Discourse and Grammar: From Sentence Types to Lexical Categories 9781614511601, 9781614512158

Table of contents :
Introduction
Part I. Semantic and Pragmatic Properties of Sentence Types
Implicatures in Discourse
Permission and Choice
Promises and Threats with Conditionals and Disjunctions
Part II. Sentence Types and Clausal Peripheries
Revisiting the CP of Clefts
Delimitation Effects and the Cartography of the Left Periphery
Sentence Types and the Japanese Right Periphery
Part III. Clausal Properties of Lexical Categories
On NPs and Clauses
Discerning Default Datives: Some Properties of the Dative Case in German
How Phase-Based Interpretations Dictate the Typology of Nominalizations
Scope and Verb Meanings

Citation preview

Discourse and Grammar

Studies in Generative Grammar 112

Editors

Henk van Riemsdijk Harry van der Hulst Jan Koster

De Gruyter Mouton

Discourse and Grammar From Sentence Types to Lexical Categories

Edited by

Günther Grewendorf Thomas Ede Zimmermann

De Gruyter Mouton

The series Studies in Generative Grammar was formerly published by Foris Publications Holland.

ISBN 978-1-61451-215-8 e-ISBN 978-1-61451-160-1 ISSN 0167-4331 Library of Congress Cataloging-in-Publication Data A CIP catalog record for this book has been applied for at the Library of Congress. Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available in the Internet at http://dnb.dnb.de. ” 2012 Walter de Gruyter, Inc., Boston/Berlin Typesetting: PTP-Berlin Protago-TeX-Production GmbH, Berlin Printing: Hubert & Co. GmbH & Co. KG, Göttingen 앝 Printed on acid-free paper 앪 Printed in Germany www.degruyter.com

Contents Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . G¨unther Grewendorf and Thomas Ede Zimmermann

1

Part I. Semantic and Pragmatic Properties of Sentence Types Implicatures in Discourse . . . . . . . . . . . . . . . . . . . . . . . . . Nicholas Asher

11

Permission and Choice . . . . . . . . . . . . . . . . . . . . . . . . . . Paul Portner

43

Promises and Threats with Conditionals and Disjunctions . . . . . . . . Robert van Rooij and Michael Franke

69

Part II. Sentence Types and Clausal Peripheries Revisiting the CP of Clefts . . . . . . . . . . . . . . . . . . . . . . . . Adriana Belletti

91

Delimitation Effects and the Cartography of the Left Periphery . . . . . 115 Luigi Rizzi Sentence Types and the Japanese Right Periphery . . . . . . . . . . . . 147 Mamoru Saito

Part III. Clausal Properties of Lexical Categories On NPs and Clauses . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 ˇ Zeljko Boˇskovi´c Discerning Default Datives: Some Properties of the Dative Case in German . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247 Henk van Riemsdijk How Phase-Based Interpretations Dictate the Typology of Nominalizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289 Leah Bauke and Tom Roeper Scope and Verb Meanings . . . . . . . . . . . . . . . . . . . . . . . . . 321 Edwin Williams

Introduction G¨unther Grewendorf and Thomas Ede Zimmermann

Bringing together papers from various subfields of theoretical linguistics, this volume gives a representative glimpse of current research on form and function in grammar. Its overarching topic is as old as it is hot: the relation between the major clause types as determined in syntax, and their canonical or idiosyncratic roles in discourse as characterized in pragmatic terms. Though none of the papers addresses this topic in its full breadth, they can all be seen to make their specific contributions to it, scrutinizing the pertinent aspects of the grammatical interfaces and elaborating detailed case studies. The way from pragmatics to syntax leads via the semantic component of grammar. The first part of this collection comprises three papers devoted to the semantics/pragmatics interface. To a large part and with Gricean pragmatics as a role model, contemporary pragmatics centers around the reconstruction of rational communicative behavior as limited by the bounds of grammar, i.e. primarily semantics. Despite this general perspective, Asher shows that the dependence of scalar implicatures on discourse structure cannot be captured by Gricean global rationality alone – correction sequences being a simple case in point: (1)

a. b.

A: John either did some of the reading or he did some of the homework. B: John did all of the reading, but you’re right, he didn’t do all of the homework.

Due to the illocutionary role of B’s utterance (1b), one of the scalar implicatures of (1a) gets cancelled, while the other one is preserved. This suggests that implicature calculation may depend on discourse structure – a hypothesis that Asher pursues and generalizes, ultimately developing a unified account of implicatures as defeasible implications that also captures more intricate cases of embedded implicatures that seemingly support ‘localist’ approaches of implicature calculation. Portner’s contribution provides an analysis of two related properties of imperatives: (i) their variation in discourse function and (ii) their licensing of free

2

G¨unther Grewendorf and Thomas Ede Zimmermann

choice inferences. With regard to (i), it is argued that imperatives are semantically uniform, and that their wide range of interpretations is explained by two factors: differences in the grounds for issuing a given imperative and the logical relationship between the imperative and other commitments of the addressee. Concerning (ii), the same ideas which are used to analyze permission, in combination with an “alternatives” semantics for disjunction and indefinites, are able to explain free choice and related phenomena, such as Ross’s Paradox and the licensing of any. Van Rooij and Franke address a puzzle concerning performative uses of declarative sentences. As the following pair illustrates, threats may be expressed by conditionals or corresponding disjunctions, in accordance with logically motivated paraphrase: (2)

a. b.

You will give me your wallet or I will punish you severely. If you don’t give me your wallet, I will punish you severely.

However, unlike its conditional counterpart (3a), a disjunctive statement like (3b) does not lend itself to use as a promise: (3)

a. b.

If you give me your wallet, I will reward you splendidly. You will not give me your wallet or I will reward you splendidly.

Using techniques from game theory to model the difference between promises and threats, the authors provide a purely pragmatic explanation of the observed distribution of use conditions. Retaining the logical equivalence between disjunctions and conditionals, the analysis exploits the difference in wording and, in particular, an asymmetry created by mentioning and thus bringing to the fore, different possibilities. The second part of this volume deals with the question of how the constitution of sentence types can be related to properties of functional categories in the clausal periphery. Based on a cartographic approach to sentence structure the contributions of this part of the collection show that in left-headed languages, the functional heads of the left clausal periphery have interpretive effects on the interfaces that determine the semantic properties of scope-discourse configurations as well as the prosodic properties associated with information structure. It is also shown that in a right-headed language like Japanese, the configuration of the functional heads that are relevant to the determination of sentence types form a right-peripheral mirror image of the left clausal periphery. Rizzi’s contribution illustrates some of the results of cartographic studies on the left clausal periphery and discusses the implications of the cartographic approach for the study of the interfaces that connect syntax with the systems

Introduction

3

of sound and meaning. In particular, Rizzi shows that the functional heads assumed by the criterial approach not only have a syntactic function in attracting elements to the periphery of the clause but also have interpretive effects on the interfaces with meaning and sound: they signal the interpretive properties of scope-discourse configurations and give instructions to the prosodic component to yield the appropriate contour associated with the various discourse functions. He then compares the cartographic approach, which assumes that complex syntactic configurations are derived by a simple computational system, with an approach that proceeds from a simpler functional lexicon, assuming only a single C-layer and admitting multiple adjunctions to TP. Rizzi argues that the cartographic approach is conceptually and empirically superior to the “simpler approach” in that the latter shifts crucial non-interpretive properties such as positional and co-occurrence restrictions (including parametric variation) to the interfaces, and is unable to deal with phenomena such as co-occurring multiple C-particles or the distributional properties of topic and focus particles. By contrast, the cartographic approach simplifies the burden of the interface systems by “syntacticizing” scope-discourse functions without enriching the computational system with specific mechanisms operative at the interfaces. Finally, Rizzi shows that the criterial positions associated with interpretive properties such as Force, Focus, Aboutness etc. have the effect of delimiting movement chains (by ‘Criterial Freezing’) and thus offer new ways of approaching the classical problem of locality. Saito investigates the complementizer system of Japanese and claims that this system is identical to the complementizer system of Spanish. He shows that Japanese has three phonetically distinct complementizers the functions of which correspond to the three (phonetically non-distinct) Spanish complementizers posited by Plann (1982): a complementizer (to) that introduces ‘paraphrases’ or ‘reports’ of direct discourse, a complementizer (ka) that selects questions, and a complementizer (no) for embedded tensed propositions, which represent events, states, or actions. Based on the observation that these complementizers, when co-occurring, always appear in the order no-ka-to, Saito argues that their hierarchical organisation correlates with the functional structure that Rizzi (1997) suggested for the Italian left clausal periphery: no represents the head of FinP, ka is located in the head position of ForceP, and to is the head of a ‘ReportP’ that is located above ForceP and is not represented in Rizzi’s system. He then shows that in Japanese, Topic heads can be (recursively) generated above Fin and below Force, which leads him to the conclusion that the peripheral clausal structure of Japanese can be represented as a right-peripheral mirror image of the Italian left periphery:

4 (4)

G¨unther Grewendorf and Thomas Ede Zimmermann

[. . . [. . . [. . . [. . . [TP . . . ] Fin (no)] (Topic*)] Force (ka)] Report (to)]

Based on a cartographic framework, according to which different syntactic positions in the functional structure of the clause overtly express different interpretations, Belletti shows that cleft constructions display interpretive variations that correlate with variation in their functional structure. Her crucial assumption is that for all clefts, the sentential complement selected by the copula represents a reduced CP that lacks the highest ForceP layer. She then points out interpretive differences between subject and non-subject clefts. While the former can be utilized as an answer to a question of information as well as an expression of contrastive focalization, non-subject clefts can only have the latter option. Belletti therefore assumes that a clefted subject that is interpreted as the (noncontrastive) focus of new information moves into the low vP-peripheral focus position of the matrix copula, passing a left-peripheral EPP-position in the reduced CP complement. By contrast, clefted constituents that are contrastively focused (subjects as well as non-subjects) move to the left-peripheral focus position in the reduced CP complement of the copula. The fact that only subject clefts can undergo movement to the internal focus position of the matrix clause is shown to be a consequence of independent locality principles: in the presence of the EPP in the left periphery of subject clefts, object movement is excluded due to relativized minimality; in non-subject clefts, where there is no left-peripheral EPP, long movement of non-subjects can be excluded by the phase impenetrability condition. Since wh-extraction from the reduced CP cannot utilize the embedded left-peripheral focus position due to criterial freezing, wh-clefts are analyzed as extraposition of the embedded FinP combined with remnant movement of the reduced CP’s entire FocP. Finally, Belletti shows that other structures such as the sentential complement of perception verbs share crucial properties with the configuration of clefts. The contributions in Part III of the volume deal with the interaction of lexical elements and clausal functional categories, each of them revealing unexpected parallels between clause structure and the internal structure in other, particularly lexical categories, including the role of their peripheries. Boˇskovi´c shows that there is a surprising interplay between the internal syntax and semantics of the traditional Noun Phrase and clause-level phenomena. He argues that there is a fundamental structural difference between languages with articles such as English, and article-less languages such as Serbo-Croatian. On the basis of several new generalizations, crucially related to the role of articles, Boˇskovi´c demonstrates that this difference can be captured by the assumption that article-less languages lack the category DP, which yields a typological distinction between DP languages and NP languages. The new generalizations that

Introduction

5

he adduces in support of this distinction involve phenomena such as negative concord with complex negative constituents, quantifier scope, radical pro-drop, number morphology, and the interpretation of possessives. He then explores consequences of the internal structure of the traditional nominal projection for the internal structure of clauses, proceeding from the assumption that the structural difference on the nominal level between languages with and without articles has its counterpart on the clausal level. His crucial claim is that just like the nominal structure is poorer in an NP language than it is in a DP language, the structure of clauses is poorer in the former than it is in the latter in the sense that the lack of DP in the nominal structure implies the lack of TP in the clausal structure. That the clausal structure of article-less languages indeed lacks TP is shown to receive independent support from crosslinguistic generalizations involving phenomena such as subject expletives, asymmetries in the locality of subject and object movement, and the Sequence of Tense phenomenon. The fundamental conclusion of this paper is that many clause-level phenomena cannot be properly understood unless close attention is paid to the internal structure and interpretation of nominal categories. Van Riemsdijk presents a novel analysis of the apparent alternation between datives and accusatives in German spatial prepositional phrases, which is known to correlate with a locative vs. directional interpretation: (5)

a.

Tisch. Peter legt das Buch auf den Peter puts the book on theACC table

b.

Tisch. Das Buch liegt auf dem the book is-lying on theDAT table

Building on semantic investigations by Joost Zwarts, van Riemsdijk decomposes directionality into source, route and goal, and goes on to argue for a domaindependent default case, which in oblique domains like PP and AP is the dative. On this analysis, the accusative in directional PPs specifying goals or paths, come out as depending on a (possibly implicit) measure phrase, whereas the dative manifests itself in the ‘elsewhere’ environments of locative or source PPs. Bauke and Roeper’s paper analyzes nominal gerunds in English and argues for a distinction between two types of nominal gerunds. Both types are shown to project verbal functional structure, associated with an aspectual projection inside the nominal compound. The distinction between the two types is argued to follow from the position of the gerundive -ing affix. In one type of gerund this affix is generated under the aspectual head, where it licenses a number of modifiers (such as aspectual modifiers, non-sentential adverbial modifiers, prepositional modifiers etc.). Further, Bauke and Roeper show that identifying the -ing affix as

6

G¨unther Grewendorf and Thomas Ede Zimmermann

an aspectual marker allows for a natural explanation of why plural morphology is absent from this type of nominal gerund. In the other type of gerund, the affix is generated under a nominal node instead. This is argued to predict that plural morphology is licensed on this type of gerund. Bauke and Roeper then illustrate that this does not imply that aspectual structure is not projected in this type of nominal gerund. On the basis of insights from Phase Theory it is shown that access to the aspectual projection inside the gerund is simply blocked by higher Phase-nodes. The authors outline further that both types of nominal gerunds allow for incorporation and that the distinction between incorporated and non-incorporated forms reduces to a symmetry-distinction according to which a symmetric c-command relation between the head of the gerund and its bare internal argument forces incorporation. In an additional step they derive further support for the distinction between two types of nominal gerunds from a comparison of the English constructions with their German counterparts, where the two types of -ing affixes are realized as -ung and -en respectively. Williams observes and investigates a surprising difference between verbs and prepositions that shows in their interaction with certain quantifying elements like only: (6)

a. b.

I only paid 5 dollars for it. ⇔ I paid only 5 dollars for it. / I bought it only 5 dollars for it. I only bought it for 5 dollars. ⇔

Generalizing these observations, Williams makes out a general pattern behind the data in (6) due to an asymmetry in the scopal behavior among lexical categories: whereas verbs adhere to (7) and thus generally “commute” with quantifiers, no such generalization holds for prepositions: (7)

No verb can have scope over its DP complement.

Williams then goes on to show how the asymmetry between verbs and prepositions is naturally accounted for within the Representation-theoretic architecture of Williams (2003): like clauses, prepositional phrases may be subject to Level Embedding. The papers collected in this volume were selected from the contributions to the conference 10 Years After, which took place in Frankfurt in June 2009, organized by the members of the Graduiertenkolleg Satzarten (Research Training Group on Sentence Types), a temporary doctoral program supporting students in theoretical and descriptive linguistics and specializing on the particular problems surrounding the distinction between the major clause types. The conference, which marked the end of the program, brought together a number of international experts who have advised and supported our students over a period of

Introduction

7

roughly 10 years (thence the title). In the name of our current and former students and colleagues, we would like to take the opportunity to thank our advisors for continuous support and encouragement, as well as the Deutsche Forschungsgemeinschaft (German Research Foundation) for their generous funding. Finally, we would like to thank Benjamin H¨ubner, Jonas Metzler, Jacob Schmidkunz, Christian Stidronski and Dina Voloshina for helping us with the preparation of the manuscript. It was our decision not to include any indices in this volume. Frankfurt, November 2012 GG & TEZ

Part I. Semantic and Pragmatic Properties of Sentence Types

Implicatures in Discourse Nicholas Asher

1.

Introduction

Implicatures are an interesting case study for the role of discourse in interpretation and in grammar generally. Scalar implicatures are the result of a defeasible inference. (1)

a. b.

John had some of the cookies. John had some of the cookies. In fact he had them all.

(1a) has the implicature that John didn’t have all the cookies, this implicature can be defeated by additional information, as in (1b). Nevertheless, scalar implicatures seem closely tied to lexical choice or structural factors, which has led some authors, most notably Chierchia (2004) but also Fox (2007) and others to incorporate the generation of implicatures within the syntax-semantics interface – i.e., clearly at the core of grammar. However, scalar implicatures share the characteristic of defeasiblility with inferences that result in the presence of discourse relations that link discourse segments together into a discourse structure for a coherent text or dialogue – call these implicatures discourse or D-implicatures. I have studied these inferences about discourse structure, their effects on content and how they are computed in the theory known Segmented Discourse Representation Theory or SDRT. This paper argues for the centrality of discourse structure in linguistic interpretation by detailing how discourse structure provides an important source of information to computing scalar implicatures. Scalar implicatures have typically received a Gricean treatment based on reasoning about the content of sentences in isolation since the work of Larry Horn (1972). I show here how the Gricean tradition has missed an important, indeed decisive, component in the calculation of implicatures. This has general implications for the way discourse structure is treated within grammar. I argue that discourse structure affects scalar implicature and that much the same procedures are operative in both. So if implicatures are part of the grammar discourse structure is too. By integrating discourse and implicature together, we get some nice consequences: at the theoretical level, we

12

Nicholas Asher

have a unified and relatively simple framework for computing all implicatures; second, we have a clear account of the interaction of D-implicatures and scalar implicatures in many cases; finally, we can capture the intuitions of so called “localist” views about scalar implicatures, while making this compatible with a broadly Neo Gricean framework; finally, since in my view discourse structure triggers scalar implicatures, this goes some way towards explaining the variability of embedded implicatures noted recently (e.g., Geurts and Pouscolous 2009). 2.

D-implicatures

Let’s examine some examples of D-implicatures. (2)

a. b. c. d.

John walked in. He poured himself a cup of coffee. John fell. Mary pushed him. We bought the apartment, *but we’ve rented it. Il commence a` dessiner et peindre en 1943 , * fr´equente les ateliers de sculpture * puis de peinture de l’ e´ cole des Beaux-Arts d’ Oran, * o`u il rencontre Guermaz (ANNODIS corpus).

A presumption of relevance leads us to infer some link between elementary discourse units or EDUs (clauses or subclausal units whose boundaries are either sentence initial or marked by * in the examples above). These links introduce relations taking these discourse units as arguments that are familiar even to the non-linguist: some units elaborate or go into more detail concerning something introduced in another constituent (these are Elaboration type relations) as in (2d); some units form a parallel or a contrast with other units (such units are linked by Parallel or Contrast), as in (2c); some units furnish explanations why something described in another unit happened (Explanation) as in (2b); and some units constitute a narrative sequence of events (Narration) as in (2a) or (2d). Other discourse relations of interest for our purposes are indirect question answer pairs (IQAP), which link responses to a prior question, Correction, where a second discourse move revises the content of a first, and Alternation, which is linked to certain uses of disjunction. Some D-implicatures are encoded grammatically through the use of certain grammatical constructions (like adverbial or purposive clauses, parentheticals or left fronted temporal or spatial adverbials)1 or through discourse connectors like as a result, because, but, . . . or the choice and sequencing of lexical items. 1. For a discussion of these, see for instance, Vieu et al. (2005)

Implicatures in Discourse

13

An example of a set of discourse relations triggered by the choice of verb and complement comes in (2d), with the use of beginning with, followed by and ending with. Sometimes, it is less clear what linguistic source triggers the inference of the discourse relation as in (2a)–(2b) – most likely, an as yet not fully understood mix of lexical semantics and world knowledge. While scalar implicatures can occur embedded under quantifiers and other operators, many people have noted that these implicatures are less robust in many contexts than the unembedded cases of scalar implicatures like (1a). Dimplicatures, on the other hand, robustly embed under quantifiers and other operators (as well as other discourse relations). (3)

a. b. c. d.

If it was late, John took off his shoes and went to bed. If it was late, John went to bed and took off his shoes. If John drank and drove, he put his passengers in danger. The CEO of Widgets & Co. doubts that the company will make a profit this year and that (as a result) there will be much in the way of dividends for shareholders this year.

In both (3a) and (3b), the D-implicature that there is a narrative sequence between the two clauses in the consequent of the conditional survives under embedding, and (3c) shows that this holds in the antecedent of a conditional as well. (3d) shows that the causal relation of result holds when embedded under a downward entailing attitude verb. Discourse relations like explanation occur between clauses under modals and other elements of discourse in a sort of quantifying in way as in (4a) and (4d). Note that the paraphrase of (4a) is (4c) not (4b). (4)

a.

John broke his leg. Sam told me (I think, It’s likely) he slipped on the ice. b. #John broke his arm because [Sam told me (I think, it’s likely) he slipped on the ice.] c. John broke his arm. Sam told me (I think, it’s likely) it’s because he slipped on the ice.

Thus, D-implicatures take scope underneath and across various syntactically realized operators.

14

Nicholas Asher

3.

Discourse Structure and D-implicatures

To investigate D-implicatures further, we need a theory about the structure, the construction and the interpretation of discourse structures with an emphasis on the computation of D-implicatures. A theory of discourse structure must answer to three tasks: – It must segment a text into EDUs. – It must provide attachment points for EDUs in a discourse structure. – It must provide one or more discourse relations between an EDU and its attachment point(s). SDRT (Asher 1993; Asher and Lascarides 2003) is a theory that does all of these things but also provides an account of how to calculate D-implicatures.An SDRT discourse structure or SDRS is a logical form for discourse with a well-defined, model-theoretic semantics that has many equivalent representations – as a sui generis, quasi first order model like structure consisting of a set of labels and assignments of formulas to labels (Asher and Lascarides 2003), as a DRS like structure (Asher 1993) or as a λ term in intensional logic (Asher and Pogodalla 2010). An SDRS is the counterpart to a formula of intensional logic in Montague Grammar for discourse interpretation. To get an idea of what SDRSs look like consider the following text (5) discussed at length in Asher and Lascarides (2003). The model-like SDRS is given in (5) . Note that some EDUs, e.g., π2 and π5 , are grouped together into complex discourse units, e.g., π6 . (5)

π1 π2 π3 π4 π5

(5)

A, F , Last, where:

John had a great evening last night. He had a great meal. He ate salmon. He devoured lots of cheese. He then won a dancing competition.

A = {π0 , π1 , π2 , π3 , π4 , π5 , π6 , π7 } F (π0 ) = Elaboration(π1 , π6 ) F (π6 ) = Narration(π2 , π5 ) ∧ Elaboration(π2 , π7 ) F (π7 ) = Narration(π3 , π4 ) Last = π5 In SDRT we can abstract away from the details of the structure to get a graph representation, which is relevant to computing discourse accessibility for

Implicatures in Discourse

15

anaphoric antecedents and sites for presupposition binding and accommodation (again for details see Asher and Lascarides (2003)). π1 Elaboration π6 Narration

π2

π5

Elaboration π7 π3

Narration

π4

Notice that some discourse relations are represented as vertical arrows in the graph whereas others are horizontal arrows; these correspond to two different types of relations – subordinating and coordinating relations, and these two types of relations affect anaphoric and attachment possibilities differently.2 The non-arrow lines show which discourse units are parts of more complex discourse units. 3.1. Inferring D-implicatures Inferring D-implicatures is a matter of defeasible and uncertain inference. Many of the features used to infer discourse relations are only good indications of a particular discourse relation or particular discourse structure; very few are in and of themselves sufficient to deductively infer the relation or structure. Many discourse connectives are for example ambiguous. In addition, many segments may bear discourse relations to other segments despite the lack of discourse connectives or known structural or lexical cues, as in (2a), (2b) or (5). To solve this problem, my colleagues and I developed a non-monotonic logic, a logic for defeasible inference, tailored to inferring D-implicatures. The task of building such a logic is not completely trivial. Integrating nonmonotonicity in discourse interpretation is problematic, especially if this integration occurs at the level of contents or what is said. Reasoning over contents non-monotonically requires finding a class P of preferred models, those with the intended discourse relations, and computing validity or logical consequence 2. For a discussion, see for example Asher 2008; Asher and Lascarides 2003.

16

Nicholas Asher

with respect to P. Given that the language of information content is at least that of first order logic, where the complexity of the computation of validity and logical consequence is only recursively enumerable and that almost all nonmonotonic logics require some sort of consistency test over the formulas one is using for the inference, the complexity of computing logical consequence with respect to the class P of preferred models is not recursively enumerable – i.e., computationally hopeless. This is not just a matter of implementation but one of principle. We cannot assume that agents, with their limited computational capacities, are able to solve a problem reliably which we can show mathematically to be incapable of having anything like what we would call an algorithm. Attributing such computational capacities to agents shows that we have mischaracterized the problem: they are not computing logical consequence over formulas of information content; either the language in which the computation is done is somehow a simplification of the language of information content, or they are computing something other than logical consequence, perhaps using some sort of heuristic. One could develop a heuristic for computing D-implicatures. In fact, Fox (2007) as well as others have proceeded to do this for scalar and free choice implicatures. The problem with taking this tack, however, lies in its verification. In order to make sure that the heuristic is doing what it is supposed to, we must check it in the way that computer programs are checked, via program verification. In most if not all instances, this means translating the problem into a logic and then checking that the result desired is in fact a logical consequence of the program’s translation and the input data. This leads us back to the task of building a non-monotonic logic for D-implicatures. SDRT’s solution to this problem is to look at non-monotonic reasoning not over contents but over logical forms. Roughly, instead of trying to compute the non-monotonic consequences of a bunch of facts about the world, facts which may be quantificationally complex and which leads to an unsolvable problem, we compute the non-monotonic consequences of a discourse logical form’s having a certain shape and of a segment’s having the lexical choices and structure that it does. This means that we are trying to solve a logical consequence problem, not in the language of information content, but in a language for describing discourse logical forms and features of discourse constituents. Asher and Lascarides (2003) develop such a language, which they call the glue language. The non-monotonic logic adapted to this language is known as the glue logic or GL. Asher and Lascarides (2003) show that the problem of logical consequence for formulas of this language is in fact decidable. This approach has an additional advantage. Computing implicatures, as we’ll see, is sensitive to the structure of discourse and of individual sentences. This

Implicatures in Discourse

17

is available at logical form, but not at the level of semantic content, which is described in terms of sets of possible worlds. This brings the calculation of implicatures into an area closer to the syntax-semantics interface, as we’ll see. GL uses axioms exploiting various resources to get the intended discourse relations to hold between discourse constituents. The general form of such axioms is this: – General Form: (?(α , β , λ ) ∧ some stuff) > R(α , β , λ ) > is a weak conditional; “some stuff ” is information about α , β and λ that’s transferred into the glue language from more expressive languages for other information sources like: compositional semantics, lexical semantics, pragmatic maxims of conversation, generalizations about agent behavior in conversation, and domain knowledge. The semantics of > was developed by Asher and Morreau (1991) in a first order non-monotonic logic known as common sense entailment. This is a logic for non-monotonic or defeasible reasoning based on a weak conditional >. Originally devised to treat generics, I have used a version of it restricted to a quantifier free description language, the glue language, to calculate D-implicatures, and it is a relatively adaptable non-monotonic logic. It has two parts: a basic, monotonic, conditional logic with a standard proof theory  and consequence relation |=, and then a defeasible inference relation |∼ and a non-monotonic consequence relation |≈ that make use of the basic logic. I will use the glue logic version of common sense entailment to model both Dand S-implicatures. Let me briefly recapitulate the basics of common sense entailment restricted to a propositional language. – A modal generic frame F = W , ∗ where W is a non empty set of worlds and ∗ : W × P(W ) → P(W ) is a selection function. A model A is constructed by adding a valuation function. – A, w |= A > B iff ∗(w, ||A||) ⊆ ||B|| – the standard clauses for the quantifiers and connectives, though the glue language itself has quantifier free logical structure. This semantics for > will be familiar to those accustomed to conditional logics, and has a complete axiomatization.3 How do we pass from a notion of monotonic consequence to a non-monotonic one? The idea of the non-monotonic consequence relation is to assume that matters are as normal as “possible” given the information in the premises and 3. The axiomatization for the first order language can be found in Asher and Morreau (1991).

18

Nicholas Asher

to then see what follows. Assuming matters are as normal as possible means making > as much like the material conditional as “possible” – moving from if p then normally q to if p then q. There is both a “proof theoretic method” for defining nonmonotonic consequence ( |∼) and a model theoretic method ( |≈ ) for which a correspondence theorem is given in Asher (1995). The proof theoretic method is less involved, and I briefly sketch it here. First, we define a A → extension of : –  ∪ {A → φ :   A > φ }, if consistent. – , if not. For each antecedent A of a > conditional derivable from , define a A → extension of  inductively relative to an ordering ρ over antecedents of > statements derivable from the previous stage in the sequence with  →ρ ,0 = . Every such sequence has a fix point. We now have the definition: Definition 1.  |∼ φ iff for all orderings ρ the fixpoint ∗ of each  →ρ extension sequence is such that ∗  φ . 4.

A Gricean Account of Scalar Implicature

With this sketch of commonsense entailment, let us now turn to Grice’s picture about implicatures. Grice’s view of implicatures is that they are calculated after compositional semantics has finished its job via his famous general maxims of conversation, quality, quantity and relevance. In principle the Gricean picture tells the beginning of an attractive story for computing scalar implicatures (Horn 1972, Schulz and van Rooij 2004, Schulz 2007, Spector 2006). Let’s take a look at how the story is supposed to go for the following example (the implicature is indicated by the ). (6)

John or Susan came to the party  John and Susan didn’t both come to the party.

The “derivation” of the implicature in (6) uses Grice’s maxims of quality and quantity via the following steps. a. The speaker has said “j or s”, so she believes that j or s (quality). b. It follows from the maxim of quantity that that the speaker does not believe more than this relative to what she could have said (j, s, j and s): in other words, she only believes j or s, i.e. she does not have the belief that j, she does not have the belief that s and she does not have the belief that j and s.

Implicatures in Discourse

19

c. Nor can she believe ¬ j or ¬ s, since given a. beliefs in these would entail a belief in s and a belief in j respectively. d. An axiom of “Expertise” allows us to infer that she believes not j and s, (it applies only in cases where the doxastic implicatures in a., b. and c. are not contradicted). I’ve said that this is only the beginning of a story, since Grice’s maxims of quality and quantity are not precise enough really to license any inferences. In addition, these maxims should clearly be formalized within a non-monotonic logic. GL provides a very simple yet precise reconstruction. Since GL works with descriptions of logical forms, we have the means to write down the fact that a certain relation holds between two formulas, namely that ψ is an alternative that could have been said in the given discourse context instead of φ . Being able to express and to define this set of alternatives is crucial to the enterprise of formalizing S-implicatures. – step a: follows defeasibly in GL if we assume the defeasible principle of Sincerity (Asher and Lascarides 2003): Sayφ > B φ – To derive step b, we need a measure of informativeness. Let’s suppose it’s logical entailment, and that Alt(φ , ψ ) holds only if ψ is a strictly stronger formula (entailing but not entailed by) than φ .  ψ → φ ,  φ → ψ  Alt(φ , ψ ) → ((Sayφ ∧ ¬Sayψ ) > ¬B ψ ) Step b will follow in GL, provided Alt( j ∨ s) = {j, s, j ∧ s}. – step c: doxastic reasoning is closed under  and so this follows. – “Expertise” about an issue gives us the stronger form of the implicature: (Alt(φ , ψ ) ∧ Sayφ ∧ ¬Sayψ ) > B(φ ∧ ¬ψ ) To get the weaker implicature from (6) that the speaker doesn’t believe that s, j or that j ∧ s, we simply turn all the relevant > statements that instantiate the schemas in steps and the appropriate instance of the schemas in steps a and b above into → statements. Without Expertise, we have only one fixed point, in which the desired conclusions hold. The stronger implicatures necessitate a slightly more complex axiom to maintain consistency of the sincerity implicature with the expertise implicatures. Expertise, together with the other axioms and the assumed facts, yields 2 different fixed points, one in which B¬j, one in which B¬s. Since non-monotonic consequence is defined as those entailments from all of the fixed points, the desired strong implicature follows: (7)

B¬(j ∧ s) ∧ B(j ∨ s)

20

Nicholas Asher

The derivation of the implicatures in GL is nonmonotonic and thus the desired implicatures follow defeasibly from the premises and the principles as I have formalized them.4 However, the GL derivation of the Gricean implicature is valid only if the alternativeness relation Alt is restricted as above. If the set of alternative contains all of the strictly stronger logical entailments, then a great many distinct fixed points arise and GL predicts no implicature can be drawn. This is a formalization of the so called symmetry problem (Kroch 1972; Jasinskaya 2002; Block 2009). – (s ∨ j) ∧ ¬(j ∧ s)  s ∨ j but it’s not the case that s ∨ j  (s ∨ j) ∧ ¬(j ∧ s). – So using the principle in step b, we get ¬B((s ∨ j) ∧ ¬(j ∧ s)). – Since by hypothesis B(s ∨ j), propositional logic and the K axiom for B yield: ¬B(¬(j ∧ s)). – Expertise cannot fire without yielding a doxastic inconsistency. Putting this derivation together with the standard one yields: ¬B(j ∧ s) ∧ ¬B(¬(j ∧ s)) – By similar reasoning, we could conclude: ¬Bj ∧ ¬B(¬j) in one fixed point ¬Bs ∧ ¬B(¬s) in another fixed point – The conclusion is that there is no particular scalar implicature. The success of the Gricean programme depends on a hidden assumption about what other more informative things could have been said in the particular discourse context. As Block notes, the Gricean programme itself says nothing about relevant sets of alternatives. We would need an additional element of the pragmatics to do this. Or it might be specified lexically and be part of the grammar (Fox and Katzir 2011). For example, we might say that alternatives are defined lexically for scalar items, for adjectives beautiful, stupendous, gorgeous,. . . , for nouns genius, idiot, . . . , for connectives (∨, ∧) (here it depends as to whether >, →, etc. are also considered connectives), and for quantifiers (no, some, many, most, all). Such lexically specified alternatives seem reasonable but we need to know how alternatives compose together. One might also try to fix the alternatives for an utterance via a question under discussion approach, but this has many difficulties of its own, not the least of which is determining what the question under discussion is.

4. In particular Sauerland’s (2004) restriction is not needed. This is desirable, since this allows for a straightforward cancellation of the implicatures if the explicit semantic content contains for example an explicit denial of one of the implicatures.

Implicatures in Discourse

21

Even though the problem of the specification of alternatives remains unresolved, we can still comment on the role of GL in the grammar. GL reasons over logical forms and so has access to the structure of the predication and of lexical choice. But more generally, reasoning over logical forms as GL does is preferable to approaches that reason directly over preferred models of information content (Van Rooij and Schulz 2004, 2006), because it leaves the semantics of the alternatives intact. Van Rooij and Schulz use a set of alternatives to induce a partial order on worlds, Information update with φ will pick those φ worlds that are minimal with respect to the ordering. Intuitively, these worlds are just those that make φ true but not any more informative response to the question under discussion.5 The problem with the model minimization technique is its side effects. Minimization tells us that only one person came to the party given this question, and that seems unwarranted. The mechanism of minimization is too powerful. Here are some other untoward consequences. (8)

Assume a problem set with 10 problems on it. How many problems did John do on the problem set? John did some of the problem set  ¬B (John did more than 1 problem of the problem set).

(9)

a. b.

Who did John kiss at the party? (Alternatively: Did John kiss all of the girls at the party?) John kissed some of the girls at the party  John kissed two girls at the party.

The minimization technique and ordering given above predicts implicatures that don’t seem born out in practice. One option is to restrict the ordering of minimal elements in the ordering to some relevant set of alternatives, but now we have injected some syntactic or at least extra-semantic restrictions into the procedure. GL’s reasoning over logical forms and explicit sets of alternatives brings the set of alternatives to the fore front. The semantic minimization in

5. More technically, we define the ordering as follows. Let φ be some sentence and P a partition. – ∀w, w ∀p ∈ P (w, w |= φ → ((w ∈ p → w ∈ p) ↔ w ≤ w)) The minimization of (6) generates a natural ordering on worlds. The worlds that are minimal w.r.t the ordering are those where just John comes to the party and no one else or where just Susan and no one else comes to the party.

22

Nicholas Asher

effect is computationally very complex and nets us nothing in addition to the simpler approach based on reasoning about logical forms.6 5.

Localist Accounts of Scalar Implicatures

The Gricean account of scalar implicatures is a rationalist one with special cognitive assumptions about cooperativity that provide a derivation of scalar implicatures. But we’ve seen that such an account cannot derive S-implicatures without the help of some sort of extra assumptions about alternatives to what was said. It seems reasonable that the set of alternatives is at least partially conventionally determined. But at what level are these alternatives computed? Griceans claim that the alternatives are computed pragmatically after truth conditional semantics has finished its job. Given a standard view of what truth conditions are, this means that Griceans do not have access to the “fine structure” of a sentence’s meaning; the particular way the truth conditions has been erased in the semantic value. A contrasting, “localist” theory claims implicatures are, like presuppositions, conventionally determined by the lexicon and computed during compositional semantic interpretation. In contrast to Griceans, localists have access to the fine structure of meaning in computing implicatures, in particular in computing the set of alternatives upon which the derivation of implicatures depends. This makes them closer to the technical apparatus of the previous section. The localist computation of implicatures goes beyond a particular take on the generation of alternatives, however. To quote Chierchia (2004): The claim is that there are situations in which (standard) implicatures are by default present and situations in which they are by default absent, and such situations are determined by structural factors.

That is, implicatures result from structural properties of the grammar, not from any pragmatic inferences based on Gricean maxims. This leads to a recalibration of the vision of pragmatics within the grammar, putting it much closer to the core. But as we will see, the notion of structure that is required for generating implicatures goes far beyond that of sentential syntax. 6. There is another technical problem with minimization. These updates need to be defined relative to very particular models that contain the “relevant worlds”. They must not bring any information extraneous to the question at hand – all the unmentioned background facts stay as they were. In the model minimization framework, without constraints on the sentences for which the models are provided, this constraint cannot be satisfied in general, and the minimization problem is unsolvable.

Implicatures in Discourse

23

The principal motivation for the localist approach is the presence of embedded implicatures, which can present problems for Neo-Gricean approaches. (10)

John did the reading or some of the homework.

This implicates John didn’t do the reading and some of the homework. It also implicates that he didn’t do all of the homework. But it doesn’t implicate that ¬ (John did the reading or all of the homework). Since Griceans compute implicatures only on whole utterances or full sentences, it’s not clear how to get the second implicature. While Chierchia’s example has impressed some linguists as decisive (for example Danny Fox), the difficulty for Griceans with (10) depends once again entirely on the set of alternatives chosen. If one of the given alternatives to (10) is in fact John did the reading or all of the homework, we have trouble using the Gricean strategy, but if we rule out somehow this alternative and have instead the set of alternatives consisting just of (11)

John did the reading and some of the homework.

(12)

John did all of the homework.

we would end up with the right predictions. A Gricean could adopt instead a localist’s computation of alternatives, while nevertheless maintaining a broadly pragmatic approach to the derivation of implicatures. Inspired by Chierchia’s localist meaning clauses, I define in the Appendix sets of alternatives using the recursive structure of the logical form and lexically stipulated alternatives. We can then use Common Sense Entailment to formalize the broadly Gricean reasoning to derive implicatures of the sort that Chierchia claims to hold. Moreover, it is an account that is compatible that Griceans can accept, as we’ve seen that the computation of the relevant set of alternatives for implicatures is not something that is forthcoming from Gricean principles alone but is rather extraneous to it. The main problem, however, is the Chierchia inspired calculation yields us implicatures that don’t fit the facts. Consider first the exhaustivity implicature based on disjunction. The epistemic reading of the disjunction is often prominent. (13)

John did the homework or the reading.

For many interpreters (13) just conveys that the speaker doesn’t know which of these two alternatives is correct, but in fact both could be. The disjointness

24

Nicholas Asher

implicature predicted by Chierchia and the account in the Appendix seems to be triggered only in certain discourse contexts. The computation of alternatives in the Appendix also undergenerates with respect to observed implicatures. For instance, it predicts that one should not get the standard implicatures (the exhaustivity implicature for or, for instance) inside the scope of downward entailing operators. But this is wrong for the antecedents of conditionals, which are downward entailing. Consider (16), and its paraphrase (17).7 (16)

If you take cheese or dessert, you pay $ 20 ; but if you take both there is a surcharge.

(17)

If you take only a cheese dish or only a dessert, the menu is 20 euros; but if you take both there is a surcharge.

or (18)

If John owns two cars, then the third one outside his house must be his girlfriend’s.

(19)

If one person reads my book, I’ll be happy.

(20)

If you want more food, you can order either the biryani or the stuffed naan.

(21)

? If you want more food, you can only order, either the biryani or the stuffed naan.

Further, the standard implicatures in the consequent of a conditional (which is upward entailing) are much less strong than predicted given the system of alternatives in the Appendix.

7. The Chierchia inspired computation of alternatives isn’t convincing either on its own for other downward entailing operators, even for the example with doubt or not believe. Prosody here I think is essential to getting the implicature: (14)

I don’t believe that John has read MANY philosophy books.

(15)

I don’t believe that JOHN has read many philosophy books. (implicature less strong)

But this issue goes beyond the scope of this paper.

Implicatures in Discourse

(22) (23)

25

If Mary comes to the party, John or Sam will be happy. ? If Mary comes to the party, only John or only Sam will be happy.

Finally the embedded implicatures which motivate the localist approach are acknowledged to be less clear cut than originally supposed. Consider (24)

Every student read some of the books.  Every student didn’t read all of the books. (No students read all of the books)

Judgements vary concerning the predicted implicature of (24), depending on how the implicature is presented. Chemla (2009) and Geurts and Pouscoulous (2007) have experimental results to show that such embedded implicatures are less strong than the simple ones. Finally, the fragility of embedded scalar implicatures contrasts with other implicatures like the free-choice implicatures or discourse based implicatures that freely embed in non-DE contexts: (25)

You may take an apple or a pear.  ♦a ∧ ♦p

(26)

Every student may take an apple or a pear.  Every student may take an apple and every student may take a pear.

(27)

John slipped on the snow and fell. (Implicature is that the falling is the result of the slipping.)

(28)

Every student slipped on the snow and fell. (Implicature is that the falling is the result of the slipping.)

6.

Structural Problems with Localist and Gricean Accounts

As Larry Horn made clear many years ago, scalar implicatures depend, at least in part, on scales associated with lexical items. The challenge is to determine how these scales create a manageable ordered set of alternatives for sentences that contain them. But while there are, most likely, scales lexically associated with determiners like some and all and modals like can and most, experimental research is less clear that adjectives like full, bald also support alternatives in arbitrary contexts. For open class words, the actual values and perhaps even the presence or the activation of the scale for the purposes of calculating implicatures is dependent on discourse context.

26 (29)

Nicholas Asher

a. b. c. d.

A: Do you like him? B: Why, yes, I do. A: I mean, do you LIKE him like him or just like him? B: Oh, I’m not sure.

In the first question-answer pair with a yes/no question, the lexical scale associated with like isn’t really operative or needed to understand the exchange to an alternative question where the scale is explicitly invoked. The dependence of implicatures on discourse contexts surfaces in other places too. Consider, the defeasibility of implicatures and their relation to ordinary semantic entailments. What a localist grammar produces is a pair of contents, the first element of which is the “narrow” semantic content of the discourse, the content given by lexical and compositional semantics, and the second element of which is the strengthened meaning containing both the narrow content and the implicature. Ordinary semantic entailments are understood as product entailments of the pair, while implicatures are understood as entailments of the second element. According to localists, the defeasibility of the implicature is done externally to the meaning computation. Those implicatures that aren’t inconsistent with established facts in the common ground or the narrow semantic content continue to be operative as the discourse content is computed. This view of implicatures gives us the wrong results. Consider the following example of a sentence (30a) generating the embedded implicature in (30b). (30)

a. b.

John either did some of the reading or he did some of the homework.  John didn’t do all of the reading; John didn’t do all of the homework; and he didn’t do some of the homework and some of the reading.

And now consider the following dialogue. (31)

a. b.

A: John either did some of the reading or he did some of the homework. B: John did all of the reading, but you’re right, he didn’t do all of the homework.

What happens to the implicature (30b) in this discourse context? To my ears, (31a), (31b) has the implicature that John did all of the reading and some of the homework. In fact this is an implicature of (31b). But this implicature overrides the implicature of (31a) that John didn’t do some of the homework and some

Implicatures in Discourse

27

of the reading. Thus, implicatures interact in important ways with discourse moves like the corrective move given by B. We need to integrate S-implicatures into discourse content and structure and then compute the appropriate update of that content after taking into account B’s corrective move. This implies that we cannot calculate implicatures merely at a sentential level, as most Griceans and localists have assumed. We need a much more finegrained view of the discourse context to calculate implicatures properly.8 Another problem with localist and Gricean accounts of implicatures is that implicatures are not always cancellable. Sometimes they are required for discourse coherence, in which case they are not cancellable. (33)

a. b. c.

John has an even number of children. He has four. (Implicature is that he has exactly 4.) #John has an even number of children (π1 ). He has three (children) (π2 ). John has an even number of children (π1 ). He has at least three (children) (π2 ). John has an even number of children. He has four (children).

A Gricean or a localist like Chierchia should predict that (33a)–(33c) are OK, since the implicature to the stronger, “exactly” meaning of three should be blocked. However, it is not, and (33b) is infelicitous. Once again, I believe this stems from an interaction of discourse structure and implicatures: there is a particular sort of elaborative move going on in the second clauses of (33), which accounts for the freezing of the implicature. Only a framework like GL even has a hope of handling such examples. A final indication that something is amiss with current accounts of implicatures is their fragility. As Chemla (2009) notes, localist theories predict that 8. We can continue this pattern with more complex embedded examples. (32)

a. b.

A: Some of the students did some of the reading or some of the homework. B: At least one student did all of the reading, but otherwise you’re right.

It would seem that B’s correction still leaves many of the implicatures of his original statement intact; he’s still committed to the implicature that Some of the students didn’t do all of the homework and that some of the students didn’t do all of the reading and some of the reading. We need a more finegrained notion of implicature revision in the face of corrections. Contrast also (32b) with (32c): (32)

c.

At least one of those students did all of the reading.

28

Nicholas Asher

(34b) should not have the implicature below, making a stark contrast between (34a) and (34b). (34)

a. b.

John didn’t read all of the books.  John read some of the books. No student read all of the books. ?  All of the students read some of the books.

For localists, the predicted implicature of (34b) is ¬ No student read some book – or some students read some of the books, which is weaker than the implicature tested by Chemla. However, (34b) is equivalent to: (34)

c.

All the students didn’t read all of the books.

and this intuitively (and on a localist theory) implicates that all the students read some of the books. So equivalent meanings seem to yield distinct implicatures! This seems to indicate strongly that implicatures depend not only on semantic content but of something else in addition. Interestingly, D-implicatures are not closed under arbitrary first order equivalences either. Consider the logical equivalence in (35a). If D-implicatures were computed on deep semantic content and hence closed under first order equivalences, we would predict no difference between (35b) and (35c) since (35b), where the relation of Explanation linking the two clauses is inferred, is perfectly coherent in contrast to (35c), where no discourse relation is inferred: (35)

a. Some one pushed him. ↔CL Not everyone didn’t push him. b. John fell. Someone pushed him. (Explanation inferred) c. #John fell. Not everyone didn’t push him.

Few people think that D-implicatures are compositionally determined at the syntax semantics interface or off pure semantics; so this failure of substitutivity, which is equivalent to a failure of compositionality, is not much of a surprise. In GL this failure comes about because D-implicatures are dependent on information about logical form and about the global discourse context, rather than just semantic content. A simple hypothesis is that S-implicatures are also dependent on logical form and the global discourse context, in particular the structure of the discourse context, in order to explain their fragility. That is the claim I shall defend in the next section.

Implicatures in Discourse

7.

29

Interactions between D-implicatures and S-implicatures

Let’s recap. I have detailed a general mechanism for computing D implicatures that also serves to capture scalar implicatures.9 I have also given arguments that call into question the empirical adequacy both of Gricean and localist accounts of implicature. Such accounts don’t explain the fragility of S-implicatures nor their occasional uncancelability. My suggestion is that Chierchia was right that implicatures, in particular the set of alternatives from which they are generated, are structurally determined; he was wrong, however, about restricting his attention to syntactic structure or the structures properties of logical forms of sentences. The relevant structure in question is discourse structure.10 This should not come as a surprise; the behavior of other sorts of non-assertoric content, like presupposed content, also has complex interactions with discourse structure (Asher and Lascarides 1998). As over 30 years of work on presupposition has shown, it is unwise to try to compute presuppositions without examining how the surrounding discourse context might affect these presuppositions. But unlike presuppositions, a theory of S-implicatures needs not a theory of accommodation or binding, but of a theory of triggering. More specifically, we must answer the question: when does a discourse context license or induce an appropriate alternative set over which to compute (scalar) implicatures? Discourse structure triggers additions to content, in particular structural relations like Contrast, Correction, Parallel, QAP, and various species of Elaboration. These relations together with prosody, which can signal a lexical choice, induce structure preserving maps that can provide a set of alternatives and license an implicature. On the other hand, implicature inferences are sometimes required to establish discourse relations. When the latter is the case, then I predict that the implicatures are not cancellable without affecting the coherence of the discourse. To build a case for my claim, I will look at examples that localists like Chierchia, Fox and Spector (2008) have put forward for the robustness of localist implicature computations. I will show that in a representative sampling of those examples, it is the discourse structure that triggers the embedded implicature,

9. Fox (2007), Kratzer and Shimoyama (2002) and Alonso-Ovalle (2005) argue that free choice implcatures should be treated with the mechanism for scalar implicatures. I have a rather different take on free choice implicatures, but that would take us too far afield here. See Asher and Bonevac (2005). 10. See Geurts (2009, 2010) for more support on this point.

30

Nicholas Asher

and in some cases the implicatures are required to maintain the discourse relation established.11 Let’s look at a class of examples in Chierchia, Fox and Spector (2008) that involve the discourse relation of Correction. (36)

a. b. c.

Joe didn’t see Mary or Sue; he saw both. (only a clear exhaustive interpretation of the disjunction). It is not just that you can write a reply. You must. I don’t expect that some students will do well, I expect that all students will.

(36a)–(36c) all are only felicitous as corrections of assertions that are echoed under the scope of the negation. The observation is that the echoic use of correction in (36) makes the embedded implicatures happen. For instance, because we take the correction move in (36a) to correct the exhaustively interpreted assertion Joe saw Mary or Sue, we have to interpret the embedded clause exhaustively as well. And then voil`a: we have an embedded implicature. Asher (2002) (written a decade earlier) provides an analysis of discourse relations in terms of a map from the source (the constituent to be linked to the discourse structure) to a target (a discourse constituent that serves as an attachment point). This map exploits prosodic cues and the logical structure of the constituents, which can be displayed in a modified embedding or ME graph, which extends the SDRS graphs I introduced in Section 3 with sentence internal logical structure (see Asher (2002) for the details, which aren’t relevant here). Such maps can also be made to serve an account of discourse triggered scalar implicatures.12 My account of Correction involves the following constraint: – Correction(α , β ) only if Kβ entails ¬Kα and there is a map μ : τβ → τα such that there is at least some element x of τβ Kβ ( μ x(x) ) > Kα . The element x is said to be the correcting element. Let’s look at how such a constraint works. In (36a), we assume an earlier constituent Kα of the form John saw Mary or Sue that is the target of the Correction move. As is often the case with Corrections, the second clause in (36a) elaborates on the first. But the Correction move and its elaboration are coherent only if we assume that an exhaustivity implicature is added to the content of the assumed, antecedent constituent. That is, Kα is in this case John saw Mary or 11. For a discussion of more of their data, see Asher (forthcoming). 12. Cf. Schwarzschild (1998) for similar ideas about the interpretation of focus.

Implicatures in Discourse

31

Sue but not both, and the exhaustivity implicature is the target of the Correction. Without the addition of the implicature, there would be nothing for (36a) to correct. The presumption of discourse coherence, and the interpretation of (36a) as a correction move triggers, indeed requires, the presence of the embedded implicature. Notice that the implicature calculated is relative to the map μ . The set of alternatives can be calculated as in the Appendix or simply using the map μ itself. Furthermore, the inference of the S-implicature is triggered by the need to establish the coherence of the discourse move, in this case Correction. Thus, D- and S-implicatures are codependent; the need to calculate a D-implicature triggers the calculation of the S-implicature, and it is the S-implicature that supports the D-implicature. Relations of Parallel and Contrast work similarly (Asher, forthcoming). Recall the example I gave above of a non-cancellable implicature. (33)

a.

John has an even number of children. He has four. (Implicature is that he has exactly 4.) b. #John has an even number of children (π1 ). He has three (children) (π2 ). c. John has an even number of children (π1 ). He has at least three (children) (π2 ). John has an even number of children. He has four (children).

These examples all exemplify the discourse relation of Elaboration. And in fact, Elaboration makes the implicature non-cancellable. Clearly it is a better Elaboration of having an even number of children to give the exact number, than it is to simply say that he has some number greater than n. The presence of the lexical choice between three and at least three together with the presence of the Elaboration relation, and the parallel structure and prosodic prominence on three which itself determines the structure preserving map μ buttresses this conclusion. The parallel structure and the instantiation of the quantifier an even number of children to a given number is now strongly a particular kind of Elaboration, Instance(π1 , π2 ), meaning that π2 furnishes an instance or witness for a quantifier in π1 . In such an Elaboration, the instance is read as a specific number (i.e. 3), which is inconsistent. The example sounds awkward, because the surface cues for a particular kind of Elaboration, Instance, and the deep semantics of the clauses come apart. When an explicit quantifier at least 3 is used, Instance is not triggered; we have a general specification of the number of John’s children, and everything remains consistent.

32

Nicholas Asher

The same observation of non-cancellability holds in fact for all the implicatures triggered by discourse structure. Consider once again (37) or (38); it would be completely incoherent to cancel the implicature of the first clause given the discourse environment provided by the second clause: (37)

#Sam or Susan came (in fact both did), or both did.

(38)

#If most (in fact all) of the students do well, I am happy; if all of them do well, I am even happier.

In general if the discourse structure requires the S-implicature for coherence, it isn’t cancellable except on pain of incoherence. On the other hand, sometimes even S-implicatures conveyed by disjunction don’t arise when not needed to verify constraints. Txurruka and Asher (2007) argue that disjunctions can play a special discourse role. They are defeasible marks of the relation Alternation. Alternation introduces alternatives that carve up the set of discourse possibilities relative to some topic, which can be introduced via a question or via a simple assertion. When the topic, however, already contains the disjunction, the constraint that is a consequence of Alternation is not met and so the relation of Alternation does not hold. Consider, for example: (39)

a. b.

Did John meet the Vice President or the President? He must have met the Vice President or the President, since he got the job.

(40)

a. b.

There’s a lot of poop on the streets here. Everyone in the village owns a horse or a donkey.

The relation in (40) is Explanation (41a), (41b). There is no alternation in either (39b) or (40b) because the topic constraint on Alternation is not met. But on the other hand, compare: (41)

a. b.

What kind of animal do people own in this village? Everyone owns a horse or a donkey. (No one owns both.)

Given the topic established by (41a), the topic constraint on Alternation is met and so we get the embedded implicature. Question-answer pairs also trigger S-implicatures. A question induces a partition on the information state (Asher 2007; Groenendijk 2008). A complete answer picks out one cell in the partition; indirect answers (which stand in the IQAP relation to the question they address) require reasoning or additional premises to infer a complete answer. Sometimes the additional information comes from

Implicatures in Discourse

33

an implicature given by a structure preserving map from the response to the question. IQAP can thus give rise to an overanswer in which we get more than just a complete answer to the question; we get a more informative subset of the element of the partition picked out. This is once again an instance of the general strategy of being as informative as possible . . . A particularly interesting case involves prosodically marked overanswers to polar questions like the following. (42)

a. b.

Did John eat all of the cookies? John ate SOME of the cookies.

(42b) is an overanswer to (42a). By itself (42b) doesn’t provide enough information to compute an answer to the question. But the prosodic marking gives rise to a structure preserving map, an ME graph in the language of Asher (1993) or Asher (2002), from the response to the question. In this case the prosodically marked some is mapped to all, and provides the relevant alternative set. The lack of a full answer also triggers the S-implicature, and including the implicature that John didn’t eat all of the cookies together with (42b) provides a complete answer to (42a). But (42b) also gives more information than just a simple no would have. A response that on its own fails to give a complete answer to a question can also trigger embedded implicatures and ones that wouldn’t be calculated from standard lexical alternatives for some. Consider, for instance, the S-implicature of (43b), which is that John believes that not many of the students passed the exam, or the implicature of (44b) which is that everyone didn’t read most of the books. These implicatures follow given the mappings on ME graphs that map some to many in (43b) and some to most in (44b). (43)

a. b.

Does John believe that many of the students passed the exam? John believes that SOME of the students passed the exam.

(44)

a. b.

Did everyone read most of the books? Everyone read SOME of the books.

Notice that once again the addition of the implicatures in these responses gives us complete answer to the questions they are paired with. My approach makes predictions about when the implicatures should not arise, even if the implicature is consistent with the information in the discourse context. (45)

Did some students go to the party?

(46)

Yes, some students went to the party.

34

Nicholas Asher

Here there is no prosodically marked element and no pairing of two distinct elements on a scale. I predict the implicature not to arise. Let’s now look briefly at Wh-questions. (47)

a. b.

Who read all of the books on the reading list? JOHN read SOME of the books.

This example follows the treatment of polar questions, except there are two prosodically prominent elements in the response. It is the second that under the ME graph mapping generates the implicature. The interaction between questions and answers for generating implicatures has an effect on how evidence for implicatures has often been gathered. As argued by Geurts (2009), introspection is a biased method for implicatures. If you’re given (48)

Some of the boys have a cough.

and you then ask yourself: does this implicate (49)

Not all of the boys have a cough.

A question is suggested: (50)

Do all of the boys have a cough?

You now get an “overanswer” or an indirect answer: IQAP (50), (48). Given the suggested question, we predict the implicature to hold in this “inference” task. The moral of this, however, is that these implicatures need not, and indeed are predicted not to hold outside of this discourse context. So how do implicatures arise in the absence of an inference task or in out of the blue contexts? Sometimes simple prosody suggests something about the discourse context. Roberts (1996) and Kadmon (2009) argue that prosodic information can tell us something about the question under discussion. – Roberts’s constraint: the focus semantic value of φ = the question under discussion addressed by φ . So (51)

Who did Larry kiss?

(52)

Larry kissed NINA.

(53)

How much of the homework did John do?

(54)

He did SOME of the homework.

Implicatures in Discourse

35

Roberts’s constraint doesn’t get us the desired implicatures. But we might imagine that something similar holds for certain lexical stressed elements. – A focussed scalar item x in an assertion φ can give rise to a QUD κ for which there exists a structure preserving map μ : φ → κ such that for some scale S, μ (x) ≥S x. Using our constraint, (55)

a. b. c.

John read SOME of the books.  the QUD: Did John read all/many/most of the books?  The implicature that John did not read all/many/most of the books.

I predict that the implicatures will be much vaguer without specifying a particular discourse context. Finally let’s go back to implicatures within downward entailing contexts. Recall that these were a problem for the more finegrained account of the set of alternatives motivated by the localists. (56)

None of the students answered all of the questions.

does not seem to have, at least in an out of the blue context, the implicature: (57)

All of the students answered some of the questions

even though (56) is equivalent to (58)

All the students didn’t answer all of the questions.

(58) appears to have the predicted implicature, at least in the right discourse context. For instance, if (58) is part of a Correction, the implicature seems to be fine: (59)

a. b.

A: All of the students answer all of the questions. B: No, all of the students DIDN’T answer ALL of the questions.

There is a natural prosodic prominence to the second occurrence all in (58) signalling a lexical choice that is the source of the disagreement, which gives rise to a structure preserving map and the relevant implicature. (56) in this context still less clearly has the implicature because the Correction is also supported by the choice of none.

36

Nicholas Asher

8.

Conclusions

My approach to S-implicatures, which links them to D-implicatures, is quite diferent from the view that implicature reasoning is triggered by default (the position of Chierchia); my approach predicts that certain out of the blue inference tasks for embedded implicatures should be quite difficult, whereas the default approach can only do so on general grounds of sentence complexity. But then the default approach can’t explain why in only certain discourse contexts embedded implicatures are inferred, why when they are inferred, they are often not cancellable. My approach is also distinct from extant Gricean approaches, as these also ignore discourse structure and its effects in their description of the reasoning mechanisms governing implicatures. They too will have troubles with the issues I have raised in this paper for a Chierchia-like position. I’ve argued that there is a unified theory of implicatures. Implicatures, understood as defeasible implications, can arise from several sources: semantics, discourse structure or prosody together with the logical form of what is said. I’ve looked at the first two sources here, and I’ve argued that S-implicatures are often parasitic on D-implicatures. The localist theory would benefit from having the left hand side of the > statements giving the implicatures restricted to those cases in which the implicatures are required for discourse coherence (though the discourse context may be quite minimal). The theory makes several predictions: implicatures can be “cancelled” even if they’re consistent with information in the discourse context; implicatures can be uncancellable even in the face of inconsistency (when mandated by discourse structure); embedded implicatures (both negative and positive) require a more elaborate discourse setting to be triggered and so should be harder to get. This line of thinking also suggests a line of empirical research: given the right discourse context, embedded implicatures should follow as easily as the unembedded ones. My argument here also has implications for the role of a theory of discourse structure in the grammar. If S-implicatures are taken to be part of the grammatical system, then discourse structure and D-implicatures will be too. Alternatively, one can throw all implicatures out of the core grammar, but then grammar becomes much more impoverished and a lot less interesting. While I’ve used S-implicatures as a case study, the effects of D-implicatures are everywhere from lexical composition and phenomena of coercion, in semantic composition and the resolution of scope ambiguities up to purely semantic structures like the temporal structure of texts. What’s also interesting is that D implicatures and their effects, if I’m right, play out at a particular level of the grammar, at the level of logical form; they complete and augment literal content, but they are best computed prior to the model-theoretic interpretation of the logical form.

Implicatures in Discourse

37

In fact, if computational tractability is an issue (and I believe that it should be), then implicatures have to be computed at the level of logical form, since the underlying non-monotonic reasoning involved at the level of full sentence contents is completely intractable.

Appendix on a Particular Calculation of Implicatures Consider a simplified language with and and or as truth functional connectives, a set of generalized quantifiers or up and downward monotonic operators and a finite set of predicates and individual constants. I assume that only predicates, operators and quantifiers are associated in the lexicon with scales, not individual constants. A scale for a predicate yields a scale for an atomic formula in an obvious way (I like John  {I love John, I adore John, I dislike John, I hate John}), and these make up alt(φ ) for φ an atomic formula. S-alt(φ ) denotes the formula that is the strongest logical alternative to φ amongst the formulas in this set, in the spirit of Chierchia. For a formula φ with a mon ↑ quantifier, S-Alt(φ ) = S-Alt(Q)φ ; if φ is of the form of a binary operator O(ψ , χ ), Salt(φ ) = S − alt(O)(ψ , χ ). The axioms below are expressed in a subjectivist semantics and can be seen as a more detailed working out of the maxim of quantity given in the paper. To have a subjective epistemic logic means that these axioms are meant to describe a space of worlds that are a subset of the doxastic possibilities of an agent, in our case the speaker. I assume that the material on the left hand side of the > statements below is a description in GL of something that has been said by the speaker. – φ > ¬S-alt(φ ), for atomic φ . – for mon ↑ operators Q: Q(ψ ) > (¬S-alt(Q)(ψ ) ∧ Q(¬(S-alt(ψ )))) – φ ∨ ψ > (¬S-alt(φ ∨ ψ ) ∧ ¬ S-alt(φ ) ∧ ¬S-alt(ψ )) – for mon ↓ operators DE: DE(φ ) > ¬(DE(S-altDE(φ ) (φ ))) – A special case of a DE operator: ¬φ > ¬(¬S-alt¬φ (φ )) – φ ∧ ψ > (¬S-alt(φ ∧ ψ )∧¬ S-alt(φ )∧¬ S-alt(ψ ))(S-alt(φ ∧ ψ ) typically is ⊥). Fact 1. If the basic scales are linear and if φ is positive, then S-alt(φ ) exists and the implicatures of φ are well defined.

38

Nicholas Asher

Proof is by induction on the complexity of φ . Base case for atomic formulas is given by our assumptions. So assume that φ , ψ have linear scales associated with them, and let Q be monotonic ↑ operator. The scale associated with Q is by assumption linear, and so there is an S-alt(Q). The induction hypothesis gives the rest. Similarly for φ ∧ ψ and φ ∨ ψ . Let’s return now to the Chierchia examples that supposedly present troubles for Griceans. (10)

John did the reading or some of the homework.

(60)

John did all of the reading or all of the homework.

This account straightforwardly yields the implicature that John didn’t do both the reading and some of the homework, and he didn’t do all of the homework for (10). Given the scales for quantifiers, no implicatures for the disjuncts are predicted; the only implicature for (60) is that John didn’t do both all of the reading and all of the homework, as desired. Multiple implicatures are in principle also not a problem. (61)

a. b.

Someone at the party will be smoking or someone will be drinking  It’s not the case that someone at the party will be smoking and that someone will be drinking and not everyone will be smoking and not everyone will be drinking.

Let’s consider (61a). Since the disjunction has wide scope, we deal with that first using the recursion, we get – s ∨ d > ¬(s ∧ d) ∧ ¬S-alt(s) ∧ ¬S-alt(d) – Since this is consistent with the context, we infer using DMP: ¬(s ∧ d) ∧ ¬S-alt(s) ∧ ¬S-alt(d) – ¬S-alt(s) = Everyone at the party will be smoking and ¬S-alt(d) = Everyone at the party will be smoking We predict the desired implicatures. Similarly an example like (62) can be shown to have the implicature in (63): (62)

Every boy did some of the homework.

(63)

Every boy didn’t do all of the homework.

Implicatures in Discourse

39

Negative implicatures work similarly: (64)

I doubt that John has read many philosophy books.  I believe that he has read some philosophy books.

Suppose we simply calculate the bits inside the doubt context via the DE rule as Chierchia suggests: (65)

I doubt John read some philosophy books. |= I doubt John read many philosophy books.

So we get as an implicature (66)

¬ I doubt (John read some philosophy books) ↔

(67)

I believe (John read some philosophy books)

References Alonso-Ovalle, Luis 2005 Distributing the disjuncts over the modal space. In: Leah Bateman and Cherlon Ussery (eds.), North east linguistics society, vol. 35. Amherst, MA. Asher, Nicholas 1993 Reference to Abstract Objects in Discourse. Dordrecht: Kluwer Academic Press. Asher, Nicholas 1995 Commonsense Entailment: A Logic for Some Conditionals. In: Gabriella Crocco, Luis Fari˜nas del Cerro and Andreas Herzig (eds.), Conditionals: From philosophy to computer science, 103–147. New York: Oxford University Press. Asher, Nicholas 2002 From Discourse Micro-structure to Macro-structure and back again: The Interpretation of Focus. In: Hans Kamp and Barbara Partee (eds.), Current Research in the Semantics/Pragmatics Interface, Vol. 11: Context Dependence in the Analysis of Linguistic Meaning. Amsterdam: Elsevier Science Ltd. Asher, Nicholas 2007 Dynamic Discourse Semantics for Embedded Speech Acts. In: Savas Tsohatzidis (ed.), John Searle’s Philosophy of Language, 211–244. Cambridge: Cambridge University Press.

40

Nicholas Asher

Asher, Nicholas 2008 Troubles on the Right Frontier. In: Peter K¨uhnlein and Anton Benz (eds.), Proceedings of Constraints in Discourse 2005. Amsterdam: John Benjamins. Asher, Nicholas 2012 Implicatures and Discourse Structure. Forthcoming in Lingua. Asher, Nicholas and Daniel Bonevac 2005 Free Choice Permission is Strong Permission. Synthese 145(3): 303– 323. Asher, Nicholas and Alex Lascarides 1998 The Semantics and Pragmatics of Presupposition. Journal of Semantics 15: 239–299. Asher, Nicholas and Alex Lascarides 2003 Logics of Conversation. Cambridge: Cambridge University Press. Asher, Nicholas and Michael Morreau 1991 Commonsense Entailment: A Modal Theory of Nonmonotonic Reasoning. On: Proceedings of the 12th International Joint Conference on Artificial Intelligence. San Francisco: Morgan Kaufmann. Asher, Nicholas and Sylvain Pogodalla 2010 SDRT and Continuation Semantics. LENLS 2010, Tokyo, Japan. Block, Eliza 2009 Gricean Implicature. Paper presented at the Michigan Pragmatics Workshop. Chemla, Emmanuel 2009 Universal Implicatures and Free Choice Effects: Experimental Data. Semantics and Pragmatics 2(2): 1–33. Chierchia, Gennaro 2004 Scalar Implicatures, Polarity Phenomena and the Syntax/Pragmatics Interface. In: Adriana Belletti (ed.), Structures and Beyond. Oxford: Oxford University Press. Chierchia, Gennaro, Danny Fox, and Benjamin Spector 2008 The Grammatical View of Scalar Implicatures and the Relationship between Semantics and Pragmatics, draft. Fox, Danny 2007 Free Choice Disjunction and the Theory of Scalar Implicatures. In: Uli Sauerland and Penka Stateva (eds.), Presupposition and Implicature in Compositional Semantics, 71–120. NewYork: Palgrave Macmillan. Fox, Danny and Roni Katzir 2011 On the characterization of alternatives. Natural Language Semantics 19(1): 87–107. Geurts, Bart 2009 Scalar Implicature and Local Pragmatics. Mind and Language 24(1): 51–79.

Implicatures in Discourse

41

Geurts, Bart 2010 Quantity Implicatures. Cambridge: Cambridge University Press. Geurts, Bart and Nausicaa Pouscoulous 2009 Embedded Implicatures?!? Semantics and Pragmatics 2(4): 1–34. Groenendijk, Jeroen 2008 Inquisitive Semantics. Proceedings of SALT XVIII The University of Massachusetts at Amherst: Amherst, MA. Horn, Laurence 1972 The semantics of logical operators in English. PhD thesis, UCLA. Horn, Laurence 2006 The Border Wars: a neo-Gricean perspective. In: Ken Turner and Klaus von Heusinger Where Semantics meets Pragmatics. Dordrecht: Elsevier. Jasinskaja, Katja 2002 Relevance and Other Constraints on the Quantification Domain of only. Proceedings of the Workshop on Information Structure in Context, IMS Stuttgart. Kadmou, Nirit 2009 Contrastive topics and the focal structure of quations. Semantics archive. URL: semanticsarchive.net/Archive/jMzOTczN/Kadmonms-2009-TOPIC-FOCUS.pdf. Kratzer, Angelika and Junko Shimoyama 2002 Indeterminate pronouns:The view from Japanese. In:Yukio Otsu (ed.), The Proceedings of the Third Tokyo Conference on Psycholinguistics, 1–25. Tokyo: Hituzi, Kroch, Anthony 1972 Lexical and inferred meanings for some time adverbs. Quarterly Progress Report of the Research Laboratory of Electronics 104: 260– 267. Roberts, Craige 1996 Information Structure in Discourse: Towards an Integrated Formal Theory of Pragmatics. In: Jae Hak Yoon and Andreas Kathol (eds.), OSU Working Papers in Linguistics 49: Papers in Semantics, 91–136. The Ohio State University Department of Linguistics. Sauerland, Uli 2004 On Embedded Implicatures. Journal of Cognitive Science 5: 107–137. Schulz, Katrin and Robert van Rooij 2006 Pragmatic Meaning and Non-Monotonic Reasoning: The Case of Exhaustive Interpretation. Linguistics and Philosophy 29(2): 205–250. Schwarzschild, Roger 1998 GIVENness, AvoidF and other Constraints on the Placement of Accent. Natural Language Semantics 7(2): 141–177.

42

Nicholas Asher

Spector, Benjamin 2006 Aspects de la pragmatique des op´erateurs logiques. PhD dissertation, Universit´e Paris 7. van Rooij, Robert and Katrin Schulz 2004 Exhaustive Interpretation of Complex Sentences. Journal of Logic, Language and Information 13(4): 491–519. Schulz, Katrin 2007 Minimal Models in Semantics and Pragmatics: Free Choice, Exhaustivity, and Conditionals. Ph.D. Dissertation, University of Amsterdam. Txurruka, Isabel and Nicholas Asher 2007 A discourse-based approach to Natural Language Disjunction (revisited). In: Michel Aurnague, Kepa Korta and Jesus Mari Larrazabal (eds.) Language, Representation and Reasoning. University of the Basque country Press. Vieu Laure, Myriam Bras, Nicholas Asher and Michel Aurnague 2005 Locating adverbials in discourse. Journal of French Language Studies 15(2): 173–193.

Permission and Choice* Paul Portner

1.

Introduction

Functions of imperatives Imperatives can be used to perform a variety of intuitively distinct speech acts, for example ordering, advising, requesting, and permitting (Schmerling 1982; Davies 1986; Han 1998; Schwager 2005a; Portner 2007; among many others): (1)

a. b. c. d.

Sit down right now! (Order) Talk to your advisor about this. (Advice) Help me! (Request) Have a piece of fruit, if you like. (Permission)

As Davies (1986) argues, it is not productive to focus on the details of the categories which happen to have names in English. Rather, we need to understand the nature of the variation: its source, limits, and effects. These distinctions are relevant to grammar. We see this, for example, with the fact that particles may limit the range of meanings available, as shown by the following data from Badiotto, due to Poletto and Zanuttini (2003), and German, from Grosz (2009a): (2)

a.

You need to eat well, so you can grow up to be big and strong. (Advice) M`ange-l ma! eat-it ma

* I have presented versions of this paper at the conference ‘10 Years After’ at the University of Frankfurt, the University of Chicago, the Ohio State University, the University of Pennsylvania, MIT, and the third conference ‘Semantics and Philosophy in Europe’ in Paris. I am thankful for the input I’ve had from many people, at these venues and elsewhere, including Chris Barker, Nate Charlow, Kai von Fintel, Irene Heim, Chris Kennedy, Tony Kroch, Jason Merchant, Peter Pagin, Miok Pak, Craige Roberts, Magdalena Schwager, and Raffaella Zanuttini.

44

(3)

Paul Portner

b.

We can’t let the food go to waste. You have to finish it, even if you don’t want to. (Order) M`ange-l mo! eat-it mo

a.

Iss *bloß/ *JA/ ruhig den Spinat! Das st¨ort mich eat BLOß JA RUHIG the spinach that disturbs me nicht. (Permission) not ‘Eat bloß/JA/ruhig the spinach! That doesn’t disturb me.’ Iss bloß/ JA/ *ruhig den Spinat! Sonst wirst du eat BLOß JA RUHIG the spinach or.else will.be you bestraft. (Order) punished ‘Eat bloß/JA/ruhig the spinach! Or else you’ll be punished.’

b.

One basic issue is whether any of these differences are semantic in nature. It is tempting to analyze the permission imperative in (1d) as different from the others, as it can be paraphrased with a possibility modal: (4)

a. b. c. d.

You must sit down right now! (Order) You should talk to your advisor about this! (Advice) Won’t you please help me? (Request) You may have a piece of fruit! (Permission)

I will refer to (1d) as a permission imperative and the others as requirement imperatives. Choice phenomena An intuitive way of describing the function of a permission sentence is to say that it offers the addressee a choice s/he didn’t have before. When a permission sentence offers more than one choice at the same time, we call it a “free choice” sentence. As is well-known, free choice sentences can be made with disjunction or an indefinite (including special indefinites like any). The free choice inference refers to the fact that a requirement or permission sentence (made with a modal or imperative, or by others means) implies that each disjunct, or each entity described by an indefinite, corresponds to a permitted option.

Permission and Choice

(5)

45

Free choice inference a. You may take an apple or an orange. ⇒ You may take an apple./You may take an orange. b. Take a piece of fruit! ⇒ You may take this apple./You may take that pear.

Related is Ross’s paradox, the lack of licit inference from a permission sentence to disjunction: (6)

Ross’s paradox a. You may take an apple.  You may take an apple or an orange. b. Take an apple!  Take an apple or an orange!

The “paradoxical” aspect can be seen from the comparison with declaratives, where p entails (p ∨ q). I lump all this together under the label choice phenomena. A key testing ground for analyses of permission sentences will be how well they fit into our understanding of choice phenomena. There has been a great deal of research on free choice in modal sentences. We may classify it into several major approaches: 1. Traditional assumptions Choice phenomena come about on the basis of fairly traditional semantic values and Gricean reasoning (Aloni and van Rooij 2004; Schulz 2005). 2. Alternatives Choice phenomena come about because the semantics introduces each alternative separately, one way (Zimmermann 2000; Geurts 2005) or another (Kratzer and Shimoyama 2002; Simons 2005; Men´endez-Benito 2005; Alonso-Ovalle 2006; Fox 2007; Aloni 2007). 3. The Andersonian reduction Choice phenomena come about because permission is defined, in the tradition of Anderson (1956), as something like ‘If p, then things are ok’ (Asher and Bonevac 2005, Barker 2010). 4. Dynamic semantics Choice phenomena come about because of the dynamic semantics associated with particular elements, for example deontic may (van Rooij 2008) or epistemic might (Ciardelli et al. 2009). We also have a divide between those who think that free choice with disjunction is a conversational implicature (Kratzer and Shimoyama; Menendez-Benito; Alonso-Ovalle; Aloni and van Rooij; Schulz), a matter of semantics (Geurts; Simons; Aloni; Barker; Ciardelli et al.), both (Fox), or something else (van Rooij).

46

Paul Portner

It will not be possible to consider all of these analyses in detail. My hope is to shed new light on the problem of free choice by focusing on choice phenomena as they occur with imperatives. While imperatives have been discussed in connection with choice phenomena, it is typically assumed that they are implicitly modal sentences, and so don’t have anything special to teach us. I will argue, in contrast, that a treatment of imperatives which doesn’t assume that they contain a modal operator serves as the basis for an insightful treatment. Main claims of this paper It is the goal of this paper to argue for the following hypotheses: – There is no semantic difference between requirement and permission imperatives. – Differences in function among imperatives mostly depend on the grounds upon which the imperative is issued. – True permission imperatives are derived from the logical relation between the imperative and the context to which it is added. – Choice phenomena with imperatives follow as a special case of the analysis of permission. 2.

Background on the Semantics of Imperatives

There are two main approaches to the semantics of imperatives: the modal theory and the dynamic theory. 1. The modal theory proposes that imperatives contain a modal operator, so that an imperative is very close in meaning to certain sentences containing must or should. (Han 1999, to appear; Schwager 2005a; Aloni 2007; Grosz 2009a). Within the overall modal approach, various authors may treat the so-called modal element proposed as more or less similar to regular modals, and at some point we might better call this the “modaloid” theory, a less attractive term to be sure, but perhaps appropriately so.1 1. If you want to treat the imperative modal as a purely dynamic modal, similarly to the treatment of epistemic modals in Groenendijk et al. (1996) and the treatment of expressions of expectation in Veltman (1996), I’d consider that an implementation of the dynamic theory. See Rooij (2008) and Portner (2009) for discussion of how this might be done. The adherents of what I call the “modal theory” assume that the modal in question falls under a standard (static) analysis of modals, such as Kratzer’s (1981, 1991, 2012).

Permission and Choice

47

2. The dynamic theory claims that the meaning of imperatives consists (entirely, or virtually so) in the way they affect the discourse context (Portner 2004, 2007; Mastop 2005; in a sense Lewis 1979). The dynamic theory of imperatives is really a part of the dynamic theory of clause types (or of sentence mood, if you prefer that terminology). Imperatives are one of the three major clause types, alongside declaratives and interrogatives (Sadock and Zwicky 1985), and we should aim for an explanation for why these three are universal (Portner 2004). Assertion is commonly analyzed in terms of Stalnaker’s concept of common ground (Stalnaker 1974, 1978), and asking a question has been analyzed in terms of a second discourse component, what Ginzburg calls the ‘Question Under Discussion Stack’ (Ginzburg 1995a, b; Roberts 1996). Parallel to these, Portner (2004) proposes that imperatives are interpreted as contributing to the addressee’s To-Do List. The central theoretical claim of this paper is that the dynamic approach can explain, in a simple and natural way, both the variation in function among imperatives and choice phenomena. Outline of the dynamic analysis of imperatives Portner (2004, 2007) argues that the meaning of imperatives can be given within a dynamic framework as follows: (7)

Pragmatic function of imperatives a. The To-Do List function T assigns to each participant α in the conversation a set of properties T (α ). b. The canonical discourse function of an imperative clause φimp is to add [[ φimp ]] to T (addressee). Where C is a context of the form CG, Q, T : C + φimp = CG, Q, T [addressee/(T (addressee) ∪ {[[ φimp ]] })]

The To-do List is similar to ideas in Lewis (1979), Han (1998), Roberts (2004), and Mastop (2005). What’s different is the Ordering pragmatics for imperatives. In particular, the To-Do List functions to impose an ordering on the worlds compatible with the Common Ground, and this ordering determines what actions an agent is committed to taking (Portner 2004):

48 (8)

Paul Portner

Ordering pragmatics for imperatives a. Ordering of worlds: For any w1 , w2 ∈ ∩CG and any participant i, w1 w6 , w4 .

Promises and Threats with Conditionals and Disjunctions

81

• A

v1

A

•

R

v2

• N

P

R

N

P

•

•

•

•

•

•

w1

w2

w3

w4

w5

w6

Figure 2. Game model for punishing and rewarding hearer actions

This captures the intuition that the speaker mostly cares about whether action A is performed. Subordinate to her preference for A, she would prefer to remain neutral over punishing and rewarding. In contrast to that, we should assume that the hearer prefers ¬A over A, but that the reward R and the punishment P that we consider are potentially efficacious, so that they outweigh the hearer’s preference about A. More precisely, the hearer’s preferences are then qualitatively given as: w4 > w1 > w5 > w2 > w6 > w3 . In other words, the hearer (is assumed by the speaker) to value most the reward, and prefers a neutral outcome over a punishment. Subordinate to these preferences is his preference of performing ¬A over performing A. It is not necessary, but also not desirable to specify the hearer’s preferences any further than that. This is because a speaker will never be able to know for sure how exactly the hearer will value a promise or a reward. Our modelling here adopts the speaker’s perspective and takes her natural uncertainty into account. In other words, the model assumes that the speaker believes the hearer’s preferences are qualitatively as specified above, but that the speaker does not know for certain how strongly, for example, w4 is preferred over w1 . Similar remarks then also apply to the speaker’s beliefs about the hearer’s beliefs. Here it is most natural to suppose that the speaker believes that the hearer expects, all else being equal, a neutral outcome. We could even go as far as saying that the hearer might not even be aware of possible punishments and threats and that it is only when pointed out to him that he accommodates these possibilities into his decision-making. To keep matters simple here, we will refrain from representing such a sequential game with possibly unaware players (c.f Feinberg 2005; Heifetz; Meier and Schipper 2009). For the present

82

Robert van Rooij and Michael Franke

purpose it suffices to assume that the speaker believes that the hearer’s beliefs are qualitatively as follows: w2 , w5  w1 , w3 , w4 , w5 . The idea is that the speaker again does not know precisely which probabilistic beliefs the hearer holds, but she does believe that, barring any speaker commitment, the hearer considers action N substantially more likely than either reward or punishment. 4.4. Risk of Strategic Inducements Given the speaker’s natural uncertainty about the hearer’s precise preferences and beliefs, it turns out that disjunctive promises are risky, and therefore suboptimal in expectation, in a sense that threats and conditional promises are not. To see what is at stake, we need to compare the statements in (1) one by one as committing strategic inducements against the background of the speaker’s uncertainty as described in the previous section. Let us look at threats and promises in turn and let us ask what update effects these statements would have on the hearer and how this affects the speaker’s assessment of her expected utility of uttering these statements. As for their semantic update effect, both the conditional threat ¬A → P in (1a) and the disjunctive threat A ∨ P in (1c) are semantically equivalent and denote, if taken as binding, the set {w1 , w2 , w3 , w6 }. If we also take conditional perfection, respectively exclusive readings of disjunctions, into account the impact of these threats is an update that leaves only outcomes {w2 , w6 }. Still, in line with our reasoning above, there should be a small difference between the conditional and the disjunctive threat. Whereas the conditional threat slightly increases the probability of w6 (in the expectation of the speaker), the disjunctive threat slightly increases the probability of w2 . This is because the conditional mentions “¬A” and so the speaker will assume a slight increase in the chance that the hearer will play this option. For the disjunctive threat rather the hearer choice A is given a slightly higher probability. That means that mentioning the speaker-desirable action A in the disjunctive threat actually has a slight increasing effect on the expected utility of that statement, as compared to the conditional threat that mentions “¬A.” However, this slightly detrimental effect of mentioning the speaker-undesirable action is relatively harmless, because the speaker believes that the hearer prefers w2 over w6 and the speaker believes that the hearer considers w2 much more likely than w6 . It is therefore not likely that the conditional threat would not be efficacious despite the fact that it might slightly increase the chance of performance of ¬A.

Promises and Threats with Conditionals and Disjunctions

83

Moreover, and more importantly, the speaker can compensate for the risk of a conditional threat by choosing a sufficiently stronger punishment. As noted in Section 3.1., this need not decrease the speaker’s expected utility, because threats are (relatively) cheap in expectation in that the (possibly costly) punishment is not (as) socially binding as in the case of a promise. This is different for disjunctive promises. Again, the conditional promise A → R in (1b) and the disjunctive promise ¬A ∨ R in (1d) are semantically equivalent. Their update effect is to eliminate outcomes w2 and w3 , and additionally, if perfection and exclusive readings are taken into account, restrict the options under consideration to {w1 , w5 }. Once more, we also attest a difference from mentioning different alternatives: whereas the conditional promise slightly increases the speaker’s expected utility because it mentions the desirable option A and thus increases the probability that w1 is realized, the disjunctive promise slightly decreases the expected utility by mentioning the undesirable option ¬A and thereby increasing the probability of w5 . However, unlike with threats, this latter decrease is more risky from the point of view of an uncertain speaker: the problem is that although w1 is assumed more hearer-desirable than w5 , the latter is naturally assumed substantially more likely. If the speaker is uncertain about the extent to which the hearer prefers w1 over w5 , mentioning the undesirable option puts the efficacy of the inducement at risk. But could the speaker not compensate this risk, as she could with conditional threats, by promising a higher reward, so as to make sure that w1 is sufficiently preferred over w5 ? She probably could, but not necessarily without sacrificing even more on expected utility. Promises are costly when efficacious and more institutionally dependent than threats. To the extent that the speaker would like to invest on the promise to compensate risk, the statement’s expected utility decreases, because any stronger reward would only be more costly and thus further decrease the speaker’s expected utility. This is then the main difference between conditional threats and disjunctive promises: although both mention a speaker-undesirable option, which is risky under uncertainty, threats, but not promises, can be “pumped up cheaply”, so to speak, to compensate for the risk. Taken together, we argue that disjunctive promises are suboptimal, because they emphasize the wrong alternative and cannot compensate for any negative effects that this might have. It is the combination of natural speaker uncertainty, priming by mentioning and the asymmetry of when punishments and rewards are speaker-costly that explains why disjunctive promises are a deficient inducement strategy.

84

Robert van Rooij and Michael Franke

5.

Conclusion

In this paper we explained promises and threats as strategic commitments of the speaker. We discussed the constraints that such commitments have to obey to be rational. We saw that from the speaker’s point of view, a threat is cheaper in expectation than a promise. The main idea presented in this paper was that disjunctions like (1d) preferably get a threat-reading because their use as a disjunctive promise is a risky, and therefore suboptimal strategic inducement. This explanation involved the idea that mere mentioning a possibility raises its salience. This helps to explain the difference in acceptability between the conditional promise A → R and the disjunctive promise ¬A ∨ R. The idea that a threat is cheaper in expectation than a promise, on the other hand, explains why the disjunctive promise is also more risky than the conditional threat ¬A → P. This explanatory strategy certainly raises a number of concerns. Perhaps the most pressing is the question whether the alleged suboptimality of a disjunctive promise is something that is checked on-the-spot, every time anew a speaker would like to influence a hearer. We emphatically do not subscribe to this obviously nonsensical view. Rather we suggest here that the suboptimality of disjunctive promises is a force that informs language organization, not ad hoc choice of formulation. Certain locutions and grammatical constructions, and not others, are conveniently and conventionally used for certain discourse functions, and not others. It is at this level of functional organization that, we suggest, evolutionary pressures have weeded out disjunctions as a viable vehicle of making promises. References Bolinger, Dwight 1979 Is the imperative an infinitive? In: Bolinger, Dwight (ed.), Meaning and Form, 152–182. New York: Longman. Brandom, Robert 1983 Asserting. Noˆus 4: 637–650. Clark, Billy 1993 Relevance and ‘pseudo-imperatives’. Linguistics and Philosophy 16: 79–121. Culicover, Peter W. and Ray Jackendoff 1997 Semantic subordination despite syntactic coordination. Linguistic Inquiry 28(2): 195–217. de Jager, Tikitu 2009 ‘Now that you mention it I wonder...’: Awareness, Attention, Assumption. PhD thesis, Universiteit van Amsterdam.

Promises and Threats with Conditionals and Disjunctions Feinberg, Yossi 2005

85

Games with incomplete awareness. Research Paper No. 1894, Stanford University. Fillenbaum, Samuel 1986 The use of conditionals in inducements and deterrents. In: E. C. Traugott, A. ter Meulen, J. S. Reilly and C. A. Ferguson (eds.), On Conditionals, 179–195. Cambridge University Press. Franke, Michael 2008 Pseudo-imperatives and other cases of conditional conjunction and conjunctive disjunction. In: C. Fabricius-Hansen, and W. Ramm (eds.), ‘Subordination’ versus ‘Coordination’ in Sentence and Text – From a Cross-Linguistic Perspective, Studies in Language Companion Series (SLCS), 255–279. Amsterdam, Philadelphia: John Benjamins. Franke, Michael and Tikitu de Jager 2011 Now that you mention it: Awareness dynamics in discourse and decisions. In: Anton Benz, Christian Ebert, Gerhard J¨ager and Robert van Rooij (eds.), Language, Games, and Evolution, 60–91, Heidelberg: Springer. Gazdar, Gerald 1979 Pragmatics: Implicature, Presupposition, and Logical Form. New York: Academic Press. Gazdar, Gerald 1981 Speech act assignment. In: Joshi, A. K., Webber, B., and Sag, I. A. (eds.), Elements of Discourse Understanding, 64–83. Cambridge University Press. Geis, Michael L. and Arnold M. Zwicky 1971 On invited inferences. Linguistic Inquiry 2(4): 561–566. Geurts, Bart 2005 Entertaining alternatives: Disjunctions as modals. Natural Language Semantics 13: 383–410. Hamblin, Charles L. 1970 Fallacies. Methuen, London. Heifetz, Aviad, Martin Meier and Burkhard C. Schipper 2009 Dynamic unawareness and rationalizable behavior. Unpublished manuscript. Horn, Laurence R. 2000 From if to iff : Conditional perfection as pragmatic strengthening. Journal of Pragmatics 32: 289–326. Hirschleifer, Jack 2001 Game-theoretic interpretations of commitment. In: R. M. Nesse (ed.), Evolution and the Capacity for Commitment, 77–93. Russell Sage Foundation.

86

Robert van Rooij and Michael Franke

Jayez, Jacques and Mathilde Dargnat 2009 One more step and you’ll get pseudo-imperatives right. In: Torgrim Solstad and Arndt Riester (eds.), Proceedings of Sinn und Bedeutung 13, 247–260, University of Stuttgart. Klein, Daniel B. and Brendan O’Flaherty 1993 A game-theoretic rendering of promises and threats. Journal of Economic Behavior and Organization 21(3): 295–314. Lawler, John M. 1975 Elliptical conditionals and / or hyperbolic imperatives: Some remarks on the inherent inadequancy of derivations. In: Papers from the 11th Regional Meeting. Chicago Linguistic Society, 371–382. Chigaco Linguistic Society. Nedungadi, Prakash 1990 Recall and consumer consideration sets: Influencing choice without altering brand evaluations. Journal of Consumer Research, 17: 263– 276. Russell, Bertrand 2007 Imperatives in conditional conjunction. Natural Language Semantics 15: 131–166. Schelling, Thomas C. 1960 The Strategy of Conflict. Harvard University Press, Cambridge, Massachusetts. Schulz, Katrin and Robert van Rooij 2006 Pragmatic meaning and non-monotonic reasoning: The case of exhaustive interpretation. Linguistics and Philosophy 29: 205–250. Searle, John 1969 Speech Acts: An Essay in the Philosophy of Language. Cambridge University Press. Searle, John and Daniel Vanderveken 1985 Foundations of Illocutionary Logic. Cambridge University Press. Schwager, Magda 2006 Interpreting Imperatives. PhD thesis, Johann-Wolfgang-Goethe Universtität zu Frankfurt am Main. Stalnaker, Robert 1978 Assertion. In: P. Cole (ed.), Syntax and Semantics, vol. 9, 315–332. New York: Academic Press. Stalnaker, Robert 1975 Indicative conditionals. Philosophia 5(3): 269–286. Stalnaker, Robert 2006 Conditional assertions and conditional propositions. Paper presented at the Formal Epistemology Workshop, UC-Berkeley.

Promises and Threats with Conditionals and Disjunctions Swanson, Eric 2006

87

Interactions with Context. PhD thesis, Massachusetts Institute ofTechnology. van Canegem-Ardijns, Ingrid and William van Belle 2008 Conditionals and types of conditional perfection. Journal of Pragmatics 40: 349–376. van der Auwera, Johan 1986 Conditionals and speech acts. In: E. C. Traugott, A. ter Meulen, J. S. Reilly, and C. A. Ferguson, (eds.), On Conditionals, 197–214. Cambridge University Press. van der Auwera, J. 1997 Conditional perfection. In: A. Athanasiadou and R. Dirven(eds.), On Conditionals Again, 169–190. John Benjamins. van Rooij, Robert 2005 A modal analysis of presupposition and modal subordination. Journal of Semantics 22: 281–305. Verbrugge, Sara, Kristien Dieussaert, Walter Schaeken and William Belle 2004 Promise is debt, threat another matter: The effect of credibility on the interpretation of conditional promises and threats. Canadian Journal of Experimental Psychology 58(2): 106–112. Zhang, Jun, Kent C. Berridge, Amy J. Tindell, Kyle S. Smith and J. Wayne Aldridge 2009 A neural computational model of incentive salience. PLoS Computational Biology 5(7): 1–14. Zimmermann, Thomas Ede 2000 Free choice disjunction and epistemic possibility. Natural Language Semantics 8: 255–290.

Part II. Sentence Types and Clausal Peripheries

Revisiting the CP of Clefts* Adriana Belletti

1.

Introduction

This paper reconsiders the analysis of the CP of cleft sentences developed in previous works (Belletti 2008, 2009), with the main aim of going deeper into some of its aspects, and widening its empirical coverage. The main issues addressed in this perspective concern in particular: the way the locality of syntactic computations conditions the syntax and interpretation of clefts; the detailed analysis of the shape of the CP domain in clefts; aspects of the interpretation of clefts in combination with closely related structures, such as the clausal complements of perception verbs. The analysis is framed in cartographic terms (Cinque 2002 ed.; Rizzi 2004 ed.; Belletti 2004 ed.; Cinque and Rizzi 2010, and related work), according to which dedicated positions in the functional structure of the clause overtly express different interpretations in different syntactic positions. This richness in the functional structure can have crucial consequences for the relevant locality principle operating in syntax, whose specific operation is considered here in the domain of clefts. A further consequence of the approach is that it also allows one to explicitly identify and characterize both the content and the size of the functional domain: the CP space of clefts is analyzed in some detail in this perspective.

* I thank here the people from whose comments and remarks this article has specially benefited, in different presentations in different occasions: Paola Beninc`a, Liliane Haegeman, David Pesetsky, Luigi Rizzi, Dominique Sportiche, Edwin Williams, Ede Zimmermann.

92

Adriana Belletti

2.

Some Bbackground

2.1. The Essential Features of the Assumed Analysis The crucial insight of the proposed analysis is twofold: on the one hand it is assumed that a general property of cleft sentences is that the copula selects as its sentential complement a reduced CP1 which lacks (at least) the highest ForceP layer, assuming the articulated analysis of CP schematized in (1) (Rizzi 1997 and much related work):2 (1)

[ForceP . . . [TopP [FocP [TopP . . . [FinP [TP

On the other hand, such reduced CP complement may have a partly different status in subject and non-subject clefts (referred to with the general term object clefts, henceforth). Only in subject clefts the reduced CP complement is (/may be) endowed with an EPP feature, the formal way to represent the predication relation which is expressed by this type of clauses. The proposal is summarized in (2): (2)

a.

b.

Subject clefts: the complement of the copula is/may be a reduced CP containing an EPP feature Object clefts: the complement of the copula (always) is a reduced CP with no EPP feature

1. See Belletti (2008, 2009) for more details. Throughout I will use the general label CP to refer to sentential (small) clausal complements. Where relevant, I will specify some of the possible labels of the reduced CP. That the copula in clefts selects a CP containing the clefted constituent is a proposal sharing similarities with classical analyses such as those in Ruwet (1975), Kayne (1994) and related work; see also Clech-Darbon, Rebuschu, Rialland (1999) for an overview of this type of approach to the analysis of clefts. The paper will only address the analysis of argument clefts, where the clefted constituent is a DP, either a subject or a direct object (with occasional reference to clefted PPs). 2. Only some of the positions assumed in the left periphery from much work in the literature are represented in (1); specifically, only the discourse related positions of Focus andTopic which will be referred to in some of the discussion below are indicated in (1). See also Rizzi (2004); Beninc`a and Poletto (2004); Haegeman (2004); Bocci (2004); Grewendorf (2005); Mioto (2003); Bianchi and Frascarelli (2009), for some items of a rich relevant literature dealing with the map of the left periphery and its interpretable positions in the sense of Chomsky (1995, 2005).

Revisiting the CP of Clefts

93

In (2) the fundamental distinction between the two types of clefts is stated. According to this proposal, subject clefts have one option more than object clefts: their reduced CP complement may be endowed with an EPP feature. As far as its size is concerned, the CP complement of the copula is thus a “small” CP in both subject and object clefts, since it lacks some layers. Technically, it can be defined as a “small clause” (Starke 1995) in the case of subject clefts, as only in this case a predication relation holds between the clefted subject and the rest of the (relative) clause following (Stowell 1983; Burzio 1986; Moro 1997; Rothstein 2000). Examples of a subject and an object cleft are given in (3a) and (3b), from French and Italian respectively: (3)

a. b.

S: C’est it is O: E’ (it) is

Marie qui – a parl´e Marie that has spoken MARIA che (i ragazzi) hanno incontrato – Maria that the boys have met

According to the proposal in (2), subject clefts in which the copula selects a small clause CP and object clefts in which the copula selects a reduced CP, correspond to the schematic structures in (4) following. For convenience, the shortcut be indicates the copula and “che” indicates the complementizer, the realization of C in Italian clefts. The crucial hypothesis in (4) concerning the complementizer, is that in clefts it is the realization of the Fin head.3 It is left open for now, the amount of reduction the CP undergoes (3.2 for relevant discussion). (4)

a. b.

Subject clefts (selecting a CP small clause): T . . . be [CP Force . . . . . . EPP. . . . . . [FinP che [TP S . . . O/(PP)]]]] Object clefts (selecting a reduced CP): T . . . be [CP Force . . . . . . . . . [FinP che [TP S . . . O/(PP)]]]]

The different analysis corresponds to the possibly different interpretations that subject clefts may allow in contrast to object clefts. The following section takes up the interpretation issue.4

3. Only finite clefts are considered here. 4. That the position occupied by a subject in the CP is different from the one occupied by an object is not a unique property of the CP of clefts. See Haegeman (1996) for similar conclusions in relation to the position of a subject and the position of an object in Germanic (Dutch) V2.

94

Adriana Belletti

2.2. The Interpretation of Subject and Object Clefts Although a cleft sentence always instantiates a form of focalization, as has generally been observed in different languages (Kiss 1998; Abels and Muringi 2005) its nature and interpretation may differ in subject and object clefts, as the different writing of the clefted constituent in (3) suggests. In the case of the subject cleft in (3a), the clefted subject can be interpreted as the focus of new information, while the clefted object in (3b) only allows for a contrastive/corrective focus interpretation, indicated with capital letters throughout (in order to make the different interpretations immediately recognizable). This is clearly revealed in the context of question-answer pairs. In languages such as French a cleft (possibly reduced, leaving the relative predicate unpronounced) can typically be utilized as the answer to a question of information on the identification of the subject. This type of answering strategy, however, cannot be utilized to identify the object (Belletti 2008, 2009 and references cited therein, bearing also on relevant L1 and L2 acquisition data): (5)

Q A

(6)

Q A

Qui (est-ce qui) a parl´e? ‘Who spoke?’ C’est Jean (qui a parl´e). ‘It is Jean (who spoke).’ Qu’est-ce que t’as achet´e ( /Qu’as-tu achet´e)? ‘What have you bought?’ ( *) C’est un livre (que j’ai achet´ e). ‘It is a book.’

Similarly for a clefted PP: Q A

Avec qui es-tu sorti? ‘With whom did you go out?’ ( *) C’est avec Jean (que je suis sorti). ‘It is with Jean.’

Of course, there is nothing wrong with the form of the clefts in (6), which are perfectly grammatical sentences in French, in particular in their non-reduced form. What is wrong, is just their use as an answer, in the question-answer context. They cannot answer a question, which is simply asking for the identification of the object, with no other presupposition implied. The clefts in (6) are perfect

Revisiting the CP of Clefts

95

in, e.g., a contrastive/corrective context as the one in (7), which illustrates the direct object case5 : (7)

a.

b.

Context: On m’a dit que hier t’as achet´e un journal. ‘They told me that yesterday you have bought a newspaper.’ Correction: Non, c’est UN LIVRE que j’ai achet´e. ‘It is a book that I have bought.’

In contrast to object clefts which only allow for a contrastive/corrective focus interpretation of the clefted constituent, subject clefts, beside the new information focus interpretation illustrated in the French example in (5), also allow for a contrastive/corrective interpretation in the appropriate context. (8) illustrates this point: (8)

a.

b.

Context: On m’a dit que Marie a parl´e. ‘They told me that Marie has spoken.’ Correction: Non, c’est JEAN qui a parl´e. ‘No, it is JEAN that/who has spoken.’

Hence, subject clefts have one interpretive option more. Ultimately, this further option should be ascribed to the possibility for subject clefts to also admit the same syntactic analysis as object clefts. How can the difference between subject and object clefts be expressed? Why do subject clefts and object clefts differ in their interpretative possibilities in the described way? Following the cartographic analysis in Belletti (2004, 2009), I assume the existence in the low part of the TP clausal map of a vP periphery containing a Focus position, characteristically dedicated in different languages to host purely new information constituents. For instance, this low Focus position is assumed to host post-verbal subjects carrying new information in a language like Italian. According to this analysis, the post-verbal subject in the answer in (9b), fills the

5. Leaving the relative predicate unpronounced is less felicitous in this case. I leave open here the issue as to what extent exactly the relative predicate of a cleft can be left unpronounced. An issue which clearly deserves attention in future research.

96

Adriana Belletti

dedicated low Focus position and it is thereby interpreted as the new information asked for in the question in (9a). This assumed analysis is schematized in (9c,d):6 (9)

a. b.

Chi ha parlato? ‘Who spoke?’ Ha parlato Gianni. has spoken Gianni

c.

[CP . . . [ TP . . . [TopP . . . [FocP [TopP . . . vP]]]]]

d.

[CP “. . . ”[ TP pro . . . ha parlato. . . [FocP Gianni . . . [vP – . . . ]]]]

Verb (or part of the verb phrase) movement, yields the verb-subject order, with the subject linearly appearing post-verbally. Going back to the interpretive issue under discussion, with the clefted constituent (possibly) new information focus in subject clefts vs contrastive/ corrective focus in object clefts, a crucial idea is that the Focus position utilized is different in the two cases. It is the vP peripheral low Focus position in the matrix clause vP periphery of the copula, in the case of new information subject clefts; it is the high left peripheral position in the reduced CP complement of the copula, in the case of contrastive/corrective object clefts – and also in subject clefts when they are interpreted/used contrastively/correctively, a possibility seen in (8). In the former case, the clefted subject is interpreted as focus of new information, and, in a language like French in which clefts are typically used in answers, the structure can be used as an answer to a question on the identification of the subject; in the latter case, the clefted object is interpreted as contrastive/ corrective focus, as is typically the case for left peripheral focalization, and a cleft cannot be used to answer a question concerning the identification of the

6. For a more detailed discussion of the various implications of this analysis in the cartographic perspective as well as of the existence of different answering strategies to the same questions across languages and their acquisition in L1 and L2, see the references of my previous work quoted. As is also discussed there, the vP periphery also containsTopic positions illustrated in (9c) in the text, along parallel lines as the CP clause external left periphery. A post-verbal subject with a Topic/given interpretation, typically associated with a downgrading intonation, should fill this low position, accordingly. I will not elaborate on this any further here, as it would take the discussion too far apart and continue to only indicate the relevant vP peripheral Focus position in the derivations. The reader is again referred to the reference quoted (in particular Belletti 2004) for detailed elaboration on this point.

Revisiting the CP of Clefts

97

object.7 Hence, coherently with cartographic guidelines, in new information subject clefts, the subject is interpreted in exactly the same position as the new information post-verbal subject in a null-subject language like Italian.8 Similarly, a contrastive/corrective focus object is interpreted in the left peripheral position, as contrastive/corrective focalized constituents typically are. Thus, a direct reason is provided as to why object clefts cannot be used as answers to questions for the identification of the object, even in a language of the French type, which widely adopts clefts as answering strategies: the new information focus position is the low focus position, where only the subject can be located, not the high left peripheral one, where the object (or a contrastive/corrective subject) is found.

7. For (classical) references on (syntactic) focalization, see Chomsky (1977); Kiss (1998) and Rizzi (1997); Cruschina (2006, 2010); Bocci (2004); Belletti (2004) in the frame of the discussion differentiating low and high focus. In the analysis of contrastive/corrective clefts assumed, these clefts share a crucial similarity with simple left peripheral focalization of the type in i., as the same focus position is utilized in both cases: i.

MARIA ho salutato (non Gianni). ‘Maria I have greeted (not Gianni).’

However, the semantics of the two structures cannot be completely assimilated. The contrastive/corrective focalization of clefts is more constrained than simple contrastive/corrective focalization as in i: a quantified expression cannot be focalized through a contrastive/corrective cleft, whereas it can be contrastively/correctively focalized in a simple sentence: ii.

a. b.

*Non e` NESSUNO / e` TUTTI che ho incontrato. ‘It is NOBODY/EVERYBODY that/whom I met.’ NESSUNO /TUTTI ho incontrato. ‘NOBODY/EVERYBODY I met.’

Thanks to P. Beninc`a and to E. Zimmermann for pointing out this type of contrasts, which seems to indicate the non quantificational status of the Focus position involved in clefts. The cartographic consequences of this conclusion will not be developed here. For ease of exposition, I am going to continue to assume that the Focus position involved is the same in both simple focalization and (object) clefts, and that the interpretive limitations shown in clefts should ultimately be derived from their overall semantics, only partly linked to the type of focalization that clefts express. 8. On the possible correlation between new information subject clefts and the nonnull subject nature of the language, see Belletti (2008, 2009), also partly based on experimental results from Brazilian Portuguese reported in Guesser (2007).

98

Adriana Belletti

The proposed analyses are given in the schematic derivations in (10). (10)

a.

Subject clefts (new information)

[TP Ce . . . T [ FocP [vP eˆtre [CP Force [CP “ . . . ” [ EPP Jean [FinP qui[ – a parl´e]]]]]]]] 9 b.

Object clefts (correction/contrast)

. . . be [CP Force . . . [FocP . . . [FinP che [TP S . . . . . . O(/PP)]]]]. . . If this type of analysis provides a straightforward characterization of the detected different interpretive possibilities of subject and object clefts, the obvious question to ask is: why is it so? Why is it that only subject clefts allow for an analysis as in (10a)? In a more explicit and formal way, the question amounts to asking: if the copula can take as its complement a small clause as in (10a), with the subject moved to the EPP position in CP and then moving further to the vP peripheral new information focus position of the copula in the matrix sentence, why couldn’t exactly the same derivation also be at work in the case of object clefts, with the object moving to the EPP position and then to the matrix vP peripheral focus position? The following section is devoted to provide an answer to this fundamental question. 2.3. A Locality Explanation There is indeed a direct principled reason as to why only subject clefts can undergo the derivation in (10a). From the perspective of the interpretation then, there is a principled reason as to why only the subject can be the focus of new information in the vP periphery of the matrix copula, as in (10a). This reason is the locality of syntactic derivations, expressed through the operation of Relativized Minimality (RM, Rizzi 1990, 2004). Ultimately, the account is in terms of intervention, along the following lines. Suppose that in the CP small clause the object were to move from its merge position within the relative predicate into the EPP position. As illustrated in (11), the subject would be crossed over, giving raise to a straight violation of RM, as both the subject and the EPP position, as well as the object, are positions of the same type with regard to the principle; simply put, in terms of the A/A’ 9. I explicitly assume that the EPP position within the CP is not a criterial position in the sense assumed for the subject position of TP, the position labeled Subj in Rizzi and Shlonsky (2007), following Cardinaletti (2004), from which movement is blocked under the criterial freezing principle motivated in their work.

Revisiting the CP of Clefts

99

distinction, they may be assumed to be all A positions. Hence, EPP cannot be satisfied by the object, as indicated in the schematic derivation in (11):

*

(11)

[TP . . . [. . . [FocP . . . [vP be [CP Force [CP EPP [FinP che [TP S . . . . . . O ]]]]]. . . This in turn implies that the object cannot move to the vP peripheral focus position, and consequently cannot be interpreted as the focus of new information. Hence, an object cleft cannot function as a possible answer to a question of information, but only a subject cleft can, as the contrasts in (8) have shown. Related to the informational/discourse issue, one may ask whether it could not be possible for the object to directly move to the vP peripheral focus position in the matrix clause, in a structure where the CP is just a reduced CP with no EPP feature, like the one in (10) b. The corresponding derivation schematically illustrated in (12) is, however, also ruled out on locality grounds: (12)

*

[ TP . . . . . . [ FocP [vP be [CP Force . . . [FocP . . . [FinP che [TP S . . . O. . .]]]“. . .” Phase theory (Chomsky 2005) explicitly rules out the possibility of such a long direct movment with no intermediate steps (within the CP), with the embedded CP sent to spell out. However, no intermediate step is possible in this case as the reduced CP complement of the copula does not contain an escape hatch edge position which could be used to exit the CP. If the CP is reduced at the level of FocP, as is likely to be the case (3.2 below), it would not contain a relevant edge position different from the criterial/interpretable Spec/FocP position, from which movement is excluded in principle through any version of criterial freezing (Rizzi 2006; Boˇskovi´c 2007). Furthermore, should the vP peripheral Focus position be considered an A position, the long movement would again be ruled out as an intervention effect induced by the embedded subject. In conclusion, in object clefts the copula cannot select for a CP small clause with an active EPP feature to be checked by the object for principled locality reasons linked to intervention, nor can the object of the relative predicate contained within the reduced CP move directly to the vP peripheral new information focus position in the matrix clause, again for ultimately locality reasons related to phase theory and, possibly, intervention again. The result is that it is for principled reasons that the analysis in (10a) may only concern subject clefts, whereas object clefts solely correspond to the an analysis along the lines in (10b). This

100

Adriana Belletti

has the desired consequences for the different discourse value of subject and object clefts discussed. In the next section new issues and some further consequences derivable from the proposed account are presented. 3.

Some Further Theoretical and Descriptive Issues

3.1. On Extraction from the Reduced CP Is extraction from the reduced CP complement of the copula always excluded or are there other movements possible? For instance, could movement affect the object in the relative predicate and move it into the matrix CP? The answer to this question is straightforwardly no: as in the case just discussed in (12), direct movement from inside the relative predicate of the cleft into the matrix clause is excluded on locality grounds, no matter where the moved object lands, be it inside the matrix clause (e.g. into the vP peripheral focus position as discussed in relation to (12)) or in its left periphery. The question, however, deserves special further attention, as different kind of sentences apparently implicating movement of the object appear to be possible. In sentences such as (13), movement of the object seems able to affect the object from the clefted position, the left peripheral focus position of the reduced CP. In the well-formed Italian sentences in (13), the clefted object seems to be able to undergo wh-movement (13a) or focus movement (13b) into the left periphery of the matrix clause, from the focus position of the cleft: (13)

a. b.

Chi e` Who is GIANNI Gianni

che Maria ha salutato –? that Maria has greeted – e` che Maria ha salutato – is that Maria has greeted –

At first sight, a possible derivation of (13a,b) is the one given in (14), modulo the different landing site of the moved DP in the two cases, which are identified with a single position in (14) to make the representation easier to read: (14) [CP [FocP / Wh [TP . . . [vP e` [CP Force . . . [FocP chi/GIANNI. . . [ FinP che [TP S. . . – ]]]. . . The derivation in (14) is, however, highly problematic. If, as illustrated in (14), movement can affect the peripheral focussed phrase from the reduced CP, as this CP only contains interpretable criterial positions by hypothesis, such movement should not be possible at all. Under the criterial freezing approach mentioned

Revisiting the CP of Clefts

101

above, a derivation like (14) should not be a viable option. I follow here Rizzi’s (2010) recent proposal on the question. According to this proposal, in cases like those in (13) movement does not actually affect the DP within the Focus Phrase as in (14), precisely because this movement would be in clear violation of criterial freezing. Movement to the matrix left periphery may rather target here a larger constituent corresponding not just to the DP in Focus, but the whole FocP. Moving a larger constituent than the one which satisfies a criterion (here the Focus criterion in the embedded reduced CP) is compatible with the freezing principle (Rizzi 2006, 2010 for detailed discussion). In order to make this movement possible, Rizzi assumes that first the relative predicate of the CP is extraposed, and then the whole remnant phrase containing the Focus Phrase, is moved to the relevant position into the matrix CP. The relevant steps of the assumed derivation are illustrated in (15): (15)

[CP [FocP/Wh [TP . . . [vP e` [FocP chi/GIANNI . . . [FinP che [TP S. . . – ]]] Extraposition of relative predicate10 [CP [FocP/Wh [TP . . . [vP e` [FocP chi/GIANNI . . . ] [FinP che [TP S. . . – ]

Movement of the remnant: [CP [FocP/Wh [FocP chi/GIANNI . . . ] [TP . . . [vP e` ] [FinP che [TP S. . . – ]

In conclusion, movement is possible from the reduced CP of clefts; however, movement may only apparently take place from the focus position; in compliance with criterial freezing, it is the whole Focus Phrase containing the focussed clefted object which moves. This answers the question raised at the beginning of this section. Before concluding this discussion, a further issue deserves some close attention, related to the analysis just proposed. A crucial step of this analysis is the extraposition of the relative predicate of the cleft; it is thanks to this operation that the subsequent remnant movement of the whole Focus Phrase can be performed, yielding the apparent effect of movement of just the focussed clefted object. But why should extraposition occur? Rizzi (2010) shows that the extraposition process is possible in (object) clefts, and provides examples as those in (16) in Italian to show this possibility. In the case of (16b), the adverb “oggi” is interpolated between the clefted object and the relative predicate: 10. For the sake of clarity, the extraposition operation is indicated as rightward movement. A different analysis in the spirit of Kayne (1994) can be formulated in terms of leftward movement and subsequent movement of the remnant. Various possibilities come to mind as to the possible landing site of the extraposed constituent, but they would take the discussion too far afield and are left open for future research.

102 (16)

Adriana Belletti

a. b.

E’ (it) is E’ (it) is

Gianni Gianni Gianni, Gianni,

che devo that I have oggi, che today, that

incontrare oggi. to meet today devo incontrare. I have to meet (Rizzi 2010: 75–77)

The important question to ask is whether the extraposition of the relative predicate is just a option, as (16b) shows, so that it is made appeal to when the freezing principle is at stake, or whether the process is somehow more general, possibly inherently linked to the cleft computation. The latter possibility would be much more interesting, as it would allow one to conclude that nothing special should concern structures such as those in (14) where movement from the reduced CP is performed, as far as extraposition is concerned. Converging evidence to the effect that extraposition of the relative predicate of clefts is not an isolated phenomenon is interestingly indicated by the following contrasts in West Flemish, brought to my attention by Liliane Haegeman (who has provided the examples): (17)

a.

Het is Val`ere niet geweest die dat gezegd heeft. it is Val`ere not been die that said has b. *Het is Val`ere die dat gezegd heeft niet geweest c. *Het is Val`ere niet die dat gezegd heeft geweest

The ungrammaticality of both (17b,c) indicates that, irrespective of the position of the negation, the relative clause predicate cannot be left in situ in the cleft (here a subject cleft), but must be extraposed to a position which ends up following the clause final past participle of the copula. It is tempting to interpret these West Flemish data as a clear indication of the widespread occurrence of extraposition in clefts, possibly a process which always occurs in these structures. I leave open to further elaboration the question as to why this should be the case and what information/discourse or purely formal factors tightly linked to clefts should trigger the extraposition process.11

11. That extraposition of the relative predicate occurs in clefts is proposed on semantic grounds in Percus (1996), see also Hedberg (2000). Relevant in this regard is also the consideration of the nature of the subject of the matrix clause in clefts – it in English – whose expletive nature has been recently put into question in various respects in Reeve (2010); see also Moro (1997) for relevant converging reasoning on French ce of French clefts. It is tempting to somehow relate, at least in part, the necessity of extraposition to the presence of the “quasi”-expletive in clefts. A research program that I will undertake in future work.

Revisiting the CP of Clefts

103

3.2. How Much Reduced is the CP? In order to answer the question of the amount of reduction the CP of clefts manifests, the discussion will be centered on object clefts, that it is to say on contrastive/corrective clefts.12 As for the reduced CP of object clefts (and contrastive/corrective subject clefts), one question to ask is whether the reduced CP contains an active Topic position or not, given the assumed left peripheral map. Even if the CP is reduced at the level of FocP, still the Topic position which is located below FocP should be present in the structure. We start by considering the contrasts in (18), which clearly indicate that clefting a Topic is not an available option. Not even if the Topic is a contrastive Topic as in (18b) (in italics, for convenience). Only a corrective Focus ((18a)) ca be expressed in this structure. Interestingly, in the very same discourse/pragmatic context, a contrastive Topic (in the sense of Bocci (2009), Beninc`a (2001), realized in a Clitic Left Dislocation structure) is possible as shown in (18c), which although similar to (18b), crucially, does not contain the copula: (18)

A. B.

Gianni ha comprato un tavolo e una sedia. ‘Gianni has bought a table and a chair.’ a. No, e` il TAVOLO che ha comprato; (la sedia l’ha avuta in regalo). ‘No, it is the TABLE that he has bought (the chair, they offered it to him).’ b. *No, e` il tavolo che l’ha comprato, la sedia l’ha avuta in regalo. ‘No, it is the table that he he has bought it, the chair, they offered it to him.’ c. No, il tavolo l’ha comprato, la sedia l’ha avuta in regalo. ‘No, the table he has bought it, the chair, they offered it to him.’

Hence, despite the possible availability of the position in the structural space, the Topic position in the reduced CP selected by the copula cannot host a clefted constituent. This is coherent with the often observed fact that clefting strictly 12. The natural assumption is made that the CP of subject clefts has the same shape, modulo presence of the EPP feature. As both subject and object clefts can be contrastive/corrective as discussed, all other things being equal, reference to contrastive/ corrective clefts makes the discussion wider in principle (than just considering new information subject clefts only).

104

Adriana Belletti

correlates with focalization, as it ultimately is a form of focalization. In order to express this crucial property of clefts, it can be proposed that the copula in clefts selects the embedded Focus head, which is then active and needs to be realized/checked; a constituent must then be attracted into its Spec. Hence, in the reduced CP of the copula one constituent must be focalized. This is precisely the clefted constituent.13 Once the selection requirement is fulfilled, one may wonder whether some other constituent can realize the Topic position, which should be present below the Focus head.A sentence like (19) indicates that this possibility can be realized: (19)

E’MARIA che il libro l’ha comprato (non Gianni). ‘It is MARIA that the book she has bought it (not Gianni).’

Let us comment on (19) a bit further and consider first, the respective order of the constituents in Focus and Topic, and second, the position of the complementizer in the CP. As far as the respective order of the Focalized and theTopicalized constituents is concerned, (19) displays the order with Focus higher than Topic. The opposite order is totally excluded, as indicated in (20): (20)

*E’ Il libro, MARIA che l’ha comprato (non Gianni). ‘It is the book, MARIA that she has bought it (not Gianni).’

The total exclusion of (20) strongly suggests that the Topic position above Focus is not present at all, thus arguing for the hypothesis that the reduction of the CP complement of the copula does not include positions higher than Focus. This conclusion is coherent with the idea just discussed that the copula selects Focus. As for the co-occurrence of the corrective/contrastive focus and the left dislocated topic in the left periphery, (19) corresponds to a sentence like (21), where the same corrective/contrastive focalization does not involve clefting but takes place in a simple declarative: (21)

MARIA il libro l’ha comprato (non Gianni). ‘MARIA the book she has bought it (not Gianni).’

13. In new information subject clefts, in which the relevant focus head is the vP peripheral one in the matrix clause, it can be proposed that the selection relation is re-established by the movement of the copula into the matrix T, as the embedded CP does not contain an active Focus head in this case, but rather the active head is the one with the EPP feature.

Revisiting the CP of Clefts

105

Notice however, that in (21) the respective order of the corrective/contrastive focalized phrase and the left dislocated topic phrase can be reversed as illustrated by the perfect status of (22): (22)

Il libro, MARIA l’ha comprato (non Gianni). ‘The book, MARIA has bought it (not Gianni).’

The possibility of (22) contrasts with the complete ungrammaticality of (20). The strong impossibility of the order Topic Focus in (20), can then indeed be considered a direct consequence of the reduced space available in the reduced CP of clefts. Thus, given the two Top positions surrounding the Focus head in a full fledged left periphery, we can conclude that the reduced CP of clefts does not contain the highest one above Focus. This in turn strongly suggests that the CP is reduced at the level of the projection of the left peripheral Focus.14 Let us now turn to the second point raised. There is an obvious question which needs to be addressed in connection with (19): if on the one hand, the fact that the order Focus-Topic is instantiated in the CP of clefts entitles us to assume that the CP is reduced at the level of the Foc layer, as suggested on the basis of the contrast between (19) and (20), on the other hand, the position of the complementizer is not the one expected in the well-formed (19). As “che” has been analyzed here as the realization of Fin, all things being equal, the complementizer should be the last element in the CP, linearly following both the Focalized and the Topicalized constituents; it is however situated between Foc and Top in (19). Although a fully structured answer to this descriptive question would go beyond the aim of this article, the following considerations can be offered to indicate a line of interpretation for this otherwise unexpected word order. Building on the proposal in Belletti (2008), it is tempting to suggest that the unexpected position of the complementizer in (19) follows from the fact that the complementizer moves from the Fin head into the highest head in the CP; as proposed, this movement is assumed to always affect complementizers of this type.15 In a full fledged CP, the highest head position is the Force head. 14. Reduction can be expressed in terms of the truncation hypothesis first developed in Rizzi (1993/94). In Belletti (2009: Chapter 11) I have proposed, following Kayne (2005) and his extension of Rizzi’s truncation idea, that elements filling the highest Spec position in the clause, – the edge – are left unpronounced. If the Focalized element in the reduced CP of clefts fills indeed the highest Spec position, hence the edge, one should conclude that the principle of unpronounceability could not affect elements in Focus; this might follow from some version of full interpretation. 15. This movement should hold in languages where the same C element realizes both the content of Fin and the content of Force. See the reference quoted for further details.

106

Adriana Belletti

In the reduced CP of clefts that we are investigating, it corresponds to the Focus head. Hence, once moved into the focus head, the complementizer ends up in the position between Focus and Topic. It can be speculated that movement from Fin into the highest head is generally triggered by the necessity to check selection: in a declarative full fledged CP, the selectional properties are generally expressed by the Force head; in the CP of clefts they are expressed by the Focus head, as the copula selects Focus, as proposed above. Converging evidence that C always fills the highest head, thus locally expressing the selectional properties of the matrix verb, is provided by the following paradigm in (23): (23)

a.

Penso che il libro, MARIA, a Gianni non glielo abbia ancora dato (non Francesca). ‘I think that the book MARIA, to Gianni has not yet given it to him (not Francesca).’ b. ?Penso il libro, che MARIA, a Gianni non glielo abbia ancora dato (non Francesca). ‘I think the book that MARIA, to Gianni has not yet given it to him (not Francesca).’ c. *Penso il libro, MARIA che a Gianni non glielo abbia ancora dato (non Francesca). ‘I think the book, MARIA that to Gianni has not yet given it to him (not Francesca).’ d. *Penso il libro, MARIA a Gianni che non glielo abbia ancora dato (non Francesca). ‘I think the book, MARIA to Gianni that has not yet given it to him (not Francesca).’

In all the sentences in (23), the two Topic positions and the Focus position in the left periphery of the complement of the verb pensare (think) are realized. (23a) is a perfect sentence where the expected word order with che in the highest Force is realized. (23b) is possible, although slightly marginal, presumably because the (declarative) force of the complement clause is realized in a non canonical head (the Topic head). (23c), however, where the complementizer follows both the high Topic and the contrastive/corrective peripheral Focus, is not just marginal, it is plainly ungrammatical. (23d), where the complementizer follows lowest Topic as well, presumably remaining in Fin, is also excluded. It is tempting and reasonable to interpret this ungrammaticality as due to selection: the complementizer is located too far from its selecting verb in (23c,d). If this interpretation is on the right track, nothing special to clefts needs to be said on the position of the complementizer in (19), unexpected at first sight, but actually

Revisiting the CP of Clefts

107

in line with the regular syntax of the complemenizer, in combination with the selectional properties of the copula, and the reduced size of the CP. 3.3. The Similarity with Other Structures 3.3.1.The CP Complement of Perception Verbs: Pseudorelatives and Clefts The shape of the reduced CP complement attributed here to subject clefts finds a close analogue with the one of pseudorelatives, the sentential complement of perception verbs (Belletti 2008, 2009). Here, I will just review the main features of the similarity shared by the two structures and then point out some interpretive constraints which follow directly from the general account proposed. The interpretive constraints manifest themselves when new information subject clefts and pseudorelatives are combined in a Question-Answer exchange. The analysis proposed for pseudorelatives, essentially updates the analysis originally proposed by Guasti (1993), under the articulated conception of the CP domain adopted here. According to this analysis, a sentence like (24a) contains a pseudorelative complement of the perception verb “vedere” (see), which is a reduced CP with an EPP feature, the CP small clause illustrated in (24b): (24)

a. b.

Ho visto Maria che parlava con Gianni. ‘I have seen Maria that spoke to Gianni.’ Ho visto [CP . . . [EPP Maria [FinP che [TP (pro) parlava (–) con Gianni]]]].

For concreteness, in (24) I assume that “Maria” is moved to the EPP position in the CP from inside the TP of the predicate, with movement taking place from the postverbal position (as is always the case in Italian, see Rizzi and Shlonsky 2007 and references cited therein), indicated with “–” in (24). As pointed out in the references quoted, a sentence like (24a) can constitute an answer to both question (25a) and question (25b) below; this is so because, differently from subject clefts, pseudorelatives do not necessarily require the focalization of the subject argument, as the focus of new information can also be constituted by the whole clause in pseudorelatives. (25c), illustrates the interpretation of the pseudorelative as an answer to (25a). It is clear from (25c), the close similarity of this type of pseudorelative with subject clefts: (25)

a. b. c.

Chi hai visto (che parlava con Gianni)? ‘Whom have you seen (that spoke to Gianni).’ Che cosa hai visto? ‘What have you seen?’ Ho visto [Foc Maria [vP – [CP . . . [EPP [FinP che [TP (pro) parlava (–) con Gianni]]]]]]

108

Adriana Belletti

As discussed in Guasti (1993) and Rizzi (2000), in the pseudorelative in (25c) there is direct perception of “Maria”. A cleft, on the other side, implies a peculiar semantics which provides a unique identification explicitly expressed by the focussed argument. In subject clefts, the (uniquely) identified argument is the subject. One may now ask: can a subject cleft constitute an answer to a question containing a pseudorelative? In looking for an answer to this question, the language to look at is, once again, French, where a cleft constitutes a privileged way to answer a question on the subject, as discussed in 2.1. The relevant question-answer pair is given in (26): (26)

Q: Qui as-tu vu qui pleurait? ‘Whom have you seen who was crying?’ A: *C’est Marie (que j’ai vu qui pleurait) ‘It is Marie (whom/that I saw who was crying)’

The answer in (26) is not possible, with the indicated meaning16. This is expected under the proposed analysis, as the subject cleft answer implies the vP peripheral focalization in the matrix clause containing the copula, which in turn inevitably yields the violation of Relativized Minimality illustrated in the derivation in (27), due to the presence of the intervening subject of the perception verb (“je”, in the example). * (27)

[TP . . . [FocP [vP be[C [EPP Maria] [FinP que[TP j’ai vu [CP [EPP ] [FinP qui [TP “. . . ”pleurait“. . . ”]]]]]]]]]

In contrast, in a context different from the question-answer one, a sentence like (28), where a contrastive subject cleft is present, is perfectly well-formed. This is also expected, as focalization is here contrastive left peripheral focalization, so that intervention by the subject of the perception verb does not matter. (28)

E’ MARIA che lui ha visto che piangeva. ‘(It) is MARIA that he has seen that was crying.’ [TP . . . [vP be[CP [FOC MARIA] [FinP che [TP lui ha visto [CP [EPP < Maria> [FinP che [TP piangeva. . . ]]]]]]]]]

16. Thanks to Dominique Sportiche for providing the judgment and also for clarifying his intuition in the following terms: (26) is not just the answer to the question asked; a different prosody would make the sentence possible, but this would mean changing the presuppositions, which is not the situation the question-answer pair in (26) refers to.

Revisiting the CP of Clefts

109

3.3.2.A Possible Different Fin: -ing Let us now ask whether there is any analogue of the small clause CP of (subject) clefts, in which an element different from a complementizer is present under Fin. I submit here the proposal that a possible candidate be the -ing gerundival ending found in the complement of perception verbs in English examples like those in (29) (thanks to D. Pesetsky for pointing out the possible connection): (29)

I saw John working with his sister.

Interestingly, a -ing clause is also possible as a subject cleft of the type illustrated in (30). A cleft like (30) can possibly function as a new information subject cleft: (30)

(Who was making this noise?) It was John working with his sister.

Once again the complement of a perception verb and subject clefts appear to be strictly related. If the proposal is on the right track, the derivation of the -ing cleft in (30) should involve both V-to-T-to-C/Fin and the familiar movement of the subject into the EPP position of the CP small clause, and, from there, into the vP peripheral Focus position of the matrix copula (plus the movement of the copula into the matrix T), as in the derivation 1 in (31) below. Alternatively, in the case of a reduced CP with no EPP feature, the sentence should involve direct focalization in the reduced left periphery, the derivation 2 in (31)17 : (31)

1 [. . . [ FocP . . . [vP be [CP FOC // EPP [FinP -ing [ TP John T [work “ . . . ” ]]]]]]] 2

4.

Conclusion

This article has focussed on two main points: the investigation of the way in which the locality of syntactic computations conditions the syntax and the interpretation of clefts; and the way in which the CP domain is shaped in clefts. 17. A possible third alternative analysis exists, in which the whole -ing clause moves into the new information focus position. This would correspond to a reading in which the whole -ing clause, not just the subject is the focus of new information, a possibility available for this type of clefts, similarly to pseudorelatives, as noted in the discussion surrounding (25).

110

Adriana Belletti

The two aspects have been shown to correlate closely, as the locality ban against intervention according to Relativized Minimality, derives the different interpretations of subject and object clefts, under the crucial assumption that the CP space, syntactically analyzed in cartographic terms, is reduced in clefts. The possibly different discourse value of clefts is thus ultimately a consequence of the operation of the general locality principle and of the specific shape of the syntactic structure of clefts. The shape of the CP space of clefts is not isolated, other structures such as the sentential complement of perception verbs share the same type of configuration; furthermore, in both the complement of perception verbs and in clefts the Fin head can host different type of material beside the complementizer; we have speculated that the -ing ending be a possible realization of this head which closes up the CP domain. Potentially problematic derivations have also been discussed, among which a prominent one is the apparent possibility of extraction out of the interpretable criterial positions of the CP of clefts, for which an account making reference to extraposition of the cleft sentence has been suggested. Rather than being just an option made use of in the special extraction conditions, the hypothesis has been entertained that the extraposition process may turn out to be inherently linked to clefts. This hypothesis, which has also sometimes been proposed on semantic grounds, will be further closely investigated in future work. References Abels, Klaus and Muriungi, Peter 2005 The Focus particle in Kˆıˆıtharaka: Syntax and Semantics. Lingua 118: 687–731. Belletti, Adriana 2009 Structures and Strategies (“Answering strategies: New information subjects and the nature of clefts”, Chapter 10). New York: Routledge. Belletti, Adriana 2008 The CP of Clefts. Rivista di Grammatica Generativa 33: 191–204. Belletti, Adriana 2004 Aspects of the low IP area. In: Luigi Rizzi (ed.), The Structure of CP and IP. The Cartography of Syntactic Structures, Vol. 2, 16–51. Oxford: Oxford University Press. Belletti, Adriana 2004 ed. Structures and Beyond. The Cartography of Syntactic Structures, Vol. 3. New York: Oxford University Press.

Revisiting the CP of Clefts Beninc`a, Paola 2001

111

Syntactic Focus and Intonational Focus in the left periphery. In: Guglielmo Cinque and Giampaolo Salvi (eds.), Current Studies in Italian Syntax Offered to Lorenzo Renzi, 39–64. Amsterdam: Elsevier. Beninc`a, Paola and Poletto, Cecilia 2004 Topic, Focus and V2: Defining the CP Sublayers. In: Luigi Rizzi (ed.), The Structure of CP and IP. The Cartography of Syntactic Structures, Vol. 2, 52–75. New York: Oxford University Press. Bocci, Giuliano 2004 Contrastive focalization on topics and preverbal subjects in Italian. Rivista di Grammatica Generativa 29: 3–59. Bocci, Giuliano 2009 On Syntax and Prosody in Italian. PhD dissertation. University of Siena. Bianchi, Valentina and Mara Frascarelli 2009 Is Topic a root phenomenon. Iberia. ˇ Boˇskovi´c, Zeljko 2007 On Successive Cyclic Movement and the Freezing Effect of Feature Checking. In: Jutta Hartmann, Veronika Hegedus and Henk van Riemsdjik (eds.), The Sounds of Silence. North Holland: Elsevier. Burzio, Luigi 1986 Italian Syntax. A Government-Binding Approach. Dordrecht: Reidel. Cardinaletti, Anna 2004 Towards a Cartography of Subject Positions. In: Luigi Rizzi (ed.), The Structure of IP and CP. The Cartography of Syntactic Structures, Vol. 2, 115–165. New York: Oxford University Press. Chomsky, Noam 1977 On Wh-Movement. In: Adrian Akmajian, Peter Culicover and Thomas Wasow (eds.), Formal Syntax, 71–132. Academic Press. Chomsky, Noam 1995 The Minimalist Program. Cambridge Mass: MIT Press. Chomsky, Noam 2005 On Phases. In: Robert Freidin, Carlos Peregr´ın Otero and Maria Luisa Zubizarreta (eds.), Fundational Issues in Linguistic Theory. Cambridge, Mass: MIT Press. Cinque, Guglielmo 2002 ed. Functional Structure in DP and IP. The Cartography of Syntactic Structures, Vol. 1, New York: Oxford University Press. Cinque, Guglielmo and Luigi Rizzi 2010 In: Bernd Heine and Heiko Narrog (eds.), The Cartography of Syntactic Structures. The Oxford Handbook of Linguistic Analysis, 51–65. Oxford/New York: Oxford University Press.

112

Adriana Belletti

Clech-Darbon, Anne, Georges Rebuschi and Annie Rialland 1999 Are there cleft sentences in French? In: Georges Rebuschi and Laurice Tuller (eds.), The Grammar of Focus. Amsterdam/Philadelphia: John Benjamins. Cruschina, Silvio 2006 Informational focus in Sicilian and the left periphery. In: Mara Frascarelli (ed.), Phases of Interpretation, 363–385. Berlin: Mouton de Gruyter. Cruschina, Silvio 2010 Fronting as Focalisation in Sicilian. In: D’Alessandro Roberta, Ledgeway Adam and Ian Roberts (eds.), Syntactic Variation. The Dialects of Italy, 247–260. Cambridge: Cambridge University Press. Grewendorf, G¨unther 2005 The discourse configurationality of scrambling. In: Joachim Sabel and Mamoru Saito (eds.), The Free Word Order Phenomena, 75–136. Berlin: Mouton de Gruyter. Guasti, Maria Teresa 1993 Causative and Perception Verbs. A Comparative Study. Torino: Rosenberg and Sellier. Guesser, S. 2007 Soggetto nullo e focalizzazione del soggetto in Portoghese Brasiliano. Master thesis; University of Siena. Haegeman, Liliane 1996 Verb second, the split CP and null subjects in Dutch finite clauses. GenGenP 4.2: 133–175. Haegeman, Liliane 2004 Topicalization, CLLD and the Left Periphery. In: Benjamin Shaer, Werner Frey and Claudia Maienborn (eds.), Proceedings of the Dislocated Elements Workshop ZAS Berlin, November 2003, 158–192. Berlin: ZAS. Hedberg, N. 2000 The referential status of clefts. Language 76(4): 891–920. Kayne, Richard 1994 The Antisymmetry of Syntax. MIT Press, Cambridge Mass. Kayne, Richard 2005 On Parameters and on Principles of Pronounciation. In: Hans Broekhuis, Norbert Corver, Riny Huybregts, Ursula Kleinhenz and Jan Koster (eds.), Organizing Grammar. Linguistic Studies in Honor of Henk van Riemsdijk, 289–299. Berlin: Mouton De Gruyter. ´ Kiss, Katalin E. 1998 Identificational Focus versus Information Focus. Language 74(2): 245–273.

Revisiting the CP of Clefts Mioto, Carlos 2003 Moro, Andrea 1997 Percus, Orin 1997 Reeve, M. 2010 Rizzi, Luigi 1990 Rizzi, Luigi 1993/94 Rizzi, Luigi 1997 Rizzi, Luigi 2000

Rizzi, Luigi 2004

Rizzi, Luigi 2004 ed. Rizzi, Luigi 2006

Rizzi, Luigi 2010

113

Focaliza¸ca˜ o e quantifica¸ca˜ o. Revista Letras (Curtiba) 61: 169–189. The Raising of Predicates. Cambridge: Cambridge University Press. Prying open the cleft. In: Kiyomi Kusumoto (ed.), Proceedings of NELS 27, 337–351. Amherst, MA: GLSA. The syntactic structure of English clefts. Lingua, doi: 10.1016/j. Relativized Minimality. Cambridge, MA: MIT Press. Some notes on linguistic theory and language development: The case of root infinitives. Language Acquisition 3: 371–393. The Fine Structure of the Left Periphery. In: Liliane Haegeman (ed.), Elements of Grammar, 281–337. Dordrecht: Kluwer. Direct perception, government and thematic sharing. In: Luigi Rizzi Comparative Syntax and Language Acquisition, 189–210. London/ New York: Routledge. Locality and the Left Periphery. In: Adriana Belletti (ed.), Structures and Beyond. The Cartography of Syntactic Structures, Vol. 3. New York: Oxford University Press. The Structure of CP and IP. The Cartography of Syntactic Structures, Vol. 2. New York: Oxford University Press. On the Form of Chains: Criterial Positions and ECP Fffects. In: Lisa Lai-Shen Cheng and Norbert Corver (eds.), Wh-Movement: Moving on, 97–134. MIT Press.

Some consequences of Criterial Freezing. In: Peter Svenonius (ed.), Functional Structure from Top to Toe. New York: Oxford University Press. Rizzi, Luigi and Ur Shlonsky 2007 Strategies of Subject Extraction. In: Uli Sauerland and Hans-Martin G¨artner (eds). Interfaces + Recursion = Language?, 115–160. Mouton De Gruyter. Rothstein, Susan 2000 Predicates and Their Subjects. Kluwer.

114

Adriana Belletti

Ruwet, Nicolas 1975 Starke, Michal 1995

Stowell, Tim 1983

Les phrases copulatives en fran¸cais. Recherches Linguistiques 3: 143– 191. On the Format of Small Clauses. In: Anna Cardinaletti and Maria Teresa Guasti (eds.), Small Clauses, Syntax and Semantics, Vol. 28, 237–269. New York: Academic Press. Subjects across categories. The Linguistic Review 2.3: 285–312.

Delimitation Effects and the Cartography of the Left Periphery Luigi Rizzi

1.

Introduction

Half a century of formal syntactic studies has brought to light the complexity and richness of syntactic structures. The cartography of syntactic structures is the line of research which addresses this topic: it is the attempt to draw maps as precise and detailed as possible of syntactic configurations. The cartographic projects started, over a decade ago, as an attempt to provide fine descriptions of certain zones of the syntactic tree in some Romance and Germanic languages, but they immediately showed a universal dimension, and were quickly extended to many other language families. In this paper I would like to illustrate some of the results of the cartographic studies in connection with the left periphery of the clause, and discuss the implications of this line of research for the study of the interfaces connecting syntax with the systems of sound and meaning. In the last part I will show how cartographic maps of the left periphery interact with classical topics of syntactic research such as the theory of locality and the freezing effects. 2.

The Criterial Approach to Scope-Discourse Semantics

Economy conditions on movement in the Minimalist Program limit the application of movement to configurations in which it yields some effect on the interpretive interface (Reinhart 2005). This is rather straightforward for one important class of cases of movement, displacing elements to the initial periphery of the clauses, and yielding A -movement chains: Consider some typical A -constructions in English, such as the following clauses: (1)

a. b.

Which book should you read ? The book which you should read

is here.

116

Luigi Rizzi

c. d.

This book, you should read . (It is) THIS BOOK (that) you should read

(not Bill’s book).

It is intuitively clear what movement to the front does in these cases. In (2a) it puts an interrogative operator, the wh-phrase, in its proper scope position to yield the appropriate logical form “for which x, x a book, [you should read x]”; similarly, the moved element is interpreted as a relative operator (the book x such that [you should read x]) in (2b), and analogous operator-variable structures are assigned to exclamatives, comparatives and other constructions involving the left periphery of the clause. So, in these cases the relevant syntactic element which book, etc. receives two interpretive properties: the argumental thematic role of patient of read, and the role of interrogative operator taking scope over the whole clause. Cases (2c) and (2d) are slightly different: here the preposed element expresses the discourse-related property of being the Topic, or the (contrastive) Focus of the structure, respectively, properties relevant for the “information packaging” of the expression (Vallduv´ı 1992), and its usability in discourse. Putting together all these cases under a synthetic label, I will follow Chomsky (2004) and call the interpretive properties associated to the initial position properties of “scope-discourse” semantics: the scope of operators and the discourse-related properties expressing the informational articulation of the structure. So, these cases of movement connect two positions: one dedicated to argumental semantics, the thematic role, or, more generally, the position in which an element is semantically selected (or S-selected: for instance, a time adverbial is S-selected by a T head, an aspectual adverbial is S-selected by the appropriate Asp head, etc. in an approach to adverb syntax like the one developed by Cinque 1999); and one dedicate to scope-discourse semantics. What does it mean for a position to be “dedicated” to a certain interpretation? In the case of argumental semantics this is straightforward and uncontroversial: argumental roles are assigned by certain lexical heads, typically verbs, to their immediate dependents. So, the verb read assigns the role “patient” to its complement and the role “agent” to its specifier (perhaps through the mediation of a “light verb” v). As for scope-discourse semantics, I would like to assume an approach which generalizes the same basic mechanism to it: there is a dedicated system of functional heads, typically in the left periphery of the clause, which assigns to its dependents such properties as “scope position and scope domain of such and such type of operator”, “Topic and Comment”, “Focus and Presupposition”, etc. So the assignment of both kinds of interpretive properties is uniformly a matter of head-dependent relations. This structural view of scope, topicality and focus is sometimes called the “criterial approach” to scope-discourse semantics:

Delimitation Effects and the Cartography of the Left Periphery

(2)

117

The criterial approach: there is a system of dedicated functional heads, typically expressed in the initial periphery of the clause, attracting elements to their Spec’s and signalling to the interface components the basic scope-discourse properties. (Rizzi 1991, 1997)

This amounts to saying that sentences (1) should have representations like the following, with heads such as Q, Top, etc., which attract the wh-operator, the Topic, etc., to their specifier (in these representations I have expressed the gap left by movement as a full unpronounced copy of the moved phrase, as in the “copy theory” of traces): (3)

a. b. c. d.

Which book The book which This book THIS BOOK

Q R Top Foc

should you read ? you should read you should read you should read not Bill’s book

In English these heads are silent, they do not correspond to any overt morpheme (except for the fact that Q in main questions attracts the functional verb, in a case like (3a) the modal, so that the position ends up being filled in this case). There is straightforward comparative evidence which makes the hypothesis expressed by (2)–(3) immediately plausible: in many languages this system of left peripheral heads is pronounced, expressed by overt particles: (4)

a.

b.

(5)

a.

b.

gezien heeft]]. Ik weet niet [wie of [Jan seen has I know not who Q Jan (Dutch varieties, Haegeman 1996) gfundn hot]]. Der Mantl [den wo [dea Hons found has The coat which R the Hans (Bavarian, Bayer 1984) ` s`e [ Un d`o [ d`an I heard that snake (Gungbe, Aboh 2004) ` s`e Un [d`o [d`an I heard that snake (Gungbe, Aboh 2004)

l´o y`a [K`of´ı h`u `ı]]]. the TOP Kofi killed it l´o w`e [K`of´ı h`u ]]]. the FOC Kofi killed

(4a) shows that the Q head is overtly expressed as of (if) in certain (non-standard) varieties of Dutch. Relative clauses are marked in Bavarian by the special complementizer wo, homophonous to the wh-element for where, and which can co-occur with the relative operator (the D-pronoun den in (4b) its Spec.The

118

Luigi Rizzi

topic and focus heads are expressed by the particles y`a and w`e , respectively, in Gungbe, as shown in (5). The A criteria were originally formulated as representational principles in terms of the GB framework: a phrase endowed with the appropriate A feature must be in a Spec-head configuration with a left-peripheral head endowed with the same feature; the configuration must be met at LF uniformly across languages as it is crucial for the interpretation of such structures; particular languages may require its satisfaction already at S-structure, a matter of parametric variation; in these cases, movement to the left periphery will have to be overt (Rizzi 1991). In derivational terms more congenial to the minimalist conception of syntactic computations, the syntactic role of the criteria can be recast as the capacity that criterial heads have of attracting phrases endowed with matching criterial features (see Aboh 2007 for an implementation along these lines): (6)

XF attracts XPF to its Spec, for F = Q, R, Top, Foc, Excl, . . .

I leave open here the question of whether attraction involves a system of uninterpretable features, as in the detailed implementation proposed by Pesetsky and Torrego (2001) (see Rizzi 2006, 2009 for a discussion of the possible role of purely formal, uninterpretable counterparts of criterial features in an approach to A syntax based on criteria). I will also leave open here the issue of the in situ strategy for wh and other A constructions, which may involve covert movement or the establishment of a pure Agree relation from the criterial head to the relevant criterial phrase. In addition to their strict syntactic role of triggering movement, criterial heads have another and equally important function: they are visible to the interface systems of sound and meaning to signal in a transparent manner the interpretive properties of scope-discourse configurations. So, for instance, the criterial head Top goes with an interpretive instruction of the following kind: (7)

[XP [Top YP ]] Topic Comment

I.e., “my Spec is to be interpreted as the topic, and my complement as the comment made about it”, where a topic expresses a referent selected among those familiar from context, and about which a comment is made (with the possibility of further subspecifications, characterizing a possible family of topics: Beninc`a and Poletto 2004; Frascarelli and Hinterhoelzl 2007; Bianchi and Frascarelli 2009; see Rizzi 2010b for a more detailed discussion of the interpretive properties of Topics). Similarly for the Foc head:

Delimitation Effects and the Cartography of the Left Periphery

(8)

119

[XP [ Foc YP ]] Focus Presupposition

I.e., “my Spec is to be interpreted as focus and my complement is the presupposition”, where focus means that XP provides the value of a variable in the presupposition, a value that is assumed to be new information for the interlocutor. Here too, there can be additional specifications which define different types of foci, for instance contrastive focus as opposed to simple new information focus. E.g., when I utter (1d), I assume not only that you know that you should read something (presupposition), and I convey the information that the thing you should read is this book (new information), but I also assume that this new information falls outside what I take to be your natural expectations, with which the new information I am providing is contrasted (so, this contrastive focal structure is very naturally concluded by a negative tag explicitly excluding the information imputed to the interlocutor’s system of beliefs, i.e., in the case of (1d), Bill’s book). Clearly, languages assign distinct focal properties to different positions, in ways that are parametrised in part. In Romance languages like Standard Italian, the clause-initial focal position is dedicated to contrastive focus, while in the Sicilian dialects and in the regional varieties of Italian spoken in Sicily the initial focus position can be used for simple new information focus (Cruschina 2008), hence be used to answer a simple wh-question. Criterial heads also guide the interpretation of the sentence at the interface with the sound system, much as they guide the interpretation at the interface with semantics and pragmatics. Consider for instance the careful experimental study of the pitch contour of Italian Topic and Focus constructions conducted by Bocci (2009). The following figures (Giuliano Bocci’s courtesy) illustrate the two contours: (9)

Pitch contour of Topic – Comment (from Bocci 2009) A Michelangelo (Top), Germanico vorrebbe presentare Pierangela.) ‘To Michelangelo (Top), Germanico would want to introduce Pierangela.’

450 400

Pitch (Hz)

300 200 100 H+ a mi he

+L* lan

L-L%

A Michelangelo 0

H+ +L*

de lo der

ma

H+ +L* ni ho vo

Germanico Time (s)

H+

rε be pre zen

vorrebbe

+L* ta

presentare

H+ +L* re pje

L-L%

ran de la

Pierangela 3.7942

120

Luigi Rizzi

(10)

Pitch contour of Focus – Presupposition (from Bocci 2009) A MICHELANGELO (Foc) Germanico vorrebbe presentare Pierangela. ‘TO MICHELANGELO (Foc) Germanico would want to introduce Pierangela.’

550 500 400

Pitch (Hz)

300 200 100 L+ +H* a

mi

he

lan

Lde lo

A MICHELANGELO 0

L* der

ma

L* ni ho

Germanico

vo rεb be pre

vorrebbe Time (s)

zen

ta

presentare

L* re pje

ran

L-L% de la

Pierangela 3.61823

The pitch contours of Topic and Focus are distinct (with contrastive Focus receiving a higher prominence than Topic, Bocci 2009), and the contours of Comment and Presupposition are sharply different, with a complete flattening of the contour of the presupposition in (9), which contrasts with the highly articulated contour of the comment in (8). Again, we can think of the criterial heads as giving instructions to the prosodic component to yield the different types of contours associated with the different discourse functions, much as they do in the interpretive components on the meaning side, as in the model that Bocci (2009) adopts and develops. In conclusion, criterial heads have a syntactic function, attracting elements to the periphery of the clause, and interpretive functions on both interfaces with sound and meaning, triggering certain semantic-pragmatic routines expressing the informational organization of the clause, and the assignment of specific pitch contours which will make the relevant interpretive properties salient and immediately detectable from the speech signal (see also the discussion in Frascarelli 2000). This way of looking at the expression of scope-discourse properties has several advantages over imaginable alternatives. On the one hand, rather than assuming a proliferation of different devices for the expression of scope-discourse semantics, it assumes a uniform mechanism to hold across languages: scope-discourse properties are expressed by a system of dedicated functional heads acting upon their immediate dependents, much as thematic properties are expressed by a system of lexical heads (or low functional heads of the v type). An elementary but salient parametric property here has to do with whether such heads are overt, expressed by pronounced morphemes,

Delimitation Effects and the Cartography of the Left Periphery

121

or not (essentially the same kind of elementary spell-out parametrisation which has been assumed for Case systems ever since the Chomsky-Vergnaud theory of syntactic Case in GB); there are of course other parameters of familiar kinds (possibly, overt or covert movement to the Spec’s of criterial heads, certain permissible variations in the hierarchical order, etc.), but the fundamental computational mechanisms and configurations dealing with the syntax-pragmatic interface can be assumed to be essentially uniform across languages. Moreover, movement to a Topic or Focus position is like other kinds of movement, internal Merge triggered by a search activated by certain morphosyntactic features, like object movement in passive, wh-movement, clitic movement and the like. There is no need to postulate a separate, special category of “prosodically driven” movement, hence no reason to enlarge the class of computational options admitted by Universal Grammar. And there is no need to postulate a direct link between intonational structure and pragmatics: the connection is fully mediated by syntax, through the interface role of criterial heads. The criterial approach makes things fully parallel for scope-discourse semantics and argumental semantics: in the latter case, it is generally assumed that the connection between the thematic role that a nominal expression receives and the position in which it is pronounced are mediated by syntax, without the need of establishing any additional, syntax-independent, link between interpretive roles at LF and positions of pronunciation at PF. So, in terms of the criterial approach we can stick to the simplest assumption for both argumental and scope-discourse semantics: representations of sounds and meanings are fully mediated by syntax, no independent and extra-syntactic connecting path needs to be postulated. 3.

The Cartography of Syntactic Structures

In many languages, Topic and Focus can co-occur, often in a fixed order, illustrated by the following Gungbe example: (11)

. . . d`o K`of´ı y`a g`ankp´a m`e w`e kp`on`on l´e s´u- `ı . . . that Kofi Top PRISON IN Foc policemen Pl shut him d´o. there (Gungbe: Aboh 2004)

Hence, cartographic issues arise at this point: we want to know what global configurations the left periphery of the clause can assume, what properties remain constant across languages and what other properties are submitted to parametric variation.

122

Luigi Rizzi

The observation that dedicated scope-discourse positions can co-occur with ordering restrictions was one of the triggering factors of the cartographic projects. Around the mid-nineteeneighties the clausal structure was assumed to result from the hierarchical organisation of three X-bar layers headed by the verb, the inflection node, and the complementizer, respectively: (12)

[CP . . . C . . . [IP . . . I . . . [VP . . . V . . . ]]] (Chomsky 1986)

But it soon became very clear that this picture was oversimplified. Examples like (11) show that more elements can co-occur in the left peripheral pre-subject position of the clause; this fact per se may still be expressed with a simple representation like (12) through the assumption that multiple adjunctions to the IP are allowed, as in a traditional line of analysis of topicalisation. But this assumption is at odds with various properties: a. the rigidity of certain orders as in (11), while a multiple adjunction analysis would typically predict free ordering; b. the presence in many languages of overt Top and Foc heads, whose distributional properties must be deferred to the interface systems in an adjunction analysis; c. the uniqueness of Topic and Focus in cases like (11): again, under an adjunction analysis one would expect free recursion, unless some special constraint is added to the adjunction mechanism, and/or to the interface systems. Similar considerations of word order, co-occurrence of different kinds of elements (adverbials and other elements), and properties of the morpho-syntax had already led to the splitting of the I node into more elementary components (Pollock 1989, Belletti 1990). The special attention to such constraints led to the conclusion that each layer of (12) is an abbreviation for a much richer structural zone. The cartography of syntactic structures is the attempt to build detailed maps of each structural zone. If the structural zones may be complex, the syntactic atoms are remarkably simple and uniform: the structures are built through successive applications of Merge, through which a head is combined with a complement and a specifier, thus projecting a phrase. If the fundamental geometry of the building block is always the same, the rich articulation of the structures is due to the richness of the inventory of functional heads: the functional heads are much more numerous than one would have assumed twenty five years ago, and the possible functional structures, resulting from the combinations of functional heads according to their selectional properties, are correspondingly more articulated.

Delimitation Effects and the Cartography of the Left Periphery

123

So, if we take the magnifier and draw maps expressing the fine details of structures, we find the same basic configurations and shapes that we observe through a first-pass, “naked eye” survey. In this sense, syntactic structures are like crystals, with the same shapes showing up at the macroscopic and microscopic levels. Cartographic projects started with detailed descriptions of the IP and CP systems in some Romance and Germanic languages (Rizzi 1997, 2004b; Cinque 1999, 2002; Belletti 2004, 2009; Grewendorf 2002; Haegeman 2003, 2006), but they quickly showed a general dimension, triggering much work on different languages and language families: Finno-Ugric (Puskas 2000), Semitic (Shlonsky 2000), Slavic (Krapova and Cinque 2004), West African (Aboh 2004), Bantu (Biloa 2008), Creole (Durrleman 2008), East-Asian (Tsai 2007; Endo 2007; Saito 2012), Dravidian (Jayaseelan 2008), Austronesian (Pearce 1999), Classical languages (Salvi 2005), etc. 4.

The Fine Structure of the Left Periphery

Going back to the CP system, the motivations for splitting it into a sequence of functional heads were twofold. First, different C-like particles occupy distinct syntactic positions, as is shown by their ordering with respect to other elements: for instance, in my variety of Italian the finite declarative complementizer che must precede topics, while the infinitival (prepositional) complementizer di necessarily follows topics (che Top /Top di). This suggests that che and di occupy distinct positions in the C space (Rizzi 1997), while both being part of the complementizer system (the complementizer status of di is well motivated for reasons discussed in Kayne (1983), Rizzi (1982) and much subsequent work). (13)

a. *Penso, a Gianni, che gli dovrei parlare. ‘I think, to Gianni, that I should speak to him.’ b. Penso che, a Gianni, gli dovrei parlare. ‘I think that, to Gianni, I should speak to him.’

(14)

a.

Penso, a Gianni, di dovergli parlare. ‘I think, to Gianni, ‘of ’ to have to speak to him.’ b. *Penso di, a Gianni, dovergli parlare. ‘I think ‘of ’, to Gianni, to have to speak to him.’

Che and di cannot co-occur, as they are specialized for finite and non-finite clauses, respectively; but sometimes delimiting complementizer particles can

124

Luigi Rizzi

co-occur, before and after the topic-focus field, as particles mai and a in Welsh, according to Roberts’ (2004) analysis: (15)

Dywedais i [ mai ‘r dynion fel arfer a Said I C the men as usual C [werthith y ci ]]. will-sell the dog (Welsh: Roberts 2004)

Back to the Italian periphery, it can be observed that the interrogative complementizer se (if), marking yes/no embedded questions, seems to occupy an intermediate position between che and di in that it can be both preceded and followed by a topic: (16)

a. b.

Non so, a Gianni, se gli potremo parlare. ‘I don’t know, to Gianni, if we could speak to him.’ Non so se, a Gianni, gli potremo parlare. ‘I don’t know if, to Gianni, we could speak.’

On the other hand, se must be high enough to obligatorily precede a focus position: (17)

a.

Non so se proprio QUESTO volessero dire (e non qualcos’altro). ‘I don’t know if exactly THIS they wanted to say (and not something else).’ b. *Non so proprio QUESTO se volessero dire (e non qualcos’altro) ‘I don’t know exactly THIS if they wanted to say (and not something else).’

These simple distributional facts are hard to reconcile with a “single layer” hypothesis for the C system, whereas they can be immediately accommodated if a richer system is assumed: che delimits the complementizer system upwards, di delimits it downwards, and se can appear in the middle, possibly preceded and followed by topics (but necessarily higher than focus): (18)

che . . . Top . . . se . . . Foc . . . Top . . . di

Much evidence of this sort can be gathered by pairwise comparisons of elements (if A > B and B > C, then A must be in a higher position than C even if the two elements cannot be found simultaneously in the same structure for independent reasons, as che and di in Romance, specialised for finite and non-finite clauses, respectively). But natural languages sometimes offer more direct evidence for

Delimitation Effects and the Cartography of the Left Periphery

125

complex hierarchical structure of functional heads: in some cases, distinct C-like particles co-occur in rigid orders, e.g. que si (that if) in certain kinds of Spanish indirect questions (Plann 1982; Su˜ner 1994, etc.) in cases like the following: (19)

Me preguntaron que si tus amigos ya te visitaron en Granada. ‘They asked me that if your friends had already visited you in Granada.’ (Su˜ner 1994: 349)

Cases like (19) straighforwardly confirm the higher part of map (18), arrived at through indirect comparisons like (13)–(16). The opposite order if that is also found in some languages, e.g. Dutch varieties in which the sequence wie of dat alternates with of in cases like (4a) (see (20a)). Clearly, (the equivalent of) that is an unmarked, versatile complementizer form, capable of occurring in the highest C position, and also, in cross-linguistically variable manners, in lower positions: as the head hosting a preposed adverbial clause in English varieties admitting (20b) (McCloskey 1992; Rizzi 2010b), as the head hosting various types of left peripheral elements including topics in the old southern Italian dialects discussed by Ledgeway (2003), as in (20c); it can also be a focus head in Brazilian Portuguese (Mioto 1999), as in (20d), and a marker of wh-exclamatives in Italian, as in (20e), etc. (20)

a.

b. c.

d.

e.

gezien heeft]]. Ik weet niet [wie of dat [Jan seen has I know not who Q that Jan (Dutch varieties, Haegeman 1994) I think that, if they arrive on time, that they will be greeted. Le mand`o a dire che tutte quille dinare che le voleva dare re de Franza per l’armata. ‘He sent (someone) to tell him that all this money that the king of France wanted to give him for the army.’ (Ledgeway 2003: 131) A Joana acha que A MARIA (que) o Jo˜ao encontrou no cinema. ‘Joana thinks that MARIA Jo˜ao met in the cinema.’ (Mioto 1999) Che bel libro che ho letto! ‘What a nice book that I have read!’

Clearly, the unmarked form of the complementizer can occur in different positions, sometimes simultaneously at the edge and within the complementizer zone (Ledgeway 2003 argues that in cases of co-occurrence like (20c) the lower occurrence can be analyzed as the spell out of the trace of che moving from Fin to Force across intermediate head positions).

126

Luigi Rizzi

A striking case of multiple C-like particles allowed to co-occur is the no ka to sequence uncovered in Japanese by Saito (2012) where to marks the “reported” character of the indirect question (much as in the Spanish case (19)), ka expresses interrogative force, and no is analyzed by Saito as a finiteness marker: (21)

Taroo-wa [ CP kare-no imooto-ga soko-ni ita (no) ka (to)] T.-top he-gen sister-nom there-in was no ka to minna-ni tazuneta. all-dat inquired ‘Taroo asked everyone if his sister was there.’ (Saito 2012: (41))

Given the head final character of the language the hierarchical order will be . . . no] ka] to] which mirrors the hierarchical order Declarative/report – interrogative force – finiteness that is postulated for Romance. Saito (2012) also provides evidence for a recursive topic field sandwiched in between Force and Finiteness in Japanese: (22)

[. . . [. . . [. . . [. . . [TP . . . ] Fin] (Topic*)] Force] Report] (Saito 2012)

This structure is very close to the one originally proposed in Rizzi (1997) for Italian/Romance, modulo headedness and other parametric differences such as, in Saito’s analysis, the non-occurrence of a peripheral focus position in Japanese. All these cases of distinct C-like elements occurring in different positions and sometimes allowed to co-occur in a rigid order provide straightforward evidence for a rich cartographic representation of the C-system. 5.

Cartography vs. “Simpler Representations”

Are there alternatives to the view embodied in (18), (22) and similar representations? The main alternative amounts to insisting on more impoverished clausal representations like (12), with basic ingredients such as: a. b. c.

a single CP layer; multiple adjunctions permitted to IP/TP; interpretive systems made capable of interpreting the adjoined material as topic or focus, and of expressing general or language specific co-occurrence and ordering constraints.

Delimitation Effects and the Cartography of the Left Periphery

127

In essence, the main alternative to the cartographic representations would amount to simplifying the syntactic representations and shifting to the interface systems much of the work that cartographic syntax does. Actually, “simplifying the syntax” is probably a misnomer to refer to such an approach. In a certain sense, syntax is equally simple with richer and more impoverished representations, as it involves the same basic computational ingredients: binary merge of heads and phrases. What is simpler in this alternative approach is the functional lexicon, and the resulting representations are more impoverished, but the mechanism of syntactic computation does not vary. Such a “simple representations” approach has several drawbacks, as the previous discussion already made clear: i.

It has no natural way to express cases with co-occurring multiple C-particles (i.e., (15), (19), (21), etc.), which often can be interrupted by phrasal material, hence cannot be analyzed as single syntactic nodes: these facts are simply inconsistent with a simple C node approach; ii. In cases of distinct C-particles occurring in different positions with respect to third elements, a “simpler representations” approach must assume complex constraints to hold on the adjunction mechanism. E.g., in order to capture the pattern in (13), (14), (16) adjunction must be assumed to be permissible to a non-finite CP and to a finite interrogative CP (to permit (14a), (16a)), but not to a finite declarative CP (to exclude (13a)); and it must be allowed to apply to a finite TP (to permit (13b), (16b)), but not to a non finite TP (to exclude (14b)). No such complex system of constraints is required by the approach assuming cartographic representations, which can in fact dispense with phrasal adjunction altogether (much as in Kayne’s 1994 restrictive approach to phrase structure), and analyzes movement to topic, focus, etc. as movement to Spec. iii. In languages with overt topic and focus particles, a “simpler representations” approach must introduce special mechanisms in the interface systems to account for their distributional properties, ordering constraints, etc. iv. More generally, it must endow the interpretive systems of mechanisms capable of capturing the observed variation: unique topics in some languages, recursive topics in other languages. In other words, such an approach must admit the possibility of expressing a parametrisation in the postsyntactic interpretive system. An approach assuming cartographic representations can capture the different co-occurrence and ordering properties, as well as the observed variation, via a fundamental, independently necessary mechanism: head-complement selection, admitting a certain amount of parametrisation: the Force head can select a Topic

128

Luigi Rizzi

Phrase, which can be recursive or not (i.e., Top can in turn select a Topic Phrase, as in Italian, or not, as in Gungbe); in many languages a Top head can select a Focus Phrase, but not vice versa; in Italian a Foc head can select a Top phrase, perhaps of a special kind (a “familiarity topic”, in the typology introduced by Frascarelli and Hinterhoelzl 2007); Force, Top, Foc can all select a Fin Phrase, and Fin always selects an IP/TP, much as T can select an Aspect Phrase, v selects a VP, V can select a DP, etc.: the same fundamental formal mechanism is used across the board to express invariant and variable properties in structure building. The selecting heads are sometimes overt and sometimes null, much as, e.g., certain tense, aspect, voice or case morphemes. A related but distinct issue is the question of the “further explanation” of the functional hierarchy that is observed: why is it that we typically find certain orders, rather than others? As pointed out in Cinque and Rizzi (2010) it is unlikely that the hierarchy may be an absolute syntactic primitive, unrelated to other requirements or constraints: why should natural language syntax express such a complex and apparently unmotivated primitive? It is more plausible that the functional hierarchy may be rooted elsewhere. External factors such as interpretive requirements may be relevant in some cases. For instance, it is argued in Rizzi (1997) that the non-recursive character of FocP is linked to the interpretive properties of the Focus – Presupposition articulation: if FocP were recursive, the lower FocP would end up being in the presuppositional part of the higher FocP, and an interpretive clash would arise. No such interpretive problem arises in the case of TopP (because the comment of a higher TopP can in turn have a Topic – Comment structure), hence TopP can be recursive, an attested parametric option. Another factor which may explain certain orders is the theory of locality. As Abels (2010) argues, if in a language the TopP has island-creating properties and the FocP does not (or such properties hold relative to the two types of movement, as in the system of Starke 2001; Rizzi 2004), then the only possible order is Top – Foc because in the opposite order the focused element could not reach the Spec of the FocP across the topic island. This is a perfectly fine mode of explanation and, as far as I can see, it is fully consistent with the adoption of cartographic representations. On the one hand, it has nothing to say on cases in which movement is not involved (i.e., in the Force – Fin system in cases like (15), (19), (21)), which therefore require a rich functional system anyhow, with ordering determined on some other basis (presumably interpretive requirements, possibly the locality of selection, as suggested in Rizzi 1997). On the other hand, in the cases which involve movement of topics, foci, wh, etc., the critical issue for the divide between cartographic vs “simpler representations” has to do with the nature and properties of scope-discourse movement, not with the primitive

Delimitation Effects and the Cartography of the Left Periphery

129

or derived nature of the ordering. If the criterial approach is on the right track, as we have argued, a rich system of attracting heads is needed anyhow, quite independently from the ordering issue, and it can very naturally be combined with an approach a` la Abels to the ultimate, principled explanation of ordering. So, the “Occam’s razor” argument that Abels gives is against the assumption of a primitive templatic ordering, while, as far as I can see, it is fully consistent with the criterial approach to scope-discourse semantics, which is the crucial ingredient of the cartographic representations defended here. Locality considerations may provide a principled explanation for certain hierarchical orderings of the functional sequence, as in Abels’ approach, and also explain certain restrictions on the occurrence of left peripheral elements in certain structural environments, as argued in detail by Haegeman (2010): in her analysis, the restrictions on the occurrence of a full-fledged left periphery in various kinds of adverbial and complement clauses is a consequence of the fact that such embedded constructions necessarily involve some kind of operator movement which would be adversely affected by the intervention of other leftperipheral elements. Again, this kind of locality-based explanation is fully consistent with detailed cartographic representations, in fact it presupposes them. In this kind of approach, the syntax-interpretation interface is fully transparent: there is a single kind of structural relation, the head-dependent relation which is read off syntactic representations by the interpretive systems, in conjunction with the content of the criterial heads, to assign topicality, focus, presupposition, etc., much as thematic role assignment works. The cartographic representations exploit fundamental syntactic devices to transparently express interpretive properties, thus simplifying the burden of the interface systems. They represent an attempt to “syntacticize” scope-discourse semantics as much as possible (Cinque and Rizzi 2010) without enriching the inventory of the computational mechanisms needed on either side of the interpretive interface. 6.

Criterial Positions Delimit Chains

A criterial position delimits a movement chain in the sense that a phrase meeting a Criterion cannot undergo further movement . This freezing effect is best illustrated by cases in which the same phrase contains two criterial features, e.g., a Q feature in the wh-specifier of a nominal expression and a (contrastive) focus feature in the lexical restriction, as in: (23)

[quantiQ LIBRIFoc ] ‘How many BOOKS’

130

Luigi Rizzi

A priori one could expect the phrase to move to one criterial position, satisfying the requirement of one feature, and then continue to move to a higher criterial position to satisfy the other feature. But this never happens, and the phrase gets stuck in the lower criterial position. In the following case, the phrase (23) moves to the Q head of the embedded question, and then cannot further focus move to the main clause, e.g. in a cleft configuration: (24)

Non ho capito [[ quantiQ LIBRIFoc ] Q [ dobbiamo leggere ]], non quanti articoli. ‘I didn’t understand how many BOOKS we have to read, not how many articles.’ b. *E’ [ quantiQ LIBRIFoc ] che non ho capito [ Q [ dobbiamo leggere ]], non quanti articoli ‘It is how many BOOKS that I didn’t understand we should read, not how many articles.’ a.

The only options for a well-formed outcome, given (24a), are focalization in situ (possibly obtained by an Agree relation with a Foc head in the periphery of the main clause) as in (24a), or pied-piping of the whole indirect question to the focus cleft position: (25)

E’ [[ quantiQ LIBRIFoc ] Q dobbiamo leggere ] che non ho capito , non quanti articoli. ‘It is how many BOOKS we should read that I didn’t understand, not how many articles.’

Here movement does not undo the criterial configuration, which remains intact in the C-system of the indirect question, but the whole configuration is piedpiped to a higher criterial position, the focus position of the cleft (Belletti 2009). This kind of pattern justifies a freezing principle of the following kind (Rizzi 2006; Rizzi and Shlonsky 2007; see also Lasnik and Saito 1992 for an earlier characterisation of the freezing effect, Boˇskovi´c 2005 for an analysis in terms of “deactivation”, and Rizzi 2010a, b for discussion): (26)

Criterial Freezing: an XP meeting a Criterion is frozen in place.

One way of conceptualizing this principle is to assume that as soon as a criterial configuration is formed, the configuration is handed over to the interface systems for interpretation, and becomes unavailable to further syntactic computation. This principle, among other things, reduces the amount of reconstruction that is needed (i.e., the necessity of reconstructing a criterial phrase into a lower criterial position will never arise), thus facilitating computation by the grammar

Delimitation Effects and the Cartography of the Left Periphery

131

and the parser. Criterial Freezing can thus be looked at it as an economy principle favoring efficiency in syntactic computations. 7.

The Subject Criterion

In previous work, I have proposed that movement to subject position is criterial in nature. Classical EPP can be restated in terms of assuming a criterial head Subj(ect), obligatory part of the clausal backbone, much as T, etc. This head expresses an interpretive relation between its specifier and its complement, as criterial heads do in general. In this particular case, the specifier of Subj, the subject, is interpreted as the argument “about which” the event referred to by the clause is presented. In this sense, an active and a passive sentence differ, as the same event is presented as being about the agent or the patient, respectively: (27)

a. b.

Mario ha insultato Piero. ‘Mario insulted Piero.’ Piero e` stato insultato da Mario. ‘Piero was insulted by Mario.’

As originally observed by Calabrese (1986), a pro subject in a null subject language picks out an aboutness subject, at least in cases in which ambiguity might arise. So, consider the following continuation of discourse: (28)

Subito dopo, pro e` andato via. ‘Immediately after that, pro left.’

If (28) is uttered immediately after (27a), it is understood that Mario left; if (28) is uttered immediately after (27b), it is Piero who left (referring to the “non subject” in (27), the object or the by-phrase, would require using the overt subject pronoun lui (he) in (28)). Aboutness is a property that subjects have in common with topics (Reinhart 2005). So, if an aboutness subject and a topic are present, they can both be picked up by pro in the following discourse continuation: (29)

a. b.

Piero, Mario lo ha insultato. ‘Piero, Mario insulted him.’ Subito dopo, pro e` andato via. ‘Immediately after that, pro left.’

132

Luigi Rizzi

In (29b), pro can pick up both the topic Piero (in the Clitic Left Dislocation construction) and the subject Mario as antecedent, perhaps with a preference for the latter, possibly for proximity reasons. As for the structural expression of the Subject Criterion, I have assumed, following Cardinaletti (2004), that the Criterial head Subj is part of the obligatory structure of the IP, in a position higher than T and immediately lower than the lowest head of the C-system Fin. A plausible candidate of a language with an overt morphological realization of Subj is a language with subject clitics occurring in between the subject DP and the predicate starting with the inflected verb, such as the Northern Italian Dialects (Poletto 2000; Manzini and Savoia 2005): (30)

a.

b.

El fio el mangia l pom. The boy subj eats the apple (Milan: Rizzi 1986; Poletto 2000; Manzini and Savoia 2005, etc.) Le ragazze le son venute. ‘The girls subj have+3pl come.’ (Florence: Brandi and Cordin 1989)

The syntactic role of the Subj head is to delimit the structure of the IP zone, and attract a nominal expression to its Spec; its interface role is to trigger the “aboutness” interpretation, imposing the interpretation of the event expressed by the predicate as being about the subject. Capitalizing on the formal similarity (or identity) of the Subj head with the determiner in cases like (30), one may speculate that Subj is a D-like head which has in common with its counterpart in the extended projection of NP’s the “affinity” with a nominal projection; more specifically, both “clausal” and “nominal” D’s would have the capacity to attract a nominal expression to their Spec’s (here I am thinking in particular of the derivational processes overtly triggered in D-final languages, analyzed by Cinque (2005) as involving the attraction of the NP to the Spec of D, with or without pied-piping of additional material). Different kinds of evidence can be provided in favor of the view that movement to subject position is determined by the attraction of nominal features by the relevant functional head. This approach can provide an explanation to a curious asymmetry observed long ago by Ruwet (1972). In French, an adnominal complement of the object can be pronominalized by the clitic pronoun en, and the remnant of the object DP can be moved to subject position in passive: (31)

a.

Jean a publi´e [la premi`ere partie de ce roman] l’ann´ee pass´ee. ‘Jean published the first part of this novel last year.’

Delimitation Effects and the Cartography of the Left Periphery

b.

c.

133

Jean en a publi´e [la premi`ere partie ] l’ann´ee pass´ee. (de ce roman : en = pro-PP) ‘Jean of-it published the first part last year.’ [La premi`ere partie ] en a e´ t´e publi´ee l’ann´ee pass´ee. ‘The first part of-it was published last year.’ (Ruwet 1972)

En can also pronominalize the NP in a DP introduced by an indefinite quantifier like [trois NP] in (32a); but in this case, the remnant [trois ] cannot be further moved to subject position, as in (32c): (32)

a.

Jean a publi´e [trois romans] l’ann´ee pass´ee. ‘Jean published three novels last year.’ b. Jean en a publi´e [trois ] l’ann´ee pass´ee. (romans : en = pro-NP) ‘Jean of-them published three last year.’ c. *[Trois ] en ont e´ t´e publi´es l’ann´ee pass´ee ‘Three of-them have been published last year.’

This prohibition is limited to movement to subject position: A movement to the left periphery can move the DP remnant after en extraction: (33)

[Combien ] il en a publi´es l’ann´ee pass´ee? ‘How many he of-them published last year?’

A natural analysis of the impossibility of (32c) is that in this case, after the pro-NP en has been extracted, the DP does not contain any more an active (attractable) component specified [+N], hence the attraction to Spec Subj cannot take place (here I am assuming, with much recent work, that even if we adopt the copy theory of traces, according to which a copy of en remains expressed within the object DP in (32), a trace cannot be attracted and moved to a higher attractor). No problem arises in (31), because the remnant DP still contains a nominal part, headed by partie, which can be attracted in (31c); nor with A movement as in (33), because here the attracting feature is [+Q], which is specified in combien in the remnant DP, not [+N]. Another classical observation which is amenable to the conception of subject movement adopted here is Koster’s generalization that “subject sentences don’t exist” (Koster 1978). Koster noticed that putative sentential subject clauses as in (34) are better analyzed as sitting in topic position; that they may not be in the canonical subject position is suggested by different kind of evidence, e.g., the fact that they don’t naturally allow I to C movement:

134 (34)

Luigi Rizzi

a. That he will come proves this. b. *What does that he will come prove?

If finite clauses are not [+N] (as is suggested by the fact that they don’t bear Case in many languages), they can’t be attracted to Spec-Subj, hence they cannot appear in the canonical subject position (while they can be topicalized, focus moved, etc.). If this analysis is on the right track, the Criteria are not solely involved in the triggering of A movement: also the core case of A-movement, movement targeting the subject position, is criterial in nature. So, the notion of Criteria cuts across the A/A distinction. On the other hand, the A/A distinction cannot be entirely obliterated: among other things, A-movement clearly is more local than typical A -movement. For instance, movement to Top can easily skip intervening DP’s, while movement to Subj is strictly local: only the closest DP can be attracted by Subj: (35)

a. Mary, I believe that Bill should talk to . b. *Mary was believed that Bill should talk to

.

The difference has to do with the different nature of the attracted feature. A criterial A head such as Top or Foc attracts an expression endowed with a matching feature, so, if in (35a) neither I nor Bill are endowed with [+Top], they can be skipped by Mary moving to Top, under a featural formulation of Relativized Minimality (Starke 2001; Rizzi 2004). On the other hand, if Subj attracts a [+N] element, it will inevitably go for the closest such element, and the different nominal interveners in (35b) will block movement of Mary. Why does Subj not attract a more selective feature analogous to Top, Foc, etc.? Presumably this has to do with the inherent weakness of the interpretive effect of the Subject Criterion: Spec-Subj does not express, per se, a particular informational property (given or new information), or a particular quantificational property, it only expresses pure aboutness. As such, the notion has no specific scope-discourse featural content to operate on, it just exerts an attraction on any nominal. Because of locality, the attracted nominal will inevitably be the closest one. So, the stricter locality of movement to subject w.r.t. A movement may well be a consequence of the weakness of the interpretive content associated to the subject position. This state of affairs may also be connected to the existence of expletive subjects. An expletive is a nominal expression which cannot carry any scopediscourse featural specification (it can’t be topical, nor focal, as it doesn’t have any referential content, nor can it express any kind of quantification), so it cannot

Delimitation Effects and the Cartography of the Left Periphery

135

undergo A criterial movement, and we typically don’t find expletives in topic, focus, or Q positions (but see Beninc`a and Poletto 2004 on a possible case of expletive topic in Old Italian). But an expletive is specified [+N], and as such it can be attracted to Spec-Subj. In this case, the aboutness interpretive routine cannot be triggered, as there is no referent about which the event is presented, and the sentence receives what is sometimes called a “presentational” interpretation. 8.

Subject-Object Asymmetries: ECP Effects as Criterial Freezing

The combined effect of the Subject Criterion and Criterial Freezing provides an explanation for the well-known subject-object asymmetries in extraction processes, instantiated by the that-trace effect in English: (36)

a. *Who do you think [that [ subj will come]]? b. Who do you think [that [Mary subj will meet

]]?

The traditional analysis involves the Empty Category Principle (ECP), requiring a certain type of government relation (proper government) to be satisfied by traces. But the ECP is hard to accommodate with the guidelines of the Minimalist Program, as it does not have a natural status within the principled typology of UG principles assumed by minimalism: neither can it be naturally construed as an economy principle, nor as a principle enforced by some requirement of the interface systems. The system proposed here provides a simple alternative to an ECP-based analysis of subject-object asymmetries (Rizzi 2006; Rizzi and Shlonsky 2007). If there is a Subject Criterion, further movement of the subject will be generally blocked by Criterial Freezing. So, in the derivation of (36a), who has to move to Spec-Subj in order to satisfy the Subject Criterion in the embedded clause, as the Criterion cannot be satisfied in any other way in this case; but then who will be stuck there, and will be disallowed to move further and undergo extraction by Criterial Freezing: (37)

Subject extraction is blocked by Criterial Freezing.

Various kinds of empirical evidence have been produced to show that this alternative is more satisfactory than an ECP approach (Rizzi and Shlonsky 2007). For instance, the ECP approach does not offer a natural analysis of the fortrace effect, the fact that a subject is not extractable across the prepositional complementizer for in English:

136 (38)

Luigi Rizzi

a. *Who would you prefer [for [ b. Who do you work for ? c. *Who would you prefer [for [

to win]]? subj to win]]?

It is hard to see how the subject trace could fail to be properly governed by the complementizer for, particularly in view of the fact that the minimally different preposition for licenses extraction, as in (38b). In terms of the alternative envisaged here the ill-formedness of representation (38c) is straightforward: for is a member of the C-system (presumably a realisation of Fin: Rizzi 1997), so the whole IP system will be developed under it, including the SubjP layer. Then (38b) involves further movement of an element which satisfies the Subject Criterion, which determines a violation of Criterial Freezing, much as in the that-trace configuration in (36a). If the complementizer is dropped in (36a), (38a), extraction becomes possible, either because complementizer drop actually involves the truncation of an important structural chunk, including the Subj layer, so that no freezing effect is determined; or, more plausibly, in the more indirect and elaborate way proposed in Rizzi and Shlonsky (2007), which assimilates subject extraction to simple movement in subject questions like Who will come? Subject extraction is notoriously harder than object extraction across languages, but it is not banned altogether. Languages normally invent strategies to make it possible to form a question or other A -constructions on an embedded subject (Rizzi and Shlonsky 2007). One typical such strategy is what Rizzi and Shlonsky (2007) call a skipping strategy: the language uses an expletive-like element to formally satisfy the Subject Criterion, and this allows the thematic subject to escape the freezing effect and remain accessible to extraction. A systematic utilisation of the skipping strategy is observed in null subject languages. Consider the following comparative generalisation: (39)

Null subject languages are not sensitive to that-trace effects. (Perlmutter 1970; Taraldsen 1978; Pesetsky 1981; Rizzi 1982, 1990; Nicolis 2005, 2008).

In fact, the word by word equivalents of sentences like (36a) are fully acceptable in null subject languages like Italian, Spanish, Rumanian, etc. In my original work on the null subject parameter, I proposed that subject extraction in these languages always proceeds from a lower position, while the canonical subject position (the EPP position) is filled by an expletive occurrence of the null pronoun pro. Hence (40a) has a representation like (40b).

Delimitation Effects and the Cartography of the Left Periphery

(40)

a.

b. c.

137

Chi credi che verr`a? ‘Who do you think that will come?’ (Rizzi 1982; 1990) Chi credi [che [pro verr`a ]]? ‘Who do you think that will come?’ Chi credi [che [pro subj verr`a ]]? ‘Who do you think that will come?’

In the analysis of Rizzi (1982), pro offered a device to avoid leaving a trace in a non-properly governed position like the subject position, hence in an illegitimate position according to the ECP. The analysis can now be immediately transposed to the Criterial Freezing approach. The representation is (40c), and pro formally satisfies the Subject Criterion and is frozen there, thus making the thematic subject accessible to extraction from a lower, non-criterial position. This is just one device that languages may use to satisfy the Subject Criterion without having to move the thematic subject to the freezing position (Rizzi and Shlonsky 2007; and see Taraldsen 2001; Endo 2008 on the role of certain clause final particles in Japanese to license subject movement; and the discussion in Miyagawa 2004 on ways of satisfying the EPP). What lower position is the thematic subject extracted from in (40)? In the original analysis along the lines indicated in (40a)–(40b), subject extraction in Romance null subject languages was made contingent upon the existence of “free subject inversion”; i.e., the following, possible in Italian, Spanish, etc., was considered an intermediate step in the derivation of (40a). (41)

Credo che verr`a Gianni. ‘I believe that will come Gianni.’

But this particular assumption is unnecessary and problematic, both theoretically and empirically. Chao (1981) had already observed that this line of analysis is difficult to maintain for Brazilian Portuguese, which does not allow a free inversion process (at least not as freely as the European null subject languages), and still admits free violations of the that-trace effect. Nicolis (2008) analyzes several Creole languages which admit a free violation of that-trace and disallow subject inversion, and observes that the lack of a dependency between these two properties also emerges from Gilligan’s (1987) survey study (three languages in his sample, Basque, Papiamento and Yoruba, permit free violations of that-trace and have no subject inversion). There are also theoretical reasons for assuming that (41) and, more generally, “subject inversion” structures, are not an intermediate step in the derivation of (40c). Belletti (2001, 2004) proposed that the core case of “subject inversion” is

138

Luigi Rizzi

in fact a device for focusing the thematic subject, which is moved to a low focus position, specified in the periphery of the vP. Null subject languages not allowing “subject inversion” can thus be seen as lacking this particular focalization device. As the low focus position expresses a scope-discourse property, it is natural to analyze it as a criterial position. But then, an inverted subject would fall under Criterial Freezing, and should not be movable from the low focus position. In fact, the skipping strategy permitting a free violation of that-trace does not require a passage through the low focus position: the essential ingredient is the availability of expletive pro in the language, the device formally satisfying the Subject Criterion and permitting the A extraction of the thematic subject from a lower position which does not have to be the low focus position (and cannot be the low focus position, under Criterial Freezing). In fact, Brazilian Portuguese, and the Creole languages studied in Nicolis (2008) (Berbice Dutch, Cape Verdean, Mauritian, Papiamentu) all admit a null expletive, hence the free violation of that-trace is expected (see also Menuzzi 2000 for an argument that in Brazilian Portuguese the subject is indeed extracted from a position lower than Spec-Subj); on the other hand, the variety of Jamaican Creole described in Durrleman (2008) does not allow null expletives in embedded clauses (only in root contexts), and it also disallows that-trace violations: the connection between null expletives and subject extraction is thus confirmed by this case (the case of Haitian Creole is complicated by the residual presence of a kind of French que → qui rule in the language: see Nicolis 2008, for discussion). 9.

Conclusion

This paper focused on A -chains, the kind of chains which associate two types of interpretive properties to expressions: properties of semantic selection (thematic properties for arguments) and properties of scope-discourse semantics. I have adopted the criterial approach to scope-discourse semantics, which traces back the assignment of discourse and informational functions to familiar syntactic ingredients, head-dependent relations, and I have shown how this view naturally leads to the study of the cartography of syntactic structures: the study of the complex syntactic configurations created by a simple computational system, operating on a richly structured functional lexicon. Looking at the structure of the initial periphery of the clause, I have compared an analysis involving rich cartographic representation, and one based on simpler representations, assuming a single C-layer, and admitting multiple adjunctions to TP. The latter “simpler representations” approach shifts much of the descriptive burden to the interface systems, not only interpretation proper, but also positional and co-occurrence

Delimitation Effects and the Cartography of the Left Periphery

139

restrictions, and the related parametrisation. On the contrary, the approach based on cartographic representations stretches the role of syntax in the effort of syntacticizing as much as possible the interfaces with sound and meaning, assuming syntactic representations which carry interpretive properties “on their sleeves”. I have tried to show that the cartographic representations compare favorably to the syntactically less elaborate alternatives. I have then turned to the issue of delimitation. Movement is delimited, in the sense that it must start and finish in particular structural positions: in fact, the positions dedicated to particular interpretive properties have the effect of delimiting the movement chains, and delimitation theory is establishing itself as a new chapter of the classical topic of locality. I have tried to show that, by assuming principles of upward delimitation, or freezing, we can explore new generalizations, and envisage novel explanatory accounts for much-studied phenomena, such as the classical subject-object asymmetries in extraction processes References Abels, Klaus 2010 Aboh, Enoch 2004 Aboh, Enoch 2007

The Italian Periphery: A View from Locality. Ms., University College London. The Morphosyntax of Complement-Head Sequences. New York: Oxford University Press. Information structuring begins with numeration. Ms., ACLC, University of Amsterdam.

Bayer, Josef 1984 Comp in Bavarian. The Linguistic Review 3: 209–274. Belletti, Adriana 1990 Generalized Verb Movement. Turin: Rosenberg & Sellier. Belletti, Adriana 2001 Inversion as Focalization. In: Aafke Hulk and Jean-Yves Pollock (eds.), Subject Inversion in Romance and the Theory of Universal Grammar, 60–90. Oxford: Oxford University Press. Belletti, Adriana 2004 Aspects of the Low IP Area. In: Luigi Rizzi (ed.), The Structure of CP and IP, 16–51 (The Cartography of Syntactic Structures 2.). Oxford: Oxford University Press. Belletti, Adriana 2009 Structures and Strategies. London: Routledge.

140

Luigi Rizzi

Beninc`a, Paola and Cecilia Poletto 2004 Topic, Focus and V2: Defining the CP sublayers. In: Luigi Rizzi (ed.), The Structure of CP and IP, 52–75 (The Cartography of Syntactic Structures 3.). Oxford: Oxford University Press. Bianchi, Valentina and Mara Frascarelli 2009 Is Topic a Root Phenomenon? Ms., University of Siena, University of Roma 3. Biloa, Edmond 2008 The Cartography of the Left Periphery in Tuki. Ms., University of Yaounde I, Cameroon. Bocci, Giuliano 2009 On Syntax and Prosody in Italian. Doctoral dissertation, University of Siena. ˇ Boˇskovi´c, Zeljko 2005 On the Operator Freezing Effect. Ms., University of Connecticut. Brandi, Luciana and Patrizia Cordin 1989 Two Italian Dialects and the Null Subject Parameter. In: Osvaldo Jaeggli and Kenneth Safir (eds.), The Null Subject Parameter, 111–142. Dordrecht: Kluwer. Calabrese, Andrea 1986 Some Properties of the Italian Pronominal System: An Analysis Based on the Notion of Thema as Subject of Predication. In: Harro Stammerjohann (ed.), Tema-Rema in Italiano, 25–36. T¨ubingen: Gunter Narr. Cardinaletti, Anna 2004 Towards a Cartography of Syntactic Positions. In: Luigi Rizzi (ed.), The Structure of CP and IP, 115–165 (The Cartography of Syntactic Structures 2.). Oxford: Oxford University Press. Chao, Wynn 1981 Pro-drop languages and nonobligatory control. University of Massachusetts Occasional Papers in Linguistics 7: 46–74. Chomsky, Noam 1986 Barriers. Cambridge, Mass: MIT Press. Chomsky, Noam 2004 Beyond Explanatory Adequacy. In: Adriana Belletti (ed.), Structures and Beyond, 104–131 (The Cartography of Syntactic Structures 3.). New York: Oxford University Press. Cinque, Guglielmo 1999 Adverbs and Inflectional Heads. Oxford, NewYork: Oxford University Press. Cinque, Guglielmo 2005 Deriving Greenberg’s Universal 20 and it’s Exceptions. Linguistic Inquiry 36(3): 315–332.

Delimitation Effects and the Cartography of the Left Periphery

141

Cinque, Guglielmo and Luigi Rizzi 2010 The Cartography of Syntactic Structures. In: Bernd Heine and Heiko Narrog (eds.), The Oxford Handbook of Linguistic Analysis, 51–65. Oxford, New York: Oxford University Press. Cruschina, Silvio 2008 Discourse-related Features and the Syntax of Peripheral Positions: A Comparative Study of Sicilian and other Romance Languages. Doctoral Dissertation, University of Cambridge. Durrleman, Stephanie 2008 The Syntax of Jamaican Creole – A Cartographic Perspective. (Linguistics Today 127.) Amsterdam/Philadelphia: John Benjamins Publishing Company. Endo, Yoshio 2007 Locality and Information Structure – A Cartographic Approach to Japanese (Linguistics Today 116.) Amsterdam/Philadelphia: John Benjamins Publishing Company. Frascarelli, Mara 2000 The Syntax-Phonology Interface in Focus and Topic Constructions in Italian. Dordrecht, Kluwer Academic Publishers. Frascarelli, Mara and Roland Hinterh¨olzl 2007 Types of Topics in German and Italian. In: Susanne Winkler and Kerstin Schwabe (eds.), On Information Structure, Meaning and Form, 87–116. Amsterdam: Benjamins. Gilligan, Gary 1987 A Cross-linguistic Approach to the Pro-drop Parameter. PhD Dissertation, University of Southern California. Grewendorf, G¨unther 2002 Left Dislocation as Movement. In: Simon Manck and Jennifer Mittelstaedt (eds.), Georgetown University Working Papers in Theoretical Linguistics, 31–81. Haegeman, Liliane 1994 An Introduction to Government-Binding Theory. Oxford: Blackwell. Haegeman, Liliane 2003 Notes on long adverbial fronting in English and the left periphery. Linguistic Inquiry 31: 640–649. Haegeman, Liliane 2006 Argument Fronting in English, Romance ClLD and the Left Periphery. In: Raffalla Zanuttini, Hector Campos, Elena Herburger and Paul Portner (eds.), Negation, Tense and Clausal Architecture: Cross-linguistic Investigations, 27–52. Washington DC: Georgetown University Press. Haegeman, Liliane 2010 Adverbial Clauses, Main Clause Phenomena, and the Composition of the Left Periphery. Ms., University of Gent.

142

Luigi Rizzi

Jayaseelan, K. A. 2008 Topic, Focus and Adverb Positions in Clause Structure. Nanzan Linguistics 4: 43–68. Kayne, Richard 1983 Connectedness and Binary Branching. Dordrecht: Foris Publications. Kayne, Richard 1994 The Antisymmetry of Syntax. Cambridge, Mass.: MIT Press. Koster, Jan 1978 Why Subject Sentences don’t Exist. In: Samuel J. Keyser (ed.), Recent Transformational Studies in European Languages, 53–64. Cambridge, Mass.: MIT Press. Krapova, Iliyana and Guglielmo Cinque 2004 On the Order of wh-phrases in Bulgarian Multiple wh fronting. Ms., University of Venice. Lasnik, Howard and Mamoru Saito 1992 Move Alpha: Conditions on its Application and Output. Cambridge, Mass.: MIT Press. Ledgeway, Adam 2003 Il sistema completivo dei dialetti meridionali: la doppia serie di complementatori, Rivista Italiana di Dialettologia 27: 89–147. Manzini, Maria R. and Leonardo Savoia 2005 I dialetti Italiani e romanci. Alessandria: Edizioni dell’Orso. McCloskey, James 1992 Adjunction, Selection and Embedded Verb Second. Ms., UCSC. Menuzzi, Sergio 2000 That-trace effects in Portuguese. Forum Ling¨u´ıstico 2.1: 13–39. Florian´opolis, UFSC. Mioto, Carlost 1999 A periferia esquerda no portuguˆes brasileiro. Ms., Universidade Federal de Santa Catarina, University of Siena. Miyagawa, Shigeru 2004 On the EPP. MITWPL 49: 201–236. Nicolis, Marco 2005 On pro drop. Ph.D. dissertation. University of Siena. Nicolis, Marci 2008 The Null Subject parameter and correlating properties: the case of Creole languages. In: Theresa Biberauer (ed.), The limits of syntactic variation, 271–294. Amsterdam: John Benjamins. Pearce, Elizabeth 1999 Topic and Focus in a head-initial language: Maori. In: Carolyn Smallwood and Catherine Kitto (eds.), Proceedings of AFLA vi. The SixthMeeting of the Austronesian Formal Linguistics Association, held at

Delimitation Effects and the Cartography of the Left Periphery

143

the University of Toronto April 16–18, 249–263. Toronto: University of Toronto Dept. of Linguistics. Perlmutter, David 1970 Deep and Surface Structure Constraints in Syntax. New York: Holt Rinehart & Winston. Pesetsky, David 1982 Complementizer-trace Phenomena and the Nominative Island Condition. The Linguistic Review 1: 297–343. Pesetsky, David and Esther Torrego 2001 T-to-C Movement: Causes and Consequences. In: Michael Kenstowicz (ed.), Ken Hale:A Life in Language, 355–426. Cambridge, Mass.: MIT Press. Plann, Susan 1982 Indirect Questions in Spanish. Linguistic Inquiry 13: 297–312. Poletto, Cecilia 2000 The Higher Functional Field. Evidence from Northern Italian Dialects. New York, Oxford University Press. Pollock, Jean-Yves 1989 Verb Movement, Universal Grammar and the Structure of IP. Linguistic Inquiry 20: 365–424. Reinhart, Tanya 2005 Interface Strategies. Cambridge, Mass.: MIT Press. Rizzi, Luigi 1982 Issues in Italian Syntax. Dordrecht: Foris Publications. Rizzi, Luigi 1986 Null Objects in Italian and the Theory of pro. Linguistic Inquiry 17(3): 501–557. Rizzi, Luigi 1991 Residual Verb Second and the Wh Criterion. Geneva Working Papers on Formal and Computational Linguistics. Republished in Rizzi (2000). Rizzi, Luigi 1997 The Fine Structure of the Left Periphery. In: Liliana Haegeman (ed.), Elements of Grammar, 281–337. Dordrecht: Kluwer. Rizzi, Luigi 2000 Comparative Syntax and Language Acquisition. London: Routledge. Rizzi, Luigi 2004a Locality and Left Periphery. In: Adriana Belletti (ed.), Structures and Beyond. The Cartography of Syntactic Structures, vol. 3, 104–131. New York: Oxford University Press Rizzi, Luigi (ed.) 2004b The Structure of CP and IP (The Cartography of Syntactic Structures 2.). Oxford: Oxford University Press.

144

Luigi Rizzi

Rizzi, Luigi 2006

Rizzi, Luigi 2009

Rizzi, Luigi 2010a

Rizzi, Luigi 2010b

On the Form of Chains: Criterial Positions and ECP Effects. In: Lisa Cheng and Norbert Corver (eds.), Wh Movement: Moving on, 97–133. Cambridge, Mass.: MIT Press. Movement and Concepts of Locality. In: Massimo Piattelli-Palmarini, Juan Uriagereka and Pello Salaburu (eds.), Of Minds and Language: A Dialogue with Noam Chomsky in the Basque Country, 155–168. Oxford, New York: Oxford University Press. On some Properties of Criterial Freezing. In: E. Phoevos Panagiotidis (ed.), The Complementizer Phase: Subjects and Operations, 17–32. Oxford, New York: Oxford University Press.

Topic, Focus and the Cartography of the Left Periphery. Ms., University of Siena. Rizzi, Luigi and Ur Shlonsky 2007 Strategies of Subject Extraction. In: Hans-Martin G¨artner and Uli Sauerland (eds), Interfaces + Recursion = Language? Chomsky’s Minimalism and the View from Syntax-Semantics, 115–160. Berlin: Mouton de Gruyter. Roberts, Ian 2004 The C-system in Brythonic Celtic Languages, V2 and the EPP. In: Luigi Rizzi (ed.), The Structure of CP and IP, 297–328 (The Cartography of Syntactic Structures 2.). Oxford: Oxford University Press. Ruwet, Nicolas 1972 Th´eorie syntaxique et syntaxe du Fran¸cais. Paris: Seuil. Saito, Mamoru 2012 Sentence Types and the Japanese Right Periphery. In: this volume. Salvi, Giampaolo 2005 Some Firm Points on Latin Word Order: The Left Periphery. In: Katalin ´ Kiss (ed.), Universal Grammar and the Reconstruction of Ancient E. Languages, 429–456. Berlin: Mouton de Gruyter. Shlonsky, Ur 2000 Subject Positions and Copular Constructions. In: Hans Bennis, Martin Everaert and Eric Reuland (eds.), Interface Strategies, proceedings of the colloquium, Amsterdam, 24–26 September 1997, 325–347. Amsterdam: Royal Netherlands Academy of Arts and Sciences. Su˜ner, Margarita 1994 V-movement and the Licensing of Argumental Wh Phrases in Spanish. Natural Language and Linguistic Theory 12, 335–372.

Delimitation Effects and the Cartography of the Left Periphery Starke, Michal 2001

145

Merge Dissolves into Move. Doctoral dissertation, University of Geneva.

Taraldsen, Tarald 1978 On the NIC, Vacuous Application and the That-Trace Filter. Bloomington: Indiana University Linguistics Club. Taraldsen, Tarald 2001 Subject Extraction, the Distribution of Expletives, and Stylistic Inversion. In: Aafke Hulk and Jean-Yves Pollock (eds.), Inversion in Romance, 163–182. Oxford: Oxford University Press. Tsai, Dylan 2007 Left Periphery and Why-How Alternations. Ms., National Tsing-Hua University, Taiwan. Vallduv´ı, Enric 1992 The Informational Component. (Outstanding Dissertations in Linguistic). New York: Garland.

Sentence Types and the Japanese Right Periphery* Mamoru Saito

1.

Introduction

In this paper, I examine the complementizer system of Japanese and present a preliminary hypothesis on the structure of the Japanese right periphery. I discuss three complementizers, to, ka, and no. Examples of their occurrences are shown in (1). (1)

a.

b.

c.

Taroo-wa [CP Hanako-ga Ziroo-ni atta to] omotteiru. H.-nom Z.-dat met to think T.-top ‘Taroo thinks that Hanako met Ziroo.’ Taroo-wa [CP Hanako-ga dare-ni atta ka] tazuneta. H.-nom who-dat met ka inquired T.-top ‘Taroo asked who Hanako met.’ Taroo-wa [CP Ziroo-ni atta no]-o kookaisiteiru. Z.-dat met no-acc regret T.-top ‘Taroo regrets that he met Ziroo.’

To is often considered the complementizer for embedded propositions as it appears in the CP complements of verbs such as omou ‘think’ and yuu ‘say’. However, I first show that it is employed for ‘paraphrases of quotes’ in the sense of Plann (1982) or ‘reports of direct discourse’ in the sense of Lahiri (1991). Given this, I re-examine no and argue that it instead is the complementizer for embedded propositions. The analysis to be proposed is summarized in (2).1

* This paper was presented at the International Conference on Sentence Types: Ten Years After, held at the University of Frankfurt on June 26–28, 2009. I would like to thank the audience, especially NicholasAsher, Adriana Belletti, G¨unther Grewendorf, Paul Portner, and Peter Sells, for helpful comments. 1. This is an extension of the analysis proposed in Kuno (1973, 1988). It becomes clear in the following pages where I depart from his analysis.

148

Mamoru Saito

(2)

a. b. c.

to is the complementizer for ‘paraphrases’ or ‘reports’ of direct discourse. ka is the complementizer for questions. no is the complementizer for propositions.

A well-known peculiar property of to is that it can follow questions as in (3). (3)

kuru Taroo-wa Ziroo-ni [CP dare-ga kare-no ie-ni who-nom he-gen house-to come T.-top Z.-dat ka to] tazuneta. ka to inquired Lit. ‘Taroo asked Ziroo that who is coming to his house.’

A similar phenomenon is observed in Spanish, as discussed in Plann (1982).2 In (4a) que heads a propositional CP but it precedes embedded questions in (4b)–(4c). (4)

a.

b.

c.

Sab´ıa que coria. knew-3sg que run-3sg ‘He knew that he was running.’ Te preguntan que para qu´e quieres el pr´estamo. you ask-3pl que for what want-2sg the loan ‘They ask you what you want the loan for.’ Pens´o que cu´ales ser´ıan adecuados. thought-3sg que which ones would be appropriate ‘He wondered which ones would be appropriate.’

Examining examples of this kind in detail, Plann proposes that que is ambiguous between complementizers for propositions and for ‘paraphrases’ of quotes. According to her analysis, Spanish has three distinct complementizers as in (5). (5)

a. b. c.

que for ‘paraphrases’ of quotes Null [+Q] C for questions que for propositions

What I propose in this paper is that Japanese has an identical complementizer system. Japanese in fact provides explicit evidence for Plann’s analysis as the three complementizers have distinct phonetic realizations: To corresponds to que in (5a), ka is the [+Q] C, and no is the counterpart of que in (5c). 2. Thanks are due to Kensuke Takita for pointing out the relevance of Plann (1982) and the parallelism between the Japanese to and the Spanish que.

Sentence Types and the Japanese Right Periphery

149

In the second part of the paper, I examine the co-occurrence restrictions on the three complementizers, to, ka and no. In addition to the ka-to sequence observed in (3), there are examples with all three complementizers, as shown in (6). (6)

Taroo-wa [CP kare-no imooto-ga soko-ni ita (no) ka (to)] he-gen sister-nom there-in was no ka to T.-top minna-ni tazuneta. all-dat inquired ‘Taroo asked everyone if his sister was there.’

The three complementizers always appear in the order no-ka-to, and this suggests the recursive CP structure in (7). (7)

[CP [CP [CP [TP . . . ] no] ka] to]

Then, I consider the distribution of thematic topics and present evidence that they can appear in CPs headed by to or ka, but not in CPs headed by no. This leads to the hypothesis that there is a Topic head, located between no and ka as in (8), that hosts thematic topics in its Spec. (8)

[CP [CP [CP [CP [TP . . . ] no] e[+TOPIC] ] ka] to]

The similarity between (8) and the structure of the Italian left periphery proposed in Rizzi (1997) is evident. His proposal is shown in (9). (9)

[Force [(Topic*) [(Focus) [(Topic*) [Finite [TP . . . ]]]]]]

If ka is Force and no is Finite, Japanese is identical to Italian except for the presence of to and the absence of Focus.3 I suggest then that the Japanese periphery is comparable to Italian with the addition of the highest C, to, which is equivalent to the Spanish que as a marker of ‘paraphrase’ or ‘report’ of direct discourse. This conclusion, if correct, provides additional evidence for the universality of the structure of the left/right periphery. In the following section, I discuss the parallelism between to and the Spanish que in some detail. As noted above, they can both take question CPs as complements. Rivero (1994) presents examples where imperatives follow que in support of Plann’s (1982) analysis. Kuno (1988) notes similar facts in Japanese 3. The C heads in Japanese appear in the mirror image of Italian because of the headparameter. As discussed below in Section 4, it is argued in Kuroda (1988) and Saito (2007) that Japanese allows multiple thematic topics. The basic pattern of focusing in Japanese is like English: Any phrase with stress is interpreted as focus in situ.

150

Mamoru Saito

and argues that the complement of to can be a “blended discourse”, which starts out as a regular embedded sentence and ends with a verb that expresses a request. I re-examine those facts and show that they too provide supporting evidence for Plann’s analysis. In Section 3, I turn to no. Kuno (1973) argues that the CPs headed by this complementizer typically carry factive presuppositions in the sense of Kiparsky and Kiparsky (1970). I first show that the distribution of no is much wider than his analysis implies. Then, I argue that it should be considered the complementizer for propositional complements. Section 4 concerns the structure of the Japanese right periphery. I discuss the co-occurrence restrictions on the three complementizers, to, ka and no, and also the distribution of thematic topics. This leads to the hypothesis on the structure of the Japanese right periphery alluded to above. Section 5 concludes the paper. 2.

To as a Complementizer for Paraphrases of Direct Discourse

In this section, I examine cases where to takes questions and expressions of request as complements and argue that it is a complementizer for ‘paraphrase’ or ‘report’ of direct discourse just like que in Spanish. Section 2.1. concerns examples in which to follows questions and Section 2.2. deals with those in which it follows expressions of request. 2.1. The Parallelism of to and que with Respect to Question Complements Another well-known property of to is that it is employed for direct discourse as well as indirect discourse. It follows a direct quotation in (10a), while it marks an indirect discourse in (10b). (10)

a.

b.

Hanako-ga, “Watasi-wa tensai da,” to itta /omotta H.-nom I-top genius is to said/thought (koto) fact ‘(the fact that) Hanako said/thought, “I’m a genius.”’ Hanako-ga [CP zibun-ga tensai da to] itta /omotta self-nom genius is to said/thought H.-nom (koto) fact ‘(the fact that) Hanako said/thought that she is a genius’

A CP complement as in (10b) is usually considered to represent a proposition. This is because to heads the CP complements of typical bridge verbs, as noted

Sentence Types and the Japanese Right Periphery

151

above. On the other hand, a different complementizer ka is required for CP complements when the matrix verbs select questions. This is illustrated in (11). (11)

a.

b.

c.

kuru] Taroo-wa [CP [TP Hanako-ga kare-no ie-ni H.-nom he-gen house-to come T.-top to/*ka] omotteiru. to ka think ‘Taroo thinks that Hanako is coming to his house.’ Taroo-wa Ziroo-ni [CP [TP Hanako-ga kare-no ie-ni H.-nom he-gen house-to T.-top Z.-dat kuru] *to/ka] tazuneta. come to ka inquired ‘Taroo asked Ziroo if Hanako is coming to his house.’ kuru] Taroo-wa [CP [TP Hanako-ga kare-no ie-ni H.-nom he-gen house-to come T.-top *to/ka] siritagatteiru. to ka want to know ‘Taroo wants to know if Hanako is coming to his house.’

The matrix verbs, tazuneru ‘inquire’and siritagatteiru ‘want to know’, in (11b)– (11c) select questions, and their CP complements must contain ka. It then looks like to is a [−Q] C while ka is a [+Q] C. However, as noted above, the [+Q] C ka can be followed by to in some cases. To embeds a yes/no question in (12a) and a wh-question in (12b). (12)

a.

b.

Taroo-wa Ziroo-ni [CP [CP [TP Hanako-ga kare-no H.-nom he-gen T.-top Z.-dat ie-ni kuru] ka] to] tazuneta. house-to come ka to inquired Lit. ‘Taroo asked Ziroo that if Hanako is coming to his house.’ Taroo-wa Ziroo-ni [CP [CP [TP dare-ga kare-no ie-ni who-nom he-gen house-to T.-top Z.-dat kuru] ka] to] tazuneta. come ka to inquired ‘Lit. Taroo asked Ziroo that who is coming to his house.’

As the matrix verb tazuneru ‘inquire’selects for a question CP as shown in (11b), it appears that the verb and the question complementizer ka enter into selectional relation across to in (12). It is then tempting to hypothesize that to is unspecified with respect to [±Q] and is transparent for the purpose of selection. But this cannot be the correct analysis because not all verbs that select questions allow

152

Mamoru Saito

the ka-to sequence. (13a)–(13b) show that (12a)–(12b) become ungrammatical when siritagatteiru ‘want to know’ is substituted for tazuneta ‘inquired’. (13)

kuru] a. *Taroo-wa [CP [CP [TP Hanako-ga kare-no ie-ni H.-nom he-gen house-to come T.-top ka] to] siritagatteiru. ka to want to know Lit. ‘Taroo wants to know that if Hanako is coming to his house.’ kuru] b. *Taroo-wa [CP [CP [TP dare-ga kare-no ie-ni who-nom he-gen house-to come T.-top ka] to] siritagatteiru. ka to want to know Lit. ‘Taroo wants to know that who is coming to his house.’

It is then necessary to examine which matrix verbs allow the ka-to sequence to find out what is going on in examples such as (12). At this point, Plann’s (1982) analysis of Spanish que mentioned above becomes quite relevant. The examples in (4b)–(4c), where que takes a question CP complement, are repeated below as (14a)–(14b). (14)

a.

b.

Te preguntan que para qu´e quieres el pr´estamo. you ask-3pl que for what want-2sg the loan ‘They ask you what you want the loan for.’ Pens´o que cu´ales ser´ıan adecuados. thought-3sg que which ones would be appropriate ‘He wondered which ones would be appropriate.’

Plann notes that only a subset of those verbs that select for question CPs allow the presence of que. (15) shows some cases where que cannot occur. (15)

supieron /entendieron /recordaron (*que) Ya already found out-3pl/understood-3pl/remember-3pl que por qu´e lo hab´ıas hecho. why it had-2sg done ‘They already found out/understood/remembered why you had done it.’

Examining more relevant examples, she draws the generalization that que can take a question CP as a complement only when the matrix verb is a verb of saying or thinking, that is, a verb that is compatible with direct quotation. Based on this, she goes on to propose that the que-headed CPs express paraphrases of direct discourse in this case.

Sentence Types and the Japanese Right Periphery

153

Plann’s generalization and analysis are directly applicable to Japanese. A partial list of the matrix predicates that allow the ka-to sequence, that is, a toheaded CP with a question complement, is given in (16a). On the other hand, the predicates in (16b) are incompatible with the ka-to sequence. (16)

a.

ka-to:

b. *ka-to:

kiku ‘ask’, situmonsuru ‘question’, yuu ‘say’, sakebu ‘scream’, omou ‘think’ tyoosasuru ‘investigate’, hakkensuru ‘discover’, rikaisuru ‘understand’, siranai ‘don’t know’

The predicates in (16a) are verbs of saying and thinking, and those in (16b) are not. The former can occur with direct quotes, and the latter cannot, as illustrated in (17). (17)

a.

Taroo-wa, “Boku-ga soko-ni ikimasyoo ka” to itta. T.-top I-nom there-to shall go ka to said ‘Taroo said, “Shall I go there?”’ b. *Taroo-wa, “Dare-ga soko-ni ikimasu ka” to siranai. T.-top who-nom there-to go ka to not know Lit. ‘Taroo doesn’t know, “Who is going there?”’

Thus, to can take a question CP as a complement in exactly the same context as que. Then, it too should be analyzed as a complementizer for paraphrases of direct discourse.4 In the following subsection, I examine another peculiar property of to discussed in Kuno (1988). I show that it provides further evidence for this analysis of to, and also for the parallelism between to and que. 2.2. Kuno (1988) on blended and quasi-direct discourse Kuno (1988) examines examples such as (18a), and argues that a to-headed CP complement can represent a “blended discourse,” which starts out as indirect and shifts to direct. (18)

a.

Taroo-wa zibun-no uti-ni kite kure to Ziroo-ni T.-top self-gen home-to come for me to Z.-dat itta. said Lit. ‘Taroo said to Ziroo that come to self ’s house.’

4. It seems that complementizers of this kind are quite widespread. See, for example, Jayaseelan (2008) for relevant discussion on Malayalam, and Grewendorf and Poletto (2009) for a similar phenomenon in Cimbrian, a German dialect spoken in northeastern Italy.

154

Mamoru Saito

b.

Taroo-wa, “Boku-no uti-ni kite kure,” to Ziroo-ni T.-top I-gen home-to come for me to Z.-dat itta. said ‘Taroo said to Ziroo, “Come to my house.”’

He assumes that the predicate of the embedded clause of (18a) represents a direct discourse as it expresses a request. On the other hand, the initial part of the clause must be indirect because zibun ‘self ’ takes the matrix subject Taroo as its antecedent. If it were a direct quotation of Taroo’s utterance, the first person pronoun boku ‘I’ should occur instead of zibun as shown in (18b). In this subsection, I argue that Kuno’s “blended discourse” is indirect discourse, and that the grammaticality of examples such as (18a) is indeed predicted by the analysis of to as a complementizer for paraphrases of direct discourse. Although Kuno analyzes the embedded clause of (18a) as “blended discourse,” he also points out that the direct part cannot be a direct quotation of Taroo’s utterance. Note first that expressions of request vary in form in accordance with the degree of “politeness,” as illustrated in (19). (19)

a.

b.

#Taroo-wa, “Boku-no uti-ni kite kure,” to T.-top I-gen home-to come for me to Ito-sensei-ni itta. I.-Prof.-dat said ‘Taroo said to Prof. Ito, “Come to my house.”’ Taroo-wa, “Watasi-no uti-ni oide itadakemasu ka” T.-top I-gen home-to come for me (polite) ka to Ito-sensei-ni itta. to I.-Prof.-dat said ‘Taroo said to Prof. Ito, “Could you to come to my house?”’

(19a) is not an appropriate utterance of Taroo, a student, to his teacher, Prof. Ito, because kite kure ‘come for me’ is a non-polite, neutral expression. (19b) shows what Taroo would actually say in this context. What Kuno observes is that the judgments, interestingly, are reversed when direct discourse is turned into “blended discourse.” (20) confirms this observation. (20)

a.

kite kure to Taroo-wa zibun-no uti-ni T.-top self-gen home-to come for me to Ito-sensei-ni itta. I.-Prof.-dat said Lit. ‘Taroo said to Prof. Ito that come to self ’s house.’

Sentence Types and the Japanese Right Periphery

155

b. *Taroo-wa zibun-no uti-ni oide itadakemasu ka to T.-top self-gen home-to come for me (polite) ka to Ito-sensei-ni itta. I.-Prof.-dat said Lit. ‘Taroo said to Prof. Ito that could you come to my house?’ (20a) contains the non-polite, neutral form, kite kure ‘come for me’, and is perfectly grammatical. On the other hand, (20b) with the polite form is not. Kuno concludes then that the direct part of “blended discourse” is not precisely direct but only “quasi-direct.” Kuno goes on to discuss why polite expressions are not allowed in “blended discourse.” His answer is that this is because polite forms of verbs do not appear in embedded clauses, as shown in (21). (21)

kaimasita] hon]-o a. *Watasi-wa [NP [ kinoo yesterday bought (polite) book-acc I-top yomimasita. read (polite) ‘I read the book I bought yesterday.’ katta] hon]-o b. Watasi-wa [NP [ kinoo yesterday bought (neutral) book-acc I-top yomimasita. read (polite) ‘I read the book I bought yesterday.’

The sentences in (20) are polite expressions as the matrix verb is in the polite form. Yet, the verb in the relative clause must be in the neutral form as the contrast between (21a) and (21b) indicates. Kuno’s analysis is that the polite form of the expression of request in (20b) is excluded for the same reason. This analysis suggests that “blended discourse” is after all indirect discourse. This is so because it patterns with embedded clauses while direct discourse is known to have matrix properties. Then, the remaining question is why to can embed a sentence expressing a request. This is mysterious if a CP headed by to stands for a proposition. The following English example is totally out: (22)

*John said to Mary that (please) come to his house.

However, the answer is straightforward given the analysis of to proposed in the preceding subsection. To, unlike that, is a complementizer for paraphrases of direct discourse. The direct discourse that is paraphrased can be an expression of request as well as a question. Hence, the analysis predicts correctly that an

156

Mamoru Saito

expression of request can appear as the complement of to just as a question can. It was argued in the preceding subsection that to is just like que in Spanish. Then, it is predicted that “blended discourse” is observed in Spanish as well. This prediction is indeed borne out. Rivero (1994) presents examples in which que takes imperative complements in support of Plann’s (1982) analysis. One of her examples is shown in (23a), together with its direct discourse counterpart in (23b). (23)

a.

b.

Dijo que a no molestarle. said-3sg que to not bother-him ‘He said not to bother him.’ Dijo, “A no molestarme!’’ said-3sg to not bother-me ‘He said, “Don’t bother me!”’

In (23a), the embedded object clitic can co-refer with the matrix subject, and in this case, the embedded clause must represent indirect discourse despite the fact that it is an imperative. Rivero’s conclusion is in fact identical to the one drawn above for “blended discourse” in Japanese. As que can be a complementizer for paraphrases of direct discourse, it is not at all surprising that it can take imperatives as complements. Thus, the comparison of Kuno (1988) and Rivero (1994) leads to another parallelism between to and que. 3.

No as a Complementizer for Propositions

It was shown above that to can be a complementizer for paraphrases of direct discourse exactly like que.According to Plann (1982), que is ambiguous between a complementizer for paraphrases and a complementizer for propositions. In this section, I argue that to is specialized for the former function. More precisely, I argue that there is a division of labor between to and another complementizer no: to is for paraphrases and no is for propositions. 3.1. Clausal Complements Designating States, Events, and Actions An example of que with an embedded proposition was shown in (4a), repeated below as (24). (24)

Sab´ıa que coria. knew-3sg que run-3sg ‘He knew that he was running.’

Sentence Types and the Japanese Right Periphery

157

In a parallel context, the complementizer no is employed in Japanese, as shown in (25).5 (25)

Taroo-wa [CP [TP Hanako-ga soko-ni iru no]-o sitteita. T.-top H.-nom there-in is no-acc knew ‘Taroo knew that Hanako was there.’

This raises the possibility that to, unlike que, is unambiguously a complementizer for paraphrases of direct discourse, and that no is the complementizer for propositions. In this section, I argue that this is indeed the case. Discussing the distributions of to and no, Kuno (1973) presents a rough generalization that no is associated with a factive presupposition in the sense of Kiparsky and Kiparsky (1970) while this is never the case with to.6 However, the distribution of no is much wider than this suggests. Partial lists of the predicates that take to-headed CP complements and those that appear with no-headed CPs are shown in (26). (26)

a.

b.

c.

predicates that take to-headed CP complements omou ‘think’, kangaeru ‘consider’, sinziru ‘believe’, yuu ‘say’, sakebu ‘scream’, syutyoosuru ‘claim, insist’, tazuneru ‘inquire’, kitaisuru ‘expect’, kanziru ‘feel’ predicates that take no-headed CP complements wasureru ‘forget’, kookaisuru ‘regret’, miru ‘see’, matu ‘wait’, tamerau ‘hesitate’, kyohisuru ‘refuse’, ukeireru ‘accept’, kitaisuru ‘expect’, kanziru ‘feel’ predicates that take no-headed CP subjects akiraka ‘clear’, kanoo ‘possible’, kantan ‘easy’, muzukasii ‘difficult’, taihen ‘big deal’

Typical factive verbs such as wasureru ‘forget’ and kookaisuru ‘regret’ take CP complements headed by no. But they are clearly a minority in (26b). Then, what would be the proper characterization of the distributions of to and no? First, the predicates in (26a) are all verbs of saying and thinking. They are indeed all compatible with direct quotation. A couple of examples are given in (27).

5. No is nominal in nature and requires a Case marker when it heads a CP in an argument position. I present one of Murasugi’s (1991) arguments in the following subsection that it should still be considered a complementizer rather than a noun. 6. He also considers koto, which has a similar, though not identical, distribution as no. I do not discuss it here as it is fairly clear that it is a noun.

158 (27)

Mamoru Saito

a.

b.

Taroo-wa, “Boku-no uti-ni atumatte T.-top I-gen house-at gather sakenda. screamed ‘Taroo screamed, “Gather at my house!”’ Hanako-wa, “Watasi-ga Taroo-ni au,” H.-top I-nom T.-dat meet ‘Hanako insisted, “I will go see Taroo.”’

kure,” to for me to

to syutyoosita. to insisted

It seems then that to always serves as a complementizer for paraphrases of direct discourse when it heads a complement CP. The no-headed CPs that appear with the predicates in (26b), on the other hand, all seem to represent events, states, or actions. For example, one regrets a past event or a present state, sees a present event or state, waits for a future event or state, and hesitates to perform an action. The same can be said of those no-headed CPs in the subject position. What can be clear is the existence (or non-existence) of an event or state in the past, present, or future. Similarly, what can be easy or difficult is to perform a certain action. In short, those CPs headed by no seem to always represent propositions. This is consistent with Kuno’s observation that factive verbs take no-headed CPs and never to-headed CPs as complements. Only sentences that express propositions can be presupposed to be true. Hence, those verbs are compatible only with no-headed CPs. Then, there is a division of labor between to and no: to is the complementizer for paraphrases of direct discourse while no is the complementizer for propositions. An additional piece of evidence for this can be found in data from child language. As discussed extensively in the acquisition literature, the overgeneration of no in relative clauses is widely observed with 2–4 year olds. The following examples are from Murasugi (1991): (28)

a.

b.

[ohana motteru *no] wanwa (2;6) flower have no doggie ‘the doggie that is holding flowers’ [buta-san tataiteru *no] taiko (2;11) Mr. Pig is hitting no drum ‘the drum that the pig is playing’

These examples are ungrammatical in adult Japanese with no. Murasugi examines the properties of the overgenerated no in detail, and argues that it is a complementizer. According to her analysis, relative clauses are TPs in adult Japanese. However, children at one point hypothesize that they are CPs, just like English relative clauses, and hence, place no in their head positions. They

Sentence Types and the Japanese Right Periphery

159

only later discover that there is no position for a complementizer in Japanese relatives and cease to overgenerate no. One question that arises with this analysis is why no, and not to, is overgenerated in the head positions of relative clauses. Murasugi (2009) addresses this question, referring to Schachter’s (1973) observation that many languages employ the same complementizer in relative clauses and clefts. No appears in Japanese clefts as shown in (29). (29)

todoita no]-wa Nagoya-kara da. [CP Nimotu-ga package-nom arrived no-top N.-from is ‘It is from Nagoya that a package arrived.’

Then, given Schachter’s generalization, it is not surprising that children overgenerate no. But one may ask further why it is that no, and not to, is employed in clefts and children’s relative clauses. And for this, the analysis of to as a complementizer for paraphrases of direct discourse provides a clear answer. The subject CP of a cleft sentence expresses a proposition and is not a paraphrase of direct discourse. Hence, no must be employed. There is simply no way for to to appear in this context. Similarly, a relative clause does not paraphrase a direct discourse. Then, children could not overgenerate to in relative clauses. This account holds if to is never a complementizer for propositions and is employed exclusively as a complementizer for paraphrases of direct discourse, as argued here. 3.2. The Nominal Nature of no and its Complementizer Status I argued above that no is the complementizer for propositions in Japanese. A CP headed by no requires Case when it is in an argument position as noted in Footnote 5, and it is often called a ‘nominalizer’ in part for this reason. Although whether no is a complementizer or a noun does not affect the overall discussion in this paper, I would like to briefly comment on its complementizer status in this subsection. The Case property of no just mentioned clearly indicates that it is nominal in nature.7 However, it does not provide decisive evidence that it is a noun. First, it is known that complementizers vary with respect to their Case properties. For example, as discussed in detail in Stowell (1981), English CPs headed by that do not appear in typical Case positions like the object position of a preposition, 7. Another relevant fact is that the predicates that take no-headed CP complements correspond roughly to those in English that take gerunds as complements. See Rosenbaum (1967) for detailed discussion on the latter.

160

Mamoru Saito

but there is no such restriction with question CPs. Relevant examples are shown in (30). (30)

a. *They talked about [CP that Mary is a genius]. b. They talked about [CP whether Mary is a genius].

Thus, Stowell concludes that that-headed CPs cannot be Case marked while question CPs can be. Note that question CPs only allow Case and do not require Case. The following examples are perfectly grammatical though the embedded CPs are not in Case positions: (31)

a. b.

They wonder [CP whether Mary is a genius]. (cf. *They wonder it) It was debated [CP whether Mary is a genius]. (cf. *It was debated it)

A parallel observation can be made with to-headed CPs and question CPs in Japanese. Thus, the former cannot appear in the object position of a postposition but the latter can, as shown in (32). (32)

a. *Karera-wa [CP Hanako-ga soko-ni iku to]-nituite H.-nom there-to go to-about they-top kangaeta. considered ‘They thought about Hanako going there.’ b. Karera-wa [CP Hanako-ga soko-ni iku beki H.-nom there-to go should they-top ka]-nituite kangaeta. ka-about considered ‘They thought about whether Hanako should go there.’

However, no adds to the paradigm in the case of Japanese. That is, to resists Case, ka allows Case, and no requires Case, as shown in (33).8 (33)

a.

Karera-wa [CP Hanako-ga soko-ni iku to](*-o) omotta. they-top H.-nom there-to go to-acc thought ‘They thought that Hanako was going there.’

8. The accusative -o is often omitted in colloquial style. But the contrast between (33b) and (33c) is quite clear. The former is grammatical without -o in any register.

Sentence Types and the Japanese Right Periphery

b.

c.

Karera-wa [CP Hanako-ga doko-ni iku beki H.-nom where-to go should they-top kentoosita. discussed ‘They discussed where Hanako should go.’ Karera-wa [CP Hanako-ga soko-ni iru no]*(-o) H.-nom there-in is no-acc they-top ‘They felt that Hanako was there.’

161

ka](-o) ka-acc

kanzita. felt

Thus, the three-way distinction in (34) obtains. (34)

a. b. c.

to cannot appear in a Case position. ka can appear in a Case position. no must appear in a Case position.

It seems difficult to account for this based on the categorial difference between a complementizer and a noun. The whole paradigm instead seems to reflect the lexical properties of the specific items. Stronger arguments for the complementizer status of no are presented in Murasugi (1991). One of them is based on children’s overgeneration of no, discussed above. The relevant examples in (28) are repeated below in (35). (35)

a.

b.

[ohana motteru *no] wanwa (2;6) flower have no doggie ‘the doggie that is holding flowers’ [buta-san tataiteru *no] taiko (2;11) Mr. Pig is hitting no drum ‘the drum that the pig is playing’

In order to examine the category of the overgenerated no, Murasugi considers three possibilities; the genitive Case marker, a pronoun, and a complementizer, which are all homophonous and realized as no. The genitive no appears after any NP or PP within a nominal projection, as illustrated in (36). (36)

a.

b.

c.

Taroo-no hon T.-gen book ‘Taroo’s book’ Hanako-no yooroppa-e-no ryokoo H.-gen Europe-to-gen trip ‘Hanako’s trip to Europe’ midori-iro-no kuruma green-color-gen car ‘a green car’

162

Mamoru Saito

The pronoun no, which roughly corresponds in meaning to one in English, is observed in examples like (37a)–(37b). (37)

a.

b.

mizukasii no Difficult one ‘a difficult one’ [Taroo-ga katta] no T.-nom bought one ‘the one that Taroo bought’

Murasugi first excludes the genitive no by observing the overgeneration pattern in Toyama Dialect, where the genitive is no as in most other dialects but the pronoun and the complementizer are realized as ga. As the Toyama Dialect speaking children overgenerate ga as in (38), the no in (35) cannot be the genitive Case marker. (38)

[anpanman tuitoru *ga] koppu (2;11) ga cup (a character) attach ‘a cup that is pictured with “anpanman” ’

Then, she presents an argument that it is not a pronoun either. Note first that if the “relative clauses” in (35) and (38) are headed by the pronoun no /ga, they must be NPs. This means that the genitive no is required between those “relatives” and the head noun. Murasugi shows through an experimental study that those children who overgenerate no /ga in relative clauses never fail to insert the genitive no after an NP modifying an N. Then, if the “relative clauses” are indeed NPs, the children must insert the genitive no after those “relatives,” but they never do. She concludes then that the overgenerated no /ga cannot be a pronoun and hence must be a complementizer. This argument against the analysis of the overgenerated no /ga as a pronoun suggests simultaneously that the “complementizer no /ga” cannot be a noun. Suppose that the children overgenerate the “complementizer no /ga” in relative clauses as Murasugi argued. If the “complementizer no /ga” is a noun, then the “relative clauses” must be NPs. Then, again, the children must insert the genitive no after those “relative clauses.” Since they do not, it is clear that the children do not consider the “complementizer no /ga” a noun. What the children overgenerate must be no /ga of the category complementizer. This constitutes indirect but strong evidence that the “complementizer no /ga” is not a noun but is indeed a complementizer in adult grammar as well.

Sentence Types and the Japanese Right Periphery

4.

163

Preliminary Notes on the Japanese Right Periphery

It was argued in the preceding sections that Japanese has the three complementizers in (39). (39)

a. b. c.

to is the complementizer for paraphrases of direct discourse in the sense of Plann (1982). ka is the complementizer for CPs that represent questions. no is a complementizer for CPs that represent propositions.

In Section 4.1., I consider the hierarchical relation among those complementizers and suggest, following Hiraiwa and Ishihara (2002), that no is the Finite head. Then, in Section 4.2., I argue that there is a Topic head above the Finite no and below ka, which I consider to be a Force head. This leads to the conclusion that the Japanese right periphery is remarkably similar in structure to the Italian left periphery discussed in Rizzi (1997). 4.1. CP Recursion at the Right Periphery As noted above, an embedded CP in Japanese can contain a sequence of the complementizers ka and to. Another example is shown in (40). (40)

Taroo-wa [CP kare-no imooto-ga soko-ni ita ka (to)] he-gen sister-nom there-in was ka to T.-top minna-ni tazuneta all-dat inquired ‘Taroo asked everyone if his sister was there.’

The number of complementizers in a single CP is not limited to two. There are in fact cases where all three complementizers appear. (40), for example, can have no preceding ka, as in (41). (41)

Taroo-wa [CP kare-no imooto-ga soko-ni ita (no) ka (to)] he-gen sister-nom there-in was no ka to T.-top minna-ni tazuneta all-dat inquired ‘Taroo asked everyone if his sister was there.’

This example instantiates three kinds of complementizer sequences, no-ka, kato, and no-ka-to. Whenever there are multiple complementizers, their order is fixed in this way. This suggests that Japanese CPs can have the recursive structure in (42).

164 (42)

Mamoru Saito

[CP [CP [CP . . . no] ka] to]

The structure in (42) undoubtedly has a semantic basis. Ka can select a CP headed by no as a question can be formed on a proposition. To, in turn, can select a CP headed by ka as to embeds a paraphrase of a direct discourse and the direct discourse can be a question, as discussed in detail in Section 2. One case that (42) allows but is missing is the no-to sequence. (43) is ungrammatical with no. (43)

Taroo-wa [CP kare-no imooto-ga soko-ni ita (*no) to] he-gen sister-nom there-in was no to T.-top omotta. thought ‘Taroo thought that his sister was there.’

This too is expected because a no-headed CP represents a proposition while to embeds a paraphrase of direct discourse. Other orderings of complementizers are plausibly excluded in similar ways. For example, the ka-no sequence is illicit as the content of a proposition cannot be a question. Given the hierarchy in (42), it is tempting to compare it with the structure of the Italian left periphery proposed in Rizzi (1997). As noted at the outset of this paper, he proposes the structure in (44). (44)

[Force [(Topic*) [(Focus) [(Topic*) [Finite [TP . . . ]]]]]]

Ka, being the question marker, is plausibly a Force head. To occupies a higher C position that does not appear in (44). Let us call this C ‘Report’, following Lahiri (1991). Finally, no is analyzed as Finite in Hiraiwa and Ishihara (2002). As it occupies the lowest C position in (42), this fits the hierarchy in (44) well.9 Further, children’s overgeneration of no in relative clauses discussed above receives a straightforward interpretation under this analysis. It is unclear what force relative clauses have, and it is unlikely that children consider relative clauses ForceP. On the other hand, it is not surprising if children produce them as FiniteP. Then, they would overgenerate no in the head position. In the remainder of this subsection, I introduce another piece of suggestive evidence from Matsumoto (2010) for this anaysis of no. Matsumoto examines the types of sentential complements no can take and shows that they are more limited when compared with ka and to. In particular, 9. Thanks to Adriana Belletti for pointing out the relevance of Hiraiwa and Ishihara (2002) in this context. I do not discuss their argument here because it is based on an attractive and yet controversial analysis of Japanese clefts.

Sentence Types and the Japanese Right Periphery

165

she argues that the complement of no must be headed by a morphologically overt T. She first notes that there are modal-like words that do not inflect for tense. Daroo ‘it is probably the case that’ in (45) is one such element. (45)

a.

b.

Taroo-wa soko-ni iru daroo. T.-top there-at is daroo ‘Taroo probably is there.’ Taroo-wa soko-ni ita daroo. T.-top there-at was daroo ‘Taroo probably was there.’

The complement of daroo can be in present or past, but daroo itself does not carry tense. And interestingly, sentences headed by daroo can be embedded under ka or to, but not under no. This is shown in (46). (46)

a.

b.

c.

Taroo-wa [CP ame-ga huru (daroo) ka] kangaeta. rain-nom fall daroo ka considered T.-nom ‘Taroo considered whether it would rain.’ Taroo-wa [CP ame-ga huru (daroo) to] omotta. rain-nom fall daroo to thought T.-top ‘Taroo thought that it would rain.’ Taroo-wa [CP ame-ga huru (*daroo) no]-o kitaisita. rain-nom fall daroo no-acc expected T.-top ‘Taroo hoped that it would rain.’

The ungrammaticality of (46c) with daroo is likely to be due to its incompatibility with no. The example is fine without daroo. Further, as Matsumoto points out, it becomes grammatical also when the formal noun koto is substituted for no as in (47). (47)

Taroo-wa [ame-ga huru (daroo) koto]-o kitaisita. T.-top rain-nom fall daroo N-acc expected ‘Taroo hoped that it would rain.’

Although koto literally means ‘matter’, ‘state’ or ‘fact’, it has little semantic content in this context.10 Thus, there is basically no difference in meaning between (46c) and (47). It seems then that (46c) is out because no, specifically, cannot take a clausal complement headed by daroo. On the basis of observations like this, Matsumoto concludes that no can only take clausal complements that are headed by Tense. This is expected if 10. Kuno (1973) considers it a complementizer, as noted in Footnote 6.

166

Mamoru Saito

no is Finite, because Finite is by definition the C that is closely related with T. Matsumoto’s discussion thus provides suggestive evidence that no is in fact the Finite head. 4.2. The Position of Topic in the CP Structure The hypothesis arrived at so far on the Japanese right periphery is shown in (48). (48)

[. . . [. . . [. . . Finite (no)] Force (ka)] Report (to)]

In this subsection, I consider how Topic fits into this structure. More specifically, I argue that Topic heads can be generated above Finite and below Force as in (49). (49)

[. . . [. . . [. . . [. . . Finite (no)] (Topic*)] Force (ka)] Report (to)]

A classical analysis of topic and focus in Japanese is found in Kuno (1973). Any stressed phrase receives focus interpretation in situ in this language. But Kuno discusses one case where focus interpretation seems to arise in a specific position in the sentence. Let us consider the examples in (50) for illustration. (50)

a.

b.

Hanako-ga sono hon-o yonde ita. H.-nom that book-acc reading was ‘Hanako was reading that book.’ Hanako-ga sono hon-ga suki da. H.-nom that book-nom like ‘It is Hanako that likes that book.’

(50a) can be a neutral description of a past progressive event. On the other hand, (50b) only has the interpretation with focus on the subject Hanako.11 Examining examples of this kind extensively, Kuno proposes the generalization that a matrix-initial nominative phrase receives focus when the predicate is stative. The generalization is on matrix clauses and it does not apply to embedded clauses. Thus, Hanako need not be interpreted as focus when (50b) is embedded as in (51). (51)

katta (koto). Taroo-ga [NP [Hanako-ga suki na] hon]-o H.-nom like book-acc bought fact T.-nom ‘(the fact that) Taroo bought a book that Hanako likes.’

11. Kuno (1973) refers to this kind of focus as ‘exhaustive-listing focus’.

Sentence Types and the Japanese Right Periphery

167

Kuno (1973) also discusses topics that are marked by the particle -wa, and notes that they can receive two distinct interpretations. Taroo in (52a), for example, can be interpreted as a thematic topic or as a contrastive topic. (52)

a.

b.

Taroo-wa sono hon-o yonda. T.-top that book-acc read A. ‘As for Taroo, he read that book.’ (thematic topic) B. ‘Taroo read that book, but I don’t know about the other people.’ (contrastive topic) Taroo-ga sono hon-wa yonda. T.-nom that book-top read ‘Taroo read that book, but I don’t know about the other books.’ (contrastive topic)

He observes in addition that while the contrastive topic interpretation is always available, only a sentence-initial wa-phrase can be construed as a thematic topic. The object is marked by -wa in (52b), and it can only be a contrastive topic. It must be placed at the sentence-initial position as in (53) to receive the thematic topic interpretation. (53)

Sono hon-wa Taroo-ga yonda. that book-top T.-nom read A. ‘As for that book, Taroo read it.’ (thematic topic) B. ‘Taroo read that book, but I don’t know about the other books.’ (contrastive topic)

Kuno notes that here too, sentence-initial means matrix-initial. The topic in the initial position of the relative clause in (54) cannot be construed as a thematic topic. (54)

katta (koto). Taroo-ga [NP [Hanako-wa suki na] hon]-o H.-top like book-acc bought fact T.-nom ‘Taroo bought a book that Hanako, though probably not the others, like.’ (contrastive topic)

Building on Kuno’s observations, Heycock (1994) argues that there is no sentence-initial focus position in Japanese and proposes to analyze the focus interpretation discussed above in the mapping from syntax to information structure. This analysis is extended to thematic topics in Heycock (2008). The strongest piece of evidence for this approach is that those focus and thematic topic interpretations are matrix phenomena. If there were focus and topic positions in Japanese, then they would be expected to occur in embedded clauses as well as

168

Mamoru Saito

matrix clauses. This is indeed the case in Italian, as the following example from Rizzi (1997) shows: (55)

Credo che a Gianni, QUESTO, domani, gli dovremmo I believe that to Gianni this tomorrow we should dire. say ‘I believe that we should say this to Gianni tomorrow.’

In this example, a Gianni and domani are topics and questo is a focus in the complement CP. On the other hand, if foci and thematic topics are represented in the information structure, it is not surprising that they occur only in matrix clauses. Heycock’s argument is well taken for the obligatory focus interpretation of sentence-initial nominative phrases as it is indeed observed only in matrix clauses. It was shown in (51) that it is not observed in a relative clause. Other types of embedded clauses exhibit the same pattern as shown in (56). (56)

a.

b.

suki na no]-o Taroo-ga [CP Hanako-ga sono hon-ga H.-nom that book-nom like no-acc T.-nom wasureteita koto. forgot fact ‘The fact that Taroo forgot that Hanako likes that book.’ suki da to] Taroo-ga [CP Hanako-ga sono hon-ga H.-nom that book-nom like to T.-nom sinziteiru koto. believe fact ‘The fact that Taroo believes that Hanako likes that book.’

In these examples, focus interpretation is not forced on the embedded subject Hanako. However, it has been known that there are cases where thematic topics occur in embedded clauses. Typical examples are shown in (57). (57)

a.

kuru to] Taroo-ga [CP Hanako-wa zibun-no uti-ni H.-top self-gen home-to come to T.-nom sinziteiru koto. believe fact A. ‘(The fact that) Taroo believes that as for Hanako, she is coming to his house.’ (thematic topic) B. ‘(The fact that) Taroo believes that Hanako, though probably not the others, is coming to his house.’ (contrastive topic)

Sentence Types and the Japanese Right Periphery

b.

169

ita to] Taroo-ga [CP Hanako-wa zibun-no uti-ni H.-top self-gen home-at was to T.-nom itta koto. said fact A. ‘(The fact that) Taroo said that as for Hanako, she was at his house.’ (thematic topic) B. ‘(The fact that) Taroo said that Hanako, though probably not the others, was at his house.’ (contrastive topic)

In both of these examples, Hanako can easily be construed as a thematic topic although it is embedded within a to-headed CP. Examples like these are often just mentioned in footnotes as exceptions. But they seem to provide an important clue for the analysis of thematic topics. First, the embedded CPs in (57) clearly represent indirect discourse as the reflexive zibun ‘self ’refers or at least can refer to the matrix subject. Hence, the interpretive property of these examples cannot be attributed to the fact that to can be a marker of direct quotation. Secondly, these examples are after all very similar to the Italian (55), which shows that embedded CPs can contain positions for foci and in particular topics.12 The contrast between (54) and (57) indeed suggests that there is a specific position for thematic topics. (54) shows that thematic topics cannot occur in relative clauses, which I assume are TPs. (57), on the other hand, suggests that they can be located within CPs headed by to, which are the largest CPs according to the recursive CP structure in (48). Then, if thematic topics occupy the Spec position of a Topic head, the following structure accommodates the data discussed so far:13 (58)

[. . . [TopicP Thematic Topic [Topic’ [TP . . . ] Topic]] Report (to)]

The position for thematic topics is outside TP, and hence they cannot occur in relative clauses. But they can be present in CPs headed by to because the position is contained within to-headed CPs. Given this kind of reasoning, it should be possible to pinpoint the location of the Topic head by examining whether thematic topics are possible in other types of CPs. Although the relevant data require subtle judgment in some cases, they indicate that the Topic head is located just above Finite and just below Force. 12. Heycock (2008), for example, does take this kind of exceptions seriously and suggests that a detailed comparison with embedded verb-second in German may prove fruitful. 13. S.-Y. Kuroda assumed over the years that thematic topics are located in CP Spec. Thus, the proposal made here is a refinement of his analysis. See in particular Kuroda (1988) for relevant discussion.

170

Mamoru Saito

First, examples such as those in (59) show that thematic topics cannot be licensed within no-headed CPs. (59)

a.

b.

kuru Taroo-ga [CP Hanako-wa zibun-no uti-ni H.-top self-gen home-to come T.-nom no]-o wasureteita koto. no-acc forgot fact ‘(The fact that) Taroo had forgotten that Hanako, though probably not the others, was coming to his house.’ (contrastive topic) hairu Taroo-ga [CP Hanako-wa zibun-no uti-ni H.-top self-gen house-to enter T.-nom no]-o mita koto. no-acc saw fact ‘(The fact that) Taroo saw Hanako, though not the others, enter his house.’ (contrastive topic)

In both (59a) and (59b), the contrastive topic interpretation is forced on the embedded subject Hanako. This indicates that the Topic head is not contained within a CP headed by no, or more straightforwardly, a FiniteP. On the other hand, the following examples suggest that the thematic interpretation of topics is possible within ka-headed CPs: (60)

a.

b.

kata ka] Taroo-ga [CP Hanako-wa zibun-no hon-o H.-top self-gen book-acc bought ka T.-nom tazuneta koto. inquired fact A. ‘(The fact that) Taroo asked if as for Hanako, she bought his book.’ (thematic topic) B. ‘(The fact that) Taroo asked if Hanako, though probably not the others, bought his book.’ (contrastive topic) kuru no Taroo-ga [CP Hanako-wa zibun-no uti-ni H.-top self-gen home-to come no T.-nom ka] siritagatteiru koto. ka want-to-know fact A. ‘(The fact that) Taroo wants to know if as for Hanako, she is coming to his house.’ (thematic topic) B. ‘(The fact that) Taroo wants to know if Hanako, though probably not the others, is coming to his house.’ (contrastive topic)

Sentence Types and the Japanese Right Periphery

171

It seems then that the Topic head is located within a ka-headed CP. This leads to the more refined CP structure in (61). (61)

[CP . . . [CP . . . [CP thematic topic [C’ [CP [TP . . . ] Finite (no)] Topic]] Force (ka)] Report (to)]

(61) predicts correctly that thematic topics can occur in CPs headed by to or ka, but not in CPs headed by no or TPs. Further, there is evidence that the Topic projection is recursive just as in Italian. As noted above, Kuno (1973) proposed a generalization that only a sentence-initial wa-phrase can be construed as a thematic topic. This is consistent with (62), where only the subject can receive thematic interpretation. (62)

Hanako-wa (kyonen) Teruabibu-e-wa itta. H.-top last year Tel Aviv-to-top went A. ‘As for Hanako, she went to Tel Aviv, but I don’t know about the other places.’ (Hanako-thematic, Tel Aviv-contrastive) B. ‘Hanako went to Tel Aviv, but I don’t know about the other people and the other places.’ (Hanako-contrastive, Tel Aviv-contrastive)

Teruabibu-e ‘to Tel Aviv’ is not sentence-initial, and it can only be a contrastive topic. However, Kuroda (1988) points out that multiple thematic topics are possible when the second topic is preposed over the first. (63) confirms this observation. (63)

Teruabibu-e-wai [Hanako-wa (kyonen) ti itta] Tel Aviv-to-top H.-top last year went A. ‘As for Tel Aviv, Hanako went there, but I don’t know about the other people.’ (Tel Aviv-thematic, Hanako-contrastive) B. ‘As for Hanako, she went to Tel Aviv, but I don’t know about the other places.’ (Tel Aviv-contrastive, Hanako-thematic) C. ‘As for Tel Aviv and as for Hanako, she went there.’ (Tel Avivthematic, Hanako-thematic) D. ‘Hanako went to Tel Aviv, but I don’t know about the other places and the other people.’ (Tel Aviv-contrastive, Hanako-contrastive)

(63) is four-ways ambiguous as indicated: each of the two topics can receive thematic or contrastive interpretation. The interpretation that is important here is the one in C, where both Teruabibu-e ‘to Tel Aviv’ and Hanako are construed as thematic topics. This shows that multiple thematic topics can occur in a single

172

Mamoru Saito

clause contrary to Kuno’s generalization. Given the analysis presented above, it implies that the Topic projection can be recursive.14 The discussion above indicates that the CP system of Japanese is remarkably similar to that in Italian. Rizzi’s (1997) proposal in (44) for Italian is repeated below in (64). (64)

[ . . . Force [ . . . (Topic*) [ . . . (Focus) [ . . . (Topic*) [ . . . Finite [TP . . . ]]]]]]

The structure for Japanese in a parallel format is as in (65). (65)

[ . . . [ . . . [ . . . [ . . . [TP . . . ] Finite] (Topic*)] Force] Report]

There are only two differences aside from the linear order. One is the presence of the Report head in Japanese, as discussed in detail in Section 2. It seems clear that there is a parameter here. Spanish and Japanese have it, but Italian and English do not. The other is the absence of the Focus head in Japanese. For this also, there is likely to be a parameter. That is, languages may vary with respect to the presence/absence of the Focus head within the C system. It would be much too hasty to propose a concrete hypothesis on the possible variations in the left/right periphery just on the basis of (64) and (65). Nevertheless, the preliminary investigation in this paper suggests that the CP structure is fairly rigid across languages with the locus of variation in Report, Focus, and possibly Topic. 5.

Conclusion

In this paper, I have examined the complementizer system of Japanese and presented a preliminary hypothesis on the structure of the Japanese right periphery. I first proposed that to is not a complementizer for embedded propositions as widely assumed, but is a complementizer for ‘paraphrases’ or ‘reports’ of direct discourse just like que in Spanish. I showed that Plann’s (1982) analysis of que is directly applicable to this complementizer. I then argued that no, which Kuno 14. See Saito (2007) for detailed discussion of examples like (63). It is suggested there that thematic topics are licensed clause-initially, and the interpretation in B obtains when the contrastive topic is scrambled over the clause-initial thematic topic. One question that remains is why the two wa-phrases in (62) cannot both be in Spec positions of Top heads and be construed as thematic topics. Although I do not have a clear account for this, I suspect that some sort of crossing constraint is at work, preventing the subject topic from occupying the Spec position of the higher TopicP.

Sentence Types and the Japanese Right Periphery

173

(1973) associates with factivity, has a wide distribution and should be considered the normal complementizer for embedded propositions. As noted above, these descriptive results provide explicit evidence for Plann’s (1982) proposal on Spanish. She proposes that Spanish has three complementizers; que for paraphrases, null C for questions, and que for propositions. Those three are not only present but have distinct phonetic realizations, to, ka and no, in Japanese. In the second part of the paper, I first considered examples where to, ka and no co-occur, and suggested that the three complementizers are hierarchically organized as in (48), repeated below in (66). (66)

[. . . [. . . [. . . Finite (no)] Force (ka)] Report (to)]

I then re-examined the distribution of thematic topics, and showed that they are not limited to the matrix-initial position as widely believed. I argued that they occur not only in to-headed CPs as sometimes observed but also in ka-headed CPs. This led to the hypothesis that there is a Topic projection located above FiniteP and below ForceP. Based on Kuroda’s (1988) observation that multiple thematic topics are possible, I proposed finally that the Japanese right periphery has the structure in (67). (67)

[ . . . [ . . . [ . . . [ . . . Finite (no)] (Topic*)] Force (ka)] Report (to)]

As repeatedly noted, this is quite similar to the structure of the Italian left periphery proposed in Rizzi (1997). Further work is required to discover the precise structure of the Japanese right periphery. But this paper has demonstrated that it is quite rich, much more so than has been assumed, and that it is comparable to Spanish and Italian. Then, it seems fairly clear that its investigation can contribute fruitfully to the research project initiated by Rizzi (1997) on the universal properties and possible variations in the left/right periphery. References Grewendorf, G¨unther and Cecilia Poletto 2009 The Hybrid Complementizer System of Cimbrian. In: Vincenzo Moscati and Emilio Servidio (eds.), Studies in Linguistics Vol. 3: Proceedings XXXV Incontro di Grammatica Generativa, 181–194. Centro Interdipartimentale di Studi Cognitivi sul Linguaggio, Universit´a di Siena. Heycock, Caroline 1994 Focus Projection in Japanese. In: Merc`e Gonz`alez (ed.), Proceedings of the North East Linguistic Society 24, 157–171. Department of Linguistics, University of Massachusetts, Amherst.

174

Mamoru Saito

Heycock, Caroline 2008 Japanese -Wa, -Ga, and Information Structure. In: Shigeru Miyagawa and Mamoru Saito (eds.), The Oxford Handbook of Japanese Linguistics, 54–83. New York: Oxford University Press. Hiraiwa, Ken and Shinichiro Ishihara 2002 Missing Links: Cleft, Sluicing and ‘no da’ Construction in Japanese. In: Tania Ionin, Heejeong Ko and Andrew Nevins (eds.), The Proceedings of Humit 2001, 35–54 (MIT Working Papers in Linguistics 43.). Cambridge, Mass: Department of Linguistics and Philosophy, MIT. Jayaseelan, K. A. 2008 Topic, Focus and Adverb Positions in Clause Structure. Nanzan Linguistics 4: 43–68. Kiparsky, Paul and Carol Kiparsky 1970 Fact. In: Manfred Bierwisch and Karl E. Heidolph (eds.), Progress in Linguistics, 143–173. The Hague: Mouton. Kuno, Susumu 1973 The Structure of the Japanese Language. Cambridge, Mass.: MIT Press. Kuno, Susumu 1988 Blended Quasi-Direct Discourse in Japanese. In: William J. Poser (ed.), Papers from the Second International Workshop on Japanese Syntax, 75–102. Stanford: CSLI Publications. Kuroda, Sige-Yuki 1988 Whether We Agree or Not: A Comparative Syntax of English and Japanese. Linguisticae Investigationes 12: 1–47. Lahiri, Utpal 1991 Embedded Interrogatives and Predicates that Embed Them. Ph.D. dissertation, MIT. Matsumoto, E. 2010 Quotation Expressions and Sentential Complementation in Japanese. B.A. thesis, Nanzan University. Murasugi, Keiko 1991 Noun Phrases in Japanese and English: A Study in Syntax, Learnability, and Acquisition. Ph.D. dissertation, University of Connecticut. Murasugi, Keiko 2009 What Japanese-Speaking Children’s Errors Tell us about Syntax. Presented at GLOW in Asia VII, EFL University, Hyderabad. Plann, Susan 1982 Indirect Questions in Spanish. Linguistic Inquiry 13: 297–312. Rivero, Maria-Luisa 1994 On Indirect Questions, Commands, and Spanish Quotative Que. Linguistic Inquiry 25: 547–554.

Sentence Types and the Japanese Right Periphery Rizzi, Luigi 1997

175

The Fine Structure of the Left Periphery. In: Liliane Haegeman (ed.), Elements of Grammar, 281–337. Dordrecht: Kluwer Academic Publishers. Rosenbaum, Peter S. 1967 The Grammar of English Predicate Complement Constructions, Cambridge, Mass.: MIT Press. Saito, Mamoru 2007 Semantic and Discourse Interpretation of the Japanese Left Periphery. Presented at the Sound Patterns of Syntax Workshop, Ben-Gurion University, Beer Sheva. Schachter, Paul 1973 Focus and Relativization. Language 49: 19–46. Stowell, Timothy A. 1981 Origins of Phrase Structure. Ph.D. dissertation, MIT.

Part III. Clausal Properties of Lexical Categories

On NPs and Clauses* ˇ Zeljko Boˇskovi´c

With a few exceptions, it is standardly assumed that languages without articles have a null D; i.e. the difference between (1) and Serbo-Croatian (SC) (2) is assumed to be PF-based, the D being null in SC. (1)

The stone broke the window.

(2)

Kamen je razbio prozor. stone is broken window

(SC)

In Boˇskovi´c (2008a) I argue that there is a fundamental structural difference in the traditional Noun Phrase (TNP) of English and article-less languages like SC, which can be captured if DP is not even present in the TNPs in (2) (see also Fukui (1988); Corver (1992); Zlati´c (1997); Chierchia (1998); Cheng and Sybesma (1999); Lyons (1999); Willim (2000); Baker (2003), among others for no-DP analyses of at least some article-less languages). My main argument for a fundamental difference in the structure of TNP in languages with and those without articles concerns a number of syntactic and semantic generalizations where the presence/lack of articles in a language plays a crucial role. In this paper I will strengthen the argument by adding a number of new generalizations that were not discussed in Boˇskovi´c (2008a). Several of these generalizations involve a surprising interplay between TNP internal syntax/semantics and clause-level phenomena, which shows that many clause-level phenomena cannot be properly * This paper has benefitted a great deal from comments and help with judgments by numerous linguists. The material from the paper was presented in seminars at the University of Connecticut, and talks at Nanzan University, University of Sarajevo, University of Novi Sad, University of Frankfurt, University of S˜ao Paolo, University of Nova Gorica, University of Venice, University of Leiden, Hungarian Academy of Sciences, Utrecht University, Goethe Universit¨at, University of Paris 8, University of Michigan, FDSL 8 (University of Potsdam), Moscow Student Conference on Linguistics 5, GLOW 33 (University of Wrocław), MayFest 2010 (University of Maryland), Syntax Fest 2010 (Indiana University), and the Linguistic Summer School in the Indian Mountains 5. This work was partially supported by NSF grant BCS-0920888.

180

ˇ Zeljko Boˇskovi´c

understood without paying close attention to the internal structure and interpretation of TNPs. I will also explore consequences of the internal structure of TNP for the internal structure of clauses under the assumption that the two have parallel structure. Taking the TNP/Clause parallelism hypothesis and the NP/DP parameter seriously leads to the conclusion that just like the structure of TNP is poorer in NP languages than in DP languages, the structure of clauses should be poorer in NP languages than in DP languages. I will start by briefly summarizing the generalizations from Boˇskovi´c (2008a). I will then discuss a number of new NP/DP generalizations and then turn to the issue of TNP/Clause parallelism and consequences of the structure of TNP and TNP internal processes for the structure of clauses and clause-level processes. 1.

Generalizations from Boˇskovi´c (2008a)1

1.1. Left Branch Extraction Languages differ regarding whether they allow left-branch extractions (LBE) like (3)–(4). (3)

*Expensive /Those i

he saw

[ti cars]

(4)

Skupa /Ta i expensive/that

je vidio is seen

[ti kola] car

(SC)

(5)

Doroguju /Tu i expensive/that

on videl he saw

[ti maˇsinu] car

(Russian)

Noting a correlation with articles, Uriagereka (1988), Corver (1992) and Boˇskovi´c (2005) establish (6). (6)

Only languages without articles may allow LBE examples like (4).

To illustrate, Boˇskovi´c (2005) notes that Bulgarian and Macedonian, the only two Slavic languages with articles, differ from most other Slavic languages 1. The generalizations from Sections 1 and 2 (the reader should note that most of them are one-way correlations) are still in the process of verification against additional languages. Future research will undoubtedly discover exceptions to many of the generalizations discussed below. However, even if the generalizations turn out to be only strong tendencies, that will still call for an explanation. Note also that a weaker version of the main claim made in this paper would be that some languages without overt articles do not have DP. The stronger (and more interesting) position is that this holds for all languages without overt articles, not just those discussed in the paper.

On NPs and Clauses

181

(e.g. SC, Russian, Polish, Czech, Ukrainian, Slovenian) in disallowing LBE. Within Romance, Latin, which did not have articles, differs from Modern Romance, which has articles, in that it had LBE. Mohawk, Southern Tiwa, and Gunwinjguan languages (see Baker 1996) as well as Hindi, Bangla, Angika, and Magahi also allow LBE and lack articles.2 a. *Novata i ja prodade Petko [ti kola]. new it sells Petko car b. Novata kola ja prodade Petko.

(7)

(Macedonian)

A particularly strong confirmation of (6) is provided by Finnish. As discussed in Laury (1997), Colloquial Finnish has developed a definite article (Dal Pozzo (2007) suggests that it may also be developing an indefinite article). Significantly, Franks (2007) observes that LBE is allowed only in literary Finnish, which does not have articles. Thus, (8a) is acceptable only in literary Finnish. (8)

a.

Punaisen red-acc b. ?*Punaisen red-acc

ostin buy-pst-1sg ostin buy-pst-1sg

auton. (literary Finnish, poetic style) car-acc (sen) auton. (spoken Finnish) the car-acc

Language change can often take a good amount of time. What we are witnessing in Finnish is rather fascinating from this perspective: the emergence of the article has led to a pretty much instantaneous loss of LBE. Another argument regarding language change comes from the history of Greek. Ancient Greek underwent a change from an article-less to an article language. Thus, while Homeric Greek was an article-less language, Koine Greek was a full-blown article language. Taylor (1990) has conducted an investigation of what she refers to as split wh-phrases (involving extraction of just the wh2. I focus on adjectival LBE (demonstratives are adjectives in Slavic LBE languages, see below), ignoring possessor extraction. The reason for this is that several accounts of the ban on AP LBE in article languages leave a loophole for possessor extraction to occur in some languages of this type (see Boˇskovi´c 2005: 4). Thus, Hungarian, which has articles, allows possessor extraction, although it disallows adjectival LBE, which is what is important for our purposes (see, however, den Dikken (1999), who suggests that Hungarian possessive extraction may actually involve a left dislocationtype configuration with a resumptive pronoun). (i)

a. b.

*Magas(-ak-at) l´atott l´any-ok-at. tall-pl-acc saw-3sg girl-pl-acc cf. Magas l´any-ok-at l´atott. ‘Tall girls, he saw.’

182

ˇ Zeljko Boˇskovi´c

word out of a wh-phrase) and split NPs in the history of Ancient Greek and observed a very significant drop in the number of split wh-phrases/NPs in the Homeric and the post-Homeric period. While not all split wh-phrases and split NPs involve LBE, many of them do, which makesTaylor’s results very significant in the current context. Taylor has examined the following texts and periods for Homeric Greek and Koine Greek: 1. Homeric period: Homer – Iliad and Odyssey (8th century BC) 2. Koine period: New Testament corpus (1st century AD). Taylor’s corpus contains 68% of split wh-phrases and 25% of split NPs for the Homeric period, which, as noted above, was an article-less language. On the other hand, the corpus for Koine Greek, an article language, contains only 15% of split wh-phrases and 0% of split NPs.3 Given that many cases of split wh-phrases/NPs involve LBE, these facts strongly confirm the generalization in (6). Before proceeding, let me note that for the purpose of (6) and other generalizations below, I take articles to be unique, i.e. occur once per TNP. The i ending in SC (9) is then not considered to be an article (see Despi´c (2011) for relevant discussion of this element).4 3. The definite article came into general use in the Classical Greek period (which comes in between the Homeric and the Koine Greek period), though it was likely fully established only in the Koine Greek Period. In this respect, it is worth noting that the percentage of split wh-phrases/NPs is significantly lower in the Classical Greek Period than in the Homeric period, though higher than in the Koine Greek period (see Taylor 1990). 4. A word is in order here regarding Modern Greek. Androutsopoulou (1998) claims that Greek allows what appears to be AP LBE. My informants, however, uniformly reject examples like (i) (in fact, even Androutsopoulou notes that speakers have difficulty accepting such examples). (i)

to kokkino idha forema. the red saw dress ‘I saw the red dress.’

Notice also that to, which is traditionally considered to be an article, can appear on more than one element in a TNP (the so-called polydefinite construction), which may cast doubt on its article status. Mathieu and Sitaridou (2002) suggest that this type of “articles” in Greek are actually agreement markers. More importantly for our purposes, Lekakou and Szendr˝ oi (2008), who treat to as a true article, analyze polydefinite constructions as involving multiple full DPs with nominal ellipsis. Under the ellipsis analysis, (i) may be analyzable as involving full DP movement, not LBE, with ellipsis of the NP in the fronted constituent. The analysis makes (i) (and Greek

On NPs and Clauses

(9)

novi /nov crveni auto new-def./new-indef. red-def. car

183

(SC)

Furthermore, it should become clear from the discussion below that what is important for the generalizations given here is the presence/absence of definite, not indefinite articles in a language, given that indefinite articles have often been argued to be located below DP even in languages like English that clearly have DP (see, e.g., Bowers 1987; Stowell 1989; Chomsky 1995; Boˇskovi´c 2007c). In fact, Slovenian, which uncontroversially has indefinite but not definite article, in all relevant respects patterns with article-less languages (see Boˇskovi´c 2009a). Thus, it allows LBE. (10)

Visoke je videl sˇtudente. tall is seen students

1.2. Adjunct Extraction from NP Consider adjunct extraction from TNP, which English disallows (see Chomsky 1986a). (11)

a. Peter met [NP girls from this city] b. *From which city i did Peter meet [NP girls ti ]?

Observing that SC and Russian allow extraction of adjuncts out of TNPs while Bulgarian does not allow it, Stjepanovi´c (1998) argues for (16). Note that Slovenian, Polish, Czech, Ukrainian, Hindi, Bangla, Angika, and Magahi, all articleless languages, pattern with SC and Russian, while Spanish, Icelandic, Dutch, German, French, Arabic, and Basque, which have articles, pattern with English.5 (12)

Iz kojeg grada i je Petar sreo [djevojke ti ]? is Peter met girls from which city

(SC)

(13)

Iz kakogo goroda ty vstrechal [devushek ti ]? from which city you met girls

(Russian)

more generally) fully consistent with the LBE generalization. It is also worth noting that Androutsopoulou (1998) treats (i) in terms of remnant DP fronting. It is shown in Boˇskovi´c (2005) that such an analysis cannot be applied to AP LBE in true AP LBE languages like SC. If Androutsopoulou’s analysis of (i) is correct we may then be dealing here with a different phenomenon from AP LBE in languages like SC (recall, however, that the grammaticality status of (i) is highly controversial.) 5. See Ticio (2003) for Spanish and Fortmann (1996) for German. ((11b) is actually acceptable in Spanish, where the relevant phrase is an argument, as Ticio shows (see Ticio for relevant tests)).

184

ˇ Zeljko Boˇskovi´c

(14)

*Ot koj grad i Petko [sreˇstna momiˇceta ti ]? girls from which city Petko met (Bg, Stjepanovi´c. 1998)

(15)

a. *¿En d´onde robaron [una estatua t]? in where stole a statue (Spanish, Ticio 2003) b. *Fr´a hvaða borg s´erð þ´u stelpur? from which city see you girls (Icelandic)

(16)

Only languages without articles may allow adjunct extraction out of TNPs.

1.3. Scrambling There is also an important correlation between articles and the availability of scrambling.6 (17)

Only languages without articles may allow scrambling.

SC, Russian, Polish, Czech, Slovenian, Latin, Japanese, Korean, Turkish, Hindi, Chukchi, Chichewa, and Warlpiri all have scrambling and lack articles. Particularly interesting are Slavic and Romance. Bulgarian, e.g., has noticeably less freedom of word order than SC. Also, all modern Romance languages have articles and lack scrambling, while Latin lacked articles and had scrambling. It is also worth noting Lakhota, Mohawk, and Wichita, which are also related languages. The latter two lack articles and have more freedom of word order than Lakhota, which has articles. 6. By scrambling I mean the kind of movement referred to as scrambling in Japanese, not German, whose “scrambling” is a very different operation with very different semantic effects from Japanese scrambling. One of the defining properties of scrambling for the purpose of (17) is taken to be the existence of long-distance scrambling from finite clauses, which German lacks (for German, see also Boˇskovi´c (2004a) and Grewendorf (2005)). One needs to be careful here regarding the usage of the term scrambling in the literature, since the term is often used for ease of exposition when an author wants to remain uncommitted regarding the nature of the movement involved. From this perspective, many potential counterexamples to the scrambling generalization can be easily explained away. Consider, e.g., Albanian and Greek, which are sometimes said to have scrambling. However, an object that is fronted to a sentence initial position in these languages either has to be clitic doubled or contrastively focused and adjacent to the verb (see here Kallulli 1999). This indicates that object fronting involves either clitic left dislocation or focus movement, not what is referred to as scrambling in Japanese.

On NPs and Clauses

185

1.4. Negative Raising I now turn to a generalization regarding negative raising (NR), where negation can be taken to be either in the matrix or the embedded clause of (18). The embedded clause option is confirmed by the strict clause-mate NPIs in (21). That these items require negation is shown by (19), while (20) shows that nonNR verbs like claim disallow long-distance licensing of these items. Since they require clause-mate negation, negation must be present in the embedded clause of (21) when the NPIs are licensed. (18)

John does not believe that Mary is smart.

(19)

a. *John left until yesterday. b. John didn’t leave until yesterday. c. *John has visited her in at least two years. d. John hasn’t visited her in at least two years.

(20)

a. *John didn’t claim [ that Mary would leave [NPI until tomorrow]] b. *John doesn’t claim [that Mary has visited her [NPI in at least two years]]

(21)

a. b.

John didn’t believe [ that Mary would leave [NPI until tomorrow]] John doesn’t believe [that Mary has visited her [NPI in at least two years]]

Before establishing the NR generalization, note that for the purpose of the generalization I confine myself to negative raising from finite clauses and use as the relevant diagnostics the ability of NR to license strict-clause mate NPIs. A crosslinguistic check of the availability of NR under these conditions reveals the following: (22)

Negative raising is disallowed in languages without articles.

SC, Czech, Slovenian, Polish, Russian, Ukrainian, Turkish, Korean, Japanese, and Chinese lack articles and NR (i.e. strict clause-mate NPI licensing under NR). On the other hand, English, German, Spanish, French, Portuguese, Romanian, and Bulgarian have both articles and NR (i.e. allow strict clause-mate NPI licensing under NR, see Boˇskovi´c (2008a)). In light of this, we may in fact be dealing here with a two-way correlation, which would strengthen (22) to (23). (23)

Languages without articles disallow NR, and languages with articles allow it.

186

ˇ Zeljko Boˇskovi´c

Interestingly, even in languages where the NPI test fails negation is interpretable in the lower clause: SC (24) has the atheist (non-agnostic) meaning “Ivan believes God doesn’t exist” (the same holds for Korean, Japanese,Turkish, Chinese, Russian, Polish, and Slovenian). (24)

Ivan ne vjeruje da bog postoji. Ivan neg believes that God exists

This suggests that lower clause negation interpretation and strict NPI licensing under NR should be divorced (contrary to the standard practice, where the two are correlated), with a three-way split among verbs: (a) negation interpreted in the lower clause and strict NPIs licensed under NR (possible only for some verbs in languages with articles) (b) negation interpreted in the lower clause, strict NPIs not licensed (c) no NR at all. 1.5. Superiority and Multiple wh-Fronting MWF languages differ regarding whether they show Superiority effects (strict ordering of fronted wh-phrases) in examples like (25)–(26). It turns out that there is a correlation between Superiority effects with multiple wh-fronting (MWF) and articles, given in (27). (25)

a.

Koj kogo viˇzda? who whom sees ‘Who sees whom?’ b. *Kogo koj viˇzda? (Bulgarian)

(26)

a. b.

(27)

Ko koga vidi? who whom sees Koga ko vidi?

(SC)

MWF languages without articles do not display superiority effects in examples like (25)–(26).

MWF languages without articles do not show Superiority effects. This is the case with SC, Polish, Czech, Russian, Slovenian, Ukrainian, and Mohawk. MWF languages that show Superiority effects all have articles. This is the case with Romanian, Bulgarian, Macedonian, Basque, and Yiddish. Hungarian is an exception (it has articles and no superiority), which however does not violate (27).7 7. Interestingly, Watanabe (2003) suggests that Hungarian traditional definite article is not a D-element, which casts doubt on its DP status. (For relevant discussion of Hungarian MWF, see Boˇskovi´c (2007a).)

On NPs and Clauses

187

1.6. Clitic Doubling Another generalization concerns clitic doubling, where Slavic again gives us a useful clue. Clitic doubling is allowed only in two Slavic languages, Bulgarian and Macedonian, which also have articles.8 Slavic languages without articles, like SC, disallow it. In fact, all clitic doubling languages I am aware of (Albanian, Macedonian, Bulgarian, Greek, Somali, Spanish, French (some dialects), Catalan, Romanian, Hebrew, Dutch (some dialects)) have articles. We then have (29). (28)

(29)

a.

Ivo go napisa pismoto. Ivo it wrote letter-the ‘Ivo wrote the letter.’ b. *Ivan (*ga) napisa pismo. Ivan it wrote letter

(Bulgarian/Macedonian)

(SC)

Only languages with articles may allow clitic doubling.

1.7. Adnominal Genitive Willim (2000) notes English, Arabic, Dutch, German, and Catalan, all article languages, allow two nominal genitive arguments, i.e. both the external and the internal argument can be genitive, where the genitive is realized via a clitic/ suffix or a dummy P (30). The same holds for Portuguese, Basque, French, Greek, Hebrew, Icelandic, Macedonian, Bulgarian, Spanish, Welsh, Maltese, Maori, Samoaon, Swedish, all article languages. On the other hand, Willim notes article-less languages Polish, Czech, Russian, and Latin disallow two lexical genitives. (Note that word order does not matter in (31). In languages of this type the external argument is generally realized via a PP headed by an analogue of English by (a semantically contentful P) or an inherent oblique Case, see (32).) The same holds for SC, Ukrainian, Quechua, and Turkish. This leads to (33).9 8. Note that the doubled NP in (28a) is in situ, it is not right-dislocated; note also that true clitic doubling is associated with a definiteness/specificity effect. 9. (33) concerns only nominal arguments, not possessives, and disregards inherent Case, as in SC liˇsavanje (depriving) sina (son, gen) njegovog (his, gen) nasledstva (inheritance, gen) ‘depriving the son of his inheritance’ (see Zlati´c 1997), where the second genitive is inherent. (The second NP remains genitive even when the casemarker is a verb, as in On liˇsava sina njegovog nasledstva ‘He is depriving the son of his inheritance’. I am not concerned with this type of lexically specified cases here.) I also ignore for obvious reasons languages such as Japanese which allow multiple identical case marking constructions. (The same holds for languages like Estonian,

188 (30)

ˇ Zeljko Boˇskovi´c

a.

b.

(31)

(32)

a. *odkrycie Ameryki Kolumba discovery America-gen Columbus-gen ‘Columbus’ discovery of America’ ˇ ıma b. *zniˇcen´ı R´ barbar˚u destruction Rome-gen barbarians-gen ‘The barbarian’s destruction of Rome’ a.

b.

(33)

Hannibals Eroberung Roms Hannibal-gen conquest Rome-gen ‘Hannibal’s conquest of Rome’ (German) l’avaluaci´o de la comissi´o dels resultats the evaluation of the comitte of the results ‘the committee’s evaluation of the results’ (Catalan)

(Polish)

(Czech)

odkrycie Ameryki przez Kolumba discovery America-gen by Columbus ‘the discovery of America by Columbus’ ˇ ıma zniˇcen´ı R´ barbary discovery Rome-gen barbarians-instr ‘the destruction of Rome by the barbarians’

Languages without articles do not allow transitive nominals with two lexical genitives.

1.8. Superlatives ˇ Zivanoviˇ c (2008) notes that Slovenian (34) does not have the reading where more than half the people drink beer. It only has the reading where more people drink beer than any other drink though it could be less than half the people. (34)

Najveˇc ljudi pije pivo. most people drink beer. ‘More people drink beer than drink any other beverage.’ (Plurality reading, MR) ‘*More than half the people drink beer.’ (Majority reading, PR)

English most gives rise to both readings, though in different contexts. German MOST also has both readings.

which allows multiple genitives on adjectives that does not arise through case concord with a noun (the noun can be non-genitive in such cases).)

On NPs and Clauses

(35)

189

Die meisten Leute trinken Bier. the most people drink beer. ‘More than half the people drink beer.’ ‘More people drink beer than any other drink.’ (with focus on beer.)

ˇ Zivanoviˇ c notes English, German, Dutch, Hungarian, Romanian, Macedonian, and Bulgarian, which have articles, allow the majority reading (the same holds for Basque and Arabic). The reading is disallowed in Slovenian, Czech, Polish, SC, Chinese, Turkish, and Punjabi, which lack articles and allow only the plurality reading (the same holds for Hindi, Angika, and Magahi).10 We then have (36) (I set aside cases where the majority reading is expressed with a noun like majority). (36)

Only languages with articles allow the majority superlative reading.

1.9. Head-Internal Relatives and Locality There is a locality distinction among languages with head-internal relatives (HIR). HIRs in Japanese, Korean, Quechua, Navajo, and Mohawk are island sensitive, while those in Mojave and Lakhota are not (Basilico 1996; Watanabe 2004; Baker 1996). Interestingly, the former lack articles, while the latter have them. This leads to (37). (37)

Head-internal relatives display island-sensitivity in article-less languages, but not in languages with articles.

Grosu and Landman (1998) show that there is also a semantic difference at work here, in particular, HIRs are restrictive in languages with articles and maximalizing in those without articles. 1.10. Polysynthetic Languages Baker (1996) observes the following generalization regarding polysynthetic languages. (38)

Polysynthetic languages do not have articles.

10. The following context enforces the majority reading for (the past tense version) of (35): Suppose people at a dinner were allowed more than one beverage. 60% of the people had a beer and 75% of the people had a glass of wine.

190 2.

ˇ Zeljko Boˇskovi´c

Additional Generalizations

I now turn to new generalizations that were not discussed in Boˇskovi´c (2008a). (Anticipating the discussion in Sections 3–4, from now on I will refer to languages with articles as DP languages, and to languages without articles as NP languages.) 2.1. Focus Morphology In some languages, negative constituents have overt focus morphology (see (39)). Such morphology is often realized through the presence of focal elements like even, also, or too (SC has two series of negative constituents, a negative concord series and an NPI series, both of which contain even), and sometimes through obligatory emphatic (focus) stress, as in Greek.11 (39)

n+i+ko i+ko neg+even+who even+who ‘noone/anyone’

(SC)

While in DP languages negative constituents may but do not have to have a focus marker, in NP languages they have a focus marker. This holds for SC, Russian, Polish, Ukrainian, Lithuanian, Hindi, Chinese, Japanese, Korean, Finnish,Yakut, Lezgian, Kannada, Quechua, Mansi, Latin, Persian, Turkish, and Kazakh.12 This leads to (40). (40)

Negative constituents must be marked for focus in NP languages.

2.2. Negative Concord with Complex Negative Constituents It is well-known that in some negative concord languages, the negative concord reading is unavailable with complex negative constiutents (NCIs). This is illustrated below with examples from Italian. (41a)–(4b) show that Italian is 11. Another option, which may be realized in Slovenian, may be obligatory focus movement of the negative constituent. 12. Boˇskovi´c (2009c) argues that in languages with both negative concord and NPI series, the two are derived from the same underlying items, which means it suffices for one of these to have a focus marker to meet (40). There is a bit of a complication with Persian. Persian negative concord series contains hic. Hic is analyzable as air one+´ciy, an emphatic particle which I assume is focus related. Hic was borrowed into Turkish and Kazakh, which I assume can be analyzed in the same way. A potential counterexample to the generalization in (40) is Georgian. Georgian does have a focus marker in the existential quantifier series, but not in NCIs. I leave a detailed examination of Georgian for future research.

On NPs and Clauses

191

a negative concord language. However, the negative concord reading becomes unavailable with multiple NCIs if one of the NCIs is a complex element. (41)

a.

b.

c.

Non ho visto nessuno /nessuno studente. neg have seen nobody/no student ‘I didn’t see anybody/any students.’ (negative concord only) Nessuno ha letto niente. nobody has read nothing (negative concord or double negation) Nessuno student ha letto nessun libro /niente. no student has read no book/nothing (double negation only)

It turns out that DP languages differ with respect to whether the double negation reading is forced in examples like (41c). Thus, the reading is forced in Italian, Spanish, West Flemish, and French. However, Brazilian Portuguese, Basque, Hebrew, and Romanian still allow the negative concord reading. On the other hand, NP negative concord languages all allow the negative concord reading in examples like (41c). This is, e.g., the case with SC, Russian, Polish, Ukrainian, Japanese, Korean, and Turkish. I am in fact unaware of any negative concord NP language that would disallow the negative concord reading. We are then led to the following generalization. (42)

The negative concord reading may be absent with multiple complex negative constituents only in DP negative concord languages.

2.3. Quantifier Scope Consider now (43). (43)

Someone loves everyone.

The example is ambiguous: Everyone can take either narrow or wide scope in (43). I will refer to the latter reading as the inverse scope reading. A number of languages disallow the inverse scope reading in the unmarked order for the subject, verb, and object (SVO in SVO languages, and SOV in SOV languages). Thus, inverse scope is allowed in English, Spanish, Brazilian Portuguese, Macedonian, and Hebrew, but it is impossible in German, Basque, Dutch, Icelandic, Bulgarian, Welsh, Romanian, Japanese, Korean, Turkish, Persian, Hindi, Bangla, Chinese, Russian, Polish, Slovenian, Ukrainian, and SC. Focusing on the latter group of languages, while the first seven are DP languages, other languages in

192

ˇ Zeljko Boˇskovi´c

this group are all NP languages. In fact, I do not know of any NP language that productively allows inverse scope in such examples.13 We then have (44). (44)

Inverse scope is unavailable in NP languages in examples like (43).

2.4. Radical Pro-drop I now turn to the phenomenon of radical pro-drop, which I define as productive argumental pro-drop of both subjects and objects in the absence of rich verbal agreement. This type of pro-drop differs from pro-drop in languages like Spanish, where pro-drop is licensed by rich verbal morphology. As a result, since Spanish has subject but not object agreement, pro-drop is allowed only with subjects in Spanish. Radical pro-drop is allowed in Japanese, Chinese, Korean, Kokota, Turkish, Hindi, Wichita, Malayalam, Thai, Burmese, and Indonesian, all of which are NP languages.14 In light of this, we have the generalization in (45). (45)

Radical pro-drop is possible only in NP languages.

2.5. Number Morphology Gill (1987), who considers only a few languages, suggests a potential correlation between obligatory number morphology and the availability of articles. The phenomenon we are looking at here is the possibility of having examples like (46), where the N can be interpreted as plural in the absence of plural morphology. (47) divides languages into two groups, where one group has languages that at least optionally can lack number morphology with at least some Ns (i.e. where some or all countable Ns can receive plural interpretation without the

13. Certain quantifiers always require wide scope. Such quantifiers tend to be interpreted with wide scope even in NP languages. What I am concerned with here is quantifiers that do not require wide scope, i.e. whether the inverse scope is productively available for all quantifiers in a given language in this type of examples. 14. See also Tomioka (2003) (and Saito (2007) and Neeleman and Szendr˝ oi (2007) for different perspectives). Turkish seems to combine Spanish type and radical pro-drop. A potential problem that is being investigated is Cheke Holo. If Cheke Holo indeed turns out to be an exception, (45) would simply be a strong tendency. (It is worth noting here that Brazilian Portuguese is not classified as a radical pro-drop language, since it does not have fully productive subject drop. In fact, since pronoun objects could incorporate into V, the availability of a fully productive subject pro-drop is crucial here.)

On NPs and Clauses

193

presence of number morphology), and the other group contains languages that have obligatory plural morphology (on either D or N).15 (46)

(47)

Susumu-ga hon-o yonda. Susumu-nom book-acc bought ‘Susumu bought a/the book/books.’

(Japanese)

No obligatory number morphology: Japanese, Korean, Chinese, Hindi, Bangla, Malayalam, Mohawk, Dyirbal, Warlpiri, Warrgamay, KukuYalanji, Indonesian, Vietnamese. Obligatory number morphology: Russian, SC, Ukrainian, Hebrew, Portuguese, German, Bulgarian, Polish, Hungarian, Spanish, Romanian, French, Slovenian, Finnish, Bulgarian, Swahili, Greek, Dutch, Italian, Latin, Ossetic, Kannada, Macedonian, Somali, Estonian.

While the second group comprises both NP and DP languages, all languages in the first group are NP languages.16 We then have the generalization in (48). (48)

Number morphology may not be obligatory only in NP languages.

2.6. Focus Adjacency The next generalization deals with the question of whether languages that have focus movement require adjacency with the verb. It is well-known that some languages require movement of focalized elements. Such languages differ regarding whether the fronted focalized element has to be adjacent to the verb. I illustrate this for Bulgarian (49) and SC (50). (Capital letters indicate contrastive focus.) (49)

a. *Kartinata Ivan podari na painting-the(foc) Ivan give-as-a-present-pst-3sg to Maria. Maria ‘Ivan gave Maria the painting as a present.’ (Lambova 2004) b. Kartinata podari Ivan na Maria.

15. I ignore here TNPs involving numerals, since numerals by their very nature express number. 16. The NP/DP status of Vietnamese is somewhat controversial; see, however, Cheng (in preparation) for arguments that Vietnamese lacks true articles, hence should be classified as an NP language.

194 (50)

ˇ Zeljko Boˇskovi´c

Jovana (Petar) savjetuje. Jovan-acc Petar-nom advises ‘Petar is advising Jovan.’

(Stjepanovi´c 1999)

It turns out that Basque, Hungarian, Bulgarian, Armenian, Greek, Catalan, Romanian, Macedonian, Italian, Spanish, and Albanian are subject to the adjacency requirement. This is not the case with Slovenian, Russian, SC, Polish, Chinese, and Nupe.17 The adjacency languages are all DP languages, while the non-adjacency languages are all NP languages. We may then have here another NP/DP generalization. (51)

Elements undergoing focus movement are subject to a verb adjacency requirement only in DP languages.

2.7. Interpretation of Possessors Partee (2006) observes that English (52) presupposes that Zhangsan has exactly three sweaters. On the other hand, there is no such exhaustivity presupposition in Chinese (53).18

17. Turkish is also subject to the adjacency requirement. However, S¸ener (2006) provides convincing evidence that Turkish actually does not have focus movement. Rather, focalized elements in Turkish remain in their base position, where they are subject to a prosodic requirement that focalized elements be exhaustively parsed into the same intonational phrase as the verb. The adjacency requirement in Turkish is therefore phonological, not syntactic in nature (in fact, it affects both contrastively focused elements and elements that bear simple new information focus, a state of affairs that is not found with focus movement, which typically affects only the former in the case of focalized non-wh-phrases.) Notice also that S¸ener (2010) argues that all elements that are interpreted as old information (both topics and discourse anaphoric elements) must undergo movement out of vP in Turkish, this in fact being the only movement that the language has, which leaves only focalized elements next to the verb. Since Turkish does not have focus movement, it is irrelevant for the phenomenon under consideration here. 18. There is more than one option for word order regarding Chinese possessors. I focus here on the possessor-numeral order (in the absence of a demonstrative); see Partee (2006) for discussion of the full paradigm. (Constructions where the posessor precedes the numeral are definite, while those where the possessor follows the numeral are indefinite and roughly correspond to English three sweaters of John’s. I exclude such examples from the discussion here since I am focusing on definite possessor phrases. For this reason I also focus on article+possessor constructions in DP languages that allow the two to co-occur.)

On NPs and Clauses

(52)

Zhangsan’s three sweaters

(53)

[san jian maoxianyi] Zhangsan de Zhangsan DEPOSS three CL sweater ‘Zhangsan’s three sweaters’

195

NP languages I have checked so far, Russian, SC, Turkish, Japanese, Korean, Hindi, Bangla, Malayalam, and Magahi, all pattern with Chinese in this respect. (Partee notes this for Russian.) Spanish, Brazilian Portuguese, Italian, Basque, Hebrew, Dutch, and Arabic, on the other hand, pattern with English. This leads to (54). (54)

Possessors may induce an exhaustivity presupposition only in DP languages.

It is worth noting in this respect that Lyons (1999) argues that the DP projection is responsible for the presupposition of uniqueness/exhaustivity. It is then not surprising that the presupposition is lacking in article-less languages, which lack DP.19 2.8. Classifiers Cheng (in preparation) examines languages with obligatory classifier systems and notes a correlation with absence/presence of articles, given here in (55). (55)

Obligatory numeral classifier systems occur only in NP languages.

In other words, if a language has an obligatory classifier system, it does not have DP. What this generalization suggests is that there is an incompatibility between a classifier system, i.e. ClP, and DP. This in turn can be interpreted as support for Cheng and Sybesma’s (1999) proposal that classifiers do the job of D in languages like Chinese. In other words, since ClP and DP basically do the same job, a language cannot have both. Note, however, that we are not simply 19. Lyons in fact argues that there is no grammaticalized definiteness in Chinese, i.e. he also argues that Chinese (and other article-less languages) lack DP. Chinese (and other article-less languages) do have some definiteness effects. However, he argues definiteness effects found in such languages represent a semantico/pragmatic notion of definiteness as identifiability (which seems to correspond to what Partee calls familiarity), correlated with topichood. According to Lyons, such non-grammaticalized definiteness fails to involve the presupposition of uniqueness/exhaustivity (i.e. nongrammaticalized definiteness normally lacks the uniqueness/exhaustivity presupposition).

196

ˇ Zeljko Boˇskovi´c

dealing here with a label difference, where DP is replaced by ClP. Cheng and Sybesma show that ClP is very low in the structure. This means that the source of definiteness is lower in the structure in NP than in DP languages. In work in preparation I take advantage of this to account for the well-known fact that number (more precisely, plurality) interacts with definiteness in ClP languages like Chinese but not in DP languages like English (the reason for this being that the projection that is responsible for plurality is higher than ClP (the source of definiteness in Chinese), but lower than DP (the source of definiteness in English). 2.9. Second Position Clitics Another generalization concerns the type of clitics a language has (see also Migdalski (2010) and Runi´c (2011)). Languages typically have either verbal (i.e. V-adjacent) clitics or so-called second position clitics.20 Languages that are standardly assumed to have second position clitics include a number of Slavic languages (SC, Czech, Slovak, Slovenian, Hucul Ukrainian, and Sorbian), Latin, Ancient Greek, Pashto, Tagalog, Ngiyambaa and Warlpiri (and a number of north-central Australian languages), which interestingly all lack articles.21 This leads us to the generalization in (56). (56)

Second-position clitic systems are found only in NP languages.

Slavic and Romance are again quite informative: while a number of Slavic languages have second-position clitic systems, Bulgarian and Macedonian are glaring exceptions. As for Romance, Latin had second-position clitics, while Modern Romance languages lack them. The history of Greek provides a rather strong confirmation of (56). Thus, Taylor (1990) shows that 90% of enclitics in the Homeric period, when Greek did not have articles, were in the second position; this simple second position cliticization system broke down in the later stages (i.e. article stages), like Koine Greek. Additional generalizations will be discussed in Section 5; for additional generalizations that cut across the DP/NP line, the reader is also referred to Herdan (2008), Marelj (2008), Boeckx (2003a), Runi´c (2011), Despi´c (2011), and Boˇskovi´c (2009d). Taken together, these generalizations provide strong evidence that there is a fundamental difference between TNP in languages like English 20. I am simplifying here the actual state of affairs. Note that true second-position clitics are not simply enclitics (i.e. not all enclitics are second-position clitics). I refer the reader to Boˇskovi´c (2001) and references therein for discussion. 21. Uto-Aztecan languages are currently being investigated in this respect.

On NPs and Clauses

197

and article-less languages like SC that cannot be reduced to phonology (overt vs phonologically null articles). If DP is posited for both, we need to make a radical principled distinction between D in English and D in SC. Appealing to phonological overtness will not work since English, e.g., disallows LBE (*Fast, he likes cars), adjunct extraction from TNP, and scrambling even with null D. Moreover, the above generalizations deal with syntactic and semantic, not phonological phenomena. It is often assumed that the TNP should be treated in the same way in articleless languages and languages like English for the sake of uniformity. This argument fails on empirical grounds in light of the above generalizations: it is simply a fact that there are radical syntactic and semantic differences between the two – there is no uniformity here. Most importantly, as shown in Boˇskovi´c (2008a), Boˇskovi´c and Gajewski (in press), and Section 4, these differences (i.e. all the generalizations discussed above) can be deduced if there is DP in the TNP of English, but not languages like SC. Moreover, the NP/DP analysis provides a uniform account of these differences, where a single difference between the two types of languages is responsible for all of them. It is extremely hard to see how this can be accomplished under a uniform DP analysis. In fact, I contend that a universal DP analysis cannot even be entertained seriously until it can be shown that the analysis can also provide a principled, uniform account of the above generalizations. 3.

D-like Items in Article-less Languages

It should also be noted that traditional D-items do not exhibit the behavior that is standardly associated with D-items in article-less languages. Let us take a look at SC as a representative of NP languages in this respect. Although SC does not have articles, it does have lexical items like that, some, and possessives. However, such items behave like adjectives in SC both morphologically and syntactically (see Zlati´c 1997 and Boˇskovi´c 2008a).22 In contrast to English D-items, they clearly have the morphology of adjectives (57), occur in typical adjectival positions like the predicate position of a copula (58), allow stacking up (59), and often (though not always) fail to induce Specificity effects that English D-items induce (60). Another interesting quirk is that SC possessives 22. The point of the following discussion is to demonstrate that the SC items in question behave differently from their English counterparts; we would not necessarily expect that the items in question will exhibit the same behavior in all NP languages or rule out the possibility that in some DP languages some of the items under discussion could exhibit some of the properties of the SC items in question.

198

ˇ Zeljko Boˇskovi´c

cannot be modified by adjectives (61), which follows if adjectives cannot modify adjectives given that SC possessors are actually adjectives. (57)

a.

b.

tim thoseFEM.PL.INST djevojkama girlsFEM.PL.INST tih thoseFEM.GEN.PL djevojaka girlsFEM.GEN.PL

nekim mladim someFEM.PL.INST youngFEM.PL.INST

nekih mladih someFEM.GEN.PL youngFEM.GEN.PL

(58)

a. *This book is my. b. Ova knjiga je moja. this book is my

(59)

a. *this my picture b. ta moja slika this my picture

(60)

O kojem piscu je proˇcitao [svaku knjigu / sve knjige / about which writer is read every book / all books / (tu) tvoju knjigu ti ]? that your book ‘*About which writer did he read every book/all books/this book of yours?’

(61)

*bogati susjedov konj rich neighbor’s horse

They also have some freedom of word order. While English D-items must precede adjectives, SC allows adjectives to precede some D-items (see Boˇskovi´c 2007b for some interpretational effects regarding the order of adjectives and possessors). (62)

Jovanova skupa slika vs. skupa Jovanova slika Jovan’s expensive picture *expensive Jovan’s picture

(63)

bivˇsa Jovanova ku´ca vs. Jovanova bivˇsa ku´ca *former Jovan’s house Jovan’s former house

Notice, however, that the order of SC adjectives and D-items is not completely free. Thus, both adjectives and possessives must follow demonstratives.

On NPs and Clauses

(64)

a. b.

ova this ova this

skupa expensive Jovanova Jovan’s

199

kola /?*skupa ova kola car slika /?*Jovanova ova slika picture

These ordering restrictions follow straightforwardly from the semantics of the elements in question (see Boˇskovi´c and Hsieh 2012 regarding Chinese). Semantically, it makes sense that possessives and adjectives should be able to occur in either order. The most plausible semantics for possessives is modificational (see e.g. Partee and Borschev 1998 and Larson and Cho 1999). (65)

Partee and Borschev (1998) (Ri is a free variable) [[ Mary’s ]] = λ x.[Ri (Mary)(x)]

(66)

Larson and Cho (1999) [[ to Mary ]] = λ x.[POSS(j,x)]

Given the standard assumptions that adjectives are also of type and that there is a rule of intersective Predicate Modification, compositional semantics imposes no restrictions on the order in which possessives and adjectives may be composed. On the other hand, the situation is different with demonstratives. Kaplan (1977/1989) argues that demonstratives are markers of direct reference. In other words, demonstrative noun phrases pick out an individual of type e. The individual is picked out at least partially as a function of its predicate complement phrase. Thus, a demonstrative element like that is a function of type e,t>,e>. Once a demonstrative has mapped a nominal element to an individual, further modification by predicates of type is impossible. Hence, semantic composition requires both adjectives and possessives to be composed before demonstrative determiners. In other words, semantic composition allows possessives to be composed either before or after modifying adjectives, while demonstratives must be composed after both adjectives and possessives.23 This perfectly matches the actual facts regarding the ordering of the elements in question in SC. It is worth noting in this respect that English counterparts of the unacceptable examples in (64) are significantly worse than the SC examples. This follows if the English examples have the semantic violation we have discussed as well as a syntactic violation (violations of the requirement that DP must be projected on top of TNP and whatever is responsible for the incompatibility of articles and possessives in English).

23. Note that the above account readily extends to non-restrictive adjectives under Morzycki’s (2008) analysis, where non-restrictive adjectives are also treated as having type and required to be interpreted inside the determiners.

200 (67)

ˇ Zeljko Boˇskovi´c

a. **expensive this car b. **John’s this picture

The proponents of the universal DP analysis (Baˇsi´c 2004; Rappaport 2000; Pereltsvaig 2007) account for (64) by placing the demonstrative in DP, which is located above the projection where possessives and adjectives are located. (α P is a projection where adjectives are generated, with multiple adjectives requiring multiple α Ps.) (68)

[DP Demonstrative [PossP Possessive [α P Adjective [NP

(Baˇsi´c 2004)

(68) accounts for (64), but it fails to capture the relative freedom of the adjectives/possessives order in SC and the SC/English contrast in this respect. Furthermore, Despi´c (2009, 2011, in press) provides conclusive evidence against (68) based on the following SC/English contrasts.24 (69)

a. b.

Hisi latest movie really disappointed Kusturicai . Kusturicai ’s latest movie really disappointed himi .

(70)

ga i je zaista razoˇcarao. a. *Kusturicin i najnoviji film Kusturica’s latest movie him is really disappointed ‘Kusturicai ’s latest movie really disappointed himi .’ je zaista razoˇcarao b. *Njegov i najnoviji film his latest movie is really disappointed Kusturicu i . Kusturica ‘Hisi latest movie really disappointed Kusturicai .’

Despi´c notes that (69) can be accounted for if, as in Kayne (1994), English possessives are located in the Spec of PossP, which is immediately dominated by DP, the DP preventing the possessive from c-commanding anything outside of the subject. The contrast between English and SC then follows if the DP is missing in SC. (Following Boˇskovi´c (2005) Despi´c in fact treats SC possessives as NP adjuncts, on a par with adjectives; see Section 4.1). Significantly, Chinese and Japanese behave just like SC in the relevant respect (see Boˇskovi´c and Hsieh (2912), Cheng (in preparation) and Takahashi (2011)

24. The examples in the rest of this subsection assume a neutral (i.e. non-focused) intepretation of the relevant nouns/pronouns. (Since contrastive focus affects binding relations it is important to control for it. I have also avoided using relational nouns like father since at least for some speakers they involve irrelevant interfering factors; see Takahashi 2011).

On NPs and Clauses

201

for further discussion of Chinese and Japanese respectively), which provides strong evidence for the no-DP analysis for these languages. (71)

a. *Tai -de zuixinde dianying rang Li-An i hen shiwan. he-gen newest movie make Li-An very disappointed ‘His latest movie really disappointed Li-An.’ b. *Li-An i zuixinde dianying rang ta i hen shiwang. Li-An newest movie make he very disappointed ‘Li-An’s latest movie really disappointed him.’ c. *Kurosawai -no saisin-no eega-wa hontoo-ni karei -o Kurosawa-gen latest-gen movie-top really him-acc rakutans-ase-ta. disappoint-cause-past ‘Kurosawa’s latest movie really disappointed him’ hontoo-ni Kurosawai -o d. *Karei -no saisin-no eega-wa he-gen latest-gen movie-top really Kurosawa-acc rakutans-ase-ta. disappoint-cause-past ‘His latest movie really disappointed Kurosawa.’

Despi´c also shows that demonstratives and adjectives do not change anything in SC, which provides strong evidence that demonstratives, possessives, and adjectives should be treated as multiple adjuncts/specs of the same phrase. Since demonstratives and adjectives do not introduce an extra projection, they do not prevent the possessive from c-commanding the co-indexed elements in (72). (72)

a. *[NP Ovaj [NP Kusturicin i [NP najnoviji [N film]]]] ga i je Kusturica’s this latest movie him is zaista razoˇcarao. really disappointed ‘This latest movie of Kusturicai really disappointed himi .’ [NP Kusturicini i [NP filmovi ]]] su ga i b. *[NP Brojni Kusturica’s numerous movies are him razoˇcarali. disappointed

It should, however, be noted that the application of Despi´c’s test shows that functional structure is not completely lacking in SC TNPs. Thus, while demonstratives and adjectives do not bring in additional projections, non-adjectival numerals which assign genitive of quantification do bring in an additional projection. Despi´c observes that these elements confine the c-command domain of

202

ˇ Zeljko Boˇskovi´c

possessives, allowing them to co-refer with other elements without causing a binding violation. The contrast between (72) and (73) provides strong evidence that additional phrasal structure is present above the possessive only in (73), involving a numeral. (73)

[QP Pet [NP Dejanovih i [NP prijatelja ]]] je doˇslo na njegovo i Dejan’sGEN five friendsGEN is come to his venˇcanje wedding ‘Five of Dejan’s friends came to his wedding.’

Notice that Chinese and Japanese classifiers pattern with SC numerals; they apparently also introduce an additional projection into the structure. (74)

a.

b.

canjia tai -de You wu-ge Zhangsani -de pengyou lai come attend he-gen have 5-cl Zhangsan-gen friend hunli. wedding ‘Five of Zhangsan’s friend came to his wedding.’ Go-nin-no Johni -no tomodachi-ga karei -no five-cl-gen John-gen friends-nom he-gen kekkonsiki-ni kita. wedding-dat came ‘Five of John’s friends came to his wedding.’

It is worth noting here that Saito, Liu, Murasugi (2008) argue that only Chinese has a ClP, classifiers in Japanese being NP-adjuncts. The above data provide evidence against this conclusion. Just like Chinese classifiers, Japanese classifiers confine the c-command domain of possessives, which indicates that they also project a ClP above NP. 4.

Some Deductions of the NP/DP Generalizations

I now turn to explanations for the generalizations from Sections 1 and 2 under the DP/NP analysis. I will first briefly summarize the account of a couple of representative generalizations given in Section 1 from Boˇskovi´c (2005, 2008a) and Boˇskovi´c and Gajewski (in press), referring the reader to Boˇskovi´c (2008a) for the deduction of other generalizations from Section 1. I will focus here on the phenomena that are relevant to the clause-level syntax, namely leftbranch extraction (I will also suggest a modification of my original analysis

On NPs and Clauses

203

of the phenomenon), which involves extraction of TNP-internal elements to the clause level (two other locality phenomena, adjunct extraction out of TNPs and islandhood with HIRs, will also be addressed), and two surprising cases of an apparent interaction between TNP-internal syntax/semantics and clauselevel phenomena, namely the generalizations concerning the interpretation of superlatives and negative raising. The generalizations in question confirm the importance of TNP-internal syntax and semantics for clause-level phenomena. I will then propose a deduction of two generalizations from Section 2, namely (45) and (48) (suggestions for deductions of two other generalizations, (54) and (55), have already been given in Section 2), and explore its consequences for the clausal structure of NP languages. 4.1. Back to Left Branch Extraction: The Phase Analysis In Boˇskovi´c (2005) I gave two deductions of (6). Here, I will summarize only one of them, the one based on the Phase-Impenetrability Condition (PIC), which says only the Spec of a phase is accessible for phrasal movement outside of the phase (so, XP movement from phase YP must proceed via SpecYP). On a par with Chomsky’s (2000) claim that CP but not IP is a phase, I assumed in Boˇskovi´c (2005) that DP is a phase, but NP isn’t. Given the PIC, XP can then move from DP only if it moves to SpecDP. There are two more ingredients of the analysis: the traditional claim that AP is NP-adjoined and the anti-locality hypothesis (the ban on movement that is too short), which is deducible from independent mechanisms and argued for by many authors (e.g. Boˇskovi´c 1994; 1997; Saito and Murasugi 1999; Ishii 1999; Abels 2003; Grohmann 2003; Ticio 2003; Boeckx 2005; Jeong 2006).25 Like most other approaches, the version of anti-locality adopted in Boˇskovi´c (2005) requires Move to cross at least one full phrasal boundary (not just a segment). AP then cannot move to SpecDP in (75) due to anti-locality. Given the PIC, it cannot move directly out of DP either, as in (76). Anti-locality/PIC thus prevent AP extraction from DP, banning AP LBE in English (they don’t ban all movement from DP, who do you see [DP t[NP friends of t]] is still allowed) (75)

*[DP APi [D D [NP ti [NP

(76)

*APi [DP [D D [NP ti [NP

25. Among other things, anti-locality accounts for the ban on short subject topicalization and zero subject null operator relatives (Boˇskovi´c 1994, 1997), the that-trace effect (Ishii 1999), the ban on movement of the phase complement (Abels 2003), and the patterns of extraction of arguments out of DPs (Grohmann 2003, Ticio 2003).

204

ˇ Zeljko Boˇskovi´c

The impossibility of adjunct extraction out of TNP in English can be accounted for in the same way as the impossibility of AP LBE, given that NP adjuncts are also adjoined to NP. Moreover, the PIC/anti-locality problem does not arise in SC, which lacks DP. I would, however, like to suggest here a modification of one aspect of my earlier analysis of LBE. As discussed in Boˇskovi´c (2005), SC disallows deep LBE, i.e. LBE out of a complement of a noun (the same holds for Polish, Czech, and Russian). (77)

On cijeni [NP [N [ prijatelje [NP pametnih friends smart he appreciates [NP studenata]]]]]. students ‘He appreciates friends of smart students.’ b. ?*Pametnihi on cijeni [NP [N [ prijatelje [NP ti [NP studenata]]]]]. a.

What this shows is that an NP above an LBE-ing NP has the same effect on LBE as a DP above an LBE-ing NP does in English; they both block LBE. This can be accounted for if NP is a phase even in NP languages. (77b) can then be accounted for in exactly the same way as (3) (*Expensive, he saw cars), with the higher NP blocking LBE for the same reason DP does it in the English example. As noted in Boˇskovi´c (2010), strong evidence that this suggestion is on the right track concerns Abels’s (2003) generalization that the complement of a phase head is immobile. Thus, Abels observes that an IP that is dominated by a CP, a phase, cannot undergo movement. This in fact follows from an interaction of the PIC and anti-locality, with the PIC requiring IP movement through SpecCP, and anti-locality blocking such movement because it is too short. Now, if NP is indeed a phase in NP languages we would expect that an NP complement of a noun cannot undergo movement. Zlati´c (1997) observes genitive complements of nouns indeed cannot be extracted in SC. (78) ?*Ovog studenta sam pronaˇsla [NP knjigu ti ]. book this student-gen am found ‘Of this student I found the/a book.’ The impossibility of deep LBE and the immobility of genitive complements of nouns thus fall into place if NP is a phase in article-less languages. They are both ruled out in exactly the same way.26 26. It should be noted here that nominal complements that bear inherent case can be extracted. However, they also allow deep LBE. The correlation between the two

On NPs and Clauses

205

Notice furthermore that it is not necessary to posit crosslinguistic variation regarding phasehood. If we simply assume that the highest phrase in a TNP counts as a phase there is no real variation in the phasehood of the TNP between NP and DP languages; the real source of variation lies in the amount of structure a TNP has in DP and NP languages (see Boˇskovi´c 2012). That this analysis is on the right track is confirmed by the genitive of quantification construction, where, as discussed in Section 3, even SC projects functional structure above NP (see (73)). Significantly, nominal complement movement is possible in this context. (79)

sam pronaˇsla [QP ti mnogo /deset Ovog studenta many/ten this studentGEN am found [NP knjiga ti ]]. books

This is expected if it is the highest phrase in a TNP that functions as a phase. QP rather than NP then functions as a phase in (79). In contrast to (78), where the phasal edge is SpecNP, genitive NP in (79) can move to the phasal edge, SpecQP, without violating anti-locality. 4.2. Back to Head Internal Relatives Turning now to (37), Watanabe (2004) argues that languages differ regarding the licensing mechanism employed in HIRs. He argues that some languages employ unselective binding, which is not subject to locality, while others employ movement/feature checking, which is subject to locality, i.e. intervention effects. Given (37), the former should be employed in DP languages, and the latter in NP languages (I depart here from Watanabe). Significantly, Bonneau (1992) argues for independent reasons that the D that comes with a HIR is the unselective binder of its head (he makes the proposal for Lakhota). Since the D is missing in article-less languages, the island-insensitive binding option is unavailable in these languages. phenomena thus still holds. I discuss inherent case complements, as well as LBE out of other phrases, in Boˇskovi´c (2010). ˇ ga je [pretnja ti ] uplaˇsila? Cime i what-instr him is threat scared ‘The threat of what scared him?’ (ii) ?Kakvom i ga je pretnja [ti smr´cu] uplaˇsila? what-kind-of him is threat death-instr scared ‘Of what kind of death did a threat scare him?’ (i)

206

ˇ Zeljko Boˇskovi´c

4.3. Back to Negative Raising Boˇskovi´c and Gajewski (in press) explain (23) by highlighting a similarity in the interpretation of definite plurals and NR predicates. A common analysis of negative raising attributes to certain predicates (negative raising predicates) an excluded middle presupposition (EMP), where A believes that p presupposes A believes that p or A believes that not p. As a presupposition, the EMP survives negation. Then, in A does not believe that p the assertion and the EMP presupposition together entail A believes that not p. Now, Gajewski (2005, 2007) argues that the EMP is the hallmark of constructions that can be semantically analyzed as distributive plural definite descriptions, rather than universal quantifiers. To illustrate the EMP of definite plural NPs, Bill saw the boys implies Bill saw all the boys; Bill didn’t see the boys implies he saw no boys – not merely not all, with a universal scoping over negation, which Gajewski attributes to the EMP and which is analogous to the lower clause negation reading with negative raising (compare Bill didn’t see the boys with the negation of a universal quantifier: Bill didn’t see all the boys). Returning to negative raising, sentence-embedding predicates are standardly treated as universal quantifiers over accessible worlds. Gajewski (2005) argues that having the EMP, negative raising predicates should be treated as plural definite descriptions, which serve as arguments of the predicates contributed by their propositional complements. In Boˇskovi´c and Gajewski (in press) we assume that sentence-embedding predicates combine a modal base (set of accessible worlds) with a quantificational element. The quantificational element may be either a universal quantifier or a definite article. If a modal base combines with the definite article, the result is a negative raising predicate. Given this, if a language lacks the definite article, it lacks the necessary material to assemble a negative raising predicate. It follows that negative raising is possible only in DP languages.27

27. Recall that even languages disallowing strict NPI licensing under negative raising allow negative raising negation interpretation. Boˇskovi´c and Gajewski (in press) suggest that this is a pragmatic effect capturable in an approach like Horn (1989), who argues that the lower clause understanding is a case of ‘inference to the best interpretation’. Significantly, Gajewski (2005) shows this approach cannot explain strict NPI licensing under negative raising (more specifically, it cannot create the anti-additive environments needed for the licensing), which his semantic account can do.

On NPs and Clauses

207

4.4. Back to Superlatives Boˇskovi´c and Gajewski (in press) also give a deduction of (36) based on Hackl’s (2007) proposal that most should be analyzed as the superlative of many (most = many-est). Szabolcsi (1986) and Heim (1985, 1999) argue that -est can move independently to take scope. Hackl shows that if we allow movement of -est in most we can derive both the majority (MR) and the plurality (PR) reading. PR corresponds to the comparative superlative reading discussed by Szabolcsi and Heim and analyzed as -est taking clausal scope. Hackl shows that MR can be derived if the -est of most stays inside the DP, taking scope below the article. The ingredients of Hackl’s analysis are given below: A. many has a modificational meaning of type ,, unlike other gradable adjectives, like tall, whose denotation is type