAYC in the Sardinian varieties as well, namely the forms nachi, nanchi (anchi), naca, nanca (anca), etc.5 These forms come from the

4 The different nature of the high vowels results from a vowel harmony process involving the feature [-ATR] that spreads regressively from the end of the word. This process is triggered by the presence of a high vowel at the end of the word: the final high vowels and all the preceding high vowels become [-ATR]. The harmony process does not take place with word-final a. High vowels within the word therefore stay [+ATR] (cf. Cruschina 2007). 5 Also the use of the compond perfect at nadu is quite frequent e.g. in fairy tales; however, there is no third-hand, but only second-hand use of this form, as far as we know. Thus, we will not treat this form, which also would contradict our claims made in 2.2, as part of the evidential SAYC-system of Sardinian.


101

third person plural and the third person singular of narrere ‘to say’ plus the complementiser chi (< Lat. quid) and ca (< Lat. quia) respectively, i.e. narrant

chi/ca and narat chi/ca (cf. Wagner 1951:397-398; Pittau 1972:90; Puddu 2000: s.v. nachi, anca; Blasco Ferrer 1984:258; Jones 1993:126-127). The verbal form has been contracted (narat > nat

6, narrant > nant), phonologically reduced (the t ending has been elided) and fused with the complementiser to a single form. Sometimes we find two of the above-mentioned variants, one with the second n and one without, in one and the same speaker (AdS; Irgoli):

(9) Nanchi it una emina chi nachi teniata trer fizos… (AdS) SAYC be.IMPF.3S a woman who SAYC have.IMPF.3S three sons 'It is said that there was a woman who alledgedly had three sons...'

As a consequence of this phonological reduction and fusion no paragogical vowel, i.e. a word final copy of the vowel in the immediately preceding syllable (cf. e.g. Bolognesi 1999), can be inserted between the verbal form and the complementiser (e.g. *nanachi), in contrast to the use of the verb ‘say’ as a full verb, where the paragogical vowel appears in phrase final or clause final position:

(10) Náranta gi bénidi zu zíndiku (Blasco Ferrer 1984:258) say.PRES.3P that come.PRES.3S the mayor 'They say that the mayor is going to come.'

In Galician, the full verbal form underlying disque is not the third person singular di plus que alone, but must be dise que, that is the third person singular form plus an enclitic impersonal pronoun (se) ‘one says that’.7 The clitic in the SAYC use shows final vowel elision.

In Romanian, the SAYC-element cică is a contraction of (se zi) ce că ‘one says that’ (cf. DEX: s.v. cică) or zice că ‘he/she says that’ (Tiktin: s.v. cică). Here too, we find other regional forms and variants like şi câ, o-s că (< a zis că / o zis că, ‘he/she has said that’ in the compound perfect) (Moţei n.y.). Our data, however, will concentrate on the form cică alone.

2.2 Morphology

Travis points out two major morphological characteristics of dizque, which differ from the full verb. First, only the full verb decir can be separated from the complementiser que, whereas dizque functions as a single and indivisible unit. 6 However, these short forms of narrere like nat ‘he says’ can also be used in poetic language, cf. Puddu (2000: s.v. narrere). 7 Note that in some contexts (e.g. subordinate clauses), in Galician, the enclitic position of the impersonal pronoun would not be possible. Nevertheless disque stays invariant in these contexts, a fact that confirms that it is an amalgamate form.


102

Second, dizque is always morphologically invariable and in no case can it be inflected: (11) Sp. dizque → *diceque (PRES), *decíaque (IMPF), *dijoque (PAST) …

The reduced form dizque in Colombian Spanish does not encode any morphological information concerning person, number, tense and mood. Any attempt to assign a morphological specification to the crystallised form dizque results in ungrammaticality, even in a context with a past reference-time. Equally, the phonologically reduced forms of SAYC in the other languages also prove to be morphologically fossilised, inseparable and not compatible with tense and mood inflection. For example, Sicilian dicica can replace the full-length verbal form, irrespective of its temporal and modal specification, but it always conveys an evidential value and has to appear in its invariable form:

(12) Dicivanu ca / Dicica / *Dicivaca a mugghieri ci fa / faciva i corna c’u nutaru. say.IMPF.3P that/SAYC/SAYC+IMPF the wife to-him.CL do.PRES.3S/do.IMPF.3S the horns with the notary ‘They said / it is said that his wife is / was cheating on him with the notary.’

Sicilian SAYC, thus, cannot encode any temporal or modal specification and is incompatible with any type of inflection (cf. Menza in press). The same is true of Sardinian nachi, Galician disque and Romanian cică:

(13) a. Sic. dicica → *dicivaca (IMPF), *dissica (PAST), *dicissica (SUBJ) … b. Srd. nachi → *naratchi (PRES), *naraiatchi (IMPF), *naduchi (PP) … d. Gal. disque → *diseque (PRES), *digaseque (SUBJ), *dicíaseque (PAST) c. Ro. cică →*zicecă (PRES), *zicăcă (SUBJ), *ziceacă (IMPF), *ziscă (PP) …

The inseparability of SAYC, which behaves as a single unit not divisible into the two original elements, and the morphological invariance, resulting from the impossibility of inflecting for any verbal feature, are the main morphological properties offering primary evidence for the grammatical status of SAYC.

2.3 Syntax

We have seen that SAYC has been reanalysed as a single unit. Not only has the verb of saying lost all its morphological and syntactic traits, but furthermore the complementiser no longer functions as such and is merely part of the unit SAYC. The syntactic consequences of this reanalysis are the following:

(14) a. SAYC can be used in isolation, for instance in an answer to a question. b. It is incompatible with negation and other lower adverbs. c. It can occur with other complementiser(-like) elements.

The main function of a complementiser is to introduce an embedded clause. Complementisers are therefore incompatible with the final position of the sentence


103

(at least in SVO languages). Due to its reanalysis and grammaticalisation, SAYC is not subject to this restriction. It can appear at the end of a sentence when used as parenthetical (cf. also 4.1) and it can also appear in isolation, especially in an answer to a question:

(15) Sic. Chi jè veru ca Maria av’a partiri pi l’America? – Dicica! INT be.PRES.3S that Maria have.PRES.3S to leave.INF for the America SAYC ‘Is it true that Mary is going to leave for America? – Apparently!’

(16) Srd. Nachi muzere tua s'est illierada?! – Nachi! (Puddu 2000: s.v. nachi) and SAYC wife your REFL be.PRES.3S liberate.PP SAYC ‘Your wife is said to have given birth? – They say so!’

(17) Gal. Entón Anxo vendeu o piso? – Disque (si). so Anxo sell.PAST.3S the flat SAYC yes ‘So Anxo sold the flat? – Apparently (so).’

In Romanian, cică can be used in isolation, e.g. as an answer to a question, but it is normally reinforced by ‘yes’ or ‘no’:

(18) Ro. E adevărat că Ion pleacă la New York? – Cică da. / Cică nu.

be.PRES.3S true that Ion leave.PRES.3S to New York SAYC yes SAYC no ‘Is it true that John goes to New York? – Alledegly yes / no.’

Verbs can typically be negated and modified by lower verbal adverbs (e.g. adverbs of time, manner, etc.). By contrast, the grammatical nature of SAYC leads to its incompatibility with negation (e.g. Sic. *un dicica) as well as with any modifier (such as ‘always’, cf. Sic. *dicica sempri, or ‘often’, cf. Sic. *dicica spissu), unless they modify the verb of the clause. Unaivalability within the scope of negation is a typical property of evidential markers across languages (cf. Willett 1988, de Haan 1999), although it is not a universal property (cf. Aikhenvald 2004:96-97). Thus, negation can take scope over the speaker’s statement, but not over the source of information. The same happens with adverbs which cannot modify the evidential marker SAYC, but can, on the other hand, modify the stated information (cf. the ungrammaticality of the English expressions *always apparently, *often apparently or *already apparently).

Travis (2006:1276) provides another syntactic test suggesting that SAYC has undergone a grammaticalisation process. Given that the merged complementiser is no longer analysable as an actual complementiser (except in Sardinian and Romanian, as will be clarified below), SAYC can co-occur with a complementiser. There is variation, however, as to the function of the second occurrence of the complementiser (which in many cases is not a real, i.e. complement-introducing complementiser). SAYC can co-occur with a second instance of the complementiser in all languages considered here, performing various functions. This second (or


104

sometimes third) complementiser often introduces a subordinate causal or also consecutive clause. It can either precede or follow SAYC:

(19) Sic. Vippi troppu assà, (ca) dicica (ca) jera fattu stari na pezza. drink.PAST.1S too much that SAYC that be.IMPF.1S make.PP stay.INF a rag ‘I drank so much that apparently I was completely out of it.’

Like in Spanish, in these languages too the complementiser is never dependent on SAYC since this element is no longer able to take a complement clause. By contrast, Sardinian and Romanian present a different scenario. In these varieties the complementiser is in fact selected by SAYC, showing that it has not been fully grammaticalised and has not lost its verbal features entirely, insofar as it is able to take an object complement clause. In (20) the complementiser introduces an embedded clause that represents the object of SAYC:

(20) E nachi chi issa no b’andaiada nudda. (AdS) and SAYC that she not there.CL go.IMPF.3S nothing ‘And it is said that she didn’t go there at all.’

The example above proves that SAYC in Sardinian is able to take an argument clause. It can also take a subject, as in (21) (from Puddu 2000: s.v. nachi), showing that, as far as theta-role assignment is concerned, it can still behave as a normal verb with its own thematic-grid. It must be noted, however, that even when taking an argument, it is morphologically invariant, failing to agree with the subject:

(21) millu, cojadu, cudhu, mih: e nois nachi si faghiat preíderu! (Puddu 2000) look.IMP-him.CL married that look.IMP and we SAYC REFL do.IMPF.3S a priest ‘Look at him, married as he is, that one, look: and we always said he would become a priest!’

Since the complementiser integrated in Romanian cică is no longer analysable as a complementiser, there too it frequently occurs that că is repeated, either before or after cică, to introduce a complement clause (cf. (22) – the că in front of cică, again, has a causal meaning):8 (22) ... şi nu vine acasă de la serviciu că cică că are şedinţe. and not come.PRES.3S home from the job because SAYC that have.PRES.3S meetings ‘... and he doesn’t come home from work because he says that he has meetings – allegedly.’

If we analyse SAYC as an evidential marker, the syntactic features discussed in this section are straightforwardly accounted for. Moreover, these features prove to 8 The subjunctive particle să can also co-occur with cică to introduce an embedded complement clause.


105

be shared by other evidential adverbs, which leads us to the conclusion that SAYC has grammaticalised as an adverbial element. However, a clear distinction has to be made between languages like Sicilian and Galician, in which SAYC only exists as a grammatical marker of evidentiality, and languages in which SAYC, irrespective of its phonological reduction and its morphological invariance, can still function as a verb that assigns theta-roles. We will return to this distinction in section 3.

2.4 Semantics

The semantic bleaching (or desemanticisation) typical of elements which are on their way on a grammaticalisation path can be recognized in the following examples where SAYC is used together with the full verb ‘to say’: (23) Sp. Y dicen que diz que […] no más trabajan en el campo. (Kany 1944:172) and say.PRES.3P that SAYC not more work.PRES.3P in the field ‘And they say that apparently they don’t work on the field anymore.’

(24) Sic. Maria mi dissi ca dicica arrubbaru a machina au dutturi. Maria to-me.CL tell.PAST.3S that SAYC steal.PAST.3P the car to-the doctor ‘Maria told me that apparently the doctor’s car has been stolen.’

(25) Srd … e an cominzadu a faeddare, e nachi ana nadu: … (AdS) and have.PRES.3P start.PP to talk.INF and SAYC have.PRES.3P say.PP

‘… and they started to talk, and allegedly they said: ...’

(26) Ro. Se zice că cică ziua se cunoaşte de dimineaţă. (I. Creangă) REFL say.PRES.3S that SAYC the-day REFL know.PRES.3S from the-morning ‘It is said that apparently the day can be judged by its morning.’

(27) Gal. Un estado aconfesional como parece que disque di a Constitución non é lóxico. (www) a state nonconfessional like seem.PRES.3S that SAYC say.PRES.3S the Constitution not be.PRES.3S logical ‘A nonconfessional state as, apparently, the Constitution seems to say, is not logical.’

Since the lexical meaning of the SAYC-element is partially lost and now extended to contexts of more functional usage, the lexical verb ‘to say’ can be inserted again, as a form of “semantic strengthening” (cf. Travis 2007:1277) or “lexical reinforcement” (cf. Aikhenvald 2004). In (23) and (24), a form of lexical ‘say’ plus C has been inserted, in (25) and (27) without C, and in (26) an impersonal verb form is used to lexically reinforce the meaning of SAYC. As for its interpretation, SAYC in its prototypical use can no longer function as a verb, but functions instead as an adverbial with an evidential meaning, cf. the following


106

examples from Sardinian, which clearly represent a minimal pair and show that naki has lost its original meaning as a declarative verb:

(28) a. Juanne naki si k’est mortu.9 (Sa-Limba 1999-2007) Juanne SAYC REFL there.CL be.PRES.3S die.PP

‘Juanne is said to have died.’

b. Juanne narat ki si k’est mortu. (Sa-Limba 1999-2007) Juanne say.PRES.3S that REFL there.CL be.PRES.3S die.PP

‘Juanne says that he has died.’

(28)a clearly has an interpretable evidential meaning, whereas (28)b is semantically odd, because of the impossibility of a dead person to speak.

3. Functions of SAYC and degrees of grammaticalisation

As has been shown in 1.1 reportative evidentiality markers can be used to express either second-, or third-hand evidence, or knowledge based on a common ground of traditions, mythology or folklore. As far as the evidential marker at issue here, SAYC, is concerned, a more finegrained differentiation might be needed in some cases. Travis, for example, identifies four functions of dizque in Colombian Spanish (cf. Travis 2006:1278), namely a) reported speech (= second-hand, in Willett’s terms), b) hearsay (= third-hand and folklore, in Willett’s terms), c) labelling, and d) the dubitative use.

As explained in (6), based on Willett (1988), reported speech exemplifies the source of the information (second-hand), whereas hearsay does not (third-hand). Travis’ further distinction between reportative dizque with direct and indirect speech will be important for the Romance varieties under investigation. Following Travis, dizque with direct speech serves also to mark authenticity, whereas it might imply a flavour of doubt when used with indirect speech (Travis 2006:1280-1281). The hearsay function of dizque, where there is no identifiable source of information, is the most common. The labelling function is found when SAYC does not represent an adverbial but a modifier of a noun phrase (or even other phrases); this can sometimes be translated by ‘so-called’, ‘supposed’, or perhaps ‘so to speak’ (Travis 2006:1287):

(29) Se presentò como, dizque narcotraficante. (Travis 2006:1279) REFL present.PAST.3S as SAYC drug-dealer ‘He presented himself as a so-called drug dealer.’

9 In Sardinian, morrere, ‘to die’, is inherently reflexive (cf. Spanish morirse); furthermore, it is important to distinguish the complementiser ki from the second ki (in this example k'), which is a locative clitic (cf. Italian ci).


107

The dubitative use of disque can be “marking falsity”, but also “unintentional consequences” involving an element of surprise (Travis 2006:1291-2), cf. (30). This is certainly the most difficult function of dizque to identify, since it remains quite vague in definition. However, the common feature here is that there is no longer a recognizable speech act (Travis 2006:1293).

(30) Yo, por Dios, dizque a limpiar baños. (Travis 2006:1292) I by God SAYC to clean.INF toilets ‘And me, by God, I was there cleaning toilets!’

Following Travis (2006), all these functions also reflect different aspects of progressive grammaticalisation. Consequently, the range of possible functions varies from language to language, and not all of them can be found in every language at issue here: For example, Sicilian dicica can only mark reported speech in indirect discourse whereas Sardinian nachi can also mark direct reported speech; both languages can, of course, encode hearsay, and Sardinian in particular has a very extended use of SAYC in folklore; by contrast, the labelling function is less easy to find – perhaps even impossible – in the varieties under discussion. It is also not yet clear whether each variety has a genuine dubitative use of SAYC or if the dubitative interpretation is due to pragmatic implicatures.

Since the use of SAYC in folklore also seems to exist in Latin American Spanish (cf. Kany 1944:140), in addition to being very common in the Romance varieties at issue, we will add this function (observed by Willett 1988) to Travis’ classification. Thus we will use the following refined typology for SAYC (integrating Willett 1988 and Travis 2006) as a point of reference:

(31) I. second-hand10 Ia. direct reported speech Ib. indirect reported speech II. third-hand IIa. non specified IIb.folklore III. Labelling IV. dubitative

In what follows, we examine the functions given in (31) with respect to the Romance varieties at issue.

In Sardinian, nachi is typically used as a reportative evidential, which does not give the specific source of information, thus encoding third-hand evidence (function

10 One could make a further distinction here with respect to the explicit or anaphoric expression of the subject.


108

II).11 It is also extremely common in fairy tales, where it may appear after nearly every sentence, sometimes every constituent (thus function IIb). Besides this third-hand use, nachi can be used in direct (Ia) and indirect speech (Ib). As a direct discourse marker it often occurs with an explicit subject (not necessarily in the third person singular):

(32) “Eh!” nachi Gesù Crilthu “tando bi benzo abberu, azzetto s’invitu.” (AdS) eh SAYC Jesus Christ then there.CL come.PRES.1S really accept.PRES.1S the invitation. ‘“Eh!” said Jesus Christ “in this case I will indeed come with you, I accept the invitation.”’

(33) Gesù Crilthu nachi “Pedru” nachi “abbaida: pro te chelzo faghere una cosa totta noa...” (AdS) Jesus Christ SAYC Peter SAYC look.IMPER for you.CL want.PRES.1S do a thing all new... ‘Jesus Christ said “Peter” he said “look: For you I will do something completely new…”’

As far as the dubitative function (IV) is concerned, nachi may express surprise or wonder, but this use does not coincide with the dubitative function described in Travis (2006). The labelling function (III) is not found. Sardinian nachi has certainly reached a certain stage of grammaticalisation, as witnessed by its main function (II) to express third-hand evidence. However, in its second-hand use, especially when used as a direct speech marker with an explicit subject, it seems to be still analysable as a verb (which has strangely enough morphologically incorporated the complementiser).

The primary function of Sicilian dicica is that of a reported speech marker. It can refer to an explicit subject, identifying second-hand evidence (I). In this case, the source of information never occurs in the same sentence containing dicica, which is not able to take a syntactic subject. Rather, it coincides with the subject ‘under discussion’, established as such in the previous discourse. In fact, unless inferable from the context, any specific reference to or indication of the actual author of information is impossible:

11 See the characterization of naki given by Jones (1993:168-9): “Also within this class of modal adverbs we may include the 'hearsay' adverb naki […] which is frequently interspersed in reported speech […], but can also have a modal value roughly equivalent to ‘apparently’. It is also frequently used to prefix sentences with which broach a new topic of conversation; e.g. Naki ses cojuande. ‘I hear that you are getting married’.”


109

(34) (*Maria) dicica (*Maria) iddu un jiva d’accordu cu a soggira. Maria SAYC Maria he not went of agreement with the mother-in-law Intended meaning: ‘Mary says he didn’t get on well with his mother-in-law.’

If no subject is present in the discourse or salient in the context, dicica indicates that the speaker is reporting third-hand evidence (IIa). Dicica is sometimes used in sentences expressing surprise or false beliefs, but it is always related to the notion of speech. Therefore, it does not perform the dubitative function (IV). Dicica only rarely reports information as folklore (proverbs, stories, riddles, and songs, IIb). Another property of dicica, confirming its adverbial status, is that it can only take a full clause within its scope, never a noun or a prepositional phrase in the labelling function (III).

Galician disque performs more or less the same functions as Sicilian dicica. It is mainly used in second- (Ib) and third-hand (IIa) indirect reportative speech. As in Sicilian, Galician SAYC cannot take a subject and the exact source of second-hand information can only be gathered from the context. Consequently, it does not function as a direct speech marker, and cannot attribute the utterance to a specific author of the speech act. Unlike Sicilian, disque is also common as a folklore evidence marker (IIb) in fairy tales, stories, and proverbs. Moreover, the meaning of disque has not generalised to cover functions completely unrelated with the notion of speech. Hence, it cannot perform the conventionalised dubitative function (IV). Finally, the labelling function (III) is not found in Galician.

Also Romanian cică can be used in a range of functions. First, it can be used impersonally as an adverb, corresponding to “reportedly”, “allegedly”, as a third-hand evidence marker (IIa). Cică, then, is extremely frequent in fairy tales and folklore (IIb). It can also be used with an explicit indication of the source of information given by the context; in this case, cică does not appear in the same clause with the given source. Thus, it can still be interpreted as a third-hand evidence marker (IIa). Unlike Sicilian dicica and Galician disque, Romanian cică can also appear with an explicit subject in the third person singular in indirect discourse (Ib) (ex. from Delavrancea, following Macrea 1955-1957: s.v. cică):

(35) Moş popa, cînd spune de evanghelie, cică să rabzi şi iar să rabzi. old priest-the when speak.PRES.3S of gospel SAYC that be-patient.PRES.2S and again that be-patient.PRES.2S

‘The old priest when he preaches always says to be patient and to be patient again.’

However, in Romanian, in contrast to Sardinian, cică cannot appear as a direct discourse marker (Ia), neither with an overt nor with an implicit subject. The labelling and the dubitative functions seem not to be available with cică in Romanian. Thus, Romanian cică does not have all the functions that Sardinian nachi has. In particular, it cannot appear in direct discourse. Yet, it also has no labelling


110

function. Thus, as far as grammaticalisation is concerned, it can be said to lie somewhere in between Latin American Spanish (more grammaticalised) and Sardinian (less grammaticalised).

The functions of SAYC in Romance are summarised in the following table:

(36) The functions of SAYC in Romance

INDIRECT/REPORTATIVE EVIDENTIALITY LABELLING DUBITATIVE

second-hand (I) third-hand (II)

direct speech

(Ia)

indirect speech

(Ib)

hearsay

(IIa)

folklore

(IIb)

(III)

(IV)

LA Spanish + + + + + +

Sardinian + + + + - -

Romanian - + + + - -

Galician - + + + - -

Sicilian - + + - - -

4. SAYC and the Evidential Mood Phrase

We have already hypothesised that SAYC should show syntactic properties typical of evidential adverbs (cf. 2.3). This section sets out to identify the position of this use of SAYC within the sentence. Adopting a cartographic approach and in particular the hierarchy of functional projections proposed by Cinque (1999), we show that SAYC is generated in the specifier position of the Evidential Mood projection. This position is typically occupied by evidential adverbs, such as, in English, allegedly, reportedly, apparently, obviously, clearly, and evidently (Cinque 1999:86). Evidence for our claim is given with respect to the unmarked position of

SAYC within the sentence, adopting some ordering tests within the higher adverbial field.

4.1 The position of SAYC within the clause

In compliance with the syntax of adverbs, SAYC can appear in different positions of the sentence, especially in parenthetical uses, including the final position. SAYC, however, can be identified as a higher sentence adverb which takes the whole sentence within its scope, and its natural position is thus sentence initial. In the unmarked word order, it precedes the subject:

(37) Sic. Dicica Maria jè malata.

(38) Gal. Disque María estaba enferma.


111

(39) Srd. Nachi Juanne est malàidu.

(40) Ro. Cică Ion e bolnav. SAYC Mary/John be.PRES.3S ill ‘Apparently Mary/John is ill.’

On the other hand, it can be preceded by a dislocated topic of the sentence, typically (but not exclusively) coinciding with the subject. In this case, the topicalised constituent does not fall under the scope of the evidential marker:

(41) Sic Maria dicica jè malata.

(42) Sp. María disque estaba enferma.

(43) Srd. Juanne nachi est malàidu.

(44) Ro. Ion cică e bolnav. Mary/John SAYC be.IMPF.3S ill ‘As for Mary/John, apparently s/he is ill.’

Examples (37)-(40) offer minimal pairs with the sentences in (41)-(44). The propositional meaning of the sentence pairs is the same. In both word orders SAYC simply means that someone said that Mary/John is ill, but it does not specify the source of this information. We know that the constituent preceding SAYC is the subject of the sentence, which has been left-dislocated in association with a marked information structure of the utterance. The order SAYC-Subject therefore represents the unmarked word order, which can be altered for discourse-related purposes. The topic usually coincides with the subject, but it does not have to be necessarily the case, as shown in the following examples containing a dislocated direct object and a dislocated locative complement, respectively:

(45) Sic. I robbi dicica si l’accatta ’ntra na putìa fina di Palermu. the clothes SAYC REFL them.CL buy.PRES.3S at a shop posh of Palermo ‘As for her clothes, apparently she buys them in a posh shop in Palermo.’

(46) Srd. ... idende unu castello. In custu castello nachi bi istaiada su fizu ’e su re. see.GER a castle in this castle SAYC there.CL stay.IMPF.3S the son of the king ‘and there they saw a castle. Apparently the son of the king was living in this castle.’ (AdS)

The order SAYC-Subj and Top-SAYC are the most common orders. This syntactic distribution reflects the status of the SAYC elements as sentence modifiers, which occur in sentence-peripheral positions, usually at the beginning of the sentence. Other orders, however, are possible as a consequence of the fact that adverbs typically allow for parenthetical uses. It has been observed that higher adverbs generally cannot occur after a finite verb (cf. Cinque 1999:31). This


112

constraint, however, is overridden whenever the adverb is used parenthetically. As a parenthetical, an adverb can occur at major constituent boundaries (e.g. between the subject and the verb or, more rarely, between the verb and peripheral elements within the VP such as PPs or clauses), and also in final position. Like adverbs that appear at the end of the sentence, SAYC is typically deaccented and prosodically separated from the preceding sentence by a pause (indicated by a comma in the following examples):

(47) Sic. Maria jera malata, dicica

(48) Gal. María estaba enferma, disque. Mary was ill SAYC ‘Mary was ill, apparently’

As pointed out by Belletti (1990:130, n. 29) and Cinque (1999), a higher adverb can occur after a finite verb also when another constituent follows the adverb: (49) Sic. Pinu si sciarrià cu Turiddu, dicica pi sordi Pinu REFL fight.PAST.3S with Turiddu, SAYC for money ‘Pinu had an argument with Turiddu, apparently for money’

In these examples, SAYC does not have scope over the preceding sentence, but just over the following constituent. However, the frequent placement of SayC at the beginning of the sentence, together with its positional mobility, confirms the claim that SAYC behaves as modal adverb.

4.2 The adverbial hierarchy and ordering constraints

Following Cinque (1999), we assume that adverbs occupy the specifier positions of a rich and articulated series of functional projections obeying a rigid hierarchical order. If SAYC is merged into the specifier of the Evidential Mood Projection then some ordering constraints with respect to other verbs are expected, giving rise to the following predictions:

(50) a. SAYC should not be compatible with other evidential adverbs competing for the same position; b. it should be followed, but not be preceded, by epistemic and possibility adverbs; c. it should be preceded, but not followed, by speech act and evaluative adverbs.

Now, corresponding to these predictions, Cinque’s hierarchical articulation of the (higher) adverbs indeed implies the following ordering constraints:

(51) a. *apparently + allegedly/clearly/evidently/obviously b. apparently + probably/possibly; *probably/possibly + apparently


113

c. frankly/honestly/unfortunately + apparently, *apparently + frankly/ honestly/ unfortunately

In Sicilian, SAYC is not the only modal adverb derived from the merging of a verb and a complementiser. Other adverbs show the same historical development: The adverb parica, ‘seemingly’, ‘apparently’, also functions as an evidential marker reporting third-hand evidence, and derives from the third person singular of the verb pariri ‘to seem’ plus the complementiser ca. The epistemic adverb pènzica expressing probability comes from a form of the verb pinzari ‘to think’, to which the complementiser has been incorporated. This grammaticalisation process peculiar of Sicilian modal adverbs has also caused the reanalysis of the sequence copula + adjective + complementiser: the locution jè capaci ca (lit. ‘it is capable/possible/ likely that’) has been grammaticalised to the form capacica, which now represents a single unit. The hierarchical system of the Sicilian higher adverbs, equivalent to Cinque’s (1999) system, is illustrated in (52) (cf. also Ledgeway and Lombardi 2005:81 for Calabrian):

(52) Sic. [onestamenti Moodspeech act [pi furtuna Moodevaluative [dicica, parica Mood evidential [pènzica, probabilmenti Moodepistemic [na vota, tannu T(past) [pua T(future) [forsi Modirrealis [pi forza Modnecessity [capacica Modpossibility [...Asp...

The predictions in (50), obeying the ordering constraints in (51), are indeed born out by our data. Dicica proves incompatible with other evidential adverbs (e.g. pàrica), but it can co-occur with epistemic adverbs (pènzica, probabilmenti), as shown by the following examples:

(53) Sic. *Dicica parica / *Parica dicica ora si senti bonu. SAYC SEEMC / SEEMC SAYC now REFL feel.PRES.3S good ‘Apparently s/he is feeling better.’

(54) Sic. a. Dicica pènzica/probabilmenti s’accattà na casa nova. SAYC THINKC probably REFL buy.PAST.3S a house new ‘Apparently s/he probably bought a new house.’ b. *Pènzica/probabilmenti dicica…

The Evidential Mood Projection is higher than the position of possibility adverbs, such as capacica. Dicica can thus precede, but not follow capacica:

(55) Sic. a. Dicica capacica ci avivanu finitu i sordi. SAYC possibly to-him/her.CL have.IMPF.3P finish.PP the money ‘Apparently s/he possibly ran out of money.’ b. *Capacica dicica...

The opposite ordering applies with speech act adverbs, like onestamenti, which can only precde dicica:


114

(56) Sic. a. Onestamenti dicica iddu si scantava. honestly SAYC he REFL be-scared.IMPF.3S

‘Frankly/honestly they say that he was scared.’ b. *Dicica onestamenti...

The same ordering contraints occur in Galician, where epistemic (probablemente) and possibility (posiblemente) adverbs cannot precede disque, while the same sentences prove grammatical if they follow it. Along the same lines, evaluative adverbs (such as desafortunadamente) need to be placed before disque. Romanian cică cannot be combined with parcă ‘seemingly’, another evidential marker stemming from se pare că ‘it seems that’ (cf. Macrea 1955-1957: s.v. parcă; Berceanu 1971:55-56; DEX s.v. parcă) in the same clause, etc. This impossibility is readily accounted for under the analysis that the two evidential adverbs compete for the same position, namely the specifier of the Evidential Mood Projection.

5. Conclusions

Based on comparative evidence, our paper offers a systematic analysis of the different functions and the varying degrees of grammaticalisation of SAYC in Romance. Firstly, we show that SAYC is found in many Romance varieties, and not just in Latin American Spanish. Secondly, we prove that SAYC has different functions and is characterised by varying degrees of grammaticalisation according to each language. In their prototypical, unmarked use, the SAYC-elements sit in the specifier position of the Evidential Mood Projection. Like adverbs, they can also appear in all kinds of parenthetical positions in the sentence. Sardinian and Romanian also allow for a (defective) verbal use of SAYC. In Latin American Romance, when performing the labelling function, SAYC can also occur within the DP.

Acknowledgements

We would like to thank Beatriz Vizcaíno Pena and Doina Vacarciuc for their grammatical judgements on Galician and Romanian, respectively. For Sardinian, we would like to thank the members of the mailing list Sa-Limba (cf. Sa-Limba 1999-2007). Further help with the extraction of the data came from Florian Scheib (Romanian), Heidrun Bohnet (Spanish) and Christian Ferraro (Sardinian). Part of the research of Eva-Maria Remberger concerning this paper has been supported by the DFG with the project “Auxiliary verbs – status, development, variation” within the SFB 471 “Variation and development in the lexicon” at the University of Konstanz as well as by a Feodor-Lynen-Fellowship granted by the Alexander von Humboldt-Foundation for a one year stay in Cambridge, which made this joint paper possible. We would also like to thank Adam Ledgeway for insightful comments.


115

References

Aikhenwald, Alexandra (2004): Evidentiality. Oxford: Oxford University Press. Anderson, Lloyd B. (1982): Evidentials, Paths of Change, and Mental Maps: Typologically

regular Asymmetries. In: Wallace Chafe & Johanna Nichols (eds.): Evidentiality: The

linguistic encoding of epistemology. Norwood: Ablex, 273-312. AdS = Archivi del Sud (1996): Contami unu Contu. Racconti popolari della Sardegna. Vol. I

Logudoro. Archivi del Sud: Alghero. Berceanu, Barbu B. (1971): Sistemul gramatical al limbii române. Bucureşti: Editura

Ştiinţifică. Blasco Ferrer, Eduardo (1984): Storia linguistica della Sardegna. Tübingen: Niemeyer. Bolognesi, Roberto (1999): Per un approccio sincronico alla linguistica e alla standar-

dizzazione del sardo. In: Roberto Bolognesi & Karen Helsloot (eds.): La lingua sarda. Atti del II Convegno del Sardinian Language Group. Cagliari: Condaghes.

Calaresu, Emilia (2004): Testuali parole: la dimensione pragmatica e testuale del discorso riportato. Milano: Franco Angeli.

Cinque, Guglielmo (1999): Adverbs as functional heads. Oxford: Oxford University Press. Dendale, Patrick & Liliane Tasmowski (2001): Introduction: Evidentiality and related

notions. In: Journal of Pragmatics 33, 339-348. DEX = Coteanu, Ion (ed.) (1998): Dicţionarul explicativ al limbii române. Bucureşti: Univers

Enciclopedic. Heine, Bernd (1993): Auxiliaries. Cognitive forces and grammaticalization. New York:

Oxford University Press. Jones, Michael Allan (1993): Sardinian Syntax. London, New York: Routledge. Kany, C. E. (1944): Impersonal dizque and its variants in American Spanish. Hispanic Review

12, 168-177. Ledgeway, Adam & Alessandra Lombardi (2005): Verb movement, adverbs and clitic

positions in Romance. Probus 17, 79-113. Macrea, Dimitrie (1955-1957) (ed.): Dicţionarul limbii romîne literare contemporane.

Bucureşti: Editura Academiei Republicii Populare Romîne. Menza, Salvatore (in press): I paraverbi del siciliano. In: Studi in onore di Nicolò Mineo.

Facoltà di Lettere, Università di Catania. Moţei, Anamaria Iulia (no year): Le verbe dire et ses dérivés – entre unités lexicales à sens

plein et marqueurs de reformulation paraphrastique (MRP). Essay. Universitatea Babeş – Bolyai, Cluj <http://lett.ubbcluj.ro/rtf-uri/Motei_AnaMaria.htm> [30.01.2008]).

Palmer, F. R. (2001): Mood and Modality. Cambridge: Cambridge University Press. Piccitto, Giorgio & Giovanni Tropea (1977-2002): Vocabolario Siciliano. 5 Vol. Catania:

Centro di studi filologici e linguistici siciliani, Opera del Vocabolario siciliano. Pittau, Massimo (1972): Grammatica del sardo-nuorese: il più conservativo dei parlari

neolatini. Bologna: Patron.


116

Plungian, Vladimir A. (2001): The place of evidentiality within the universal grammatical space. Studies in Language 12/1, 349-357.

Puddu, Mario (2000): Ditzionàriu de sa limba e de sa cultura sarda. Cagliari: Condaghes. (cf. also: http://www.ditzionariu.org).

Real Academia Española: Banco de datos (CORDE) [online]. Corpus diacrónico del español. <http://www.rae.es> [1.12.2007].

Sa-Limba (1999-2007): Electronic Corpus of the Postings to the Mailing List ‘Sa-Limba’. Köln, Berlin <http://www.lingrom.fu-berlin.de/sardu/sa-limba.html> [14.02.2008].

Squartini, Mario (2001): The internal structure of evidentiality in Romance. Studies in

Language 25(2), 297-334. Squartini, Mario (2004): Disentangling evidentiality and epistemic modality in Romance.

Lingua 114, 873-895. Squartini, Mario (2005): L’evidenzialità in rumeno e nelle altre lingue romanze. Zeitschrift

für romanische Philologie, 121(2): 246-268. Tiktin, Heimann Hariton (1903-1925): Rumänisch-deutsches Wörterbuch. Bucureşti. Travis, Catherine (2006): Dizque: a Colombian evidentiality strategy. Linguistics 44-6 (2006),

1269-1297. Van der Auwera, Johan & Vladimir A. Plungian (1998): Modality's Semantic Map. Linguistic

Typology 2, 79-124. Wagner, Max Leopold (1951): La lingua sarda. Storia, spirito e forma. Francke: Bern. Willett, Thomas (1988): “A crosslinguistic survey of the grammaticization of Evidentiality.”

In: Studies in Language 12/1, 51-97.


33 – 2008

117

A CASE OF INCOMPLETE JESPERSEN’S CYCLE

IN ROMANCE*

Jacopo Garzonio

0. Introduction

In this paper, I examine some diachronic and synchronic aspects of the syntax of negative expressions in Florentine, a dialect of Central Italy. More precisely, I describe and analyze the development of the negative adverb punto (lit. ‘point’). This element was in origin a minimizer (in the sense of Horn 1989), and in early stages of Florentine became a postverbal marker used to strengthen the negation. This change is a typical first step in the well known process called Jespersen’s cycle (Jespersen 1917), whereby the preverbal negative marker of a language is firstly weakened, then strengthened by a postverbal element and finally supplanted by it. As I will show, a crucial difference between a typical Jespersen’s cycle and the historical development of negation in Florentine is that Florentine never lost the preverbal negative marker and punto has been kept as an aspectual/intensity adverb, corresponding to ‘at all’.

While other minimizers in Northern Italian Dialects, like pa in Piedmontese, briza in Emilian or mi(n)ga in Milanese have become negative markers and have lost their quantificational meaning, punto is still a quantifier in Modern Florentine. I will argue that this is one of the reasons why Jespersen’s cycle was not completed in Florentine. Adopting Roberts & Roussou’s (2003) theory about grammaticalization, I will claim that a postverbal negative adverb can become the unique or the main negative marker only if it undergoes a process of simplification of features, i.e. the loss of quantificational or phi-features.

* I am grateful to the audience of the XXXIV IGG for the helpful discussion. Furthermore, I wish to thank all the people who gave me precious advice and support for this research: Paola Benincà, Anne Breitbarth, Anna Cardinaletti, Andrea Cattaneo, Luciano Giannelli, Giuliana Giusti, Nicoletta Penello, Diego Pescarini, Cecilia Poletto. I thank an anonymous reviewer for the helpful comments on a preliminary version of this paper.

Jacopo Garzonio

118

Moreover, I will present some data about punto that suggest that the position of negative adverbs derived from minimizers in the clause structure is the same position reached by quantified internal arguments of the predicate.

The paper has the following structure: in section 1, I present some general facts about the adverb punto and the syntax of Florentine; in section 2, I describe the syntax of punto in Old Florentine; in a parallel way, I deal with the syntax of punto in Modern Florentine in section 3; section 4 is dedicated to the analysis of the historical development of this adverb; in section 5, I propose that punto occupies a precise position in the adverbial hierarchy; section 6 concludes the paper.

1. Preliminary remarks about punto and Florentine

Punto is both an adverb and a quantifier in Florentine and in other varieties of Tuscany as well. As an adverb, it is used in postverbal position to reinforce sentential negation. In this case, punto is equivalent to ‘at all’: it can encode the completive aspect, but it can also deny an explicit or implicit assertion:

(1) a. La radio la un funziona punto. (Florentine) the radio SCL NEG works PUNTO “The radio does not work at all.” b. Questa donna umme piace punto. (Aretino) this woman NEG-to-me likes PUNTO “I do not like at all this woman.” c. Io n ce rido punto. (Cortonese) I NEG of-it laugh PUNTO “I do not laugh at all at it.” d. Nun sento punto r freddo stamattina. (Pisano) NEG feel PUNTO the cold this-morning “I do not feel at all the cold this morning.”

Thus, for instance, it is appropriate to use a sentence like (1a) to express the fact that the radio is on mute or that it does not work as it was supposed to.

On the other hand, punto is also a quantifier, equivalent to English no or Italian nessun(o) or alcun(o):

(2) a. Un vedo punti libri. (Florentine) neg see punto.m.pl. books “I see no books.” b. un l’ho mandato in punti posti. (Pisano) neg him-have sent in punto.m.pl places “I have sent him nowhere.” c. enn’ho punti soldi. (Cortonese)

A case of incomplete Jespersen’s cycle in Romance

119

neg have punto.m.pl money “I have no money.”

As a quantifier, punto displays adjectival morphology, that is it agrees in gender and number with the quantified noun.

Florentine punto is very similar to Old French point (Foulet 1990, 264f.; Roberts & Roussou 2003, 152f.): both derive from the noun for ‘point, stitch’ and are minimizers which have undergone a re-analysis process as negative adverbs. It is worth noting that, as an adverb, punto is restricted to negative contexts, while as a quantifier –especially in Old Florentine– it can be used both with negative and positive readings. In (3), for instance, the presence of punto is licensed by the interrogative operator:

(3) t’ hai trovato punti libri su i’ tavolo? SCL have found PUNTO.M.PL books on the table ‘Have you found any book on the table?’

From this point of view the quantifier punto behaves like other negative words in Romance, labeled Negative Polarity Items by Laka (1990) and N-words by Zeijlstra (2004). However, punto alone cannot express sentential negation, as the contrast in (4) shows:

(4) a. Nessuno studente è venuto. (Italian) no student is come b. *Punto studente gl’ è venuto. (Florentine) punto.m.sg student SCl is come “No student has come.”

While nessun(o) in preverbal position is sufficient to introduce a negative context, punto cannot be used in preverbal position. I will not deal here with the semantic aspects of the licensing of punto and of NPIs in general (see Zeijlstra 2004, 40f.). What is relevant here is the fact that punto always needs a (preverbal) c-commanding negative element to be licensed, i.e. a specific syntactic configuration. From this point of view the adverb punto and the quantifier punto, as I will show in the following sections, behave in a very similar way.

The sources for the Old Florentine examples are (mainly prose) texts from the XIII and XIV centuries. As for Modern Florentine, I investigated three different varieties: the variety of Florence itself (sometimes called Urban Florentine) and two peripheral (and for some aspects more conservative) varieties, the variety of Empoli (30 km west from Florence) and Pontassieve (15 km east from Florence). There are two main differences between Old and Modern Florentine: first, Modern Florentine, like Northern Italian Dialects, has developed a system of subject clitics, while the

Jacopo Garzonio

120

medieval variety had weak pronouns1; second, Old Florentine, like other medieval Romance languages, displays verb movement to the left periphery in main clauses2, while Modern Florentine has totally lost any trace of Verb Second syntax, even in root interrogatives.

2. The syntax of punto in Old Florentine

The original function of punto in Old Florentine appears to be that of a paucal quantifier or, in other words, of a minimizer. Thus, it can, but does not have to, appear in negative contexts. The examples in (5) show the use of punto in negative contexts, while the examples in (6) and (7) display non-negative contexts where the presence of punto is licensed by another type of operator: a wh element in (6) or a yes/no interrogative or hypothetical operator in (7).

(5) a. Andoronne al detto forno furiosamente e sì rubarono e portaronne went to-the said bakery angrily and so stole and took tutto pane cotto e chrudo, e grano e farina, che non ve ne all bread baked and raw and wheat and flour that NEG there of-it rimase punto. (Libro del Biadaiolo) remained PUNTO. ‘They went to the foresaid bakery and stole and took all the bread, both baked and raw, the wheat, the flour, so that there did not remain any.’ b. Il re d’Inghilterra fu a gran pericolo con sua oste […] che 8 dì the king of England was at great peril with his army that 8 days stettono, che non ebbono se non poco pane né punto di vino. stayed that NEG had but NEG little bread and-NEG PUNTO of wine (G. Villani – Nuova Cronica 13.66) ‘The king of England was in great peril with his army, since for 8 days they had not but a little of bread and had no wine.’

(6) a. El tempio d’ Ercule […] a chi ha punto delle storie romane the temple of Hercules to whom has PUNTO of-the histories roman 1 See, among others, Renzi & Vanelli (1983), Brandi & Cordin (1989), Poletto (2000). It is worth noting that the preverbal negation marker is a clitic itself and interacts with subject clitics. Usually the negation un precedes subject clitics, but follows the third person feminine clitics, both singular and plural: la un dorme ‘she does not sleep’ / le un dormano ‘they (f.) do not sleep’ (Manzini & Savoia 2005, I-140). 2 See, among others, Benincà (1984), Adams (1987), Benincà (2006). Like other medieval varieties of Italy, Old Florentine had a less rigid V2 syntax than other Romance languages. V1, V3 or V4 were very common in Old Florentine, since multiple elements had access to the preverbal portion of the CP.


121

letto, è manifesto. (Commento a Ovidio 471) read is manifest ‘The temple of Hercules is manifest to whom has read a little of roman histories.’ b. E quando hai punto di tempo quieto pensa a’ benifici di Dio.

and when have PUNTO of time quiet think about benefits of god

(G. delle Celle – Lettere 1392.10) ‘When you have a little of quiet time, think about the benefits of God.’

(7) a. Noi avemo […] parlato delle travaglie de’ nostri paesi […] e sse we have talked of-the troubles of our countries and if intra esse fosse stato punto di tempo quieto o tranquillo. between them had been PUNTO of time quiet or peaceful (M. Villani - Cronica 8.106)

‘We talked about the troubles of our countries and whether between them there had been a little of quiet or peaceful time.’ b. Se punto di gentilezza nell’animo hai […] tutta l’avresti bruttata if PUNTO of gentleness in-the soul have all her would-have ruined e guasta, costei amando. (G. Boccaccio – Il Corbaccio 511.20) and spoilt, her loving ‘If you have a little of gentleness in your soul, you’d have spoilt her with your love.’

These examples show, furthermore, that punto is used without any article (this is expected, since it is used as a quantifier and, therefore, it lacks any referential meaning; Roberts & Roussou 2003, 150). Note that the quantified nouns appear as complement PPs introduced by di ‘of’. We can assume that the quantifier punto is derived through a grammaticalization process whereby the N punto is raised to a higher position in the DP structure, which I will label NumP following Roberts & Roussou (2003) and the works cited there:

(8) punto di vino (5b) [DP [D Ø][NumP [Num punto] [NP punto [PP di vino]]]]

Beside this quantifier punto, Old Florentine had an adverb punto, used in postverbal position to strengthen negation. In the texts of XIII and XIV centuries, this element is very frequent, thus suggesting that, at this stage, the grammaticalization process –from quantifier to adverb– was complete and both these elements were present in the grammar. The adverb punto usually followed the inflected verb, always required the sentential negation and could be used with both gradable and non-gradable predicates. Some examples are given in (9):

(9) a. Di ciò non s’adirò punto, perciò ch’era molto of it NEG REFL became-angry PUNTO because that was very

Jacopo Garzonio

122

gentile cavaliere. (Novellino, 77)

gentle knight ‘He did not become angry at all about it, since he was a very gentle knight.’ b. Rivoltato il destriere […] a Diomede non parlò punto. turned the steed to Diomedes NEG spoke PUNTO (G. Boccaccio – Filostrato 5.13) ‘He turned his steed and did not say a word to Diomedes.’ c. Il quale veggendo Bruto venire, e conosciutolo, non schifò who seeing Brutus come and recognized-him NEG dodged punto il colpo, ma […] (G. Boccaccio – Esposizioni, 4(i).234) PUNTO the blow but ‘…who, seeing that Brutus was incoming and having recognized him, did not dodge the blow, but…’ d. Ma guarda del sentiero luminoso, che davanti ti vedi e per lo

but look of-the path bright that before you see and through the

quale io andrò, tu non uscissi punto. (G. Boccaccio – Corbaccio)

which I will-go you NEG exit PUNTO ‘Do not leave the bright path that you see before yourself and that I will follow.’

Cases where punto precedes the inflected verb are rare. Some examples are given below. In (10a) punto is licensed by the coordinative negation né, in (10b) it appears in a relative clause and the licenser is the negation of the main clause.

(10) a. Perch’elli vive bestialmente, ed usa con quelli che bestialmente because he lives beastly and stays with those that beastly vivono, né da loro punto si parte. (Ottimo Commento 19) live and-NEG from them PUNTO REFL separates ‘…because he lives as a beast and stays with those that live as beasts and does not separate from them.’ b. In tutta la detta oste non ebbe altra gente che punto reggesse o in all the said army NEG was other people that PUNTO resist or combattesse. (G. Villani – Nuova Cronica 11.216) fight ‘In all the foresaid army there was not anyone who did not resist or fight.’

Structures with punto and the infinitival are very often introduced by senza ‘without’, which performs the function of the negative licenser of punto. In this case, the adverb can precede or follow the verb. Some examples are given in (11).

(11) a. E là sanza parlare punto, entra nel letto, e fa’ ragione and there without speak PUNTO enter in-the bed and make reason


123

d’essere me… (Trecentonovelle 206) of be me ‘And there, without saying a word, get into the bed and think that you are me.’ b. La giovane, senza schifar punto il colpo, lui similmente cominciò a the young without dodge PUNTO the blow him similarly started to amare. (G. Boccaccio – Decameron 5.4) love ‘The girl, without dodging the hit, began to love him with the same passion.’ c. Bergamino allora, senza punto pensare, […] disse questa B. then without PUNTO think told this novella… (G. Boccaccio – Decameron 1.7) novel ‘Bergamino, without thinking at all, told the following novel.’

I will assume that this use of punto derives from the grammaticalization into an adverb of the quantificational noun presented in examples (5)-(7). I will analyze more in detail this development in section 4; for now it is sufficient to say that punto, like other minimizers in Romance, lost its DP layer and became a clausal element. However, while other elements (like point in French or mica in Italian) ceased to be used as quantifiers, punto has still this function, both in Old and Modern Florentine.

3. The Syntax of punto in Modern Florentine

The adverb punto still exists in Modern Florentine, but it has not yet supplanted preverbal negation and is used less frequently than in the medieval variety. In Modern Florentine it is used mainly with gradable predicates, and it encodes a negative completive or intensity reading (corresponding approximately to English ‘at all’). From the syntactic point of view, it has many of the properties of punto in Old Florentine. It needs the preverbal negation and cannot precede the inflected verb, as it is shown in (12a-d), unless it is contrastively focalized (12e):

(12) a. La radio la *(un) funziona punto. the radio SCL NEG works PUNTO ‘The radio does not work at all.’ b. (E) *(un) mi piace punto. SCL NEG to-me likes PUNTO ‘I do not like it at all.’ c. *La radio punto la funziona./*la radio la punto funziona. d. *Punto mi piace. e. A: Ma a te ti piace?

Jacopo Garzonio

124

but to you to-you likes ‘As for you, do you like it?’ B: PUNTO mi piace. PUNTO to-me likes ‘I do not like it AT ALL.’

In the presence of an auxiliary verb, the unmarked position of punto is between the auxiliary and the lexical verb. In this case, punto must follow other temporal/aspectual adverbs, if present:

(13) a. *Un m’è punto più piaciuto. *punto-più NEG to-me is PUNTO anymore liked b. Un m’è più punto piaciuto. più-punto ‘I have not liked it anymore at all.’ c. *Un m’è punto mai piaciuto. *punto-mai NEG to-me is PUNTO never liked d. Un m’è mai punto piaciuto. mai-punto ‘I never liked it at all.’

The examples in (13) show that punto is located in a low position in the adverbial hierarchy in IP (Cinque 1999). It should be pointed out that punto is not compatible with temporal/aspectual adverbs that do not need preverbal negation (or another type of operator) to be licensed. For instance, it is impossible to combine punto and già ‘already’.

(14) a. *Un l’ho punto già visto. *punto-già NEG it have PUNTO already seen ‘I have not already seen it.’ b. *Un l’ho già punto visto. *già-punto

So it seems that temporal/aspectual adverbs like già interfere with the licensing conditions of punto. NPI-like adverbs like mai or più do not interfere with these conditions. It should also be pointed out that punto is not compatible with tutto ‘all’, which, according to Cinque’s (1999) analysis, is located in the [Spec] of the Completive Aspect Phrase.

(15) a. *Un (l’)ho punto fatto tutto. *punto-tutto NEG it have PUNTO done all ‘I have not done (it) all.’ b. *Un (l’)ho fatto punto tutto. *punto-tutto c. *Un (l’)ho fatto tutto punto. *tutto-punto

The data presented so far show that the syntax of the adverb punto has not changed so much in its diachronic evolution and punto has not lost its licensing


125

conditions. More importantly, punto has not become a negation marker in the sense of Zeijlstra (2004, 51), i.e. a particle or affix encoding sentential negation.

Consider now the development of the quantifier punto. As I have shown in section 2, in Old Florentine quantifier punto was a noun selecting a PP as complement. In Modern Florentine, like in all the varieties of Tuscany where it is present, it is a quantifier. It displays gender and number agreement with the quantified noun. Thus, what in Modern Florentine is punta paura ‘no fear’ (16b), in Old Florentine would have been punto di paura.

(16) a. Un ho punto sonno. NEG have PUNTO.M.SG sleep ‘I do not feel sleepy.’ b. Un ho punta paura. NEG have PUNTO.F.SG fear ‘I have no fear.’ c. Un ho punti libri. NEG have PUNTO.M.PL books ‘I have no books.’ d. Un ho punte sorelle. NEG have PUNTO.F.PL sisters ‘I have no sisters.’

The agreement paradigm of punto is richer than that of similar elements in Italian and other dialects. For instance, the Italian adjective nessun(o) ‘no’ has only singular forms (nessun(o), nessuna, but *nessuni, *nessune)3.

Like the adverb, the quantifier punto needs to be licensed. While the adverb can be licensed only by negation, the quantifier can be licensed also by an interrogative or hypothetical operator. The examples in (17) show that the quantifier punto cannot appear without an appropriate licenser, and that sentential negation is preferred to a preverbal NPI like nessuno ‘nobody’ as a licenser. Notice that Negative Concord arises in these cases.

(17) a. (E) un vedo punti libri. SCL NEG see PUNTO.M.PL books

3 Differently than the quantifier alcuno of Italian, punto cannot follow its quantified DP complement: (i) a. …senza paura alcuna. without fear ALCUNO.F.SG ‘…with no fear.’ b. *…senza paura punta. This can be explained assuming that punto does not allow the movement of its complement to [Spec, QP]. I thank an anonymous reviewer for pointing out this property.

Jacopo Garzonio

126

‘I see no books.’ b. *Vedo punti libri. c. ??Nessuno ha visto punte persone. nobody has seen PUNTO.F.PL people ‘Nobody has seen anyone.’

Example (18) shows that, like the adverb, the quantifier punto can appear in preverbal position only when it is focalized. In this case, it is not compatible with sentential negation:

(18) a. PUNTE PERSONE hanno visto Mario. PUNTO.F.PL people have seen M. ‘NO ONE has seen Mario.’ b. *PUNTE PERSONE un hanno visto Mario. PUNTO.F.PL people NEG have seen M.

Example (19) shows that punto can be licensed also by an interrogative or hypothetical operator, but, contrary to Old Florentine (see examples in (6)), it cannot be licensed by a wh operator.

(19) a. T’hai trovato punti libri su i tavolo? SCL have found PUNTO.M.PL books on the table ‘Have you found any book on the table?’ b. Se tu trovi punti libri, portali. if SCL find PUNTO.M.PL book bring-them ‘If you find any books, bring them. c. *Chi c’ha punti libri, li porti. who there has PUNTO.M.PL books them bring ‘Who has any books should bring them.

The examples in (20) show that there is no pronominal counterpart of punto.

(20) a. *Un s’è visto punto. NEG SCL is seen PUNTO.M.SG b. *Un s’è visto punti. NEG SCL is seen PUNTO.M.PL ‘We have not seen anyone.’ c. *Punto (un) ha visto Mario. PUNTO.M.SG NEG has seen M. d. *Punti (un) hanno visto Mario. PUNTO.M.PL NEG have seen M. ‘Nobody has seen Mario.’

Finally, quantifier punto is incompatibile with the definite article and admits the extraction of the quantified noun by means of the partitive clitic ne (examples in


127

(21)). These properties are characteristic of a group of quantifiers, which, according to Giusti (1997), are Q heads that select indefinite DPs as complements.

(21) a. *Punti i libri… PUNTO.M.PL the books b. *I punti libri… c. Un ne ho visti punti. NEG of-them have seen PUNTO.M.PL ‘I have not seen any of them.’

Thus, I will assume that the internal structure of a punto QP is as represented in (22):

(22) punti libri (17a) QP 2

Q’ 2

Q° DP | 5

punti libri

4. The diachronic development

Let us summarize some relevant facts about the adverb and quantifier punto. In Old Florentine it was a minimizer which could be used in negative constructions alongside preverbal negation. This type of grammaticalization can be considered one of the first stages of a Jespersen’s cycle. Optional negative reinforcement by a postverbal marker is stage 2 (out of 7) in Jespersen’s cycle – as assumed by Zeijlstra (2004, 56; see also Willis 2005). In Modern Florentine we have both an adverb punto, which is optional, and a quantifier, which, contrary to the Old Florentine quantifier, is a Quantificational head and not a DP. All these elements, in both varieties, are not free choice items, but need a licensing configuration. So, it seems that punto has failed to become the unique negative marker in Florentine. In other Italian dialects, elements which were in origin minimizers very similar to punto have undergone a complete cycle. Take, for instance, Milanese mi(n)ga. In Old Milanese it was a noun expressing a small quantity and could be used in both positive and negative contexts, like in (23):

(23) a. on sté de scisceri e miga de vin d’intrà (Lancino Curti 6.14) one staio of chickpeas and MIGA of wine of income ‘One staio (20 l) of chickpeas and a little of wine as income…’

Jacopo Garzonio

128

b. là no se sente miga de male (Barsegapé 2430) there NEG REFL feels MIGA of pain ‘There one does not feel any pain.’

In Modern Milanese, mi(n)ga is only employed as sentential negation (see Vai 1996; Rohlfs 1969, 304):

(24) a. I credun che mi sia minga bun. SCL believe that SCL am NEG good ‘They think that I am not able (to do that).’ b. L’idea che nquidun al sia dizunest l’è minga nova. the idea that someone SCL is dishonest SCL is NEG new ‘The idea that someone is dishonest is not new.’

In this section, I will account for the fact that punto did not undergo the same process of other minimizers in Italo-Romance. I will follow some of the claims made by Roberts & Roussou (2003) and Roberts (2007) about negative Agree and minimizers in Jespersen’s cycle. As we have seen, the quantifier punto is not intrinsically negative, since it can appear in non-negative contexts, where it is licensed by different operators. This means that it bears no negative feature, but an uninterpretable operator feature (like n-words in Old French; see Roberts 2007). This is why punto must be licensed: bearing an uninterpretable operator feature, it must be in an Agree relation with an appropriate c-commanding licenser. When the licenser is an interrogative or wh operator, punto has a positive polarity interpretation. When the licenser is negation, punto acquires a negative polarity interpretation. Notice that punto, contrary to Italian n-words, cannot function as a negative operator (i.e. it cannot be used in preverbal position).

Initially, punto was a DP, with the structure proposed in (8). At the stage described in section 2, it could be used as an adverb, without any PP (or NP) as its complement. I will assume that this grammaticalization process corresponds to a transition from DP to QP4, which can agree only with the preverbal negation (the adverb punto can be licensed only by negation):

4 This is in accordance with the Head Preference Principle (HPP) ‘Be a head, rather than a phrase’ – as formulated by van Gelderen (2004) – which is a general principle of language acquisition and change.


129

(25) DP QP 2 2

NumP Q’ 2 → 2

punto NP Q° 2 | punto PP punto

Willis (2005) claims that elements which become sentential negation markers undergo a change of interpretability of features: the uninterpretable negative feature on the postverbal negative head becomes interpretable, while the interpretable feature on the preverbal one becomes uninterpretable. Following this analysis, I would suggest that the negative feature on punto has not become interpretable. Furthermore, I propose that the adverb and the quantifier punto are the same element in syntax, even in Modern Florentine. The only difference between the structure of the adverb punto (as in (25)) and that of the quantifier punto (as in (22)) is that the former lacks a DP complement. However, notice that in the more archaic peripheral varieties of Empoli and Pontassieve, the adverb can optionally show morphological agreement with the direct object or the subject of an unaccusative verb5:

(26) a La un mi piace punta, questa storia. SCL NEG to-me likes PUNTO.F.SG this story ‘I do not like at all this story.’ a’ ?Questa storia la un mi piace punta. b Un l’ho punta sentita, questa storia. NEG her have PUNTO.F.SG heard this story ‘I have not heard at all about this story.’ b’ ??un l’ho sentita punta, questa storia. b’’??un ho punta sentita questa storia.

5 This construction seems to be more frequent with psychological verbs. For instance, the following example has been considered ungrammatical by all the informants: (i) *La tu radio la un funziona punta. the your radio scl neg works punto.f.sg ‘Your radio does not work at all.’ The phenomenon was already attested in the literature. For instance the AIS atlas (Sprach und Sachatlas Italiens und der Südschweitz, 1678) reports the following sentence for Incisa in Val d’Arno (20 kms southeast of Florence): (ii) La ommi garba punta, chesta donna. scl neg-to-me likes punto.f.sg this woman ‘I do not like at all this woman.’

Jacopo Garzonio

130

(26) also shows that right dislocation is preferred in these constructions ((26b’) shows that punto can marginally follow the past participle). I consider these examples evidence supporting the idea that the two puntos are to be reduced to a unique lexical element, that is a quantificational head. I will return/come back to these examples in the following section.

Thus, the main difference between Old and Modern Florentine lies in the selectional properties of punto: it either selects a PP in Old Florentine, or a DP in Modern Florentine. Examples of adjectival agreement across a preposition like di ‘of’ can be found in other contemporary Italo-Romance varieties. Some examples, taken from Cinque (1997) are given in (27):

(27) a. Un poca de pasta. (Veneto) a little.F.SG of pasta b. Un pochi de bizi. (Veneto) a little.M.PL of peas c. Tante de onge. (Bellunese) many.F.PL of nails d. Tante de petre. (Salentino) many.F.PL of rocks

The agreement morphology of punto is likely to have derived from similar structures. The presence of adjectival agreement on quantifiers can be seen as the development of new agreement features on the quantificational head. In Modern Florentine punto, as other adjectival-like quantifiers, is the probe of an Agree relation of number and gender phi-features, which are valued by the phi-features of the complement DP. I argue that the presence of these phi-features on punto inhibits the development of the following stage of the Jespersen’s cycle. In other words, punto acquires new features (gender and number) but loses the capability to bear an interpretable negative feature. All the minimizers that have become sentential negations have lost any referential content and phi-features (mica in Standard Italian, mi(n)ga in Milanese, briza in Emilian, pa in Piedmontese, bitʃ in Romansh, and so on)6. Note that all the negative quantifiers that have become sentential negations are pronouns and not adjectives (nen and nente in Piedmontese, nylia in Romansh, and so on). Since punto does not become a sentential negation marker, it must be the case that it keeps its quantificational interpretation as well.

6 As pointed out by an anonymous reviewer, Catalan has two comparable elements: cap (< caput), which is both adverb and quantifier (and like modern Florentine punto does not take a de complement, but can be used pronominally), and gens (< gens) which again is both adverb and quantifier.


131

5. The position of punto

In the preceding section, I have proposed that punto is a quantificational head that can be used either as a polarity quantifier (in this case it selects a DP as its complement) or as a postverbal reinforcer of negation. The main piece of evidence for this claim is that punto, even when it is used as a reinforcer of negation, can optionally display morphology of agreement with the internal argument. Phenomena suggesting that there is a relation between (some) argumental positions and sentential negation are not rare. Manzini & Savoia (2005, III-206), reporting a fact already noted by Meyer-Lübke (1889), show that in Old French, the postverbal negative element mie (in origin a minimizer, meaning ‘crumble’ like the Italian mica), sometimes triggered a partitive direct object, even if it was not quantifiable at all:

(28) De s’espee ne volt mie guerpir. (Chanson de Roland 465)

of his sword NEG wanted MIE leave ‘He did not want to leave his sword.’

Similar structures can be found in modern Italo-Romance varieties. The following example is reported by Manzini & Savoia (2005, III-279f.) for the Piedmontese variety of Quarna Sotto:

(29) Nə caman mia d əu te frial. NEG-SCL call MIA of the your brother ‘They do not call your brother.’

In this section, I will propose an analysis for the fact that the “adverb” punto agrees with the internal object.

Firstly, it must be said that punto seems to occupy a position which is lower than the positions Neg2 and Neg3 individuated by Zanuttini (1997) for postverbal negation in Romance. In fact, as it was shown in the examples in (13), punto must follow temporal and aspectual adverbs. In (30) I show the position of punto in the adverbial hierarchy of Cinque (1999):

(30) […[T Anterior *già [Asp Terminative più [Asp Perfective sempre/mai [ punto [Asp Completive *tutto]]]]]]

I argue that punto is higher than tutto ‘all’ because tutto, contrary to punto, cannot appear between the auxiliary and the past participle in Italian and in Modern Florentine.

(31) a. Un ho punto capito. NEG have PUNTO understood ‘I have not understood at all.’ b. *Un ho tutto capito.

Jacopo Garzonio

132

‘I have not understood all.’ c. Un ho capito tutto.

Neg2 and Neg3 are higher than the position of punto. According to Zanuttini (1997), Neg2 is higher than T Anterior (già), while Neg3 is between T Anterior and Asp Terminative (più). As we have seen, punto is not compatible with già but follows più. Neg2 is the position occupied by postverbal negations which derive from minimizers, while Neg3 is the position occupied by postverbal negations which derive from negative quantifiers (Poletto 2008). Thus, punto has not become a true negative marker and, therefore, occupies a different position from true negative markers.

In (30) I have proposed that punto is higher than tutto because the latter cannot precede the past participle. Notice, however, that punto and tutto seem to encode the same aspect: a Completive one. Cinque (1999, 104) already noted that positive and negative adverbs encoding the same aspect seem to occupy different positions. We can assume that punto is in some sense the negative counterpart of tutto. Furthermore, it must be pointed out that tutto is a quantifier and that, even if it can be analyzed as an aspectual adverb, at the same time it is one of the arguments of the predicate. I will not analyze further the properties of tutto and the similarities with punto, but it is evident that the encoding of Completive aspect interacts with the argumental structure of the predicate.

Following a proposal by Poletto (2007) about dedicated positions for quantifiers in the clause structure, I propose the analysis represented in (32):

(32) Un ho punta sentita questa storia (26b’’) NegP 2

un TP 2

ho CompletiveP 2

punta vP 2

sentita QP 2

Q° DP punta storia

Punto is merged in a position for universal quantifiers of the internal argument. The object DP is moved to this position and the phi-features of the DP are copied onto the quantifier. From this position, the quantifier is moved to the IP layer where


133

it checks its aspectual feature. When there is no internal argument, we can assume that no element is moved to the quantifier position, and punto agrees with a silent element with unmarked phi-features7. From a diachronic point of view, this configuration is the starting point of the grammaticalization of minimizers and negative quantifiers into negation markers. As proposed by Poletto (2008), negation has an articulated structure and always involves the presence of a scalar component (the minimizer). However, as I have argued in the preceding section, a minimizer can get a negative feature only if it has no phi-features. While this was the case of minimizers selecting a PP as complement, punto’s grammaticalization was inhibited by its adjectival morphology.

6. Conclusions

In this article I have examined the diachronic development of the element punto in Florentine. In Old Florentine it was a DP used at first as a minimizer and then as a reinforcer of sentential negation. In Modern Florentine it is a polarity quantifier and an adverbial element meaning ‘at all’. I have proposed that it has failed to become a true postverbal negation marker because it has developed agreement morphology. If a head has phi-features, it cannot grammaticalize into negation. In other words, it cannot acquire a(n interpretable) negative feature if it has phi-features that need to be valued. This analysis entails that in the grammar of Florentine there has always been a unique punto, i.e. the quantifier and the adverb are not separate elements. Furthermore, I have proposed that punto (and possibly similar elements in Romance) occupies a precise position in the adverbial hierarchy, which is between Perfective Aspect (sempre/mai) and Completive Aspect (tutto). I argue that this position is for negative adverbs encoding Completive Aspect.

Many facts await for further refinement of this analysis. Among these, the fact that adverbial punto is not compatible with any other aspectual adverb, but it can only combine with adverbs that have to be licensed by an operator (for instance negation). It seems that adverbs which do not need to be licensed block the licensing process. Similar cases of NPIs with “licensing chains” are described by den Dikken (2002).

7 Notice that the order punto-past participle is possible only if punto is used alone (i.e. if it does not quantify over a DP). It seems, thus, that further movement of the participle is obligatory if some element moves to the punto phrase.

Jacopo Garzonio

134

References

Adams, M. (1987). Old French, null subjects and Verb Second phenomena, Ph.D. dissertation, UCLA.

Benincà, P. (1984) ‘Un’ipotesi sulla sintassi delle lingue romanze medievali’, Quaderni

Patavini di Linguistica 4: 3-19. Benincà, P. (2006) ‘A Detailed Map of the Left Periphery of Medieval Romance’, in: R.

Zanuttini, H. Campos, E. Herburger & P. Portner (eds.) Crosslinguistic Research in

Syntax and Semantics: Negation, Tense, and Clausal Architecture, Washington, DC: Georgetown University Press: 53-86.

Brandi, L. & P. Cordin (1989) ‘Two Italian Dialects and the Null Subject Parameter’, in: O. Jaeggli & K. Safir (eds.) The Null Subject Parameter, Dordrecht: Kluwer: 111-142.

Cinque, G. (1997) ‘Quantifiers’, in: M. Maiden & M. Parry (eds.) The Dialects of Italy, New York: Routledge: 186-189.

Cinque, G. (1999) Adverbs and Functional Heads: A Cross-Linguistic Perspective, New York/Oxford: Oxford University Press.

Dikken, M. den (2002) ‘Direct and parasitic polarity item licensing’, Journal of Comparative

Germanic Linguistics 5: 35-66. Foulet, L. (1990) Petite syntaxe de l’ancien français, Paris: Librairie Honoré Champion. Gelderen, E. van (2004) Grammaticalization as Economy, Amsterdam: John Benjamins. Giusti, G. (1997) ‘The Categorial Status of Determiners’, in: L. Haegeman (ed.) The New

Comparative Syntax, London: Longman: 95-123. Horn, L. (1989) A Natural History of Negation, Chicago: The University of Chicago Press. Jespersen, O. (1917) Negation in English and other Languages, Copenhagen: A.F. Høst. Laka, I. (1990) Negation in Syntax: On the Nature of Functional Categories and Projections.

Ph.D. dissertation, MIT, Cambridge. Manzini, M. R. & L. M. Savoia (2005) I dialetti italiani e romanci: morfosintassi generativa,

Alessandria: dell’Orso. Meyer-Lübke W. (1899) Grammatik der Romanischen Sprachen. 3, Syntax, Leipzig:

Reisland. Poletto, C. (2000) The Higher Functional Field: Evidence from Northern Italian Dialects,

New York/Oxford: Oxford University Press. Poletto, C. (2007) ‘Parallel Phases: a Study on the High and Low Left Periphery of Old

Italian’, in: M. Frascarelli (ed.) Phases of Interpretation, Berlin: Mouton de Gruyter: 261-294.

Poletto, C. (2008) ‘On Negation Splitting’.Talk given at the III Cambridge Italian Dialect Syntax Meeting.

Renzi, L. & L. Vanelli (1983) ‘I pronomi soggetto in alcune varietà romanze’, in: Scritti

linguistici in onore di Giovan Battista Pellegrini, Pisa: Pacini: 121-145. Roberts, I. (2007) Diachronic Syntax, Oxford: Oxford University Press.


135

Roberts, I. & A. Roussou (2003) Syntactic Change: A minimalist Approach to

Grammaticalization, Cambridge: Cambridge University Press. Rohlfs, G. (1969) Grammatica storica della lingua italiana e dei suoi dialetti III: Sintassi,

Torino: Einaudi. Vai, M. (1996) ‘Per una storia della negazione in milanese in comparazione con altre varietà

altoitaliane’, Annali della Facoltà di Lettere e Filosofia dell’Università degli Studi di

Milano 49: 57-98. Willis, D. (2005) ‘A Minimalist Approach to Jespersen’s Cycle in Welsh’. Ms, University of

Cambridge. Zanuttini, R. (1997) Negation and Clausal Structure. Oxford: Oxford University Press. Zeijlstra, H. H. (2004) Sentential Negation and Negative Concord, Utrecht: LOT

Publications.


33 – 2008

137

THE SPLIT DP HYPOTHESIS

EVIDENCE FROM ANCIENT GREEK

Daniela Isac, Allison Kirk

1. Introduction

1.1 DPs in Ancient Greek

In Ancient Greek one may find elements of a DP split from the noun, as in (1). Adjectives (A), quantifiers (Q), demonstratives (Dem), wh-elements (WH-) may precede or follow nouns (N), with some syntactic material intervening.

(1) a. A/Q/Dem/WH-......... N b. N ......... A/Q/Dem

(2) ep’ andras strateuometha agathou (Her. 7.53.2) against men.acc.pl. fight.1pl.pres.ind.mid. valiant.acc.pl. ‘we are fighting against valiant men’

(3) tou gar dê Ludôn dêmou hai

the.gen.sg. for PART Lydians.gen.pl. people.gen.sg the.nom.pl. thugateres porneuonta pasai (Her.1.93.4) daughter.nom.pl. prostitute.3pl.pres.ind.act. all.nom.pl. ‘all the daughters of the common people of Lydia prostitute themselves’

(4) tênde echô gnômên(Her.2.27.1) this.acc.sg. hold.1sg.pres.ind.act. opinion.acc.sg. ‘I hold this opinion’

(5) basileu koion ephthenxao epos? (Her.5.106.3) king.voc.sg. what-kind.acc.sg. utter.2sg.aor.ind.mid word.acc.sg. ‘king, what word do you utter?’

In this paper we will refer to such instances as ‘discontinuous DPs’.


138

Each of the items that show up split from the noun in the examples above may also occur as part of the same DP (in what we will call a ‘continuous DP’ configuration), as in the following examples:

(6) ton nautikon straton kinduneusei the-acc.sg. of-sea.acc.sg. army-acc.sg. peril-3sg.fut.ind.act. basileus apobalein (Her.8.65.3) king-nom.sg lose-aor.inf. ‘The king will be in danger of losing his sea army’

(7) hai men nun echidnai kata pasan

the.nom.pl. PART now viper.nom.pl. on all.acc.sg tên gên eisi (Her.3.109.3) the.acc.sg land.acc.sg. be.3pl.pres.ind.act. ‘vipers are in every land’

(8) tês de gês tês peri hode

the.gen.sg. PART land.gen.sg. which.gen.sg. about this.nom.sg. ho logos hormêtai legesthai

the.nom.sg. word.nom.sg. begin.3.sg.perf.ind.mid. say.pres.inf.mid. ‘As for the land of which my history has begun to speak’ (Her.4.16.1)

(9) tas teôi tropôi perêsomen; them.acc.pl. what.dat.sg. way.dat.sg. go-past.1pl.fut.ind.act. ‘how will we get past them?’ (the guards) (Her.3.72.1)

In spite of the fact that Ancient Greek was characterized as a relatively free word order language, the orders that are attested (either in continuous or discontinuous DPs) are subject to certain restrictions. In particular, even though adjectives, for instance, can be split from the nouns, they can be split in a configuration like: [D N… D A], but not in a configuration like [D A… D N]. Similarly, there are restrictions on the availability of ‘continuous’ configurations involving adjectives: [D N D A] is a possible order, as are [N D A] or [D A N], but orders such as [D A D N], or [D N A], or [N A D] for instance, are unattested.

Below is a sum up of all the orders we have found of the DP internal words (both continuous and discontinuous). All the orders that are not in this table are unattested in the corpus we looked at (Herodotus’ Histories

1).

1 The Loeb edition (based on Stein’s text) is used as a main source, with consultation with Hude’s OCT edition on the examples taken.

The split DP hypothesis evidence from Ancient Greek

139

(10)

Attested orders

Adjective Strong Quantifier

Weak Quantifier

Demonstrative wh-

continuous N(D)A (D)AN DNDA

QDN DNQ

QN NQ DNQ

(D)NDem Dem(D)N

wh-N

discontinuous N...(D)A (D)A...N DN...DA

Q...D(A)N DN...Q N...DQ

Q...N N...Q

(D)N...Dem Dem...(D)N

wh-...N

Unattested Orders: all other order apart from above

Given the lack of negative evidence in a dead language like Ancient Greek, one cannot be certain if absent constructions were ungrammatical or just not used in the text(s) from which one draws the data. The strategy we have adopted is to let the theory that we have built based on the positive evidence decide whether these constructions were ungrammatical or just not employed in the text.

This paper aims at providing an analysis that could be instrumental in teasing apart ungrammatical orders from orders that are simply unattested in the corpus we have used. In this paper we will focus on DPs modified by adjectives. We found these DPs particularly challenging, as they pose problems not only connected to the possible continuous and discontinuous orders, but also problems related to the so-called Determiner spreading phenomenon. However, even if we do not discuss the other types of DPs that occur in Table (10), we tentatively submit that an analysis along similar lines can be extended to such DPs as well. We leave it for further research to verify this. For an analysis of all of these DPs, both continuous and discontinuous, which is similar to ours in that it is based on a Topic and a Focus projection within the DP, but which differs from ours in that it does not assume a split DP structure, see Kirk 2007.

1.2 Proposal in a nutshell

In a nutshell, we propose that the word order in continuous and discontinuous DPs in Ancient Greek can be accounted for by assuming that (i) nominal constituents have two DP layers, (ii) both DPs layers are phases in the sense of Chomsky 2001, 2005; and (iii) the periphery of each phase contains Topic and Focus positions. None of these claims, taken in isolation, is new. What is new is the attempt to put them together in order to account for the syntax of both continuous and discontinuous DPs and for their distributional restrictions. The interaction of these three factors guarantees that the material within the DP is prepackaged informationally into Topic-comment or Focus-presupposition. This information


140

structure within the DP can be further exploited by the left periphery of the clause – possibly, material that has been focused within the nominal will move to the clausal Focus, and material that has been topicalized within the nominal can move up to the clausal Topic.2

The paper is organized as follows: in section 2, we discuss the theoretical assumptions of this paper, in section 3 we present the analysis of modified DPs, first modified DPs with two overt Ds, followed by the analysis of modified DPs with only one overt D; and in section 4 we present the conclusions.

2. The Split DP Hypothesis

In this section we will provide details about our theoretical assumptions. As stated above, our analysis relies on three claims: (i) nominal constituents have two DP layers, (ii) both DPs layers are phases; and (iii) the periphery of each phase contains Topic and Focus positions. In what follows we discuss each of these in more detail.

2.1 Two DP layers

There is by now a well established line of research on the structure of DPs that points to the existence of two layers within nominal constituents (the Split-DP hypothesis), each corresponding to different semantic content. (Giusti 1993, Zamparelli 1995, Aboh 2004, Ishane and Puskas 2001, Kariaeva 2001, Laenzlinger 2005, and others). In line with most of these proposals, we will assume that the higher DP (DPexternal) is the locus of pragmatic information, such as referentiality and deixis, while the lower DP (DPinternal) is the locus of information such as definiteness and indefiniteness. Lexical material is assumed to originate in the DPinternal and to then move up to DPexternal to check pragmatic features.

(11) [DP-ext Dext [DP-int Dint [NP N ] ] ]

That the two features hosted by the outer D and the inner D are different is supported by examples like the following:

(12) J’ai pris le train (Ishane and Puskas 2001:120) I have taken the train

2 We make no claims about the exact interraction between the discourse oriented projections within the DP and the analogue ones at the clausal level. It is possible that material that has been ‘branded’ as focus or topic within the DP can only be recuperated as such at the clausal level. But it is not out of the question that material that is topicalized within the DP for instance is contrastively focused at the clausal level. The decision is, we believe, largely an empirical matter, although there are probably some theoretical factors as well that will turn out to play a role, that concern the dynamics of focus-topic interraction.


141

The DP le train/ ‘the train’ is definite, but it can be interpreted either as a specific or as a non-specific DP. If specific, the DP le train/ refers to a pre-established train in the context; if non-specific, the DP le train fails to refer to a particular train and simply refers to any train in the set of trains, as opposed to, for instance the set of buses, or boats.

2.2 Two phases

Apart from assuming a double layered DP structure as above, we will also assume that both DPs (DPexternal and DPinternal) are phases, in the sense of Chomsky 2001, 2005. Given the parallelism between CPs and DPs initially proposed by Abney 1987, Szabolcsi 1987, this is not surprising: the higher D would be the analogue of C, while the lower D would be parallel to the little v in a clause3.

One consequence of this assumption is that both DP layers are subject to the Phase Impenetrability Condition/ PIC (Chomsky 2001): what this means for our purposes is that only syntactic material that has previously been moved to the edge of the DP phase (either the lower one or the higher one) can further be subject to movement. As will become apparent below, this property of syntactic phases is crucial for accounting for the existing restrictions on the available orderings of elements inside a DP.

2.3 Topic and Focus inside the DP

Syntactic research on the phrasal architecture of the clause has led to the identification of a left peripheral area (the CP domain) that is taken to encode discourse-related features such as topic and focus. In Rizzi 1997, information structural notions such as topic and focus are associated with specific positions in the syntactic architecture of the CP layer. A number of linguists, starting with Hallman 1997, 2000, Beghelli and Stowell 1997, Starke 2001, Belletti 2003, Jayaseelan 2001, etc. have argued that the layer of discourse related functional structure that sits on top of the clause is reiterated lower down in the clause, on the top of the vP. The emerging picture contains two phases – the CP and the vP – each with its own periphery containing discourse oriented projections like Topic and Focus.

(13) [CP [FocP/TopP [TP [vP [FocP/TopP [vP …

Moving on to the nominal domain, we expect a similar situation, i.e. the existence of discourse oriented projections like Topic and Focus at the periphery of

3 See also Gutierrez-Rexach & Mallen’s 2001, who, on the basis of optional adjectival placement, also propose that DPs may express two separate phases: a predicative D phase, and a separate ‘propositional’ D phase.


142

both DP layers. This expectation is induced by the parallelism between the CP domain and the DP domain, as well as by a view on phases as syntactic domains with a periphery containing quantificational and discourse oriented properties (Matthewson 2001, Butler 2004, Svenonius 2004, and others).

Several authors have shown that the DP includes a left periphery of discourse related projections (Bernstein 1997, 2001, Dimitrova-Vulchanova & Giusti 1996, Aboh 2004, Ishane & Puskas 2001, Giusti 1996, 2002, Gutierrez-Rexach & Mallen 2001, Haegeman 2004, etc). Under a split DP assumption, the existence of discourse oriented projections like Topic and Focus is to be expected at the periphery of both DP layers, as in (14).

(14) [DP-ext Dext [FocP/TopP [DP-int Dint [FocP/TopP [NumP [NP]]]]]]

The structure in (14) is thus completely parallel to the one in (13) in having a middle discourse oriented field in addition to the left periphery one.

Two types of evidence are discussed in the literature for the existence of these projections: the existence of specialized NP-internal morphological expression of focus and topic, or DP-internal displacement phenomena involved in the expression of topic and focus. The type of evidence that Ancient Greek provides is of the second type. Moreover, Ancient Greek shows displacement phenomena to the Topic and Focus projections in the higher DP, rather than the lower one.

The focus position inside the Ancient Greek DP is thus analogous to a contrastive focus position in the clause, along the lines discussed in Szabolcsi 1981 and Kiss 1998, among others.

On the other hand, given that the evidence discussed by the above mentioned authors points to the existence of Topic and Focus projections which are lower than the overt definite D, one can assimilate the latter to the left periphery of the lower DP in (14) above.

2.4 Features

More particular to Ancient Greek, we propose that the Topic/ Focus heads bear [EPP] features in addition to the [top] and [foc] feature, and thus always trigger movement of nominal subconstituents to their Specifier. Moreover, we propose that one discourse related projection is sufficient to partition the DP into two informational chunks: Topic-Comment or Focus-Background. What gets projected is thus either the TopicP or the FocusP. If the TopicP is projected, the material attracted to its Spec will be interpreted as Topic while the material in the complement of the Top head will be interpreted as Comment. The notion of Comment can be equated with the notion of Rheme or Focus, so the resulting partition is between Topic and (rhematic) focus. If, on the other hand, FocusP is projected, the material attracted in its Spec will be interpreted as (contrastive) focus,


143

while the material in the complement of the Focus head will be interpreted as Presupposition, i.e. old information, i.e. topic-like information. Thus, the syntactic material within the DP will always be partitioned informationally in a topic-like and a focus-like part and this, we claim, will be further exploited by the left periphery of the clause. The result will be either nominals that are displaced in their entirety to the left periphery of the clause, or nominals that will be divided into one part that is attracted to either the TopicP or the FocusP in the left periphery of the clause and another part that will stay in situ.

Let us now go back to the structure in (14). In what follows we will ignore the lower Topic and Focus projections, as they are not active in Ancient Greek. The structure in (14) raises the question of defining the feature content of the syntactic heads involved. To begin with, we assume that overt definite Ds are D(iscourse)-linked (see Comorovski 1996, Pesetsky 1987 for D-linked WH- phrases), in the sense of being linked to a familiar individual in the discourse. In order to capture this, we propose that higher D hosts a [+fam] feature. In this, we follow Heim 1982, who defines definiteness in terms of familiarity, or more formally, in terms of identity of the indices of card files for NPs, as in (15):

(15) The Familiarity Condition (Heim 1982, 369f) An NPi in a sentence φ with respect to file F and the Domain of filenames

Dom(F) is (i) [+definite] if i ∈ Dom(F), and (ii) [-definite] if i ∉ Dom(F)

What (15) says is that every NP comes with an index i, which represents the discourse referent (or Heim’s ‘file card’) associated with that NP. If the discourse referent i is already introduced in the discourse- or more formally, if the index i is an element of the set of all established discourse referents Dom(D), then the NP must be definite. If, however, the discourse referent i is not among the already established discourse referents, i.e. if i ∉ Dom(D), then the NP must be indefinite. Definiteness signals the familiarity of the discourse referent associated with the NP.4

(16) [DP-ext Dext [FocP/TopP [DP-int Dint [NumP [NP]]]

[+fam] [foc]/[top] [+def]

Null Ds, on the other hand, that occur with the so called bare nouns, are not assumed to be necessarily D-linked, and thus to have a [+fam] feature. We will

4 Notice that in positing a [+fam] feature on the higher D head of overt definites, we are not excluding the possibility that the higher D also hosts a [specificity] feature, along the lines proposed by Ishane and Puskas 2001. The specificity feature of a definite may be valued either positively or negatively, as illustrated in (12), but what we are proposing is that the [fam] feature of a definite will always be set as [+fam].


144

assume that Dint that are phonologically null are matched with a higher D that lacks

specification for familiarity, i.e. they are [α-fam].

(17) [DP-ext Dext [FocP/TopP [DP-int Dint [NumP [NP]]]

[α-fam] [foc]/[top]

The underspecification of the [fam] feature in the case of null Ds captures the fact that crosslinguistically the value of the [α-fam] feature can vary. In languages like Chinese, Russian and Hindi (Dayal, 2004) null Ds can be interpreted as [+ familiar], and this contrasts with English, in which null Ds are interpreted as [-familiar].

(18) Kuch bacce andar aaye. Bacce bahut khush the (Hindi) Some children inside came Children very happy were “Some children came in. They children were very happy.”

The [α-fam] feature should be understood in terms of Rooryck’s 1994 proposal. Rooryck distinguishes between two types of underspecified features: non-variable or [0-features], and variable, or [α-features]. [0-features] are neutral in the sense that they have no positive or negative value for a given feature, but they do have an Attribute specification of the respective feature. [0-features] can be equated with uninterpretable features. In Chomsky’s 2001 terms: “the uninterpretable features, and only these, enter the derivation without values, and are distinguished from interpretable features by virtue of this property”. In contrast, [α-features], are completely underspecified – both for the Attribute and for the value. The latter type of features should be kept apart from unspecified features which are distinguished from underspecified [α-features] in that they play no grammatical role whatsoever. This distinction will turn out to play a crucial role in our analysis.

Let us now go back to the structure in (17) and the features involved, more specifically to the question of whether the lower D of bare nouns has any definiteness feature. We posit that there is such a feature on the lower D and that can be valued either positively or negatively. As illustrated in (18) with Hindi, bare nouns can refer back to a familiar individual. Given Heim’s Condition in (15), it follows that these bare nouns must be definite. On the other hand, bare nouns can also be [-definite] as in the following Italian example.

(19) Gianni lavora con cani. (Zamparelli 1995) Gianni works with dogs.

Let us now define the notion of [topic] in relation to the feature [fam]. It seems at first glance that topic material corresponds to material which is [+familiar], and focus to material which is [-familiar]. This would lead to the conclusion that full definite Ds would correspond to Topic and null Ds to Topic or Focus, since full definite Ds are [+familiar], while null Ds are [α-fam]. However, despite the


145

similarities between the notions of familiarity and Topic, the two are not exactly the same. For one thing, the notion of familiarity is tied to definiteness, while Topic is not. This is proved by the fact that Topic material does not need to be a definite DP. This is true cross-linguistically and in Ancient Greek as well. To illustrate, consider the following examples from English and Japanese in which what has been topicalized is an indefinite DP.

(20) As for apples, red ones are my favourite.

(21) Sakana wa tai ga ii (Japanese, Krifka, 2005) fish TOP red snapper NOM excellent ‘As for fish, red snapper is excellent’

Similarly, in Ancient Greek, we see many other items besides definite DPs in Topic positions. For example, nouns without definite articles may be topicalized, as in (22).

(22) hosa de anthrópéia prégmata, hóde elegon how-much PART human.acc.pl. matters.acc.pl. thus say.3.pl.impf. ‘As far as human affairs, they said thus’ (Her.2.4.1)

Likewise, a [+fam] discourse individual does not necessarily have to be interpreted as Topic. Familiar discourse individuals may be contrastively focused for example, as the example in (24) shows.

(23) A man met a woman. It was THE WOMAN who introduced herself first.

To sum up on the feature specification of the Topic head, we will assume that this head bears two features: (i) a [fam] feature, which will capture the fact that all Topics denote familiar individuals (although the reverse is not true, as discussed above), and (ii) a [top] feature, to capture the fact that topics refer not only to familiar individuals, but to individuals that are ‘prominent’ in the discourse. Using Reinhart’s 1982 file cards metaphor for modelling information in a discourse,5 what this means is that Topics can only refer to cards that are on top of the stack of cards that constitute the set of discourse individuals, i.e. to cards that refer to what Reinhart calls ‘prominent’ individuals. Correlatively, in the Focus head we will posit an [α-fam] feature, meant to capture the fact that any discourse individual can be focused- be it familiar or unfamiliar. If familiar, an individual can be contrastively focused, as illustrated above in (23), and if unfamiliar, it is focus simply by virtue of being new information.

5 According to Reinhart 1982, the context set of a given discourse at a given point consists of a set of file cards which represent existing discourse referents, each of which contains information about a discourse individual: its properties and its relations to other discourse individuals.


146

Before we conclude, let us propose one more feature that will turn out to be important in accounting for the syntax of DPs in Ancient Greek, namely an [ufam] feature on the lower D. This feature can be checked by a matching [fam] feature born by any other element in the DP in principle: so far we have identified this feature in the Topic head and in the higher D head. If the Topic phrase is present, it will be closer to the lower D than the higher D is, so it will be the [fam] feature on Top that will check the uninterpretable feature on the lower D. If Topic is not projected, then the only other element that bears a matching [fam] feature is the higher D and the [ufam] feature on the lower D will check against it.

The configuration in (24) offers a sum up of the features involved in the structures we assume, both for DPs that contain an overt definite D and for DPs with a null D.

(24) a. Overt definite Ds: [DP-ext Dext [ FocP / TopP [DP-int Dint [NumP [NP]]] [+fam] [foc] [EPP] / [top] [EPP] [+def] / [+fam] [ufam] b. Null Ds: [DP-ext Dext [ FocP / TopP [DP-int Dint [NumP [NP]]]

[α-fam] [foc] [EPP] / [top] [EPP] [±def] / [+fam] [ufam]

In what follows we will show how the three assumptions we have discussed above correctly predict the observed restrictions of co-occurrence and on discontinuous modified DPs noted in Table (10).

3. Analysis. DPs modified by adjectives

This section offers an illustration of the mechanics of our proposal as applied to DPs modified by adjectives.

3.1 Adjectives in Ancient Greek

Adjectives in Ancient Greek can occur on either side of the N, as shown below.

(25) stêlas estêse duo ep’ autou

pillar.acc.pl set.3sg.aor.ind.act. two beside it.gen.sg. lithou leukou (Her.4.87.1) stone.gen.sg white.gen.sg. ‘he set up two pillars beside it of white stone’

(26) ta de alla hêmiplinthia leukou

the.nom.pl PART other.nom.pl half-brick.nom.pl white.gen.sg chrusou (Her.1.50.2)


147

gold.gen.sg. ‘the others were alloys of white gold’

Given that there is no detectable difference in the interpretation of the adjective in (25) vs. (26), we adopt a unified base position for all adjectives, as proposed by Kayne 1994:

(27) [DP D [CP C0 [IP DP I0 AP]]] (Kayne 1994)

The adjective in (27) is part of a reduced relative clause, in which the adjective enters a predication relation with a DP in the Specifier position of the reduced clause. We will refer to this DP as DPsubj, short for the DP subject of predication.

In Kayne’s analysis, the prenominal position of adjectives is derived by raising the AP, the predicate, across the subject of predication, to Spec-CP, while the postnominal position of adjectives is derived by raising the DP subject of predication to SpecCP. The two possibilities are represented in (28).

(28) a. [DP D [CP APi C0 [IP DP I0 APi]]]

b. [DP D [CP DPi C0 [IP DPi I0 AP]]]

In contrast to Kayne 1994, we will propose that in Ancient Greek all adjectives, be they prenominal or postnominal at Spell-out, involve a syntax like that in (28a) above. In other words, we will assume that all adjectives raise from their base predicative position within the reduced relative clause to SpecCP. The prenominal and postnominal positions of adjectives will be derived as a consequence of the other functional heads that have identified within the DP in Ancient Greek – the lower D head, the Topic head, the Focus head, and the higher D head. What is important for us is that the structure in (28a) contains a D position in DPsubj which is in addition to the already existing D position in the main DP, and that this allows for determiner doubling.6

3.2 Double D structures

Let us now go through the steps of the derivation of a DP modified by an adjective in Ancient Greek, in a bottom-up fashion. The complete structure of modified DPs will result from putting together (28a) above with the higher

6 We believe that the occurrence of two Ds is sensitive to the type of adjective that is involved. In particular, we assume that only non-intensional adjectives can be affected by D spreading. It is well-known that intensional adjectives combine with common nouns, i.e. nominal predicates, rather than with DPs. Thus the nominal within the reduced relative clause cannot be a DP when the adjective is intensional. Rather, it must be an NP. This accounts for the fact that ‘proteros’/ ‘former’ was not found in sequences of DNDA.


148

functional structure we have argued for in the previous section. Below we have represented the complete DP structure for DPs in which Topic is projected –see (29)– and for DPs in which Focus is projected- see (30).

(29) DPext

Dext TopP [+fam] [EPP] Top DPint [top] [EPP] Dint CP [+fam] [+def] [EPP] APi C0 IP [ufam] DPsubj I0 APi [+fam] [+fam]

(30) DPext

Dext FocP [+fam] [EPP] Foc DPint [foc] [EPP] Dint CP [α-fam] [+def] [EPP] APi C0 IP [ufam] DPsubj I0 APi [+fam] [+fam] There are two important observations about these structures. First, the DPsubj

should be thought of as a complete DP structure. This means that if the DPsubj is definite, its higher D will bear a [+fam] feature. Second, notice that Dint in (29) and (30) above has different selectional properties as compared to Dint in (24a)- the former an [EPP] feature, while the latter does not. This is not surprising, given that the two Ds have different selectional features (one selects a CP while the other


149

selects a NumP) and thus could be considered as two different lexical instantiations of the category Dint. This [EPP] feature on the Dint in (29) and (30) will trigger movement of the DPsubj to the SpecDint. Once this happens, the [ufam] feature on Dint will check against the [+fam] feature of the moved DP. Notice that this is different from what happens in unmodified definite DPs, where the [ufam] feature of the lower D gets checked by the [fam] feature of the upper D. This implies that DPs that contain a modifier will always ‘borrow’ the value of the [fam] feature from the DP subject of predication.

The next step in the derivation is the merging of a Topic or a Focus head. Remember that we assume that DPs contain only one discourse related projection above the definite DP: either a TopicP or a FocusP. If a Topic head is merged, as in (29), its [EPP] feature will trigger movement of an XP bearing a [top] feature to its Specifier. There are two possible XPs that can be attracted: either the DPsubj that had been raised to SpecDint, or the whole DPint. These are the only possibilities because DPint is a phase and hence subject to the PIC. However, we will assume, together with Kayne 2005:54 that movement of the complement of a given syntactic head to the Specifier of that same head is illicit.7 The only XP that can be attracted to SpecTop is thus either the DPsubj that had been raised to SpecDint. If a Focus head is merged instead of a Topic head, as represented in (30), then its [EPP] feature will essentially trigger movement of the same DPsubj to its Spec as the [EPP] feature on the Topic head above. The only difference from the structure in (29) will be that the [α-fam] feature on the Focus head will in addition be valued as [+fam] as a consequence of entering a checking relation with the [+fam] feature on the DP moved to its Spec. There is no such valuation taking place in the case of a Topic head, because the Topic head has a [+fam] feature already.

The next step in the derivation is Merge of the outer D head. Given its [EPP] feature, Dext will attract an XP with a [+fam] feature. The only possible candidate is the DP in Spec Top/Foc. No other option is available; DPint is a phase and hence subject to the PIC and in addition, the option of moving the whole TopP/FocP to SpecDext is out for independent reasons (see Kayne 2005). The resulting word order under (29) and (30) is always [DNDA] and never [DADN].8 We thus predict that DADN structures are not simply unattested but actually ungrammatical and hence that it is not just a coincidence that such orders were not found in our corpus.

7 In order to derive this constraint, Kayne 2005 assumes that the maximal set of matching features must be checked at the point where H and its complement are Merged. In other words, when H merges with its complement, their relationship is ‘consummed’; there is thus no reason for movement to Spec H to satisfy this relationship. 8 Remember that it is Dint that has phonological content under our assumptions, not Dext.


150

(29) and (30) can also explain why the discontinuous [DN… DA] is grammatical, while [DA… DN] is not. (29) and (30) are the springboard for the split of the DP in the sense that only material in the SpecDext will be allowed to move to a higher clausal discourse oriented projection. This is because DPext is a phase in our view, and hence subject to the PIC. The only other available option would be movement of the whole DPext to the clausal Top or Foc position, but this would obviously not result in a discontinuous DP configuration.

3.3 Single D structures

An additional property of the structures in (29) and (30) is that they can also account for modified DPs with only one overt D. As mentioned in section 1, the only attested orders of this type are the continuous NDA, DAN, and the discontinuous N… DA, DA… N. Crucially, in our view, the unique overt D in these constructions is the main Dint. In other words, the DPsubj is a so-called `bare NP’ in these instances. Recall from section 2.4 that the higher D of `bare NPs’ is underspecified for familiarity. The consequence of this is that it allows for an order in which the adjective precedes the noun, which is never possible if the DP subject of predication contains an overtly definite D, since the latter will bear a [+fam] feature. To be more precise, the structure of the DPint will be exactly the same as the DPint above with the exception of the fact that the [ufam] feature on the Dint head cannot be checked against the [α-fam] feature of the DP moved to its Spec. This is because an [α-feature] is by definition unvalued and thus cannot value an uninterpretable matching feature. Now, if Top is projected, the DP in SpecDint will be attracted to its Spec and the [α-fam] feature of the DPsubj, as well the [ufam] feature of the Dint will be valued as [+fam] as a consequence of entering a checking relation with the matching [+fam] feature on the Topic head. Once DPext is projected, the DP in the SpecTop (DPsubj) will be attracted to SpecDext, since by now the DPsubj bears a valued [+fam] feature. The resulting order will always be NDA. This is represented in (31).


151

(31) DPext

Dext TopP [+fam] [EPP] Top DPint [top] [EPP] Dint CP [+fam] [+def] [EPP] APi C0 IP [ufam] DPsubj I0 APi [α-fam]

If Focus is projected instead of Topic, it will attract the material in the Spec, Dint, just as when the Topic is projected. Similar to the situation above, the [ufam] feature on the Dint head cannot be checked against the [α-fam] feature of the DPsubj moved to its Spec. Merging the Foc head is not going to contribute anything to the checking or valuing of these features, since Foc bears an [α-fam] feature itself.

(32) DPext

Dext FocP

[+fam] [EPP] DPsubj Foc DPint

[α- fam] [foc] [EPP] Dint CP [α-fam] [def] [EPP] APi C0 IP [ufam] DPsubj I0 APi [α-fam]

Once DPext is projected, the closest DP with a [fam] feature will be attracted. But now notice that the DPsubj in the SpecFoc is not a suitable candidate since its


152

[fam] feature is underspecified. FocP is not a suitable candidate either given that FocP is the complement of the attracting head. The only possibility is thus for the DPint to be attracted to SpecDext, since DPint is specified for familiarity and it does not violate any condition. The resulting order is DAN. Summing up the two situations (DPs with TopicP and DPs with FocusP) our analysis predicts that the only grammatical orders are NDA.

Crucially for our analysis, we are relying on Rooryck’s 1994 distinction between [α-features] and [0-features]. Both are underspecified but to different drgrees. In terms of an Attribute-Value system for the representation of features, [0-features] are not specified for their Value, but they are specified for their Attribute. In this sense, uninterpretable features are [0-features]. On the other hand, [α-features] exhibit a complete absence of specification, both in the Attribute and in the Value. The implicit assumption we are making is that only features whose Attribute is specified count as suitable Goal for a Probe searching for a feature (in this particular case, the [EPP] feature on Dext is the Probe). Crucially, not only features that are valued as [+feature] or [- feature] count as specified, but also uninterpretable ones. To have an uninterpretable [ufeature] means to be specified for that feature (i.e. for the Attribute) but not to have a value for that feature yet. To be underspecified for an Attribute means simply to be compatible with having a feature with that Attribute, but not to actually be specified for it.

4. Conclusions

In this paper we have argued that the restrictions on the possible word orders of continuous DPs in Ancient Greek, as well as the restrictions on the possible discontinuous DPs follow from (i) assuming a double layered DP syntax; (ii) assuming that each of the two DP layers is a phase and hence subject to the PIC, and (iii) assuming that DP phases have a left periphery including discourse oriented projections like TopicP and FocusP, in the same way as clauses do. Unlike existing proposals on the left periphery of clauses, we have proposed that one discourse related projection –either TopP or FocP– is sufficient to partition the DP into two informational chunks: Topic-comment or Focus-background. Informational subparts that are at the edge of the DP can be further computed as independent constituents by the clause level syntax.

The mechanics of our proposal crucially made use of four features within the DP domain: definiteness, familiarity, topic and focus. Crucially, apart from the interpretable/non-interpretable distinction within features, we also assumed a distinction between specified and underspecified features. Specified features are either interpretable ones, which are valued, or uninterpretable ones, which are unvalued. Thus we proposed to keep apart uninterpretable features, which we assumed to be specified but unvalued, from underspecified features, which we


153

assumed to be not specified and unvalued. This distinction allowed us to have a Probe with an uninterpretable feature that could only be checked by a matching specified feature. What we wanted to rule out was a situation in which an uninterpretable feature on a Probe gets matched by a matching feature that is underspecified.

Last but not least, the theory we have built predicts that all the orders of modified DPs that were unattested in our corpus are in fact ungrammatical. This does not rule out in principle the possibility of reaching a different result for the other types of DPs that we haven’t analyzed in this paper (i.e. DPs with Demonstratives, with quantifiers, and with wh-elements. As mentioned before, we leave the analysis of the ordering restrictions on these other types of DPs for further research, but we hope that we have provided a line a thought that will prove to be instrumental for these DPs as well.

References

Abney, Paul S. 1987. The English Noun Phrase in its Sentential Aspect. Ph.D. dissertation, MIT.

Aboh, Enoch O. 2004. ‘Topic and focus within D’. Linguistics in the Netherlands 2004 21: 1-12.

Belletti, Adriana. 2003. ‘Aspects of the low IP area’. In Luigi Rizzi (ed.) The structure of IP

and CP. The Cartography of Syntactic Structures, Oxford: OUP. Bernstein, Judy. 1997. ‘Demonstratives and Reinforcers in Romance and Germanic

languages.’ Lingua 102: 87-113. Bernstein, Judy 2001. ‘Focusing the ‘Right’ Way in Romance Determiner Phrases.’ Probus

13-1: 1-29. Butler, Jonny. 2004. Phase Structure, Phrase Structure, and Quantification. Ph.D. thesis,

University of York. Chomsky, Noam. 2001. ‘Derivation by Phase’. In: Kenstowicz, M. (ed.). Ken Hale: A life in

language. Cambridge, Mass.: MIT Press, 1-52. Chomsky, Noam. 2005. ‘On Phases’. Unpublished ms.: MIT. [Available on-line at:

http://www.phon.ucl.ac.uk/home/hans/mrg/chomsky_onphases_1204.pdf] Comorovski, Ileana. 1996. Interrogative Phrses and the Syntax-Semantics Interface.

Dordrecht: Kluwer Academic Publishers. Dayal, Vaneeta. 2004. ‘Number marking and (in)definiteness in kind terms’. Linguistics &

Philosophy 27, vol. 4, p. 393-450. Dimitrova-Vulchanova, M., and G. Giusti. 1996. ‘Quantified noun phrase structure in

Bulgarian’. In Formal Approaches to Slavic Linguistics, ed. by J. Toman. 123-44. Giusti, Giuliana. 1996.‘Is there a Top/Foc in the NP?’ In University of Venice WPL 6: 105-

128. Giusti, Giuliana. 2002. ‘The functional structure of noun phrases: A bare phrase structure

approach’. In Guigliermo, Cinque (ed.) Functional Structure in DP and IP, 54-90.


154

Giusti, Giuliana. 1993. La sintassi dei determinanti, Padova: Unipress. Gutierrez-Rexach, J., Mallen, E., 2001. ‘NP movement and adjective position in the DP

phases.’ In Herschensohn, J., Mallen, E., Zagona, K. (eds.), Features and Interface. John Benjamins, Amsterdam/Philadephia, pp. 107-132.

Hallman, Peter. 1997. ‘Reiterative Syntax’. In James R. Black & Virginia Motapanyane (eds.) Clitics, Pronouns and Movement, Amsterdam: John Benjamins, 87-131.

Hallman, Peter. 2000. The Structure of Predicates: Interactions of Derivation, Case and

Quantification. PhD diss., UCLA. Haegeman, L. 2004. ‘DP periphery and clausal periphery: possessor doubling in West

Flemish? Relating nominal periphery to clausal periphery’. In D. Adger, C. de Cat & G. Tsoulas. (eds.), Peripheries. Dordrecht: Kluwer, 211-23.

Heim, Irene. 1982. The Semantics of Definite and Indefinite Noun Phrases. PhD diss., University of Massachusetts at Amherst.

Ishane, T., Puskas, G., 2001. ‘Specific is Not Definite’, GG@G, vol. 2. University of Geneva, pp. 39-54.

Jayaseelan, K.A. 2001. ‘IP-internal Topic and Focus Phrases’. Studia Linguistica, 55: 39-75. Kariaeva, Natalia. 2001. ‘Determiner Spreading in Modern Greek: Split DP Hypothesis’.

Rutgers University. Kayne, Richard S. 1994. The Antisymmetry of Syntax. Cambridge, Massachusetts, London,

England: MIT Press. Kayne, Richard S. 2005. ‘Some notes on comparative syntax, with special reference to

English and French.’ In The Oxford handbook of comparative syntax, ed. Guglielmo Cinque and Richard S. Kayne, 3-69. Oxford: OUP.

Kirk, Allison. 2007. A syntactic account of split DPs in Herodotus. MA thesis. Concordia University.

Kiss, K. E. 1998. ‘Identificational Focus versus Informational Focus’, Language 74 (2): 245-273.

Krifka, Manfred 2005. ‘Focus and/or Context: A second look at second occurrence expressions.’ In H. Kamp & B. Partee (eds.), Context-Dependence in the Analysis of

Linguistic Meaning, Elsevier Publications, 2005. Laenzlinger, Christopher. 2005. ‘French adjective ordering: perspectives on DP internal

movement types’. In Lingua 115: 645-689. Matthewson, Lisa. 2001. ‘Quantification and the nature of crosslinguistic variation’. Natural

Language Semantics 9(2): 145-189. Pesetsky, David. 1987. ‘Wh-in-situ, movement, and unselective binding.’ In Reuland & ter

Meulen (eds.) The representation of (in)definiteness. MIT Press: Cambridge. Reinhart, Tanya. 1982. Pragmatics and linguistics: An analysis of sentence topics. Indiana

University Linguistics Club, Bloomington, Indiana. Rizzi, Luigi. 1997. ‘The Fine Structure of the Left Periphery.’ In Liliane Haegeman, ed.,

Elements of Grammar. Kluwer, Dordrecht, Boston, London. Rooryck, Johan. 1994. ‘On two types of underspecification: Towards a feature theory shared

by syntax and phonology’, Probus 6: 207-233.


155

Starke, Michal. 2001. Move Dissolves into Merge: a theory of Locality. PhD diss., University of Geneva.

Svenonius, P. 2004.’On the Edge’, in D. Adger, C. de Cat & G. Tsoulas (eds.) Peripheries:

Syntactic Edges and their Effects. Dordrecht: Kluwer, 259-287. Szabolcsi, A. 1981 ‘The semantics of Topic-Focus articulation’. In Formal Methods in the

Study of Language, Jan Groendijk, Theo Janssen, and Martin Stokhof (eds), 513-541. Szabolcsi, Anna. 1987. ‘Functional Categories in the Noun Phrase.’ In István Kenesei (ed.)

Approaches to Hungarian. JATE, Szeged. Zamparelli, Roberto. 1995. Layers in the Determiner Phrase, Ph.D. Diss.,University of

Rochester.


33 – 2008

157

FROM MIHI EST TO ‘HAVE’

ACROSS BRETON DIALECTS*

Melanie Jouitteau, Milan Rezac

1. From mihi est to I have

The verb ‘have’ has unique properties in the grammatical system of Breton. All other verbs are subject to the Complementarity Effect on agreement:

(1) Complementarity Effect: Agreement morphology is restricted to cross-referencing pro-dropped subjects of transitives and intransitives, whereas in the context of phonologically overt subjects the 3SG form of the verb shows up (Jouitteau & Rezac 2006 and references there).

(2) a. …e oar/*ouzont/*ouz ar merc'hed. b. …e ouzont. that know.3SG/3PL/know the girls … that know.3PL ‘… that the girls know.’ ‘… that they know.’

Jouitteau and Rezac (2006), henceforth J&R, propose that the effect is due to the intervention of a 3SG element between the phi-probe of T and the overt subject in the (extended) vP. The element intercepts the probe by relativized minimality and controls the phi-Agree of T to give verbs their 3SG form (3a). Pro-dropped arguments which cliticize / incorporate into T bypass the intervener and the agreement morphology of the verb, shown in Table 1a, spells out phi-Agree and/or pro itself (3b).

(3) a. Tuphi=3SG … 3SG [vP … subject …] (overt subject) b. Tuphi=i + proi … 3SG [vP … tpro …] (null subject)

J&R take the 3SG intervener to be specifically the boundary of the extended vP, transitive and intransitive alike, which has nominal properties in Breton. This extended verbal structure could be identified as a small clause. J&R show that overt

* We thank the audiences of the Fifth Celtic Linguistics Conference at Plas Gregynog conference, seminar La Bretagne Linguistique in Brest 2008, and Incontro di Grammatica

Generativa 34 at Padua, especially Luigi Rizzi and Ian Roberts, as well as Jacqueline Guéron. Thanks also to an anonymous reviewer.


158

subject Case licensing occurs low within it. The Case licensing of the direct objects of transitives occurs still lower in the vP. Case appears overtly only on postverbal weak pronouns. In Table 1b, objects appear as verbal proclitics, and subjects as pro-drop agreement on the verb, for example (we return to other variants).

Table 1: Agreement in Breton

X a: X knows b: breaks X c: X has d: X has (variant) 1SG gouz-on a m dorr a m e(u)s Meus 2SG gouz-out a z torr

(ha torr) a c'h eus (ha (t)es)

p/f/teus

3SGM en dorr en de(u)s Neus 3SGF

oar he dorr he de(u)s Deus

1PL gouz-omp hon dorr hon e(u)s m/neu(so)mp 2PL gouz-oc'h ho torr hoc'h eus peus/t/c'h, neusoc'h 3PL gouz-ont o dorr o deus neu(so)nt

The verb ‘have’ in Table 1c is a morphological anomaly in Breton. Its agreement is a prefix, not a suffix, and it recalls strongly the direct object proclitics of transitive verbs (1b) although not synchronically identifiable with them (1d). This prefix is tacked onto a root whose initial segment varies, recalling again the variation of the initial segment of the root torr-/dorr in function of the object proclitics, followed by eus. Eus is independently the existential form corresponding to French il y a “there is” but lit. “there has” and English “there is”. The rest of the paradigm, not given here, makes it clear that ‘have’ is mostly built by prefixes on the same root on which be is built by suffixes, e.g. am bez ‘I have (habitually)’, bez-an ‘I am (habitually)’. Moreover, ‘have’ simply recruits the participle bet and mostly the infinitive bout/bezañ of be for its own participle and infinitive1. That at least is the situation in the more conservative varieties. As Table 1d shows, innovation in others has gone a long way towards regularizing the morphology to suffixal exponence.

More importantly still for us, the verb ‘have’ is also anomalous syntactically: it is the sole verb in Breton that does not obey the Complementarity Effect, agreeing with an overt subject, (4)a. J&R present an account where ‘have’ is built on an applicative structure that introduces the possessor above the 3SG intervener for agreement with other subjects. The point of departure is the hypothesis that the structure underlying the verb is the copula be and a prepositional element (Benveniste 1960, Freeze 1992, Kayne 1993, Guéron 1986, 1995, Harley 2002). On

1 The infinitive of ‘have’ can in some varieties also be kaout, coopted from kavout ‘find’, but with a restricted syntactic distribution: it cannot be used for the perfect auxiliary have, where the infinitive of be always steps in. Like the verb ‘be’, ‘have’ can also not be targeted by ‘long head movement’.

From mihi est to ‘have’ across Breton dialects

159

this view, there is an intimate connection between be in (4)b, with a small clause complement relating the possessor and the theme similar to a theme and a location, and ‘have’ in (4)a.

(4) a. Daou varc’h-houarn o deus ar merc'hed, … two horses-iron have.3PL the girls ‘The girls have (own) two bicycles,

b. … met hemañ ‘zo gante bremañ ‘zo din. but this.one is with.them now is to me but the one that they have (lit. is with them) now is mine.’

The analytic copular constructions (4)b are attributed the structure (5)a, and they are the only possibility in the other Celtic languages for expressing possession. Breton too uses them for accompaniment and possession or attribution in (4)b. For possession however, Breton may also use the synthetic verb ‘have’ (4)a. According to the proposals cited, this is another way of distributing the same basic elements, suggested internally to Breton by the use of the root of be in ‘have’ noted above. The typical execution relates analytic ‘be’ to synthetic ‘have’ through movement, deriving Errore. L'origine riferimento non è stata trovata. from (5)a, corresponding to the derivational theory of applicative constructions where double object Errore. L'origine riferimento non è stata trovata. originates as prepositional (6)a (Baker 1988).

(5) a. [VP be [SC possessor theme]] b. [TP possessork (…) [vP be+Pi [tk ti theme]]] c. [PP possessor [P/Appl+bei [vP ti theme]]]

(6) a. she [√gave [apples [P/Appl John]]] = She gave apples to John b. she [John [P/Appl [√gave apples]]] = She gave+P John apples

J&R propose rather to base-generate the elements of the analytic construction in a different configuration to give the synthetic one, Errore. L'origine riferimento

non è stata trovata.. Doing so takes cue from the current nonderivational theory of applicative structures, which relates prepositional (6)a and double object Errore.

L'origine riferimento non è stata trovata. by base-generating the same elements in different configurations, rather than by movement (Pylkkänen 2002). Starting with the applicative Errore. L'origine riferimento non è stata trovata. permits base-generating the possessor outside the extended vP with its 3SG intervener, explaining why it is the sole argument immune to the Complementarity Effect, insofar as ‘have’ is the sole verb with this structure. However, we may leave it open whether Errore.

L'origine riferimento non è stata trovata. can first derive from (5)a, along the lines of Ormazabal and Romero (1998); it is only important that prior to T phi-Agree the structure Errore. L'origine riferimento non è stata trovata. obtains. J&R suppose that the same structure is also a possibility for the auxiliary ‘have’,


160

replacing possessor by agent and the complement of be by the past participle phrase; this converges with the same parallelism proposed in Mahajan (1997), Kayne (1993), Guéron (1986).2,3

In this paper, we turn to the variation of ‘have’ across Breton dialects, which goes far beyond the morphological regularization note in J&R and indicated in Table 1d. The variation instantiates key points in the development from the analytic construction of the other Celtic languages to a fully transitive ‘have’ where full assimilation to regular transitive verbs has taken place, tracing an evolutionary path in UG parameter space. We follow this route, noting that it maps reasonably onto a procession among Breton dialects from the conservative peripheral regions to the innovative centre. The starting point is a surprisingly conservative variety where ‘have’ is essentially an intransitive applicative, the ending point a full transitive, and the syntactic stage proposed in J&R in between.

(7) Evolution of Breton ‘have’ a. Analytic prepositional ‘be’: theme nominative subject + PP possessor b. Synthetic applicative ‘be’: oblique applicative possessor as subject + nonagreeeing nominative theme as object (Gwenedeg). c. Unique transitive: nonoblique possessor subject above be-vP, unique in visibility to phi-Agree + regular transitive object. (KLT / Central) d. Regular transitive: possessor becomes agent in [Spec, vP]. (Marginal)

2. Gwenedeg: Applicative intransitive

The origin of the Breton synthetic ‘have’ is in an applicative construction regularly and literally expressing X is to Y, where the theme X is the subject of the intransitive copula be and the possessor Y is the indirect object. As a pronoun, the indirect object could be coded as an infix enclitic on an obligatory preverbal particle, 2 We note that Guéron (1986: 166f.) proposes essentially the same two structures, one for possesor have with have+P selecting the thematic possessor in [Spec, IP], one for relational have where be+P select a small clause to the nonthematic subject of which P assigns a possessor reading. While we can connect Breton have to her proposal about theta-roles, we cannot do so for another hallmark of the relational construction for her, indefiniteness of the object. Among the uses of English have, Breton has only the possession and recipient (‘get’) reading, and the perfect auxiliary function; unavailable are the locational reading and as in French the experiencer and causative readings. However, the definiteness restriction seems to be about the same as in French or English, The town has the resource / the fame #(to do this), with the attributive X is to Y construction used for definites when infelicitous here. 3 A reviewer asks if our hypothesis predicts that all ditransitive verbs like ‘give’ should pattern with ‘have’, due to the presence of an Applicative structure. This is not the case. We predict that only ditransitive verbs whose applicative structure is higher than the extended vP should.


161

glossed R. In (8), we see this with the indirect object of give, and in (9), that of be. Here the theme clearly behaves as the intransitive subject in controlling the agreement morphology of be.

(8) y gwr a-m roddes i gwin the man R-me gave the wine (Middle Welsh, Evans 1964: 57, Fleuriot 2002: 23, Book of Taliesin 40)

(9) Gueisson a-m bu-yint. servants R-1SG be-3PL.IMPF.HAB Servants were to me = ‘I had servants.’ (Middle Welsh, Black Book of Carmarthen 96.2)

In Breton a reanalysis of this regular be + indirect object changed its structure to ‘have’. In the first step, the theme ceased to be treated as the subject of the verb for agreement, and be froze in the 3SG form. It is tempting to link this to the Complementarity Effect, whereby phonologically overt subjects do not agree, for it was fully in vigor by Middle Breton when synthetic ‘have’ appears and the agreement seen in Middle Welsh (9) would be impossible.4 As the theme lost its subjecthood for agreement, the possessor gained it, by coming to double the indirect object clitic. Thus (10) has two parses, as be with a theme subject, with no agreement because of the Complementarity Effect, or ‘have’ with a possessor theme. In (11), the latter structure is favored, because the possessor controls a covariant verbal morphology or agreement, a typological typical correlate of subjecthood. (11) extends doubling to a configuration where it was not available to indirect objects, of the independent pronoun in the preverbal focus position, creating a new possessor subject.

(10) Chwiriodd a-m bu hefyd (Middle Welsh) C'hoarezed a-m boe ivez (Modern Breton) sisters R-1SG be(PT).3SG also Sisters were to me / I had sisters. (Fleuriot 2002: 24)

(11) Me a-m boe c'hoarezed I R-1SG had sisters

The switch in agreement controllers from theme to possessor was a watershed in the transition from ‘be’ to ‘have’ (cf. Even 1987: 125). From the earliest robust attestations of synthetic ‘have’ in Middle Breton, the case morphology and word

4 Notably the Complementarity Effect comes into force much later in Irish, and it is often violated with postverbal subjects in Middle Welsh and systematically with nonclefted preverbal subjects in the ‘abnormal sentence’ seen in (9) (Willis 1998, vs. clefted ‘mixed sentence’). These languages did not develop syntetic have although they have the antecedent analytic structure.


162

order make it clear that the possessor is the subject and the theme is not. However, ‘have’ now comes with a unique agreement pattern in the language: an overt subject that agrees, and an agreement / pro-drop morphology that is prefixal rather than suffixal.

The Gwenedeg dialect shows what appears to us to be pretty much this system (11), and establishing this point permits us to present diagnostics of the intransitive applicative structure. The possessor is the subject, albeit with the unique property of not obeying the Complementarity Effect. As a lexical noun or focused pronoun, it has direct case rather than the prepositional shell of indirect objects; as a weak pronoun, it is coded by the prefixal agreement of ‘have’ which has become distinct from particle + object proclitic, while in the meanwhile indirect objects have lost the ability to be coded by proclitics entirely; and it occurs in the structural position of the subject, preceding the theme. The theme, however, has not adopted the manner of transitive objects at all. Direct objects in Gwenedeg are coded by proclitics (O) on the finite verb, the er of (12)a. In contrast, the objects of ‘have’ are marked by a distinct series, the absolute enclitics (A) like ean of (12)b, and moreover they are restricted to 3rd person. The absolute pronouns had historically the same distribution as DPs, but by Middle Breton came to be used (i) in the preverbal focus position for focalized subject or object, (ii) in the postverbal position only for subjects, (iii) as a partly distinct echoic pronominal series to double clitics and pro-drop agreement (Hemon 2000; for echoics, Stump 1989).5 The object of ‘have’ is the unique non-subject that uses the independent absolute pronouns rather than the clitic/agreement-dependent echoics in the postverbal field. In this, it looks on the surface like a subject, which it continues historically. Unlike all other absolute and echoic pronouns however, it is restricted to third person. So the theme of ‘have’ presents two mysteries: a subject-like morphological marking not found elsewhere for objects, and its restriction to 3rd person.

(12) a. M' er/hur guel b. M' em es ean / *te I him/us(O) see.3SG I have.1SG him(A)/you(A) (Gwenedeg, based on Guillevic and Le Goff 1986)

This behavior of the objects of ‘have’ is brought into relation with the distribution of and restrictions on “nominative objects” in Finnish in Rezac (2004:

5 An example of a postverbal non-echoic, absolute subject is (16)a; evit bout me koh ‘despite being me old’ is another. Postverbal absolutes have some other uses, mostly subject-like (setu

huy ‘there you are’), except for objects of imperatives. Objects of imperatives fit the same story as for objects of have below, as discussed in Rezac (2004) for both Breton and Finnish, but we note that they are unexpectedly not always constrained to 3rd person (in Gwenedeg). Independently of this constraint, Central Breton (Humphreys 1995) and Low-Gwenedeg (Cheveau 2007) use a still further pronominal series for objects of imperatives.


163

305-17). Finnish expresses ‘have’ by the be construction in (13), with an oblique (allative) possessor as the structural subject and the theme as the object (Kiparsky 2001). The theme is nominative if and only if it is 3rd person, and accusative like regular transitive objects otherwise. This person-based nominative-accusative alternation is general in Finnish for a certain class of structures. They fit the cross-linguistic Person Case Constraint, by which Case assignment and phi-agreement are out across an oblique intervener, (14) (Anagnostopoulou 2003). In Finnish the constraint restricts Case-licensing of the theme by T to 3rd person if there is a c-commanding oblique subject, which in (13) is the possessor. So (14) is instantiated as (15). Finnish has an alternative way of Case-marking the theme if it is 1st/2nd person, by the accusative lower down in the structure; other languages that show the same restriction may not, like Icelandic (Rezac 2007).

Finnish

(13) Hänella on kirja / se / *sinut // *kirjan / *sen / sinä him.ALL is book/ it /*you.NOM // *book / *it /√you.ACC ‘He has the book / it / you’.

(14) AGR/Case … (*oblique) … XP [Person Case Constraint]

(15) possessor-obliquei T-AGR/CaseNOM … ti … theme=3 Finnish

The Gwenedeg ‘have’ is strikingly reminiscent of Finnish. The historical origins of ‘have’ are basically (13)/(15), and here and only here in the language do we find a non-subject postverbal absolute pronoun, and an absolute pronoun restricted to 3rd person. We take this to diagnose the constraint (14) and attribute to Gwenedeg the structure (15). The possessor is a Quirky subject that receives inherent direct Case. The Case of T thus can go to the theme. Restriction to 3rd person results from the possessor that intervenes between T and the pronoun that receives direct Case.6

Our approach thus stands on the property of the origin of the possessor as an indirect object to treat it as the oblique intervener in the structure (15). This is a theoretical prediction of our account for a grammar with “nominative” objects restricted to 3rd person, since a nonoblique element would absorb the Case licensing capacity of T and leave the object with only the Case of v, like transitive objects. Within Gwenedeg itself, the behavior of the possessor is consistent with its being

6 Weak subject pronouns are generally required to incorporate into T/V to trigger pro-drop agreement. The object of have does not have this option, along with the other postverbal absolutes of note 5. This is unsurprising, although it needs more work. First, there is already agreement with the possessor of have, below analysed as clitic doubling, arguably pre-empting the pro-drop agreement of the theme. Second, it is cross-linguistically common for 3rd person intransitive subjects to control agreement if postverbal rather than preverbal, as is indeed the case in Finnish independently of oblique-subjects (Kiparsky 2001).


164

oblique, unlike in the later stages of the development of ‘have’. The verb ‘have’ in Breton is the signal exception to the Complementarity Effect in having an agreeing subject, but in fact in Gwenedeg this is not quite so: the possessor only controls prefixal morphology on the verb if 1st/2nd person. A 3rd person DP co-occurs with the 3SGM prefix instead of the proper 3SGF or 3PL prefix, which shows up only under pro-drop, (16). Gwenedeg varieties differ in treating independent 3rd person pronouns as 1st/2nd person or as 3rd person lexical DPs.

(16) a. ar peh en des ind / er sent groeit what have.3SG they / the saints done (Larboulette, Buhe er sent) b. m' am-eus me glinet I have.1SG I slipped ‘I have slipped.’ (Herrieu, Kammdro an Ankoù)

Person and pronominality based restrictions are familiar characteristics of clitic

doubling, not of the phi-Agree underlying phi-agreement in English and French. Clitic doubling may be unselective, but it frequently picks targets high on the scale of personhood and pronominality (Poletto 2000)7. Clitic doubling differs from phi-Agree in another way: it can occur with arguments with or without structural Case, while phi-Agree is restricted to the latter (Rezac 2008). We conclude that the morphological covariance of ‘have’ with an overt possessor in Gwenedeg is not due to phi-Agree but to clitic doubling, making it possible for the possessor to be analysed as an oblique. The clitic that doubles the subject is 3SG and incorporates into the verbal complex. Being an oblique, it is invisible for Agree.

Historically, clitic doubling may represent a very early stage in the transition of ‘have’ to a possessor subject: a point where agreement of be with the theme has been lost, but its agreement with the possessor is relatively close to what is available to proclitics. This step may well have been aided because the proclitic in all uses (direct and indirect object, oblique subject of be) could be doubled by a following ‘echoic’ pronoun: a-m boe me (Fleuriot 2002: 19(2), 23(3)). Such postverbal “echoic” pronouns, non-argumental but of the same form as absolute pronouns, could have helped the rise of possessor-’have’ agreement (16b).8 In Gwenedeg we seem to glimpse the first step of possessor agreement by clitic doubling of an absolute pronoun9.

7 There is no variation in this matter according to the lexical or auxiliary status of have, and the preverbal or postverbal position of the subject. 8 Echoic doubling is sometimes restricted to 1st/2nd person, Trépos (2001), Kervella (1995). 9 We find a pronoun doubled by a clitic in two contexts: (i) clitic doubling of a preverbal independent pronoun by the oblique proclitic of ‘be’, and (ii) doubling of such proclitics by a following ‘echoic’. The two cases could have a common origin in the doubling of


165

Clitic doubling for ‘have’ agreement is suggested by another phenomenon unique to Gwenedeg in Breton: agreeing infinitives. If the agreement of ‘have’ arises from the clitic doubling of oblique arguments, it has nothing to do with phi-Agree and is not expected to be restricted to the canonical domain of phi-Agree, finite structures.10 We briefly note three points about Gwenedeg agreeing infinitives here. First, they are obligatory in some varieties, in alternation with a nonagreeing infinitive in others, and absent in some other (Gwenedeg izel). Second, they occur in all contexts where other Breton dialects use an infinitive, including full CP control complements, reduced complements of modals, and even the X° infinitives fronted by long head movement in the ober conjugation (q.v. Borsley et al. 1996, Jouitteau 2005). Third, they co-occur with absolute objects. We get:

(17) a. Me / hi garehe am bout / hi devout bara/ean. I / she would.like 1SG / 3SGF have.INF bread/him ‘I / she would like to have bread / him.’ (Gwenedeg, based on Guillevic and Le Goff 1986) b. Em bout a ran plijadur o heuli va gwall dechoù. 1SG have.INF R do.I pleasure at follow my bad tendencies ‘I have pleasure to follow my bad tendencies.’ (Hemon 1958: 721) c. em bout ean cheleuet 1SG have.INF him(A) listened ‘(I have had great good) having listened to him.’ (Gwenedeg, Guillome 1836: 44)

Clitic doubling need not be subject to the above illustrated 1st/2nd person restriction11. Extension of clitic doubling in Gwenedeg to all possessors would set up a situation identical to what phi-Agree would produce, creating a point of transition to a system where the possessor is analysed as a nonoblique goal of phi-Agree. As a consequence, the theme loses the ability to be Case-licensed by T which must license the possessor, and can only survive by become a regular transitive object. This will be the next stage in the evolution of ‘have’.

independent absolute pronouns in preverbal and postverbal position alike by the proclitics. In Modern Breton however, they are to be distinguished: the preverbal pronouns have yielded the absolute series that are generally preverbal (minor examples mentioned in note 5) and only ever doubled by the prefix (< proclitic) of ‘have’, while the postverbal pronouns have become the echoics that double a variety of preceding agreement morphemes (including the proclitics). 10 Crosslinguistically, there do exist agreeing infinitives (Portuguese), but they are rare. 11 Dialectal variation in subject doubling is not surprising. See Poletto (2002) for the range of subject doubling variation across related dialects.


166

Gwenedeg provides a second point of transition towards a transitive structure: auxiliary ‘have’ is found beside lexical ‘have’ in Gwenedeg and as far back in the history of Breton as synthetic ‘have’ goes. Analytic tense with ‘have’ auxiliary provide the sole context where ‘have’ can code the theme either by absolute pronouns, ean, or by the same proclitic series as regular direct objects, er, (18) (cf. partly Finnish (13)), and suggests free variation in the presence of the transitive Case-licensing system above the theme: Case-licensing by T is available only to 3rd person theme, by v to all.

(18) a. Te ha tes ean/*ni perpet karet. b. Te ha tes perpet er/hur haret. you have.2SG him/us (A) always him/us (O) loved. ‘You have always loved him/us.’ (cf. Guillevic & Le Goff 1986: 30-1)

The proclitic object option is available for all lexical verbs (cf. above or ‘see’ in (12)a). ‘Have’ stands out in the paradigm once again and cannot host the proclitic. Le Roux (1957: 202) and Kervella (1995: §428(k)) attribute complementary distribution of object proclitics and the prefix agreement morphology of tensed ‘have’ to their common origin. We do not disagree but it leaves unexplained the lack of object proclitics in non-finite lexical ‘have’. We note that historically it is the participle in auxiliary construction that licenses the direct object proclitic. The participle either does or does not distribute case to its internal argument. In Gwenedeg, ‘have’ is intransitive and licenses no internal argument12.

Gwenedeg therefore has a synthetic ‘have’ which is still close to the original analytic X is to Y structure. It is an applicative construction rather than a prepositional construction, mapping the possessor above the theme. The theme is still Case-licensed by the system Case-licensing subjects rather than direct objects, although the latter system has already obtained its foothold in auxiliary ‘have’. The possessor shows an agreement pattern that makes ‘have’ the unique agreeing verb in Breton, but for now it is a clitic doubling rather than a phi-Agree pattern and it extends to infinitives.

In the next stage, the traits of an intransitive applicative vanish and a full if unique transitive has arisen, for which we have identified two points of transition.

12 Proclitics of tensed verbs and proclitics of non-tensed verbs were morphologically distinct in earlier Breton, where the proclitics of the participle and infinitive belong to the distinct possessive series (Hemon 2000: §53(3)). Gwenedeg no longer keeps this distinction (Guillevic and Le Goff 1986: 154).


167

3. KLT: Special transitive

In other Breton dialects, the transition of ‘have’ to a full transitive structure incipient in Gwenedeg has been completed. The result is a verb identical in case marking and infinitive formation to any other transitive, yet distinct in the syntax of agreement by displaying no Complementarity Effect: overt subjects agree. This is the pattern of the standard language, but not an artifact of it. It seemed earlier on to be the dominant pattern across the three non-Gwenedeg dialects which form a unit beside it for other isoglosses: Kerneveg, Leoneg, and Tregerieg (or KLT). In the central area at the Kerne-Treger boundary a new system has arisen and expanded across much of Kerne and Treger, relegating the older pattern to the periphery. We call the latter the KLT pattern, beside the newer Central pattern.13

When we left Gwenedeg, neither the possessor nor the theme of be behaved like the arguments of a transitive. In KLT, the theme has become a plain transitive object, (19). KLT dialects do not code weak pronoun objects as proclitics like Gwenedeg does. The older system survives in parts of Leon, beside a new system that replaces them elsewhere. The new a-forms consist of a preposition a- ‘from’, on which the pronoun incorporates. Contrary to the proclitics, the a-forms have no restrictions. The absolute and echoic pronoun series exist but are not used for objects of ‘have’.

(19) a. Me oar anezho b. Me am eus (gouezet) anezho I know them(O) I 1SG-have known them(O)

The treatment of the possessor also differs from Gwenedeg. All possessors, overt or null, control the same prefix agreement morphology, (20). On the surface, this makes ‘have’ more different from other verbs of the language than in Gwenedeg, because for the other verbs no overt subjects agree. However, in a deeper way the possessor comes closer to the plain subject of transitives: all possessors are treated alike regardless of person and type, and all control one of the typologically common traits of transitive subjects, phi-agreement, even if other subjects do not by the Complementarity Effect. Like other verbs, ‘have’ also has no agreeing infinitives.

(20) a. A-wechoù o deus ar re yaouank seiz tonnel da droc'hañ. sometimes 3PL have the ones young seven barrel to cut ‘Sometimes the young had 7 barrels to cut.’

13 The traditional nomenclature of Breton dialects follows the old diocesal lines: Gwenedeg in the South-East (uhel ‘low’ in E and izel ‘low’ in W), Kerneveg SW (uhel inland, izel maritime), Leoneg NW, Tregerieg NE. Overlaid on it is a newer pattern where the innovative center at the Kerne-Treger boundary creates a Central continuum spreading out chiefly SW and NE, leaving Gwenedeg and Leoneg as peripheral conservative regions, beside over very distinctive though non-conservative pockets like Bigouden in Kerne.


168

(Leon, Blaz an douar, p. 175) b. Ar re-ze o-doa komprenet pell 'oa. these 3PL had understood long.ago These ones had understood long ago. (Kerne izel, Marvaillou 11: 22)

It is this stage that is the focus of Jouitteau and Rezac (2006). They interpret it as a modification of the applicative structure (15) of Gwenedeg by the loss of oblique case on the possessor. It becomes a plain DP and as such participates in structural Case assignment and phi-Agree relations of T, (21). This simple change derives the dramatic differences of KLT from Gwenedeg. Loss of oblique case assimilates the possessor to transitive subjects and it gets the Case they get. However, its origin in the specifier of the applicative head P/Appl places it in a position where it avoids the intervention of the 3SG (extended) vP boundary that creates the Complementarity Effect for other subjects, which are within it. The result is a unique configuration where overt DPs control phi-Agree. The structural Case of the possessor has repercussions farther down for the theme, because the possessor absorbs the Case of T and it is not available for the theme. The latter must be Case-licensed by the same mechanism as transitive objects, so its absolute form disappears (it resorts to an a-form). Consistently with the idea that infinitives are not the locus for phi-Agree, in dialects where the agreement morphology of ‘have’ comes from phi-Agree, we find no ‘agreeing’ infinitives for ‘have’.

(21) Tuphi [ApplP/PP possiphi Appl/P [vP=3SG be [vP vtrans … theme]]] --phi-Agree/Case-- -----------

There are two telling variations on the KLT pattern that fit into this proposal. The first is a regularization in agreement morphology of ‘have’. Its historical origin (opaque in synchrony) as an indirect object proclitic makes it a unique proclitic subject agreement. In some KLT varieties a partial regularization to that used by other verbs occurs. In (22) for example, the 3PL future def-i-nt [have-FUT.3PL] replaces o def-o [3PL have-FUT.3SG] by taking on to the perceived stem def- of ‘have’ the regular future 3PL suffix -i-nt. Such regularization never seems to occur within a Gwenedeg-type system were prefixes reflect clitic doubling, not phi-Agree14.

14 J&R propose that the pro-drop morphology of regular verbs is reflecting phi-Agree with pro rather than the subject pro itself, so it is unsurprising that phi-Agree with the overt subject of ‘have’ undergoes phi-Agree starts using this morphology, although we will see a different road for a dialect to take in the next section.


169

(22) Ma faotred def-i-nt… [defint for regular o defo] my sons have-FUT-3PL ‘My sons will have…’ (Kerne, Ernault 1888: 265 < Barzaz Breiz)

The second change within much of KLT is the loss of gender distinctions in 3SG. Within Breton, pronouns but not verb agreement make the M(asc)/F(em) distinction in 3SG.15 Supposing phi-Agree to be insensitive to gender distinctions, J&R suggest that the lack of gender in verb agreement comes from its spelling out phi-Agree with pro rather than pro itself. On ‘have’, the Gwenedeg systems shows gender distinctions, reflecting the nature of the prefix as an object proclitic. If in KLT, ‘have’ agreement with overt subjects is phi-Agree, it should neutralize gender distinctions. Many varieties do so, (23): some only for agreement with overt subjects, others for pro-dropped subjects of ‘have’ as well.16 However, contrary to the proposal, it seems relatively common for a KLT system to reflect gender distinctions even with overt subjects of ‘have’, suggesting that phi-Agree does make gender distinctions and they do not appear on the pro-drop agreement of regular verbs by diachronic accident. We leave this for future research.

(23) Marharid ‘n eus laret din n’eus ket ‘ vara … Marharid 3SG.M has told to-me NEG is NEG P bread ‘Marharid told me there is no bread.’ (Leon, Amañ 'z eus plijadur, p. 66)17

4. The Central innovation

In the “Central” dialectal area at the Kernev-Treger-Gwened border, the principal source of innovation in Breton dialects, the (arguably antecedent) KLT pattern is modified by one that at first sight looks like a mixture of KLT and Gwenedeg properties. Mostly, it is KLT: the object is coded fully as the object of transitives (mostly with a-forms in this area), and there are no agreeing infinitives. As in KLT, a regularization of the agreement morphology of ‘have’ occurs, but generally goes considerably further than in KLT. The Gwenedeg-like characteristic is the apparent restriction of agreement with overt subjects to 1st/2nd person, as in the paradigm in Table 2.

15 Independent and clitic pronouns, as well as the prosodically fused ones that surface as suffixes on prepositions with the behavior of strong pronouns: Jouitteau and Rezac (2006). 16 In Le Roux (1977: map 84), the latter occurs in most of Kerne but is only rare in Leon and Treger. 17 Jean-Pierre le Goff, Plounévez-Lochrist, Amañ ‘z eus plijadur:66.


170

Table 2: ‘have’ paradigm in Poher (Kerne uhel), Trevidig (1987)

SUBJ ‘have’ a�noued 'a cold' ‘have’ + pro SUBJ

1SG mé meus a�noued meus

2SG té peus a�noued peus

3SGM,F he�w, hi neus a�noued neus

1PL ni �(m) meum a�noued meum

2PL houi peus a�noued peus

3PL i/hè neus a�noued ne�u �int

Morphologically, we see that the old prefix paradigm has been reinterpreted to show only person distinctions (1 m, 2 p, 3 n), and plurality in 1st/3rd person is marked by recruiting the suffixes of the regular verb (1PL m(p), 3PL int). However, the 3PL marking is only used for pro-dropped 3PL subjects; overt subjects use the 3SG form neus. Although this looks like Gwenedeg agreement with 1st/2nd person, the Central dialects do not show the properties of Gwenedeg that go with oblique possessors: absolute objects and agreeing infinitives.

There is another interpretation of the pattern in Table 2: phi-Agree occurs but does not affect suffix morphology. Other dialects in this area furnish evidence for it. In them, the 3PL overt/pro alternation i neus – pro neuint extends to 1PL as ni

beus/meus – pro meu(so)mp/neu(so)mp (Favereau 1997: §424), showing that the prefix and suffix reflect distinct systems that correlate with overt and pro-dropped subjects respectively. We posit that the prefix reflects phi-Agree, and because it tends to make only person distinctions, 3rd person is neutralized, but 1-2-3 contrasts remain. Rarely, phi-Agree is also shown by the suffix, as in Table 2 ni meum. Typically however, the suffix reflects only pro, alone or in combination with the prefix for phi-Agree: 1PL meu(so)mp/neu(so)mp, 2PL peuc'h, 3PL neuint.

5. Plain transitive

The final stage in the evolution of Breton takes the KLT system one step further by removing the sole syntactic difference between it and regular verbs. Agreement with all overt subjects disappears, and ‘have’ falls under the Complementarity Effect, whereby it agrees only with pro-dropped subjects. All other properties of the KLT ‘have’ remain, which already are those of plain transitives. The system appears rarely.18

18 While the pattern is commonly cited, it seems hard verify. The loci classici give eighteenth century Treger sources where 2nd/3rd persons use the 3SGM form: Le Roux (Le Roux 1957: 186), Hemon (2000: §174). Le Roux (1977: maps 82ff.) finds (only) nearly complete


171

(24) ni/te/huy/ey (a) neus I/you(SG)/you(PL)/he R have.3SG ‘We/You have.’ or ‘He has.’

This is the logical conclusion of the transitivization of the verb, the creation of a regular transitive. We suppose at this step to occur simply the loss of the applicative structure and thus to result a plain transitive like Czech mít, or Basque eduki ‘have, hold’. Chiefly possessive, in combination with a small clause they also express the perfect (25) and so could serve as the perfect auxiliary. For Breton, this stage remains to be documented and investigated.

(25) Mám tu chybu opravenou (Czech) I.have the error corrected ‘I’ve got the error corrected’ ≠ ‘I have the error which is corrected.’

6. Conclusion and prospects

We have followed the development of the verb ‘have’ from an applicative intransitive to a plain transitive, tracing the path of its diachronic evolution through its incarnations across the spectrum of Modern Breton dialects. Along the road, various tools have been used to probe the ‘have’ construction in order to ascertain its syntactic character:

(26) a. Morphology and agreement of the theme as canonical intransitive subject, intransitive subject in object position, and regular transitive object. b. The Person Case Constraint restriction on the theme, revealing a c-commanding oblique intervener for T-theme Case licensing. c. Sensitivity to person and pronominality for agreement with overt subjects, indicating clitic doubling rather than phi-Agree. d. Agreeing infinitives, indicating clitic doubling rather than phi-Agree. e. Loss of gender distinctions, perhaps indicating phi-Agree.

There are many theoretical predictions here to be set against further data from Breton and elsewhere. The various aspects of the syntax of ‘have’ do not define a bi-polar opposition but a collection of combinable properties, some of which have not been met with. Gwenedeg has oblique possessors with restricted clitic doubling and absolute themes Case-licensed by T; eliminating the restriction on clitic doubling would produce a Gwenedeg system with fully “agreeing” possessors, while adding the transitive v would produce a Gwenedeg system with themes optionally and

regularization only at St. Fiacre (Treger). We note however that the posited evolution of ‘have’ has happened in the history of Breton to another, much rarer verb: teurvezout ‘daign’, which started out with the same history of the proclitic indirect object to become subject agreement and finished by full regularization before the modern period (Hemon 2000: §151).


172

obligatorily marked as direct objects. However, other aspects of the system are fixed; absolute objects restricted to 3rd person indicate Case-licensing by T and a c-commanding intervener, rather than Case licensing by v or free Case licensing by T. Beside predictions, partial and partly contradictory syntax-morphology correlations emerge, whose status is unknown and that are to be explored in future work. Gwenedeg-type systems with clitic doubling show no tendency to regularize the prefix morphology of ‘have’, which might be explained if the suffix morphology of regular verbs indicates phi-Agree with pro rather than incorporated pro itself. However, in the Central area the suffix morphology seems recruited specifically to spell out pro and the prefix the person dimension of phi-Agree, in contrast.

Turning to diachronic development, the forms of ‘have’ across Breton instantiate key stages along the path of transformation from an analytic X is to Y to transitive Y

possesses X construction (Le Roux 1957), along with the likely points of transition between adjacent stages. Gwenedeg proves surprisingly archaic, arguably a simple intransitive applicative, accounting for most of the properties of its ‘have’ construction that diverge from Standard Breton: quirky subjects, absolute objects, their restriction to 3rd person, non-agreement with 3rd person overt subjects, agreeing infinitives. Yet already as in the earliest Breton sources for ‘have’, the verb is in use in auxiliary constructions where the object of the participle has regular transitive marking. It facilitates transition to the KLT pattern with its loss of inherent case on the possessor and a consequent unique pattern of agreement with an overt subject, revealing its syntactic origin in a high applicative above the extended vP barrier that envelops all other arguments. In KLT the anomalous prefixal agreement morphology tends to get fixed, and nowhere more so than in the innovative central zone whose pattern is rapidly spreading out across Breton and has relegated the classical KLT pattern to the western periphery, meeting the archaic Gwenedeg system at its South Eastern edge. At the end of the evolutionary road, in some North Eastern Treger varieties ‘have’ seems to be on its way to become a regular transitive with no applicativity to it anymore.

We have barely touched here on the factors facilitating the first steps of this transition and the difference between Breton (and Cornish), with stable synthetic ‘have’ as possession and auxiliary verb, and Old Irish and Middle Welsh which began but aborted its formation. The Complementarity Effect seems to us one answer: present from the beginning of Middle Breton and earlier, it is far more finicky in Middle Welsh for postverbal subjects and absent for preverbal ones (Willis 2005), and arises in Modern Irish by convoluted routes during the Middle Irish period (McCone 1987). As we have noted, the Complementarity Effect would have made it easy in Breton to reanalyse the theme as not a canonical subject for agreement, giving it its structural object status. Case also has a role, at least in Irish where dative distinguishes indirect objects and might have prevented their reanalysis as subjects. We leave the details for future work.


173

References

Anagnostopoulou, E. 2003. The syntax of ditransitives. de Gruyter: Berlin. Benveniste, E., 1960. ‘Etre et Avoir dans leurs fonctions linguistiques’, In: Bulletin de la

société Linguistique de Paris, LV, reproduced In: E. Benveniste, 1966, Problèmes de linguistique générale. Gallimard, Paris, 187-207.

Borsley, R. D., M.-L. Rivero, and J. Stephens. 1996. Long head movement in Breton. In The

syntax of the Celtic languages, ed. R. D. Borsley and I. Roberts, 53-74. Cambridge: Cambridge University Press.

Cheveau, L. 2007. Approche phonologique, morphologique et syntaxique du breton du Grand Lorient (bas-vannetais). Doctoral dissertation, Rennes 2.

Ernault, E. 1888. Etudes bretonnes VI. Revue Celtique 9: 245-266. Evans, D. S. 1964. A grammar of Middle Welsh. Dublin: Dublin Institute for Advanced

Studies. Even, A. 1987. Istor ar yezhoù keltiek, vol.I. Lesneven: Hor Yezh. Favereau, F. 1997. Grammaire du breton contemporain. Morlaix. Fleuriot, L. 2002. Ar raganvioù-gour er yezhoù predenek. Hor Yezh 229: 17-26. Freeze, R. 1992. Existentials and other locatives. Language 68: 553-595. Guéron, J. 1986. Le verbe avoir. In Forme (Hommages à Mitsou Ronat), ed. Pierre Pica,

Recherches Linguistiques de Vincennes 14-15: 155-188. Guéron, J. 1995. On HAVE and BE. NELS 26, 1: 191-206. Amherst: GLSA. Guilliome, J. 1836. Grammaire français-bretonne. J.-M. Galles. Harley, H. 2002. Possession and the double object construction. In The linguistic Variation

Yearbook 2: 29-68. Hemon, R. 1958. Dafar geriadur historel ar brezhoneg. Brest: Al Liamm. Hemon, R. 2000. Yezhadur istorel ar brezhoneg. Plañvour: Hor Yezh. Humphreys, H. L. 1995. Phonologie et morphosyntaxe du parler breton de Bothoa. Brest:

Emglev Breizh. Guillevic, A., Le Goff, P. 1986 [1902]. Grammaire Bretonne du Dialecte de Vannes. Brest:

Ar Skol Vrezhoneg – Emgleo Breiz. Jouitteau, M. 2005. La syntaxe comparée du breton. Doctoral dissertation, Université de

Nantes. Jouitteau, M., and M. Rezac. 2006. Deriving the Complementarity Effect: Relativized

Minimality in Breton agreement. Lingua 116: 1915-1945. Kayne, R. 1993. Towards a modular theory of auxiliary selection. Studia Linguistica 47: 3–31. Kervella, F. 1995. Yezhadur bras ar brezhoneg. La Baule: Skridoù Breizh. Kiparsky, P. 2001. Structural Case in Finnish. Lingua 111: 315-376. Mahajan, A. K. 1997. Universal Grammar and the typology of ergative languages. In Studies

on Universal Grammar and typological variation, ed. A. Alexiadou and T. Hall, 35-57. Amsterdam: John Benjamins.

McCone, K. 1987. The Early Irish Verb. Maynooth: An Sagart.


174

Pylkkänen, L. 2002. Introducing arguments. Doctoral dissertation, MIT. Rezac, M. 2004. Elements of cyclic syntax: Agree and Merge. Doctoral dissertation,

University of Toronto. Rezac, M. 2007. Escaping the Person Case Constraint: Referential computation in the φ-

system. Linguistic Variation Yearbook 6: 97-138. Rezac, M. 2008. Φ-Agree and θ-related Case. In Φ theory: Φ-features across interfaces and

modules, ed. D. Harbour, D. Adger, and S. Béjar, 83-129. Oxford: Oxford University Press.

Le Roux, P. 1977. Atlas Linguistique de la Basse-Bretagne. Brud Nevez. Le Roux, P. 1957. Le Verbe breton. Rennes: Librairie Plihon. Poletto, C. 2000. The higher functional field: Evidence from Northern Italian dialects. Oxford:

Oxford University Press. Stump, G. T. 1989. Further remarks on Breton agreement. Natural Language and Linguistic

Theory 7: 429-471. Trépos, P. 2001. Grammaire bretonne. Brest: Brud Nevez. Willis, D. 2005. Historical syntax. In The syntax of Welsh, ed. R.D. Borsley, M. Tallerman,

and D. Willis. Cambridge: Cambridge University Press.


33 – 2008

175

ARGUING AGAINST OBLIGATORY

FEATURE INHERITANCE:

EVIDENCE FROM FRENCH TRANSITIVE

PARTICIPLE AGREEMENT

Michèle Vincent, Andrew Radford

1. Introduction

Chomsky (2005) and Richards (2007) argue that the uninterpretable case/agreement features carried by T and V originate on the phase head which selects them (viz. C and v respectively). This claim gains empirical support from the phenomenon of complementiser agreement (See Rizzi 1990, Haegeman 2002, Boeckx 2003, Carstens 2003, Kornfilt 2004, Miyagawa 2005), which can be illustrated by the following data from West Flemish, showing that ‘the complementiser of the finite clause agrees in person and number with the grammatical subject of the sentence it introduces’ (Haegeman 1992, p. 47):

(1) a. Kpeinzen dank ik morgen goan

I.think that1.Sg I tomorrow go (‘I think that I’ll go tomorrow) b. Kpeinzen daj gie morgen goat I.think that2.Sg (you) tomorrow go (‘I think that you will go tomorrow’) c. Kpeinzen dan Valère en Pol morgen goan I.think that3.Pl Valère and Pol tomorrow go (‘I think that Valere and Paul will go tomorrow’)

In these examples, the (italicised) complementiser overtly inflects for agreement in person and number with the (bold-printed) subject of its clause, lending plausibility to the claim that C is the locus of the agreement features on T1.

1 As pointed out by Radford (in press), the claim that the agreement features carried by T are inherited from C is potentially problematic for defective T (Tdef) which Chomsky (2005) takes to carry person but not to be selected by C.


176

The claim that C is the locus of the case assigned to subjects is argued for in Radford and Vincent (2007) and Radford (in press) on the basis of sentences such as (2) where the difference in the case of the italicised subjects in the two bracketed infinitive clauses would appear to correlate with the (accusative or nominative) case-assigning properties of the underlined complementisers heading the relevant CP phases. (2) a. [For me to stand as a candidate in the elections] would require a lot of money

b. [Para yo presentarme a las elecciones] sería necesario mucho dinero For I present.myself at the elections would.be necessary much money

‘For me (literally: ‘I’) to stand as a candidate in the elections

would require a lot of money (Mensching 2000: 7)

A further conclusion suggested by the contrast in (2) is that the case assigned by a particular complementiser is a lexical property of the complementiser. More generally, we can suppose that each phase head carries a case assignment feature which determines what case it assigns to its goal.

A key additional claim made by Chomsky and Richards is that uninterpretable case/agreement features on a phase head obligatorily percolate down from the phase head onto the selected head (e.g. from C onto T, and from v onto V) in order to ensure that they are deleted as part of Transfer, since only the domain of a phase head undergoes Transfer at the end of a phase: let us call this the Percolation

Hypothesis. If percolation did not take place, the relevant uninterpretable features would fail to be deleted on the next phase by virtue of having already been assigned a value (e.g. via agreement) and thereby having become indistinguishable from interpretable features.

In this article, we accept the view that the relevant type of case/agreement features originate on phase heads, but argue against a strong view of the Percolation Hypothesis on which uninterpretable features obligatorily percolate down from a phase head onto a selected head: on the contrary, we maintain that there are structures in which uninterpretable case/agreement features remain on the phase head throughout the derivation. The main empirical evidence we adduce in support of our claim comes from a novel analysis of French past participle agreement which builds on earlier work by Radford and Vincent (2007) and Vincent (2007). In section 2, we briefly characterise French past participle agreement, and outline the key assumptions which our analysis makes. We show how our analysis handles past participle agreement with a local direct object in section 3, and go on to show how it correctly specifies when (and why) agreement can take place with the subject of an embedded infinitive complement in section 4. In section 5, we present further empirical evidence against the Percolation Hypothesis from a range of independent phenomena, and highlight some theoretical inadequacies of the hypothesis, as well

Arguing against obligatory feature inheritance

177

as reconsidering the motivation for feature percolation. Finally, in section 6 we summarize our overall conclusions.

2. A brief overview of French past participle agreement

In French transitive clauses containing the auxiliary avoirhave and a complement headed by a past participle, the participle (in bold) optionally agrees in number and gender with an (underlined) preceding specific local direct object in structures like (3) below, so that the participle can either be spelled out with the same number/gender properties as the object, or be spelled out in the default/Def (masculine singular) form:

(3) Quelles femmes a-t-il a mis/mises en prison? Which womenF.Pl has-he put.Def /putF.Pl in prison ‘Which women did he put in prison?’

However, there are a number of conditions which govern participial agreement in structures with avoirhave. For instance, a participle can only agree with a preceding

direct object (as in (3) above), and not with a following (e.g. in situ) object like that underlined in (4) or with a preceding non-specific object as in (5)2:

(4) Il a mis/*mises ces femmes en prison He has putDef /*putF.Pl thesePl womenF.Pl in prison ‘He put these women in prison’

(5) Des femmes pareilles, il en a souvent mis/*mises en prison SomePl womenF.Pl similarF.Pl, he some has often putDef /*putF.Pl in prison ‘Similar women, he has often put in prison’

A participle can also agree with a subject extracted out of an embedded infinitive complement in a structure such as (6) below:

(6) Quelles femmes a-t-il jugé/jugées être complices? WhichF.Pl womenF.Pl has-he judgedDef /*judgedF.Pl be accomplices? ‘Which women did he judge to be accomplices?’

However, while participle agreement is possible with a matrix verb like jugerjudge, it is not possible with a matrix verb like diresay, as we see from:

(7) Quelles femmes a-t-il dit/*dites être complices? WhichF.Pl womenF.Pl has he saidDef /*saidF.Pl be accomplices? ‘Which women did he claim to be accomplices?’

A key assumption which we will make in order to account for the agreement patterns illustrated in (2-7) above is that an item only carries a given feature if the

2 See Obenauer (1994), Richards (1997), and Déprez (1998) for relevant discussion.


178

latter is detectable, in the sense that it has an overt reflex on X or on some Y agreeing with X (Detectability Criterion)3. In other words, when a participle does not inflect for number and gender, we assume that it carries no such features. This assumption is at odds with Chomsky (2000, 2005) for whom case assignment is a reflex of Agree whether the agreement features of the case assigner are detectable or not. In other words, for Chomsky, the transitive participle mis ‘put’ assigns accusative case to the object quelles femmes ‘which women’ in (3) via agreement even when the agreement features of the participle are not detectable. However, the claim that agreement is a necessary correlate of case assignment has been argued against by authors such as Iatridou (1993) who maintains that in Modern Greek, it is tense, not agreement, which is responsible for nominative case assignment4. In a similar spirit, Radford and Vincent (2007) and Vincent (2007) have proposed that accusative case is assigned to a (pro)nominal expression by an accusative-case-assigning feature on v (see also Chomsky 1995, Radford 1997 and Adger 2003) which v may only carry if it has a thematic external argument.

In order to account for the correlation between agreement, movement and specificity, we propose the following. We follow Chomsky (2005) in assuming that movement to the edge of a phase is driven by an edge feature (EF) on the phase head. However, unlike Chomsky who argues that EF is always unselective and can attract any kind of expression to the edge of a phase, we propose instead that EF can be specified for a value. More specifically, we propose that an interrogative wh-expression is attracted to the edge of a phase by an uninterpretable [wh-edge] feature on the phase head5, and that a specific (i.e. definite/D-linked/topicalised) expression is attracted to the edge of a phase by an uninterpretable feature [sp(ecific)-edge] feature on the phase head. The uninterpretable edge feature is deleted via movement of a

3 For present purposes, we take a feature to be detectable if it has a phonetic or orthographic reflex. Thus, the feminine plural participle jugées ‘judgedF.Pl’ has detectable gender and number features by virtue of carrying the orthographic feminine gender and plural number suffixes -e and -s respectively, although these do not normally receive a phonetic spell-out. 4 Case and agreement have also been argued to be dissociated in Ancient Hebrew (Mensching 2000), Bantu (Carstens 2001; Henderson 2006), and Lithuanian (Franks and Lavine 2006). 5 An alternative possibility (not pursued here) is that movement to the edge of a phase can either be selective or unselective, and that movement of a specific expression to the edge of a phase is driven by a selective edge feature [sp-edge] working in conjunction with agreement, while unselective movement of any wh-constituent α to the edge of a phase is driven by an unselective edge feature [α-edge] without agreement. There could then be parallels with EPP, which can either be a selective feature driving movement of a specific expression to the edge of TP (and working in conjunction with agreement), or unselective (and working without agreement) in e.g. locative inversion structures like ‘He could see that [out of a tiny hole in

the ground had emerged an army of red ants].’


179

matching expression to the edge of the phase. We further assume that v can carry an accusative case feature if it has a thematic external argument, and that if v has accusative case and [sp-edge] features, it will also carry a set of (number and gender) agreement features. The fact that number and gender features are obligatorily present when v carries [sp-edge] and case features ensures that these last two features have an overt morphophonological exponent, since specificity and case are not directly spelled out on v. A further (standard) assumption we make is that when (person, number, or gender) agreement features probe, they can only ‘see’ a constituent in an A-position, not one in an A-bar position. Finally, we assume that case/agreement features on a phase head H remain on H if they locate an accessible goal within the immediate domain of H, but otherwise percolate down onto the head selected by H.

We summarize our overall assumptions in (8) below:

(8) In a transitive vP headed by a past participle serving as the complement of avoirhave

(i) The lexical verb originates in V and subsequently raises to adjoin to a participial light verb v, so ensuring that the verb is spelled out in a participial form.

(ii) v can carry an edge feature/EF, which may be valued either as [wh-

edge] or [sp-edge]. If v has a [sp-edge] feature, it attracts a specific (i.e. definite/D-linked/topicalised) goal to move to the edge of vP; if v has a [wh-edge] feature, it attracts an interrogative wh-expression (whether specific or non-specific) to move to the edge of vP

(iii) If v has a thematic external argument, v can carry a structural (accusative) case feature which enables it to value an unvalued case feature on a goal as accusative.

(iv) If v has accusative case and [sp-edge] features, it also carries a set of (number and gender) concord/agreement features.

(v) A-agreement features (= person-/number-/gender-agreement features) can only ‘see’ a goal in an A-position (i.e. in a thematic argument position, or in the specifier position of an A-head like T: Visibility Condition)

(vi) A phase head H retains its case/agreement features just in case it locates a visible goal within its immediate domain (i.e. a goal which is within the domain of H, but not within the domain of any head c-commanded by H), but otherwise the case/agreement features of H percolate down onto the head of its complement (Percolation Condition)


180

In the context of the key theoretical question addressed here of whether phase heads obligatorily pass on their uninterpretable case/agreement features to the head of the complement they select, it should be noted that (8vi) posits that case/agreement features in French past participle structures remain on the phase head if they are able to locate a goal in their immediate domain. In the next two sections, we show how these assumptions help us provide a principled account of past participle agreement with direct objects and embedded subjects.

3. Past participle agreement with a direct object

The first example which illustrates that the uninterpretable agreement features carried by a phase head must sometimes remain on the phase head is (3) repeated in (9) below:

(9) Quelles femmes il a mis/mises en prison? Which womenF.Pl he has put.Def /putF.Pl in prison ‘Which women did he put in prison?’

Consider first the derivation of the agreeing structure in (9) and assume that we have reached the stage of derivation represented by the simplified structure shown below6 (where uninterpretable features are italicised, and interpretable features are in non-italic print):

(10) [vP ilhe [v ø] [VP [QP quelleswhich femmeswomen] [V misput] enin prison]] [sp-edge] [3-Per] [u-Num] [Pl-Num] [u-Gen] [F-Gen] [Acc-Case] [u-Case]

The order of merger of the arguments here is in accordance with the Merger

Hypothesis of Radford (in press), which specifies that nominal arguments are the last internal arguments to be merged with V, and hence occupy the highest position within VP.

In (10), the light verb (by hypothesis) has a [sp-edge] feature in conformity with (8ii). Since it has a thematic external argument, the light verb can also carry an accusative case feature in accordance with (8iii), and if so will also carry uninterpretable number and gender features via (8iv), and these will look for a goal with matching valued features in order to value and delete them. When v probes in (10), it can ‘see’ the QP goal on the edge of VP in accordance with (8v), since QP is

6 In order to avoid visual clutter, we simplify structures throughout by showing only uninterpretable features and the corresponding interpretable features which serve to value them. This means, inter alia, that we do not show the interpretable features marking QP as interrogative (or otherwise) and as specific (or otherwise).


181

in an A-position (by virtue of being in a thematic argument position): let us suppose that QP (as used here) is specific and interrogative in interpretation, albeit these features of QP are not shown in the simplified structural representation in (10). Since v locates a suitable goal within its own immediate domain, it retains its case/agreement features via (8vi), and these do not percolate down onto V. Accordingly, v assigns accusative case to the QP and the case features of both v and of the QP get deleted. The number and gender features of the QP value those of v as feminine plural and delete them. In addition, the [sp-edge] feature of v gets deleted by attracting the specific QP quelles femmes ‘which women’ to spec-v, ultimately deriving the structure in (11):

(11) [vP [QP qu.which fe.women] ilhe [v ø] [VP [QP t] [V misput] en prison]] [3-Per] [sp-edge]

[Pl-Num] [Pl.Num] [F-Gen] [F.Gen] [Acc-Case] [Acc-Case]

The derivation proceeds by merging the T auxiliary avoir ‘have’ and subsequently a null C complementiser carrying an uninterpretable [wh-edge] feature as in (12) (where we omit the features carried by the QP and v whose uninterpretable features have all been deleted)7:

(12) [CP [C ø] [TP [T avoirhave] [vP [QP qu.fe.] ilhe [v mis] en pr]] [wh-edge] [Past-Tns] [M-Gen] [u-Pers] [3-Pers] [u-Num] [Sg-Num] [Nom-Case] [u-Case]

[EPP]

Given the Visibility Condition (8v), the T-probe will not be able to ‘see’ the QP quelles femmes (because this is in an A-bar position), but will be able to see the pronoun il (because this is in an A-position by virtue of occupying a thematic argument position): this is consonant with the assumption made by Boeckx (2007, p.83) that a wh-moved QP on the edge of vP is ‘transparent’ to T. Consequently T agrees with, case-marks and (by virtue of its EPP feature) attracts the pronoun il to

7 We assume that, in the same way as accusative case is assigned by an uninterpretable accusative case feature on v, so too nominative case is assigned by a nominative case feature which T inherits from C. As should be obvious, the nature of such features challenges the claim that uninterpretable features enter the derivation unvalued. The EPP feature on T can be treated as an [A-edge] feature which attracts the closest constituent with an active A-feature to move to the edge of TP. It makes little difference for our purposes whether C is assumed to carry a [wh-edge] or a [sp-edge] feature.


182

move to the edge of TP. The [wh-edge] feature on C in turn attracts the interrogative QP to move to the edge of CP, ultimately deriving Quelles femmes il a mises en

prison? ‘Which women did he put in prison?’. A key premise of the analysis outlined above is that the uninterpretable

case/agreement features on the light verb remain on v and do not percolate down onto V - contra the Chomsky/Richards Percolation Hypothesis. Let us now examine what would happen if (contrary to what we claim here) the case and agreement features of the light verb were to percolate onto V. In such a case, the structure in (10) would be replaced by that in (13):

(13) [vP ilhe [v ø] [VP [QP quelleswhich femmeswomen] [V misput] en prison]] [sp-edge] [Pl-Num] [u-Num]

[F-Gen] [u-Gen] [u-case] [Acc-case]

[3-Per]

In (13), QP is not an accessible goal for V, since QP lies outside the c-command domain of V: this means that the number/gender features on V remain unvalued and undeleted (causing a crash at the PF interface), and the uninterpretable number/ gender/case features on V remain undeleted (causing a crash at the semantics interface). The conclusion we therefore reach is that the agreement features on v in (the agreeing counterpart of) sentences like (9) cannot percolate down onto V but rather must remain on v.

Now consider how we handle the agreementless counterpart of (9). Suppose we have reached the stage of derivation shown in (14) below:

(14) [vP ilhe [v ø] [VP [QP quelleswhich femmeswomen] [V misput] en pr.]] [wh-edge] [3-Per] [Acc-Case] [u-Case] [F-Gen] [Pl-Num]

The difference between the agreementless structure in (14) and its agreeing counterpart in (10) is that this time v carries a [wh-edge] feature in accordance with (8ii), and therefore cannot carry agreement features (because 8iv specifies that v only has agreement if it also has a [sp-edge] feature). The case feature on v will locate QP as a visible local goal, and value its case feature as accusative (with the case features on v and Q both being deleted thereby). The [sp-edge] feature on v will likewise target the specificity feature on QP (not shown above), and thereby attract QP to move to the edge of vP. The derivation will then continue in familiar ways,


183

ultimately yielding the convergent structure associated with the agreementless sentence Quelles femmes il a mis en prison?

8 The key conclusion to be drawn from our discussion in this section is that the

agreement features of the participle in agreeing structures like (1) Quelles femmes il

a mises en prison? must remain on the light-verb heading the phase if the derivation is not to crash.

4. Participle agreement with an embedded subject

Let us now consider the derivation of a sentence like (6) repeated in (15):

(15) Quelles femmes a-t-il jugé/jugées être complices? WhichF.Pl womenF.Pl has-he judgedDef /*judgedF.Pl be accomplices? ‘Which women did he judge to be accomplices?’

Sentences such as (16) below suggest that jugerjudge can function as an ECM (Exceptional Case Marking) verb selecting a TP complement:

(16) Il a jugé l’attitude des syndicalistes être la cause directe des grèves He has judged the attitude of the unionists be the cause direct of the strike ‘He judged the attitude of the trade union members to be the direct cause of the strikes’

Let us suppose that we have reached the stage shown in (17) below in the derivation of the agreeing counterpart of sentence (15):

(17) [vP ilhe [v ø][VP [V jugéjudged] [TP quelleswhich femmeswomen [T être] compl.]]] [sp-edge] [3-Per] [u-Num] [Pl-Num] [u-Gen] [F-Gen] [Acc-Case] [u-Case]

v carries an uninterpretable [sp-edge] feature via (8ii), and so attracts a specific expression to the edge of the vP projection it heads. Since matrix v has a thematic external argument, it can also carry an accusative assigning case feature via (8iii). By virtue of carrying case and [sp-edge] features, v also carries number and gender agreement features, in accordance with (8iv).

However, the case/agreement features on the phase head v cannot target QP, because this is not within the immediate domain of v (rather, QP is within the immediate domain of V: see 8vi). Thus, in accordance with the Percolation

8 Although we lack the space to discuss this here, it should be noted that a past participle can only agree with its own object, not with the object of a subordinate verb: this is because a subordinate object will have been rendered inactive for agreement with a matrix verb after being case-marked by the subordinate verb.


184

Hypothesis (8vi), the case/agreement features on v percolate down onto V, so deriving the structure shown below:

(18) [vP ilhe [v ø] [VP [V jugéjudged] [TP quwhich femmeswomen [T être] compl.]]] [sp-edge] [u-Num] [Pl-Num] [u-Gen] [F-Gen] [Acc-Case] [u-Case] [3-Per]

V then agrees with and assigns accusative case to QP9, with the relevant uninterpretable case/agreement features being valued and deleted thereby. The [spec-edge] feature on v triggers movement of QP to the edge of vP, and the derivation then continues in familiar ways and ultimately leads to a convergent outcome, thereby accounting for the possibility of agreement between a past participle and an extracted ECM subject.

Although we lack the space to go into full details here, the agreementless counterpart of (15) will have a parallel derivation to that sketched above, save that v will carry a [wh-edge] feature in place of the [sp-edge] feature in (17), and hence will lack agreement features in consequence of (8iv).

Having shown how agreement between a past participle and an extracted ECM subject can be dealt with, let us now consider why agreement is not possible in infinitival structures such as (7), repeated as (19) below:

(19) Quelles femmes a-t-il dit/*dites être complices? WhichF.Pl womenF.Pl has he saidDef /*saidF.Pl be accomplices? ‘Which women did he claim to be accomplices?’

A significant property of verbs like dire is that they cannot be used in ECM structures like (20):

(20) *Il a dit ces femmes être complices He has said those women be accomplices ‘He said those women to be accomplices’

This can be accounted for if we follow Kayne (1984) in taking dire to be a verb which selects an infinitival CP complement. Let us therefore assume that a series of merger operations have formed the agreeing structure in (21) (with outline font marking constituents which have already undergone Transfer):

(21) [vP ilhe [v ø] [VP [V ditsaid] [CP quelleswhich femmeswomen [C ø] [TP ...]]]] [sp-edge] [3-Per]

9 A descriptive detail which we set aside here is the possibility that V may also attract QP to move to the edge of VP.


185

[u-Num] [Pl-Num] [u-Gen] [F-Gen] [Acc-Case] [u-Case]

We have assumed here that v has an accusative-assigning case feature by virtue of having a thematic external argument, and carries agreement features by virtue of having a [sp-edge] feature. However, the Visibility Condition (8v) means that the number/gender agreement features on v cannot ‘see’ QP, because QP is in an A-bar position by virtue of being in a nonthematic position on the edge of a phase. Moreover, the Percolation Condition (8vi) bars the case/agreement features on v from targeting QP as a goal, because QP is not within the immediate domain of v (rather, QP lies within the immediate domain of V). Accordingly, the uninterpretable number/gender agreement features on v (and the uninterpretable case feature on QP) remain unvalued and undeleted, so causing the derivation to crash.

Because they cannot locate a visible goal in (21), the case/agreement features on v will percolate down onto V in accordance with the Percolation Condition (8vi), so deriving the structure (22) below:

(22) [vP ilhe [v ø] [VP [V ditsaid] [CP quelleswhich femmeswomen [C ø] [TP ...]] [sp-edge] [u-Num] [Pl-Num]

[u-Gen] [F-Gen] [Acc-Case] [u-Case] [3-Per]

However, the Visibility Condition (8v) prevents the number/gender features on V from locating QP as a goal, since QP is in an A-bar position. Accordingly, the derivation will once more crash, for familiar reasons.

But now consider what happens if the light verb carries a [wh-edge] feature and so lacks agreement features, as in (23) below:

(23) [vP ilhe [v ø] [VP [V ditsaid] [CP quelleswhich femmeswomen [C ø] [TP ...]]] [wh-edge] [3-Per] [Acc-Case] [u-Case] [F-Gen] [Pl-Num]

As before, QP is invisible to the case feature on v, because QP is not in the immediate domain of v. Consequently, the case feature on v percolates onto V, as in (24) below:

(24) [vP ilhe [v ø] [VP [V ditsaid] [CP quwhich femmeswomen [C ø] [TP ...]]] [wh-edge] [Acc-Case] [u-Case] [3-Pers] [F-Gen] [Pl-Num]


186

Nothing now prevents the case feature which has percolated down onto V from serving as a probe which locates QP as its goal (since QP is within the immediate domain of V), thereby valuing the unvalued case feature on QP as accusative, and resulting in the uninterpretable case feature of both constituents being deleted. Likewise, the [wh-edge] feature on v can target the interrogative QP as its goal, thereby attracting QP to move to the edge of vP (and concomitantly deleting the edge feature on v). Thus, the analysis outlined here correctly specifies that the derivation can only converge in structures like (19) if the participle is unspecified for agreement.

5. Implications and Issues

Thus far, we have presented evidence from past participle agreement in French that phase heads must sometimes retain uninterpretable case/agreement features. There is independent cross-linguistic evidence in support of this conclusion from the phenomenon of complementiser-subject agreement found in a number of languages, and illustrated earlier in relation to the West Flemish data in (1). What is significant about the data in (1) is that the (italicised) complementiser overtly inflects for agreement in person and number with the (bold-printed) subject of its clause. However, since the finite verb in the clause also inflects for agreement with the subject, it would seem that the agreement features of C do not percolate onto T but rather are copied from C onto T so that both C and T end up bearing a copy of the relevant agreement features.

A parallel phenomenon of complementiser agreement is found in a non-standard variety of English spoken by people who come from the Boston area of Massachusetts, in which C agrees with a preposed (italicised) wh-expression that it attracts to move to spec-C, e.g. in structures such as the following (from Kimball and Aissen 1971, p.246):

(25) Where are the boys [who Tom think [Dick believe [Harry expect to be late]]]

Here, the head C constituent of each of the bracketed CPs attracts the italicised relative pronoun who (which is plural by virtue of having the plural noun boys as its antecedent) to move to spec-C. C ultimately ‘hands over’ these agreement features to the bold-printed verbs in the head v position of vP (perhaps via a downward feature-percolation operation in the PF component, of which Affix Lowering may be a particular instance).

A further challenge to the claim that the uninterpretable A-features of v must always percolate onto V comes from our claim that phase heads carry a case-assigning feature which enters the derivation valued but uninterpretable. Recall that a crucial assumption made by Chomsky (2005) and Richards (2007) is that value and interpretability are indissociable. For this reason, uninterpretable features must


187

be deleted at the point where they are valued. Moreover, because the edge of a phase and its complement are transferred separately, the uninterpretable features of a phase head must percolate onto the non-phase head they select in order to ensure that uninterpretable features are valued and transferred simultaneously. However, if value and interpretability are not necessarily two faces of the same coin, Chomsky and Richards’ argumentation falls apart.

We note that support for the claim that phase heads carry a case assigning feature comes from the data in (2) above which suggests, as pointed out by Radford (in press), that ‘the case assignment properties of complementisers are language-specific, and hence unpredictable’. Radford further concludes that ‘complementisers need to carry some uninterpretable feature in their lexical entry specifying what case they (hand over to T to) assign to an active goal within their domain’.

Additional support for the dissociation between value and interpretability comes from the morphosyntax of expletive pronouns. Under the analysis of English expletives in Radford (in press), expletive there enters the derivation carrying an uninterpretable (lexically specified and intrinsically valued) third-person feature, and expletive it likewise enters the derivation carrying uninterpretable (but intrinsically valued) third-person and singular-number features. Likewise, under the analysis of French expletives in Vincent (2007), expletive ilit enters the derivation carrying uninterpretable (but lexically specified and intrinsically valued) third-person, singular-number and masculine-gender features. If the analysis of English/French expletives outlined in the relevant works is correct in essence, it provides further evidence that the Chomsky-Richards correlation between the interpretability and valuation of features cannot be maintained.

It may be that we can reach a similar conclusion in relation to the gender feature carried by nouns in languages like French. French has two genders: masculine and feminine. In some cases, these two genders correspond to the biological property of an entity being male or female respectively, but this is true only of a small number of nouns. In fact, most French nouns carry a purely arbitrary gender which does not relate to any biological property of the entities the nouns refer to. This might lead us to the conclusion that it is more appropriate to look at grammatical gender as a purely formal grammatical, uninterpretable feature. Another property of gender is that it is a lexical feature of nouns (lexical features being inherent features), so that (e.g.) the noun chaise ‘chair’ in French is inherently feminine and can never be masculine (la/*le chaise ‘theF/*M chairF’). Although it is uninterpretable, gender is an inherent feature of nouns which are necessarily masculine or feminine when entering the derivation. We might therefore conclude that French nominals enter the derivation with a valued but uninterpretable gender feature, so reinforcing the claim


188

that value and interpretability do not necessarily work hand in hand10. As we pointed out earlier in this section, if value and interpretability are dissociable notions, the kind of motivation for feature inheritance argued for by Richards is heavily compromised. But if this is so, we need an alternative way of motivating feature percolation.

Ouali (2007) suggests that feature percolation takes place because of closeness. More specifically, he claims that the goal valuing the uninterpretable features carried by the phase head is closer - in terms of the c-command path - to the head selected by the phase head than to the phase head itself. As a result, it is ‘more computationally efficient’ (p. 4) for uFs to percolate onto the head selected by the phase than to remain on the phase head. However it is far from clear whether adding a feature transmission operation to shorten the c-command path of a probe by one node only is more computationally efficient than the opposite: moreover, Chomsky (1995) explicitly rejected node-counting views of economy.

Nonetheless, the intuition that economy and computational efficiency lie behind percolation seems to us to be right. In our version of the Percolation Condition in (8vi), we suppose that (in the relevant class of participial structures in French), case/agreement features on a phase head only percolate down onto the head they select when they fail to locate a visible goal within their immediate domain. Thus, percolation is a form of crash-avoidance strategy designed to avoid unnecessary operations (e.g. avoiding percolation where this is unnecessary, as economy considerations would lead us to expect), while at the same time serving as a means for rescuing derivations which would otherwise fail (as considerations of computational efficiency would lead us to expect).

6. Conclusions

In this article, we started by outlining the rationale behind Chomsky and Richards’ claim that the uninterpretable features of a phase head must obligatorily percolate onto the head it selects. After providing the reader with a summary of the assumptions we make concerning the mechanisms of agreement and case assignment in French transitive participial clauses, we showed in sections 3 and 4 how our assumptions correctly specify that agreement can take place between a past participle and an extracted direct object or ECM subject, but not between a past 10 However, the issue of whether gender features in French are indeed uninterpretable (and get deleted in the course of the derivation) is a vexed one - as can be illustrated by a question-answer dialogue such as the following: (i) Cette maison a été vendue? - Non, elle est toujours en vente This house has been sold? - No, it (= she) is still for sale If the gender feature of the DP cette maison is deleted in the course of the derivation, how are we to know that it is the antecedent of the feminine pronoun elle?


189

participle subject and the extracted subject of an infinitival CP complement. A key assumption in our analysis was that case/agreement features remain on v when they locate a goal within the immediate domain of v. The claim that agreement features sometimes remain on a phase head was supported in section 5 by the observation that in languages like West Flemish, finite complementisers like those in (1) must also retain their agreement features. In addition to providing empirical evidence for the claim that uninterpretable features may remain on the phase head, we pointed to the existence of inherently valued uninterpretable features which falsify the argumentation put forward by Richards. If the notions of value and interpretability are not indissociable, the need for uninterpretable features to be deleted at the point where they are transferred no longer holds, as a result of which uninterpretable features are no longer required to obligatorily percolate onto the head selected by the phase head. We proposed instead that percolation is motivated by considerations of economy and computational efficiency.

7. References

Adger, D. (2003) Core Syntax: A Minimalist Approach, Oxford University Press, Oxford. Boeckx, C. (2003) Islands and Chains: Resumption as stranding, Benjamins, Amsterdam. Boeckx, C. (2007) Understanding Minimalist Syntax, Blackwell, Oxford. Carstens, V. (2001) ‘Multiple agreement and case deletion: Against φ-(in)completeness’,

Syntax 4: 147-163. Carstens, V. (2003) ‘Rethinking complementiser agreement: agree with a case-checked goal’,

LinguisticInquiry 34: 393-412. Chomsky, N. (1995) The Minimalist Program, MIT Press, Cambridge Mass. Chomsky, N. (2000) Minimalist Inquiries: The Framework, in R. Martin, D. Michaels and J

Uriagereka (eds) Step by Step: Essays on Minimalism in Honor of Howard Lasnik, MIT Press, Cambridge Mass., pp. 89-155.

Chomsky, N. (2001) Derivation by Phase, in M. Kenstowicz (ed) Ken Hale: A Life in

Language, MIT Press, Cambridge Mass., pp.1-52. Chomsky, N. (2005b) ‘On Phases’, unpublished paper, MIT (to appear in r. Freidin, C. P.

Otero and M.-L. Zubizaretta (eds) Foundational Issues in Linguistic Theory, MIT Press, Cambridge Mass.).

Déprez, V. (1998). ‘Semantics effects of agreement: The case of French past participle agreement’, Probus 10: 1-65.

Franks, S. & Lavine, J.E.(2006) ‘Case and word order in Lithuanian’, Journal of Linguistics 42: 239-288.

Haegeman, L. (1992) Theory and Description in Generative Syntax: A Case Study of West

Flemish, Cambridge University Press, Cambridge. Haegeman, L. (2000) ‘Inversion, non-adjacent inversion and adjuncts in CP’, in Transactions

of the Philological Society 98: 121-160.


190

Henderson, B. (2006) ‘Multiple agreement and inversion in Bantu’, Syntax 9: 275-289. Iatridou, S. (1993) ‘On nominative case assignment and a few related things’, MIT Working

Papers in Linguistics 19: 175-196. Kayne, R.S. (1984) Connectedness and Binary Branching, Dordrecht, Foris. Kimball, & Aissen, J (1971) ‘I think, you think, he think’, Linguistic Inquiry 2: 241-246. Kornfilt, J. (2004) ‘Unmasking covert complementiser agreement’, unpublished paper

presented at LSA conference, Boston, Mass. January 2004. Mensching, G. (2000) Infinitive Constructions with Specified Subjects: A Syntactic Analysis

of the Romance Languages, Oxford University Press, Oxford. Miyagawa, S. (2005) ‘On the EPP’, Proceedings of the EPP/Phase Workshop, MIT Working

Papers in Linguistics. Obenauer, H.G. (1994) Aspects de la syntaxe A-barre, Thèse de doctorat d’état, Université de

Paris VIII. Ouali, H. (2007) ‘On C-to-T Φ-Feature Transfer: the Nature of Agreement and Anti-

Agreement in Berber’, to appear in Agreement Restrictions, R. D’Alessandro, G. H. Hrafnbjargarson and S. Fischer (eds), Mouton de Gruyter, Berlin.

Radford, A. (1997) Syntactic Theory and the Structure of English, Cambridge University Press, London.

Radford, A. (in press) Analysing English Sentences: A Minimalist Analysis, Cambridge University Press, London.

Radford, A. & Vincent, M. (2007) ‘On transitive past participle agreement in French’, in Proceedings of the 33° Incontro di Grammatica Generativa, http://amsacta. cib.unibo.it/-archive/00002397/01/PROCEEDINGS_IGG33.pdf.

Richards, M.D. (2007) ‘On Feature Inheritance: An argument from the Phase Impenetrability Condition’, Linguistic Inquiry 38: 563-572.

Richards, N. (1997) What Moves Where When in Which Language? PhD diss., MIT. Rizzi, L. (1990) Relativised Minimality, MIT Press, Cambridge MA. Vincent, M. (2007) Aspects of French Participle Agreement: a Minimalist Approach, PhD

diss., University of Essex.


33 – 2008

191

THE CP OF CLEFTS

Adriana Belletti

1. Introduction

I have argued in recent work (Belletti 2008) that the CP of cleft sentences has some peculiar properties, the crucial ones being the following two:

i. it is a reduced CP;

ii. the reduced CP may or may not be endowed with an EPP feature.

In the pages that follow I would like to review the main arguments for i.-ii. in light of the different kinds of focalization that the two CP structures lead to in clefts. The main emphasis of the discussion here will be on the particular shape of the articulated CP projection in clefts. Throughout I will limit my attention to structures where the clefted constituent is a DP (or a PP), either a subject or a direct /indirect object.

2. The CP small clause

Consider the basic shape of the split CP projection. According to current analyses (Rizzi (1997), Benincà & Poletto (2004), Haegeman (2006), Bocci (2004) Grewendorf (2005), Mioto (2003), and related work), the Fin head is found at the bottom of the projection selecting the inflectional functional system of the following clause, while the Force head sits at the top of the projection expressing the illocutionary content of the clause, e.g. whether it is a declarative or an interrogative; the Force head is selected by the matrix verb when CP is embedded. Given this familiar background of assumptions, the question of the status of the CP of clefts naturally arises. Take the following two cleft sentences in (1) from Italian. I will use Italian throughout to illustrate different properties, unless other languages, in particular French, are needed to draw relevant distinctions.

(1) a (subject cleft) È Gianni che ha parlato it is Gianni that has spoken b (object cleft) È Gianni che i ragazzi hanno salutato it is Gianni that the boys habe greeted

Adriana Belletti

192

Two main questions should be asked: i. where in the clause structure is the clefted constituent located? ii. Is the shape of the CP the same in the two cases?

Assume a vP periphery along the lines I have argued for in previous work (Belletti 2004, 2005); assume the classical hypothesis according to which the copula – be as a shortcut – takes as its complement a small clause (a long standing hypothesis, dating back at least to Burzio (1986), Stowell (1983), and thoroughly developed in Moro (1997), Rothstein (2000)). It can be proposed that, in the case of clefts, the small clause of the copula is a CP, as schematized in (2) (Belletti (2008) and references cited there):

(2) ………. Be [ CP ………..]

The dots above (left of) be contain a vP periphery, where a new information focus head is present whose specifier is ready to host a new information constituent (see the references quoted for details). We have now to make explicit what the dots in CP correspond to. It seems correct to assume that they differ in part in subject vs non subject clefts.

2.1 The small CP of subject clefts

As clearly evidenced by different languages, e.g. French, the postcopular subject DP of subject clefts can be the focus of new information. Typically, a subject cleft (with an often deleted/unpronounced predicate) can provide the answer to a question on the identification of the subject of the clause:

(3) a Qui (est-ce que qui) a parlé? who spoke b C’est Jean (qui a parlé) it is Jean (who spoke)

As I have discussed in detail in the references quoted, this characteristic answering strategy of French share a crucial property with the inversion strategy characteristically adopted in similar contexts in languages allowing for post verbal new information subjects, such as, e.g., Italian:

(4) a Chi ha parlato? who spoke b Ha parlato Gianni has spoken Gianni

In a cartographic perspective, in both cases the subject fills the same position: the specifier of the low vP peripheral new information focus position. It is in this position that it is interpreted as the constituent carrying the required new

The CP of clefts

193

information. According to this analysis, the concealed/disguised inversion of subject clefts like (3)b is attributed the analysis in (5), details omitted1:

(5) [TPCe … [ FocP [ ……. [vP être [ Jean [ CP qui a parlé] ]]]]]]

The subject of the small clause complement of the copula raises to the vP peripheral focus where it is interpreted; movement of the copula to a high functional head is also indicated in (5). The CP predicate of the small clause is/can be left unpronounced/deleted. If the hypothesis in (2) is adopted, the whole small clause complement of the copula is a CP in turn.2 This idea can be naturally expressed in a split conception of the CP: there can be room within CP for both the subject of the small clause and the CP predicate. The subject of the CP small clause is the DP about which the CP introduced by the relative complementizer predicates some property. If we take the idea that a small clause is any constituent where a predication relation obtains – close in spirit to Stowell’s (1983) subject across categories original proposal – if we equate the presence of a predication relation with the formal property “having an EPP feature”, we can formally characterize a small clause as any categorical projection endowed with an EPP feature. The small clause of the copula in clefts is thus a CP with an EPP feature. Let us refer to it as a small CP. The DP about which the following CP predicates some property, generally referred to as the subject of the small clause, is the constituent which then moves to be associated with new information focus. Thus, we can make (5) more precise, by attributing the label CP to the whole small clause, as in (5’), for the same French sentence: 1 See Belletti (2008). The vP periphery of the copula may also contain Top-type positions (within the dots in (5)) as is the case with the vP periphery of lexical verbs. However, a peculiarity of the copula, in particular in clefts, seems to be that it necessarily involves focalization. This could be expressed by the idea that the vP periphery of the copula is reduced and it solely contains the focus head. I will not develop this idea in detail here, but just note that one could go further and assume that the copula itself is a realization of the focus head in clefts. This idea would at the same time account for the necessary focalization involved in clefts and provide a natural characterization of the frequently observed fact across languages according to which the copula tends to grammaticalize into a focus particle. See Haraiwa & Ishiara (2002) who attribute this observation to Chris Collins and Frascarelli and Puglielli (2005) for discussing the relation between the focus particle of Somali and the copula in similar terms. 2 On a first proposal that CPs can be small clauses and its generalization that all small clauses may be CPs, see Starke (1995). See below for reference to Guasti’s work on pseudorelatives in closely related terms.

Adriana Belletti

194

(5’) [TPCe … ……[ FocP [ ……. [vP être [CP Jean [ CP qui a parlé] ]]]]]]

It is time now to make precise what the two CP labels in (5’) correspond to in a split-articulated conception of the CP projection. I would like to propose that the low CP corresponds to the projection of the Fin head, while the high CP corresponds to some head lower than Force. Thus crucially, in this proposal the CP of a subject cleft is a reduced CP which does not contain the highest part of a CP projection, the projection of Force. The proposal is schematized in (6), with reference to the same French example; the highest head projection, lower than Force, is left unlabeled in (6), and it is again indicated with the neutral label CP:

(6)

[TPCe … …[ FocP [ . [vPêtre[CP…[EPPJean [FinPqui [a parlé]]]]]]]]

In (6) the subject of the small CP fills the specifier of the head carrying the EPP

feature, which, by assumption, is active within the small CP of subject clefts. In terms of the A/A’ distinction, the EPP position of the small clause is an A type position, much as the subject position of regular TP clauses where a predication relation is established with the verbal predicate. The same analysis can be attributed to the small CP complement of perception verbs in pseudorelatives, of the type illustrated in (7)a, b in Italian, thus essentially updating the proposal originally due to Guasti (1993), keeping the main insight unchanged:

(7) a Ho visto Maria che parlava con Gianni I have seen Maria that spoke to Giani b Ho visto [CP….[EPPMaria [ FinPche [(pro) parlava (-) con Gianni]]]]

In (7), “Maria” is either directly merged in the EPP position of the small CP and a small pro related to it is present in the subject position of the following clause, or it is moved to the EPP position from the position where it is merged in the clause. In the latter derivation, extraction should take place from the vP-internal postverbal position indicated as “-“ in (7)b, and an expletive pro should sit in the preverbal high subject position (Rizzi & Shlonsky 2006, Cardinaletti 2004)). The former derivation is the one which most directly represents an update of the one assumed in Guasti

The CP of clefts

195

(1993).3 This issue aside, the small CP of (7)b has exactly the same shape as the one of a subject cleft in (6).

Pseudorelatives of the type in (7) differ from a subject cleft in one respect: while the cleft requires focalization of the small clause subject, focalization in the pseudorelative can either affect the subject or the entire small clause. This is witnessed by the possibility of using the same sentence (7)a as an answer to the following two questions:

(8) a Chi hai visto (che parlava con Gianni)? whom have you seen (that spoke to Gianni) b Che cosa hai visto? what have you seen

As discussed in Guasti (1993) and Rizzi (2000), in both (8)a and b there is direct perception of “Maria”, but in (8)b it is the whole small CP which the question focuses on. Clefts, on the other hand, imply a peculiar semantics which provides a unique identification explicitly expressed by the focussed argument. In subject clefts, the (uniquely) identified argument is precisely the subject.4

Given the A status of the EPP position of the small CP complement of the copula that we have assumed, a prediction is directly derived: this position can be filled by the subject of the following clause, but it cannot be filled by a DP corresponding to the direct or indirect object of the following clause. This is so for locality reasons: Relativized Minimalty (RM) would be violated in moving an object to the EPP position of the small CP, crossing over the intervening subject. The relevant part of the derivation is illustrated in the following schema in (9):

3 A third derivation is possible, where “Maria” may be moved to the EPP position within CP from a “big DP” inside the clause, stranding a silent (referential, Belletti (2005)) small pro in the relevant EPP position within the clause. I leave a resolution of the various technical implementations open, which should optimally be decided on both empirical and theoretical independent grounds. 4 Exchanges like the following, discussed in Rialland, Doetjes & Rebuschi (2002) are possible in French: i. Q. Qu’est-ce qui se passe? what happens A. C’est le petit qui est tombé dans l’escalier (qui se passe) it is the kid who has fallen on the stairs (that happens) In cases like this, as suggested by the possibly unpronounced intended predicate in i.A, the uniquely identified argument is present in the cleft answer to the general question of information, modified by a (restrictive) relative clause.

Adriana Belletti

196

(9) *

…. […[FocP …[vP be [CP EPP [FinP che [TP S … O/PP]]]]]…

Hence, a direct consequence of the proposed analysis of subject clefts is that only the subject of the (TP) clause can reach the EPP position – or be directly merged there – in the small CP, for principled reasons.5 Indeed, this is precisely what happens in pseudorelatives. Only the subject of the clause can be the head of a pseudorelative. This is a well known fact accounted for in similar locality terms in Guasti (1993). The ungrammaticality of (10)a,b, minimally contrasting with the wellformedness of (7)a, repeated in (10)c, illustrates the relevant contrast:

(10) a *Ho visto Maria che Gianni/i ragazzi salutava/salutavano I have seen Maria that Gianni /the boys greeted b *Ho visto con Maria che Gianni parlava I have seen with Maria that Gianni spoke c Ho visto Maria che parlava con Gianni I have seen Maria that spoke with Gianni

In contrast, in non subject clefts, to which we turn in the following section, the unique identification implied by the semantics of clefts may also be brought about by a focussed non subject argument. The kind of focalization, however, is not the same in subject and non-subject clefts as we argue in 2.2. And this is the key of the contrast with pseudorelatives.

2.2 The reduced/truncated CP of non subject clefts

Suppose that the CP complement of the copula has the same shape as in (9), but that no active EPP feature is present. In this case the CP is not a small CP, in the technical sense defined above. Nevertheless, it is a similarly reduced CP, where the Force head is lacking. We may see this CP as a truncated CP, in Rizzi’s (2005) sense. We can propose that the CP complement of the copula in non subject clefts is precisely a reduced/truncated CP of this sort. If there is no EPP to be satisfied, this has the consequence that there should be no restriction for non-subject arguments to

5 This in turn has the consequence that only the subject can then reach the new information focus position in the vP periphery of the copula. See the discussion in 2.2. Note that for a direct/indirect object the option of being directly merged in the EPP position with a related pro sitting in the argument position of the clause , is not an available option (due to the licensing constraints on object pro). Hence, the intervention problem necessarily arises in this case. Furthermore, the movement of a PP to the EPP position would be different from a PP pre-posing operation , which is an A’ type operation.

The CP of clefts

197

move into the reduced/truncated CP crossing over the intervening subject argument in TP; the movement implemented in this case would be an A’ type movement, hence no intervention effect should be produced in this derivation.

To the extent that clefts involve a form of focalization, the natural proposal can then be made that non subject arguments move to the focus position of the reduced/ truncated CP. Indeed, this kind of left peripheral focalization within the CP complement of the copula is the only focalization admitted for non subject arguments. Specifically, an alternative direct long movement of an (direct or indirect) object from the embedded TP into the new information focus position in the vP-periphery of the matrix copula is excluded on principled locality grounds. Phase theory (Chomsky (2005)) explicitly rules out the possibility of such long direct moment with no intermediate steps (within the CP), with the embedded CP sent to spell out. But no intermediate step is possible in this case as the reduced/truncated CP complement of the copula does not contain any escape hatch edge position, given its reduced/truncated nature. In particular, it does not contain any position different from the criterial interpretable ones, such as e.g., the focus position, from which movement is excluded in principle, through any version of criterial freezing (Rizzi (2006)). The consequence of all this is that while the focalization of subject clefts can occur in the vP periphery of the copula and correspond to the new information focalization expressed by this position, the focalization of non subject arguments necessarily corresponds to left peripheral focalization.

As discussed in detail in various works (Belletti (2004, 2008; Bocci (2004), Rizzi (1997), a.o.), left peripheral focalization involves more than just new information. It typically is contrastive/corrective focalization. If this is the case, then, one direct consequence of the proposal is that although clefts constitute a form of focalization in general, the focalization of subjects can be new information focalization, while the focalization of non subject arguments is contrastive/ corrective focalization6. I have proposed in the quoted references that a direct reflex of this different way of focalization may be found in the fact that although a cleft (with an often deleted/unpronounced predicate) can be used as a felicitous answer to a question of information on the identification of the subject, the same possibility is not available for a question of information on the identification of the object. This is visible in those languages where clefts are used as a most suitable answering strategy, as in the case of French. The contrast in (11) in French can be taken as an illustration of this important distinction: 6 Of course, a subject cleft can also instantiate contrastive/corrective focalization, implementing the same movement in the reduced/truncated CP left periphery as non subject arguments. Nothing excludes this possibility. “S” can thus be involved in the same derivation illustrated in (12) following, for non subject clefts the only available derivational option.

Adriana Belletti

198

(11) Q Qu’est-ce-que t’as acheté (/Qu’as-tu acheté)? what have you bought A *C’est un livre it is a book

Q Avec qui es-tu sorti? with whom did you get out A *C’est avec Jean it is with Jean

On the other hand, it may be speculated that this clear distinction should not hold in languages where both new information focus and contrastive focus are realized in the left periphery of the clause. Should languages of this type exist, all other things being equal, in these languages both subject and non subject clefts should qualify as possible answering strategies to questions of information. I leave the development of this parametrical option open for further study.7

The proposed derivation of non subject clefts is schematized in (12):

(12) ok

….. be [CP Force … [FocP … [ FinP che [TP S …O/PP]]]]..

For the sake of explicitness, in (12) the reduced status of the CP complement of the copula is illustrated in terms of the truncation idea.

Summarizing, in both subject and non subject clefts the copula be takes a reduced/ truncated CP as complement. The reduced/truncated CP may or may not contain an active EPP feature. When it does, it is a small clause CP (a small CP, as we have called it) where a predication type relation holds between the subject of the small clause and the (rest of the) CP predicate, the same kind of relation instantiated in the pseudorelative complement of perception verbs. For principled locality reasons, only the subject of the CP predicate can check the EPP feature. One consequence of this is that only the subject can realize the new information focus in the vP periphery of the matrix copula. If, in contrast, the reduced/truncated CP complement of the copula does not contain any EPP feature, then the focalization implemented by the cleft is left peripheral focalization within the reduced CP complement of the copula.

7 Hungarian and Sicilian (Belletti (2008)) may be two languages worth looking at in this perspective. Furthermore, other languages may also more or less parasitically exploit this UG option. One first thing to determine, however, is whether a cleft is an answering strategy normally adopted in the relevant language. I am not in a position to provide structured data in this domain for the moment, so I leave open to future investigation this intriguing comparative issue.

The CP of clefts

199

Such focalization is thus contrastive/corrective focalization and not simple new information focalization and, crucially, may also affect non-subject arguments.

3. The position of “che” and the nature of the CP complement of “be”

Given the general analysis presented in the preceding paragraphs, I would now like to look more closely at the shape of the reduced CP complement of the copula, concentrating more specifically on the position and nature of the complementizer present in clefts, che in Italian in the examples we will consider8.

According to the proposed analysis che is not the realization of the Force head in clefts; rather, it is the realization of finiteness, the lowest Fin head in the articulated CP. This amounts to claiming that the (reduced) clausal complement of the copula is not a full fledged declarative sentence in any case.

Clear distributional evidence in favour of this distinction comes from the contrast in (13). In (13)a left peripheral focalization is implemented within the declarative CP complement of the verb “dire” (say); as the ungrammaticality of (13)b shows, the mandatory respective order of the complementizer and the focalized argument, here a direct object, is C – Foc and cannot be Foc – C. This is expected given the shape of the articulated CP, the very nature of the Force head, and the respective order of the Force and Focus heads, with Force the highest head of the articulated CP.

(13) a Ho detto che GIANNI avrebbero assunto (non Maria) I have said that GIANNI they would have hired (not Maria) b *Ho detto GIANNI che avrebbero assunto (non Maria) I have said GIANNI that they would have hired (not Maria)

Similarly, whenever a perception verb like “vedere”(see) is used in its epistemic reading (and not in its perception reading) the order, as expected, is once again C – Foc and not Foc – C, as illustrated in (14).

(14) a Ho visto che GIANNI avrebbero assunto (non Maria) I have seen that GIANNI they would have hired (not Maria) b *Ho visto GIANNI che avrebbero assunto (non Maria) I have seen GIANNI that they would hae hired (not Maria)

8 A regular restrictive relative introduced by a clear relative pronoun does not seem to be possible in Italian, as witnessed by the strong marginality of sentences like i.a which contrast with i.b: i. a*? È (a) Gianni a cui parlerò di questo problema it is (to) Gianni to whom I will speak of this problem b È a Gianni che parlerò di questo problema it is to Gianni that I will speak of this problem

Adriana Belletti

200

In the CP complement of the copula in clefts, which in the proposal we have developed is reduced/truncated under Force, the order is rather Foc – C, as (15) reminds.

(15) a È GIANNI che assumeranno (non Maria) it is GIANNI that they will hire (not Maria) b È con GIANNI che parleranno del problema (non con Maria) it is with GIANNI that they will speak of the problem (not with Maria) c È GIANNI che ha parlato (non Maria) it is GIANNI that spoke (not Maria)

The respective order of Foc and C in (15) is directly obtained if che is not here the realization of Force, but rather the realization of Fin, as assumed. (15)c is an instance of left peripheral focalization of the subject in the reduced CP complement of the copula, an option available for all kinds of arguments, direct and PP complements included as in (15)a, b (see footnote 6, and the discussion in 2.2).

If che is not the expression of the declarative Force of the clause in clefts, this comes close to claiming that clefts like those in (15), which instantiate left peripheral focalization, are not that different from root left peripheral focalization in sentences like (16).

(16) a GIANNI assumeranno (non Maria) GIANNI they will hire (not Maria) b Con GIANNI parleranno del problema (non con Maria) with GIANNI they will speak of the problem (not with Maria) c GIANNI ha parlato (non Maria) GIANNI spoke (not Maria)

Under cartographic assumptions, in (16) the contrastively/correctively focalized phrase fills the specifier of the high focus position within the articulated CP; this is exactly the same position occupied by the embedded focalized argument in the clefts of (15).

However, even though no declarative force is expressed by the complementizer in clefts like (15), this is not equivalent to saying that focalizing by means of a cleft as in (15) amounts to exactly the same kind of focalization as root left peripheral focalization of the kind in (16). As noted above, a cleft is not just a way of focalizing a phrase. Even if this may be (one of) the most salient property of clefts in general, other semantic/discourse values are implied by use of a cleft. In particular, a cleft also implies a unique identification of the focussed element (Kiss (1988); Abels & Muriungi (2005) for more recent discussion). Furthermore, there is in clefts what we may call a presupposition of existence, likely to be induced by the very presence

The CP of clefts

201

of the copula.9 A similar presupposition is not necessarily implied in (root) left peripheral focalization. The following contrast in (17), brought to my attention by Paola Benincà, suggests exactly this kind of distinction between the two focalization procedures. The contrast identifies one context where use of one of the two structures is not just infelicitous, it is plainly ungrammatical.

(17) a NESSUNO ho incontrato (non tutti) nobody I met (not everybody) b.*(Non) È NESSNO che ho incontrato (non tutti) it is (not) nobody that I met (not everybody)

In (17) the indefinite negative quantifier “nessuno” (nobody), corresponding to the direct object of the clause, is (contrastively) focalized in the left periphery. The clear ungrammaticality of (17)b in contrast with the possibility of (17)a indicates that such focalization can be done by means of a plain left peripheral operation, but not through a cleft.10 We interpret this as due to the special semantic-discourse value implied by use of a cleft, which, as a first approximation, we have identified in a presupposition of existence of the uniquely identified argument, linked to the very presence of the copula. Thus, although the kind of focalization can be the same in the two structures, involving the left peripheral focus position in both cases, the two structures are not equivalent in their overall discourse value, with the consequence that, in pairs like (17), one is just plainly ungrammatical.

It thus seems that there are good reasons to believe that the complementizer present in clefts is not the one found in embedded declarative clauses expressing the declarative illocutionary force of the sentence. Since, however, it is the same word, at least in Italian (che) and several other languages, it would be most welcome if this coincidence could be expressed in some form. Let us then hypothesize in this connection that a complementizer like che always originates in Fin. Let us further assume that in a full fledged selected subordinate CP it raises up to the Force head to check the interpretive illocutionary declarative force of the clause. It would be through a mechanism of this sort that che at the same time expresses the finite nature of the clause and its declarative force. Given these assumptions, should the Force head not be present, che could solely express finiteness. Our proposal has been that

9 An interesting comparative question that I am not in a position to properly address here, concerns the status of clefts (or rather their equivalents) in languages without the copula. This is left open to future investigation. 10 The left peripheral focalization of the indefinite quantifier can also be located in an embedded CP, with the decarative complementizer preceding the focalized phrase, as always. This is illustrated in i. a Ho detto che NESSUNO assumeranno (non tutti) I have said that NOBODY they will hire (not everybody)

Adriana Belletti

202

this is precisely the case of clefts, where che remains in Fin and there expresses the finite nature of the embedded clause.

It is known that there are languages where more than one complementizer is/can be expressed in complementation.11 If the derivational mechanism just described is on the right track and if it has a general application, it would provide a direct reason why this possibility should arise: given the view that movement is copying (Chomsky (1995) and subsequent work), the two instances of the complementizer could just be seen as two spell-outs of different copies. The copies would be located in the distinct Force and Fin heads, with the complementizer originally externally merged in Fin, and subsequently internally re-merged in Force. 12 While it is generally just the highest copy the one which is sent to spell out, it is in principle conceivable that, under defined conditions whose identification is beyond the aims of the present work, both copies be phonetically realized. This may be a non trivial general consequence of the idea that che is primarily the realization of Fin. A property that cleft sentences appear to overtly realize in a reduced/truncated CP, where che remains in Fin.

References

Abels, K. ad Muriungi, P. (2005) ‘The Focus particle in Kîîtharaka’, ZAS Papers in

Linguistics, 42: 219-239. Belletti, A. (2008) “Answering strategies: New information subjects and the nature of clefts”,

chapter 10 of Structures and Strategies New York: Routledge, in press. Belletti, A. (2004) ‘Aspects of the low IP area’, in L. Rizzi (ed.) The structure of CP and IP.

The cartography of Syntactic Structures, vol. 2, Oxford: Oxford University Press, 16-51.

Belletti, A. (2005) ‘Extended Doubling and the vP periphery’, Probus, 17(1): 1-35. Benincà, P. and Poletto, C. (2004) ‘Topic, Focus and V2: Defining the CP Sublayers’, in

Rizzi (ed.) The structure of CP and IP. The Cartography of Syntactic Structures, vol. 2, New York: Oxford University Press, 52-75.

11 See Paoli (2007) for recent discussion of different Romance varieties manifesting this possibility, which is interpreted along lines related to ours, with the complementizers exploiting the two heads Force and Fin. Our proposal capitalizes on the nature of movement as copying as is discussed momentarily in the text. 12 As pointed out in Belletti (2008), the proposal also opens up the possibility that some language may have two different complementizers realizing Force and Fin respectively, or else that some language realizes the complementizer of clefts in a way different from the declarative complementizer. A way of characterizing these (hypothetical) languages could be that the complementizer in Fin would have the property of not moving to Force, where the relevant feature would be expressed by a complementizer directly merged in Force.

The CP of clefts

203

Bocci, G. (2004) ‘Contrastive focalization on topics and preverbal subjects in Italian’, Rivista

di Grammatica Generativa, 29, 3-59. Burzio, L. (1986) Italian Syntax. A Government-Binding Approach, Dordrecht: Reidel. Cardinaletti, A. (2004) ‘Towards a Cartography of Subject Positions’, in L.Rizzi (ed.) The

structure of IP and CP. The Cartography of Syntactic Structures, vol. 2, New York: Oxford University Press, 115-165.

Chomsky, N. (1995) The Minimalist Program, Cambridge Mass: MIT Press. Chomsky, N. (2005) ‘On Phases’, in R. Freidin, C. Otero and M.L. Zubizarreta (eds)

Fundational Issues in Linguistic Theory, Cambridge, Mass: MIT Press. Frascarelli, M. & A.R. Puglielli (2005), “The Focus System in Cushitic Languages” , in P.

Fronzaroli e P. Marrassini (eds.), Proceedings of the 10th Hamito-Semitic Congress. Afroasiatic Linguistics.

Grewendorf, G. (2005) “The discourse configurationality of scrambling”, in J. Sabel and M. Saito (eds) The Free Word Order Phenomena, Berlin: Mouton de Gruyter, 75-136.

Guasti, M.T. (1993) Causative and Perception Verbs. A Comparative Study, Torino: Rosenberg and Sellier.

Haegeman, L. (2004) ‘Topicalization, CLLD and the Left Periphery’, in B. Shaer, W. Frey and C. Maienborn (eds.) Proceedings of the Dislocated Elements Workshop ZAS Berlin, November 2003. 35:12 Berlin: ZAS.

Hiraiwa, K. & S Ishihara (2002) “Missing Links: Cleft, Sluicing, and “No da” Construction in Japanese”, MIT Working Papers in Linguistics.

Kiss, K.E. (1998) ‘Identificational Focus versus Information Focus’, Language, 74, 2: 245- 273.

Mioto, C. (2003) ‘Focalização e quantificação’, Revista Letras, 61, Curtiba, 169-189. Moro, A. (1997) The Raising of Predicates, Cambridge: Cambridge University Press. Paoli, S. (2007) “The fine structure of the left periphery: COMPs and subjects. Evidence from

Romance”, Lingua 117: 1057-1079. Rialland, A., J. Doetjes & G. Rebuschi (2002) “ What is focussed in C’est XP qui/que Cleft

sentences in French”, Speech Prosody 2002, ISCA Archive. Rizzi, L. (1997) ‘The Fine Structure of the Left Periphery’, in L. Haegeman (ed.) Elements of

Grammar, Dordrecht: Kluwer, 281-337. Rizzi, L. (2000) ‘Direct perception, government and thematic sharing’, in L. Rizzi

Comparative Syntax and Language Acquisition, London/New York:Routledge, 189-210.

Rizzi, L. (2005) ‘On the grammatical basis of language development: a case study’, in G. Cinque and Kayne R. (eds) The Oxford Handbook of Comparative Syntax. New York: Oxford University Press.

Rizzi, L. (2006) ‘On the Form of Chains: Criterial Positions and ECP Effects’ in S.Cheng and N.Corver (eds), WH-Movement Moving on, MIT Press, 97-134.

Rizzi, L. and Shlonsky, U. (2007) ‘Strategies of Subject Extraction’, in U. Sauerland and H.M. Gärtner (eds) Interfaces + Recursion = Language?, Mouton De Gruyter, 115-160.

Adriana Belletti

204

Rothstein, S. (2000) Predicates and Their Subjects, Kluwer. Starke, M. (1995) ‘On the Format of Small Clauses’, in A. Cardinaletti and M.T. Guasti (eds.)

Small Clauses, Syntax and Semantics, vol.28, New York: Academic Press 237-269. Stowell, T. (1983) ‘Subjects across categories’, The Linguistic Review, 2.3, 285-312.

I dattiloscritti devono essere inviati in duplice copia a uno dei componenti la direzione, ai seguenti indirizzi:

Guglielmo CINQUE Dipartimento di Scienze del Linguaggio

Università di Venezia

Dorsoduro 1075 - 30100 VENEZIA

Luigi RIZZI CISCL - Scienze della Comunicazione

Università di Siena

Via Roma 56

53100 Siena

Prezzi di vendita: Area Euro Privati: euro 20,00 Istituzioni: euro 30,00

Modalità di pagamento: 1. Anticipato, senza spese di spedizione, a mezzo versamento su

conto corrente postale nr. 10644359 intestato a Unipress, via Ce-sare Battisti 231 - 35121 Padova

2. Contrassegno, con addebito di spese di spedizione, inoltrando l'or-dine a Unipress, via Cesare Battisti 231 - 35121 Padova (valido so-lo per l’Italia).

3. Anticipato, senza spese di spedizione, accludendo all'ordine asse-gno bancario o circolare a favore di Unipress (valido solo per l’Italia).

4. Accettiamo VISA e MASTERCARD – comunicate numero carta, nome del titolare, codice di sicurezza e data di scadenza

Subscription price (non-Euro area):

Foreign Countries Individuals: $ 40.00 [no postage]

Institutions: $ 50.00 + 3.00 [postage] Payment Methods 1. Check enclosed

2. International Postal Money Order 3. Post Office Current Account (International Giro Ac-

count nr 10644359 Unipress, Padova) 4. We accept VISA and MASTERCARD – please

send to Unipress: number of Card, authorized sig-nature, expiry date, security code.

All private orders must be prepaid and sent directly to the Publisher and are for per-sonal use only

Rivista di Grammatica Generativalear.unive.it/jspui/bitstream/11707/420/1/2008compl.pdf · Rivista...

Documents

Transcript of Rivista di Grammatica Generativalear.unive.it/jspui/bitstream/11707/420/1/2008compl.pdf · Rivista...