The Impact of Pronominal Form on Interpretation 9781614517016, 9781614517801

Table of contents :
Contents
The impact of pronominal form on interpretation
Part I: Syntax and pronominal form
Fake form. and what it tells us about the relation between form and interpretation
Referential dependence across phase boundaries in Russian
Anaphora vs. agreement. A new kind of Anaphor Agreement Effect in Tamil
The only real pro-nouns. Comparing English one and Italian ne as Noun Phrase pro-forms
When personal pronouns compete with relative pronouns
Part II: Semantics and pronominal interpretation
The form of bound pronouns: Towards a uniform account
Demonstrative pronouns and perspective
A ‘point’ of inquiry: The case of the (non-)pronominal IX in ASL
Definite or still demonstrative?. Some ideas on the semantics of the German distal demonstrative jen-
Deferred pronouns: A sketch based on Relevance Theory and Dynamic Syntax
Part III: Experimental insights on the syntax and semantics of pronouns
Restrictions on the interpretation of null and overt subjects in Italian and Spanish and the Morphological Ambiguity hypothesis
Inter-speaker variation in Korean pronouns
L2 speakers’ processing of reflexives and personal pronouns. A self-paced reading study of German learners of Dutch
Index

Citation preview

Patrick Grosz, Pritty Patel-Grosz (Eds.) The Impact of Pronominal Form on Interpretation

Studies in Generative Grammar

| Editors Norbert Corver Harry van der Hulst Roumyana Pancheva Founding editors Jan Koster Henk van Riemsdijk

Volume 125

The Impact of Pronominal Form on Interpretation | Edited by Patrick Grosz and Pritty Patel-Grosz

ISBN 978-1-61451-780-1 e-ISBN (PDF) 978-1-61451-701-6 e-ISBN (EPUB) 978-1-5015-0077-0 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 on the Internet at http://dnb.dnb.de. © 2016 Walter de Gruyter, Inc., Boston/Berlin Typesetting: PTP-Berlin, Protago-TEX-Production GmbH, Berlin Printing and binding: CPI books GmbH, Leck ♾ Printed on acid-free paper Printed in Germany www.degruyter.com

Contents Patrick G. Grosz and Pritty Patel-Grosz The impact of pronominal form on interpretation Introduction | 1

Part I: Syntax and pronominal form Martina Wiltschko Fake form and what it tells us about the relation between form and interpretation | 13 Inna Livitz Referential dependence across phase boundaries in Russian | 53 Sandhya Sundaresan Anaphora vs. agreement A new kind of Anaphor Agreement Effect in Tamil | 77 Michelangelo Falco and Roberto Zamparelli The only real pro-nouns Comparing English one and Italian ne as Noun Phrase pro-forms | 107 Ewa Trutkowski and Helmut Weiß When personal pronouns compete with relative pronouns | 135

Part II: Semantics and pronominal interpretation Uli Sauerland The form of bound pronouns: Towards a uniform account | 169 Stefan Hinterwimmer and Peter Bosch Demonstrative pronouns and perspective | 189 Elena Koulidobrova and Diane Lillo-Martin A ‘point’ of inquiry: The case of the (non-)pronominal IX in ASL | 221

VI | Contents

Maria Averintseva-Klisch Definite or still demonstrative? Some ideas on the semantics of the German distal demonstrative jen- | 251 Thiago N Galery Deferred pronouns: A sketch based on Relevance Theory and Dynamic Syntax | 287

Part III: Experimental insights on the syntax and semantics of pronouns Francesca Filiaci Restrictions on the interpretation of null and overt subjects in Italian and Spanish and the Morphological Ambiguity hypothesis | 319 Kyeong-min Kim and Chung-hye Han Inter-speaker variation in Korean pronouns | 347 Hendrikje Ziemann and Esther Ruigendijk L2 speakers’ processing of reflexives and personal pronouns A self-paced reading study of German learners of Dutch | 373 Index | 393

Patrick G. Grosz and Pritty Patel-Grosz (University of Tübingen)

The impact of pronominal form on interpretation Introduction

1 General background Pronouns (e.g. he/she/it) are natural language expressions with meanings that have proven difficult to capture in formal linguistic theory. Pronouns vary along two dimensions: (i) their interpretation and (ii) their form. To illustrate the (‘semantic/pragmatic’) interpretation dimension, consider the difference between (1) and (2). In (1a), the pronoun he appears to refer to John, cf. (1b). By contrast, he in (2a) does not refer to any particular person; the intended reading for (2a) appears to be a reading that can be rendered in a (quasi-)logical meta-language as in (2b), where no man has a quantificational meaning and he denotes a bound variable. The general approach to the distinction between (1b) and (2b) is to assume that he in (1) is a so-called referential pronoun, whereas he in (2) is a so-called bound pronoun. This is only one of many different distinctions amongst pronominal meanings that have been introduced in the literature (cf. Büring 2011 for a recent overview). (1)

a. I yesterday met John. He can climb a vertical wall. (referential use of ‘he’) b. paraphrase: I yesterday met John. John can climb a vertical wall.

(2) a. No man believes that he can climb a vertical wall. (bound use of ‘he’) b. paraphrase: ¬∃x[x is a man & x believes that x can climb a vertical wall] To illustrate the (‘syntactic/morphosyntactic’) form dimension, consider the differences in (3). German allows for personal pronouns (ihn ‘him’) as well as demonstrative pronouns (den ‘him, that one’), as in (3a). Similarly, Spanish differentiates between unpronounced pronominal subjects (so-called null pro) and pronounced, ‘overt’ pronominal subjects (él ‘he’, ella ‘she’), as in (3b). Other distinctions on the form side include the distinction between reflexive pronouns (himself ) and non-reflexive pronouns (him), the distinction between indexical (1st /2nd person) personal pronouns and non-indexical (3rd person) personal pro-

2 | Patrick G. Grosz and Pritty Patel-Grosz

nouns, and distinctions along the lines of gender (masculine, feminine and neuter). (3) a. Maria hat { ihn / den } gesehen. Maria has him dem seen ‘Maria has seen him.’ (Wiltschko 1998:144) b. Nadie cree que { 0 / él } es inteligente. nobody believes that he he is intelligent ‘Nobody believes that he is intelligent.’ (Montalbetti 1984:26) The goal of this volume is to shed new light on the interplay of these two dimensions, i.e. how do choices along the form dimension (e.g. overt vs null) restrict the possible choices along the interpretation dimension (and vice versa)? This volume approaches phenomena of this type by drawing together different contributions, all of which concern the interaction of pronominal form and pronominal interpretation, thus forming a coherent whole. The papers in this volume aim at an interdisciplinary interface perspective, and each paper attempts to connect the syntactic ‘form’ of pronouns to their semantic ‘interpretation’. The structure of this volume groups such studies in three parts: Part I contains studies that approach the topic from a predominantly syntactic perspective, while the studies in Part II take semantics as their point of departure; finally, Part III envelopes contributions whose main focus is on experimental evidence, given that the behavior and distribution of pronouns is often inaccessible to ‘raw’ introspective intuitions.

2 Syntax and pronominal form The internal (morpho-)syntax of pronouns and their external syntactic behavior are still widely under debate. Concerning their internal syntax, influential studies such as Cardinaletti & Starke (1999) argue that all 3rd person pronouns contain a null NP, a view that has been argued for independently in the semantics literature by authors such as Elbourne (2005, 2013) and Sauerland (2007); by contrast, authors such as Wiltschko (1998) and Dechaine & Wiltschko (2002) pursue a more differentiated view. In Part I of the volume, the chapters by Wiltschko, Falco & Zamparelli, and Trutkowski & Weiß contribute new insights to our understanding of the internal build-up of pronouns.

The impact of pronominal form on interpretation

|

3

Concerning the external syntax of pronouns, we observe that there is a long tradition that aims at understanding the connection between bound pronouns and their antecedents (cf. Reuland 2011, Rooryck & vanden Wyngaerd 2011 for recent monographs on this topic). Focusing on current open questions that arise in this respect from a cross-linguistic perspective, Livitz and Sundaresan aim to account for the properties of pronouns in embedded clauses, in Russian and Tamil, respectively. Wiltschko’s paper Fake form and what it tells us about the relation between form and interpretation builds on Wiltschko (2014), and aims to shed new light on the internal composition of pronouns and how it interfaces with their interpretation. The goal is to explain how language-specific buildling blocks are recruited in pronominal paradigms, based on universal linguistic categories, and how this helps us understand phenomena such as fake indexicals (e.g. the occurrence of my in only I remember my name, with the ‘bound variable’ entailment nobody else remembers his name). Wiltschko compares pronominal paradigms from a number of different languages (including French, German, Gothic, Icelandic, Norwegian, and Russian), and observes that the morphological form of the 1st person singular nominative pronoun (e.g. German ich, Russian ja) is contained in the 3rd person reflexive pronoun (German sich, Russian sebja). This is taken to be a reflex of a reflexive [+ident] component present in both pronominal forms. Moving on to a case study of how the internal syntax of pronouns interfaces with their external syntactic behavior and with their interpretation, Livitz’s Referential dependence across phase boundaries in Russian explores the difference between overt and null subjects of Russian embedded clauses (e.g. čto vesnoj {on / 0} poedet v Pariž ‘that he will go to Paris in the spring’); to account for this difference, she assumes that null pronouns differ from overt pronouns in their internal makeup (the former being ϕPs, the latter being DPs), which in turn has consequences for how they interact with the linguistic configuration that they occur in. While null pronouns are, in a sense, deficient, and must enter a ϕagreement chain involving the matrix v and the embedded C, overt pronouns enter the derivation as fully valued, non-deficient elements, precluding them from participating in such an agreement chain. Livitz connects her analysis to phase-based approaches (Chomsky 2000, 2001, and subsequent work), arguing that the null vs. overt distinction correlates with flexible phasehood in Russian; embedded CPs are phases when their subjects are overt, but their phase status is suspended when their subjects are null. In a related spirit, Sundaresan’s Anaphora vs. agreement: a new kind of Anaphor Agreement Effect in Tamil explores the ϕ-agreement properties of nonlocally bound occurrences of the Tamil (Dravidian) reflexive taan. Based on an investigation of the contexts that allow such long-distance reflexives, Sundaresan

4 | Patrick G. Grosz and Pritty Patel-Grosz

argues that embedded contexts can contain a null pronoun in the specifier of a perspectival phrase (PerspP) in the clausal spine. It is this null pronoun that binds the long-distance reflexive and serves as the trigger of overt ϕ-agreement. These insights shed new light on how languages can avoid violations of the so-called Anaphor Agreement Effect generalization (Rizzi 1990), which states that reflexives are unaccessible for ϕ-agreement operations. Turning to different categories of possible pronominal elements, Falco & Zamparelli (The only real pro-nouns: comparing English ‘one’ and Italian ‘ne’ as Noun Phrase pro-forms) focus on the behavior of elements that they argue are core instantiations of pro-NPs (as opposed to pro-DPs), namely English one (in sentences such as their example you own four small (red) bikes and I own two big ones) and its Italian counterpart ne. Investigating the constraints on the syntactic distribution and semantic interpretation of these two elements, they point out differences as well as communalities, arguing that the differences involve a distinction between clitic ne (in Italian) and non-clitic one (in English), and a more general difference between the licensing condition of bare nominals in English vs. Italian. Trutkowski & Weiss (When personal pronouns compete with relative pronouns) revisit observations from Ito & Mester (2000) concerning the distribution of pronominal forms and verbal agreement in German relative clauses that modify a 1st or 2nd person pronoun (e.g. Ich, der ich sechzig bin, … ‘I who am sixty years old’). They provide experimental evidence showing that there are different agreement patterns that arise in such configurations, and pursue a detailed analysis of how to account for the observed patterns. For the apparently competing pronouns in relative clauses (e.g. the demonstrative pronoun der ‘who’ and the personal pronoun ich ‘I’), they assume a view where both are generated in a single DP, and their co-occurrence involves multiple spell-out of a moved element.

3 Semantics and pronominal interpretation Since most detailed investigations in the semantics of pronouns (cf. Heim & Kratzer 1998, Büring 2011 for textbook overviews) have taken English as a point of departure, investigations on how pronominal form (e.g. personal vs. demonstrative pronouns) effects their interpretation have a comparatively recent tradition. For instance, early accounts of the distinction between personal and demonstrative pronouns in German (Olsen 1991a, 1991b; Wiltschko 1998) adopt a predominantly syntactic view, without a formal semantic implementation. Similarly, the topic of how ϕ-features are actually interpreted by the semantic component

The impact of pronominal form on interpretation

|

5

(Cooper 1979), has largely been backgrounded, and only recently experienced a revival (e.g. Sauerland 2003, Sudo 2012). Sauerland’s The form of bound pronouns: towards a uniform account aims at outlining a general framework of how the presuppositions associated with ϕ-features (gender, number, person) interact with contexts in which pronouns occur in the scope of semantic operators (such as the focus particle only, attitude verbs like dream, or distributive elements such as all in the boys all were riding their bike). Sauerland works out a general account of ‘weakened projection’ in such contexts, where not all presuppositions project across the embedding operators; he shows that his proposal also carries over to a view that treats reflexivity as a presupposition, thus shedding new light on the nature of reflexive pronouns and Binding Condition A (which Sauerland implements in his novel rendering of Pollard & Sag 1992, and Reinhart & Reuland 1993). With the goal of semantically modeling the distinction between personal and demonstrative pronouns in German, Hinterwimmer & Bosch (Demonstrative pronouns and perspective) explore the co-reference and binding restrictions on demonstrative pronouns (DPros). After reviewing evidence for a topic-avoidance property of DPros, Hinterimmer & Bosch argue that the core constraint involves an inability to select the current ‘perspectival center’ as the DPro’s antecedent; in this sense, DPros can be seen as anti-logophoric elements (e.g. contrasting with logophoric uses of taan in Tamil, which depend on a locally represented perspectival center). Topic-avoidance is then reanalyzed as a special case of avoiding the perspectival center, while both notions can be subsumed under a constraint that prohibits DPros from selecting the most prominent salient antecedent. Taking a different perspective on the distinction between personal pronouns and demonstrative pronouns, Koulidobrova & Lillo-Martin’s A ‘point’ of inquiry: the case of the (non-)pronominal IX in ASL argues that the element IX in American Sign Language, which is typically treated as a personal pronoun, is more aptly analyzed as a demonstrative pronoun. They provide a number of tests, showing that IX exhibits the hallmark signs of demonstratives and should thus be analyzed as a demonstrative expression. This allows us to capture all uses of IX (definite determiner, locative adverbial, pronominal) in a uniform manner, while its surprising properties fall out naturally. While Hinterwimmer & Bosch investigate the German ‘weak’ demonstrative pronoun der/die/das, which is surface-identical to the definite determiner, Averintseva-Klisch (Definite or still demonstrative? Some ideas on the semantics of German distal demonstrative ‘jen-’) takes a closer look at the ‘strong’ distal demonstrative paradigm, involving the root jen-. This element has often been claimed to be archaic, but still occurs quite frequently in written German. Based on a corpus investigation, Averintseva-Klisch proposes a view in which jen- in

6 | Patrick G. Grosz and Pritty Patel-Grosz

Present Day German encodes definiteness at the at-issue level (cf. Potts 2005 amongst others), while conveying ‘mental distance’ at the non-at-issue level. Galery’s Deferred pronouns: a sketch based on Relevance Theory and Dynamic Syntax investigates the notorious ‘deferred’ readings of 3rd person pronouns (e.g. when pointing at St. Peter’s Basilica and saying: on average, he (= the Pope) is 64 years old when elected). Galery pursues a proposal where semantic derivations can directly access pragmatic inferences, which he implements in terms of Relevance Theory (as the pragmatic component) and Dynamic Syntax (providing the syntactic/semantic component). In doing so, he aims at shedding new light on the interface between grammar and pragmatics.

4 Experimental insights on the syntax and semantics of pronouns Overall, this volume aims at connecting formal theoretical considerations (in syntax and semantics, as well as morphology and pragmatics) to the empirical domain of pronominal form and interpretation. Given that distributional properties of pronouns (e.g. whether overt pronouns are less likely than null pronouns to occur in bound readings) do not lend themselves for direct access via a syntactician or semanticist’s introspective intuitions, experimental studies are imperative in this area of research. The three studies in this section aim at investigating core questions experimentally, thus shedding new light on the impact of pronominal form on interpretation. Filiaci’s Restrictions on the interpretation of null and overt subjects in Italian and Spanish and the Morphological Ambiguity hypothesis addresses a topic that goes back to investigations such as Montalbetti (1984) and Carminati (2002): the question of how null subjects differ from overt subjects in Romance languages such as Italian and Spanish. It seems intuitive that different languages pattern along the same lines when it comes to the distribution of null vs. overt pronominal forms, e.g. in disallowing overt pronouns that refer to the most prominent antecedent when they compete with a null pronoun. Nevertheless, Filiaci shows that there are surprising differences between Italian and Spanish in how overt pronouns are constrained. Her study thus opens new lines of inquiry concerning the nature of overt pronominal expressions in null-subject languages. On a related note, Kim & Han (Inter-speaker variation in Korean pronouns) demonstrate that even within the same language, there appears to be inter-speaker variation concerning the distributional difference between overt and null pronouns. While overt pronouns can consistently receive a bound variable interpre-

The impact of pronominal form on interpretation |

7

tation for one group of Korean speakers, such a reading is rejected by a distinct group of Korean speakers. The authors attribute this to impoverishment in first language acquisition, which gives rise to the first group construing overt pronouns as demonstrative pronouns, while the second group construes overt pronouns as personal pronouns. The former group of speakers disallows bound variable interpretations, while the latter readily allows them. Kim & Han’s proposal thus further corroborates the insight that a more nuanced perspective on null vs. overt distinctions within a language and across languages is necessary. Ziemann & Ruigendijk’s L2 speaker’s processing of reflexives and personal pronouns: a self-paced reading study of German learners of Dutch uses experimental studies of L2 acquisition to test current hypotheses concerning the difference between reflexive and non-reflexive pronouns. Their experiments are set up to test for the difference between syntactic operations (e.g. syntactic binding of reflexives) vs. discourse operations (e.g. contextual resolution of pronouns), and how these are reflected in a comparison of L1 speakers of Dutch and L2 learners of Dutch. The experimental results indicate that L2 learners rely on discourse operations, and also corroborate the idea that pronoun resolution via syntactic operations is less costly than discourse-based pronoun resolution (Reuland 2001). Acknowledgment: This edited volume is based on the workshop “The Impact of Pronominal Form on Interpretation”, held at the University of Tübingen in November 2013 and co-organized by Katrin Axel-Tober, Patrick G. Grosz and Pritty PatelGrosz. The editors would like to thank the Daimler and Benz Foundation (grant 32-01/12) for financially supporting the workshop. Pritty Patel-Grosz is funded by the Collaborative Research Center SFB 833 (projects B2 and C4) of the German Science Foundation (DFG) at the University of Tübingen. We would also like to thank Beate Starke from the Collaborative Research Center SFB 833 for organizational support, as well as our organizing team, Nadine Bade, Polina Berezovskaya, Verena Hehl, Anna Howell, Kalle Müller, Carolin Munderich, Anna Pryslopska, Anthea Schöller, Sonja Tiemann, and Sarah Zobel. Finally, we thank an anonymous reviewer for the Studies in Generative Grammar series, as well as Lena Ebert and Lara Wysong from de Gruyter and series co-editor Norbert Corver for editorial support.

8 | Patrick G. Grosz and Pritty Patel-Grosz

References Büring, Daniel. 2011. Pronouns. In Klaus von Heusinger, Claudia Maienborn and Paul Portner (eds.): Semantics: An International Handbook of Natural Language Meaning. Vol. 2. Berlin: Mouton de Gruyter, 971–995. Cardinaletti, Anna, and Michal Starke. 1999. The typology of structural deficiency: A case study of the three classes of pronouns. In Henk C. van Riemsdijk (ed.): Clitics in the Languages of Europe. Berlin: Walter de Gruyter, 145–234. Carminati, Maria Nella. 2002. The Processing of Italian Subject Pronouns. PhD thesis, University of Massachusetts Amherst, Amherst: MA. Chomsky, Noam. 2000. Minimalist inquiries: The framework. In Roger Martin, David Michaels, and Juan Uriagereka (eds.): Step by step: Essays in honor of Howard Lasnik. Cambridge, MA: MIT Press, 89–155. Chomsky, Noam. 2001. Derivation by phase. In Michael Kenstowicz (ed.): Ken Hale: A life in language. Cambridge, MA: MIT Press, 1–52. Cooper, Robin. 1979. The interpretation of pronouns. In Frank Heny and Helmut Schnelle (eds.): Selections from the third Groningen round table, syntax and semantics, volume 10. New York: Academic Press, 61–92. Déchaine, Rose-Marie, and Martina Wiltschko. 2002. Decomposing Pronouns, Linguistic Inquiry 33, 409–442. Elbourne, Paul. 2005. Situations and Individuals. Cambridge, MA: MIT Press. Elbourne, Paul. 2013. Definite Descriptions. Oxford: Oxford University Press. Heim, Irene, and Angelika Kratzer. 1998. Semantics in Generative Grammar. Oxford: Blackwell. Ito, Junko, and Armin Mester. 2000. Ich, der ich sechzig bin: An Agreement Puzzle. In Sandy Chung, Jim McCloskey, and Nathan Sanders (eds.): Jorge Hankamer WebFest. URL: http://babel.ucsc.edu/Jorge/ito_mester.html Montalbetti, Mario. 1984. After binding: On the interpretation of pronouns. Doctoral Dissertation, MIT, Cambridge, Massachusettes. Olsen, Susan. 1991a. AGR(eement) und Flexion in der deutschen Nominalphrase. In Gisbert Fanselow and Sascha W. Felix (eds.): Strukturen und Merkmale syntaktischer Kategorien. Tübingen: Narr, 51–69. Olsen, Susan. 1991b. Die deutsche Nominalphrase als Determinerphrase. In Susan Olsen and Gisbert Fanselow (eds.): DET, COMP, und INFL: Zur Syntax funktionaler Kategorien und grammatischer Funktionen. Tübingen: Niemeyer, 35–56. Pollard, Carl, and Ivan A. Sag. 1992. Anaphors in English and the scope of binding theory. Linguistic Inquiry 23, 261–305. Potts, Christopher. 2005. The logics of conventional implicatures. Oxford: Oxford University Press. Reinhart, Tanya, and Eric Reuland. 1993. Reflexivity. Linguistic Inquiry 24, 657–720. Reuland, Eric 2001. Primitives of Binding. Linguistic Inquiry 32, 439–492. Reuland, Eric. 2011. Anaphora and language design. Cambridge, Mass.: MIT Press. Rizzi, Luigi. 1990. On the anaphor-agreement effect. Rivista di Linguistica 2, 27–42. Rooryck, Johan, and Guido vanden Wyngaerd. 2011. Dissolving binding theory. Oxford Studies in Theoretical Linguistics. Oxford: Oxford University Press. Sauerland, Uli. 2003. A new semantics for number. In Robert B. Young and Yuping Zhou (eds.): The Proceedings of SALT 13. Cornell University, Ithaca, N.Y.: CLC-Publications, 258–275.

The impact of pronominal form on interpretation |

9

Sauerland, Uli. 2007. Flat Binding: Binding without Sequences. In Uli Sauerland and HansMartin Gärtner (eds.): Interfaces + Recursion = Language? Chomsky’s Minimalism and the View from Syntax-Semantics. Berlin: Mouton de Gruyter, 197–253. Sudo, Yasutada. 2012. On the semantics of phi features on pronouns. Doctoral Dissertation, Massachusetts Institute of Technology, Cambridge, Mass. Wiltschko, Martina. 2014. The universal structure of categories. Towards a formal typology. Cambridge: University of Cambridge Press. Wiltschko, Martina. 1998. On the Syntax and Semantics of (Relative) Pronouns and Determiners. Journal of Comparative Germanic Linguistics 2, 143–181.

| Part I: Syntax and pronominal form

Martina Wiltschko (The University of British Columbia)

Fake form and what it tells us about the relation between form and interpretation Abstract: This paper presents a new analysis of fake indexicals (indexical pronouns used as bound variables). It is shown that there is nothing “fake” about fake indexicals. Rather it is argued that both the indexical use and the bound variable use of pronouns are derived via the syntactic context. In this respect, the analysis developed has much in common with Iatridou’s 2000 analysis of fake past. In particular, it is argued that 1st person pronouns in German are underlyingly specified as identity predicates. It is only by virtue of associating with syntactic structure (i.e., an indexical shell) that the indexical interpretation is derived. Evidence for this analysis stems from the fact that the same form is found in 2nd person accusative pronouns as well as 3rd person reflexives. This would be unexpected if 1st person pronouns were intrinsically specified as indexical. In addition, 1st person pronouns can also be used as generic pronouns, again unexpected under current analysis of fake indexicals. The analysis developed takes the multifunctionality of a given form as significant. It suggests that the interpretation of a seemingly simplex form is mediated by syntax as suggested in Wiltschko 2014. Keywords: fake indexicals, fake past, impersonal pronouns, reflexives, multifunctionality

1 Introduction The core goal of this paper is to explore the formal underpinnings of fake form. I here use the term fake form as a descriptive term for forms that appear to have two distinct interpretations: one is typically considered the real one whereas the other one is considered fake. The main empirical domain within which I investigate this notion of fakeness is that of indexical pronouns. In particular, so-called fake indexicals have received considerable attention in the recent literature (Kratzer 2009). A crucial example, widely discussed, is presented in (1). (1)

I’m the only one around here who can take care of my children. (Kratzer 2009: 188 (1); modeled after Partee 1989: fn. 3)

14 | Martina Wiltschko

The first person pronoun my in (1) is ambiguous between an indexical (referential) reading and a bound variable reading. On the bound variable interpretation, (1) implies that nobody else around here can take care of his or her own children. It is in their use as bound variables that indexical pronouns are often referred to as fake indexicals. This suggests that the indexical interpretation is somehow considered basic and that the bound variable interpretation is not quite true to the underlying function of the pronoun. This is schematized in Figure 1, where BVA refers to Bound Variable Anaphor.

form I

basic interpretation 1st indexical

fake interpretation BVA

Figure 1: Fake indexicals.

An obvious solution to the problem of fake interpretation is simply to posit lexical ambiguity. That is, we could duplicate the lexical entries for all indexical pronouns that allow for the bound variable interpretation as schematized in Figure 2.

form I

form I

basic interpretation 1st indexical

fake interpretation BVA

Figure 2: A lexical ambiguity approach to fake indexicals.

Fake form

|

15

Since fake indexicals are a pervasive and systematic phenomenon it seems undesirable and rather costly to identify the source of the ambiguity in the lexical entry of the 1st person pronoun (Kratzer 2009). The purpose of this paper is to develop an alternative. In particular, I will consider the phenomenon of fake indexicals within a larger context. On the one hand fake forms are not restricted to pronouns, and on the other hand multifunctionality is not restricted to patterns that are labeled as fake. The central question within which such patterns of multi-functionality have to be explored is at the very core of linguistic theory. How should we model the relation between form and interpretation?

Form

???

Interpretation

Figure 3: The challenge: modeling the relation between form and interpretation.

Patterns of multi-functionality, such as the fake instance of the indexical my considered above, provide us with a unique and constructive window into this question. The solution to the problem of fake forms, which I develop in the domain of pronouns, can equally be applied to other patterns of multi-functionality. And since multi-functionality is a pervasive property of all natural languages, any attempt to model the form-meaning relation that predicts these patterns will have broader empirical coverage than those theories that have to be amended in some way to capture these patterns. The core insight we gain from fake patterns is that the relation between form and interpretation is more complicated than depicted in Figure 3. That is, the relation between a particular form and its interpretation is not always direct. Rather – at least sometimes – this relation is mediated by syntax (Wiltschko 2014) as schematized in Figure 4. It is precisely the apparent mismatches between form and interpretation considered here, i.e., fake forms, that speak to the necessity of such a model. Applied to the cases at hand, I shall argue that neither of the two interpretations associated with the pronoun is basic or fake. Rather, the 1st person pronoun (along with other indexicals) is not analyzed as being intrinsically indexical. Instead, both their indexical and their bound variable interpretation are derived by virtue of associating the form with syntactic structure such that a difference in syntactic structure results in a difference in interpretation as schematized in Figure 5.

16 | Martina Wiltschko

Syntax

Form

Interpretation

Figure 4: The relation between form and interpretation is mediated by syntax.

syntactic context 1

1st person pronoun

syntactic context 2

BVA

1st person pronoun

indexical

Figure 5: The relation between pronominal form and interpretation is mediated by syntax.

On this view, indexicality, just like anaphoricity, is a syntactic construct rather than a lexical primitive. The paper is organized as follows. I start in Section 2 with a closer look at the nature of fakeness establishing first that the phenomenon is not restricted to pronouns, and second that it is an instance of a larger pattern: linguistic forms are pervasively associated with patterns of multi-functionality. In Section 3, I introduce the analysis, which is largely based on Wiltschko 2014. In Section 4, I discuss three consequences of the claim that ich is not intrinsically indexical: (i) the bound variable interpretation is not the only non-indexical use of local pronouns; (ii) ich is a building block of 2nd and 3rd person pronouns (dich/sich) as well; and (iii) we will have a closer look at third person pronouns. In Section 5, I conclude.

Fake form

|

17

2 What is a fake form? Taking literally, the term fake implies that there is a basic form with a basic meaning. Fake forms have been discussed under this label in two empirical domains: fake indexicals (Kratzer 2009) and fake past (Iatridou 2000). In what follows, I discuss each of these phenomena in turn.

2.1 Fake indexicals The term fake indexical has been introduced in Kratzer 2009 and has become the standard way of referring to 1st and 2nd person pronouns used as bound variable anaphors.¹ In this subsection, I consider the relevant data that have lead to this term. 1st and 2nd person pronouns have typically been considered to be directly referential, indexing the speaker and addressee, respectively (Kaplan 1989). And moreover, they are typically considered indexical as a matter of their lexical specification. In this respect, they differ from 3rd person pronouns and definite descriptions, which may have an indexical use, but are not intrinsically specified as indexical (see Schlenker 2003 for discussion). This difference can be seen on the basis of the fact that the definite description the speaker in (2a) cannot be replaced by the 1st person pronoun as shown in (2b). The latter sentence makes sense only if indeed I was John at some point. No such requirement is associated with the sentence in (2a). (2) a. At some point the speaker was John. b. At some point I was John. modeled after Schlenker 2003: 30 The difference between (2a) and (2b) comes about despite the fact that the 1st person pronoun refers to the speaker – however, it does so indexically. There are, however, contexts in which the indexical interpretation of 1st and 2nd person pronouns disappears. Examples of this sort are given in (3).

1 Angelika Kratzer (p.c.) thinks that the term was around before she used it. However, I was unable to find any references to it prior to her 2006 version of Kratzer 2009.

18 | Martina Wiltschko

(3) a. I’m the only one around here who will admit that I could be wrong.² Partee 1989, fn. 3 b. Only I got a question that I understood. Kratzer 1998: 92 (4)³ c. We all think we’re smart. Rullmann 2004: 161 (2a) What all these instances of 1st person pronouns have in common is that they need not be interpreted indexically (referring to the speaker) but instead their interpretation depends on variable assignment – they behave as bound variable anaphors. Take for example the sentence in (3b). It has two readings: one indexical (4a), the other is that of a bound variable (4b). (4) a. λx [x got a question that I understood] b. λx [x got a question that x understood] It is only if we consider the forms I and we to be intrinsically indexical that the term fake indexical is appropriate. In that case, the examples in (3) would indicate that indexical forms may sometimes lose their indexical interpretation. They would only appear to be indexical (i.e., be fake indexicals) when in fact they are not. Thus, the term fake indexical suggests that the non-indexical use of the indexical is derived, such that one of the meaning components is lost under certain conditions.⁴ This type of approach is summarized in Figure 1, repeated from above. An analysis that adheres to this line of reasoning is that of von Stechow 2003. He argues that the features of a pronoun are not interpreted by the interpretive component just in case the pronoun is bound. In this case the feature checking operation results in the deletion of the relevant features yielding a fake pronoun. This is illustrated in Figure 6.

2 In (3a) the 1st person pronoun is embedded in an attitude context. Consequently it may be interpreted as a logophor and as such it can be considered to be interpreted indexically. That is, if indexicality is defined as applying to expressions whose semantic value is determined by some feature of a context of utterance then the 1st person pronoun in (3a) qualifies as indexical: it refers to the attitude holder of some context. As such it may be considered akin to pronouns found in logophoric systems such as Amharic (cf. Schlenker 1999, 2003). 3 Kratzer 1998 attributes this example to Irene Heim, class lectures. 4 See Déchaine & Wiltschko 2015 for a discussion on the conditions that allow for the bound variable interpretation of indexical pronouns.

Fake form

form I

|

19

basic interpretation 1st indexical

fake interpretation BVA

Figure 1: Fake indexicals.

Syntax feature checking: deletion Form

Interpretation

fully specified pronoun

BVA

Figure 6: von Stechow’s 2003 analysis of fake indexicals.

Here the fake indexical is derived from the fully specified form (which in the absence of binding is interpreted indexically) via feature checking. Interestingly, the analysis for fake indexicals developed in Kratzer 2009, the very paper that introduced the term fake indexical into the field, does not in fact reflect the intuition that fake forms are somehow derived from fully indexical forms. In particular, she analyzes the fake interpretation as derived from an abstract pronominal base, i.e., a minimal pronoun with a defective feature set. This defective feature set is completed in the course of the syntactic derivation. In Kratzer 2009’s particular implementation, feature valuation is achieved through binding by a λ-operator, which is in turn associated with verbal functional categories. Crucially, the process of feature transmission, which supplies a defective pronoun with its features, is invisible to the semantic component allowing for the bound variable interpretation. This is schematized in Figure 7.

20 | Martina Wiltschko

fully specified pronoun Syntax feature transmission Underlying Form

Interpretation

Minimal pronoun

BVA

Figure 7: Kratzer’s 2009 minimal pronoun approach towards BVA.

Spell out fully specified pronoun Syntax (feature transmission) Underlying Form

Interpretation

fully specified pronoun

indexical

Spell out fully specified pronoun Syntax

context shift Underlying Form

Interpretation

fully specified pronoun

long distance fake indexical

Figure 8: Kratzer’s 2009 fully specified pronouns (indexical/long distance fake indexicals).

Fake form

|

21

However, in Kratzer’s 2009 analysis, pronouns need not be born as minimal (unspecified) pronouns. Instead a pronoun can also be fully specified. If it is, the pronoun may either be indexical in which case its underlying form is directly interpreted as a fully referential (indexical) pronoun. However, it may also be bound by a context shifting λ-operator (Cable 2005) in which case the result is a long distance fake indexical. This is schematized in Figure 8. Thus, on this view there are two possible underlying forms for pronouns: one defective the other fully or partially specified. In addition, there are two ways in which fake indexicals can be derived. On the one hand, the defective minimal pronoun may be interpreted independent of its feature specification (which is supplied syntactically). On the other hand a fully specified pronoun may be interpreted via context shifting operations. Accordingly, there is no sense in which the indexical interpretation is basic and the fake form derived from it.

2.2 Fake past The term fake past was first introduced in Iatridou 2000 to refer to a past tense morpheme when it is used without denoting past time reference. In particular, in a number of unrelated languages, past tense morphology is used to express counterfactuality. This is shown on the basis of the English examples in (5) and (6) below. Within an indicative clause such as in (5), past morphology on the verb expresses a past temporal reference. This is the reason why a temporal adverb of past time (in 2003) is felicitous (5a), while a temporal adverb of present time (now) is ill-formed (5b). (5) a. I had a car in 2003. b. * I had a car now. The same morphology may however systematically be used to express counterfactuality. This is the case in counterfactuals (6a) and embedded under verbs of wishing which results in what Iatridou 2000 refers to as present counterfactual wishes (6b). Evidence that in such contexts past tense morphology is no longer associated with a past interpretation comes from the fact that in this context it is compatible with a temporal adverb of present time (now). (6) a. If I had a car now, I might drive to school. b. I wish I had a car now.

22 | Martina Wiltschko

Again, the term fake past is only really appropriate if we consider the past tense morphology to be intrinsically denoting past time reference. In that case, the examples in (6) would indicate that past tense may sometimes lose is past interpretation. It would only appear to be past (i.e., be fake past) when in fact it is not. Thus, the term fake past suggests that the non-past use of the past morphology is derived, such that one of the meaning components is lost under certain conditions. This approach is schematized in Figure 9.

form /-ed/

basic interpretation past

fake interpretation counterfactual Figure 9: Fake past.

Interestingly, the analysis developed in the very paper that introduced the term fake past, i.e., Iatridou 2000 does not reflect this intuition. Rather on Iatridou’s 2000 approach both the past and the fake past (i.e., the counterfactual) interpretation are derived from a more abstract form. The so-called past tense morpheme is analyzed as providing a skeletal meaning of the form in (7). (7) T(x) excludes C(x). where T(x) = “Topic( x )” (i. e., “the x that we are talking about”) C(x) = “the x that for all we know is the x of the speaker’ Iatridou 2000: 246 Thus, what we call the past morpheme is really the realization of an exclusion feature (ExclF). When it ranges over times it receives a temporal interpretation: the topic time excludes the utterance time, as schematized in Figure 10. When access to alternative worlds is given (such as in conditionals and under verbs of wanting), then the exclusion feature ranges over worlds and yields a counterfactual interpretation: the topic world excludes the actual world. This is schematized in Figure 11.

Fake form

|

23

Syntax

ordering times Underlying Form

Interpretation

ExclF

past

Figure 10: Deriving a past interpretation from ExclF.

Syntax

ordering worlds Underlying Form

Interpretation

ExclF

counterfactual

Figure 11: Deriving a counterfactual interpretation from ExclF.

2.3 Summary We have now seen that fake indexicals and fake past forms have a common plot: there is an intuitive sense in which one interpretation is basic while the other is derived, hence fake. This is schematized in Figure 12. More precisely, the term fake suggests that one of the meaning components of the basic interpretation is somehow lost in certain contexts. Indexicals lose their

form

basic interpretation

fake interpretation

Figure 12: Fake form.

24 | Martina Wiltschko

indexical force allowing for a bound variable interpretation, and past morphology loses its past force allowing for a counterfactual interpretation.⁵ While the terms fake indexical and fake past are at first sight insightful names for the phenomena at hand, neither Kratzer 2009 nor Iatridou 2000 ultimately adopt an analysis that would be true to the term. On Iatridou’s account, the syntactic derivation serves as a crucial mediator between the form and the interpretation. That is, both the past and the counterfactual occurrence of the same morpheme are derived from an underlying form. The syntactic configuration crucially determines the surface interpretation of ExclF either as past tense or as counterfactual. In this way, counterfactual marking is not really an instance of fake form. The intuition that past marking is basic while counterfactual marking derived through some kind of loss (i.e., fake) stems from the fact that the syntactic context for past marking is more basic than that for counterfactual marking. The former involves all matrix indicative clauses, while the latter must be embedded under an operator that allows access to worlds. The logic behind Iatridou’s analysis of fakeness is summarized in Figure 13. The logic behind Kratzer’s analysis of fake forms is somewhat different. In particular, she argues that the surface form is enriched in the syntactic component but crucially in a way that does not affect its interpretation. Moreover it differs

syntactic context 1 basic

abstract form

interpretation 1

syntactic context 2 complex

abstract form

interpretation 2

“fake form” Figure 13: Fake forms aren’t fake.

5 It is not surprising that indexical pronouns and past tense morphology behave in similar ways: in matrix clauses, tense marking is itself deictic in that it relates the event situation to the utterance situation – and is thus dependent in its interpretation on features of the context, just as indexicals. Moreover, tense marking is sometimes treated on a par with pronouns (Partee 1973, Kratzer 1998).

Fake form

|

25

from non-fake (genuine) indexical forms in that the latter are born with fully specified features. The intuition that we are dealing with a fake form may again be related to the fact that the bound variable reading requires a more complex syntactic configuration, namely one where binding takes place. This is schematized in Figure 14.

Spell out fully specified form

syntactic context 1 basic

fully specified form

interpretation 1

syntactic context 2 binding

abstract form

interpretation 2

“fake form” Figure 14: Fake forms aren’t fake.

Thus, Kratzer’s 2009 account of fake form differs crucially from Iatridou’s 2000 in that it does not view syntax as mediating between form and interpretation. Rather, the interpretation can be directly read off their underlying form and thus the difference between the two interpretations is entirely a matter of semantics (Kratzer 2009: 219). The proposal I develop here is in the spirit of Iatridou’s 2000 analysis of fake past but at the same time it also has some ingredients of Kratzer’s 2009 analysis of fake indexicals. In the next section, I introduce the basic insight as well as the necessary background assumptions.

26 | Martina Wiltschko

3 The relation between form and interpretation is mediated by syntactic category Fake forms provide us with a unique window into the relation between form and interpretation and I suggest, following Wiltschko 2014, that this relation can be mediated by syntax. In this section I first introduce the background that allows us to develop the proposal (3.1) and then I move on to introducing the proposal (3.2).

3.1 Background Wiltschko 2014 argues that the interpretation of a particular form is affected by the syntactic category (κ) it associates with, as schematized in Figure 15.

Syntactic Category κ

Form π

Interpretation Σ

Figure 15: Syntactic category mediates the form-interpretation relation.

This claim is similar in spirit to the standard view on the form-interpretation relation when it comes to complex expressions (i.e., sentences). In particular, it has long been established that syntax mediates the relation between the level of representation that feeds the component responsible for semantic interpretation (LF) and the level of representation that feeds the component responsible for phonetic interpretation (PF). Crucially, there is no direct relation between LF and PF, neither in the government and binding version of modeling the architecture of grammar – the T-model illustrated in Figure 16 (Chomsky 1981) - nor in the minimalist version of that model – the Y-model illustrated in Figure 17 (Chomsky 1995). In addition, it is also necessary to adopt some of Kratzer’s assumptions on syntactic computation. In particular, Kratzer subscribes to the so-called syntactic contextualism of the kind initiated by Borer (2005a,b). On this view, many prop-

Fake form

|

27

D-structure syntax S-structure

PF

LF

Figure 16: The T-model.

spell-out

LF

PF syntax Figure 17: The Y-model.

erties of lexical forms are contributed by functional heads, hence by the syntactic environment of the respective lexical item. This is a version of the claim that syntax plays an important role in the mediation between form and interpretation. However, Kratzer’s version of it takes it to determine the “ultimate shape of pronouns” rather than its interpretation (Kratzer 2009: 189). The way this is possible is to admit that lexical items may be inserted after syntactic computation i.e., via the process of late insertion familiar from Distributed Morphology (Halle & Marantz 1993; but see also Sproat 1985, Beard 1995 and Jackendoff 1997). Unlike most approaches that allow for late insertion however, Wiltschko 2014 argues that it is necessary to recognize that linguistic forms may be either inserted early or late. The particular framework Wiltschko 2014 develops recognizes a universal syntactic spine, i.e., a series of hierarchically organized functional categories, which come with core abstract functions (such as anchoring or classifying). As illustrated in Figure 18, each of these functions has verbal and nominal instantiations capturing the long-established parallelism between verbal and nominal extended projections (Chomsky’s 1970, Abney 1987, Cardinaletti and Starke 1999; Bernstein 2001, 2008; Grimshaw 2005; Koopman 2005; Rijkhoff 2008).

28 | Martina Wiltschko

linking

CP CP IP IP

KP KP DP DP

anchoring

point-of-view

AspP AspP

vP vP ...

AspP PhiP

classification

nP ... nP

Figure 18: The universal spine hypothesis.

What differentiates the universal spine hypothesis from most other approaches towards functional projections is the claim that the substantive content of each category is provided by language-specific lexical forms, referred to as units of language (UoL) in Wiltschko 2014 (see also Ritter & Wiltschko 2009, 2014). On this view then, the spine exists independently of the language-specific UoLs that are used to construct sentences. Wiltschko 2014 further assumes that all categories along the spine have in common that they relate two abstract arguments to each other, as in (8) (cf. Speas 2010). (8) The universal structure of categories κP arg

κ κ

arg

In particular, verbal categories relate situation arguments (argsit ) (9a), (see Section 3.2.1) while nominal categories relate individual arguments (argind ) (9b) (see Section 3.2.2; see also Gruber 2013). Moreover, these arguments are related to each other via an abstract unvalued feature intrinsically associated with the head of the category κ. For verbal categories this feature is [coincidence] in the sense of Hale 1983 and for nominal categories this feature is [identity].

Fake form

(9) a. verbal categories κP arg-sit

29

b. nominal categories κP κ

κ [ucoin]

|

arg-ind arg-sit

κ κ [uident]

arg-ind

This view allows for a novel formal typology for linguistic categories fully explored in Wiltschko 2014. There are three dimension across which the association of UoLs with the spine may differ: (i) place, (ii) manner, (iii) and timing. That is, any particular UoL may associate with the spine in different places, i.e., at different layers within the spine as illustrated in Figure 19. κ: linking

κ: anchoring

κ: point-of-view UoL κ: classification

Figure 19: Variable place of association.

The second parameter relative to which language-specific UoLs may differ has to do with the manner of association. UoLs differ as to whether they associate with the spine via the is-a relation in which case they display the distribution of syntactic heads. Alternatively UoLs may associate as modifiers. The difference in the manner of association is schematized in Figure 20. Finally, the third parameter relative to which language-specific UoLs may differ has to do with the timing of association. UoLs differ as to whether they are associated early or late, as schematized in Figure 21. The arrow pointing from UoL towards κ indicates early insertion (i.e, before syntactic computation); the arrow pointing from κ towards UoL indicates late insertion (i.e., after syntactic computation).

30 | Martina Wiltschko κ

κ

[UoL]κ

arg

κ

arg UoL

κ

Figure 20: Variable manner of association.

κ

κ

UoL

κ

arg

κ

arg

UoL

Figure 21: Variable timing of association.

With these assumptions in place, the universal spine hypothesis allows for syntactic contextualism in two ways. For late association UoLs, syntactic computation will affect the form of a given UoL, just as in Kratzer’s 2009 analysis of fake indexicals. However, for early association UoLs, syntactic computation will affect the interpretation of a given UoL just as in Iatridou’s 2000 analysis of fake past. The effect on interpretation follows from the assumption that each layer in the spine is associated with a unique function as a matter of UG. Hence it follows that a given UoL may acquire some meaning simply by virtue of being associated with the syntactic spine. In this way, the relation between form and interpretation is mediated by syntax. While a given, early association UoL is associated with a particular meaning, its interpretation may change depending on the syntactic context it associates with. The universal spine hypothesis thus predicts the existence of multi-functionality. And indeed patterns of multi-functionality such as the apparent fake interpretations discussed here are ubiquitous in the languages of the world. In particular, there are several phenomena that require a model of grammar where the relation between form and interpretation is mediated by syntactic structure. This is because a given form is interpreted in different ways depending on the syntactic context. Such phenomena include not only fake past and fake indexicals, but also expletives, polysemy, conversion, as well as syncretism. In what follows, I develop an analysis of fake indexicals within the framework of the universal spine hypothesis.

Fake form

|

31

3.2 Proposal In this section I review the analysis of fake past developed in Ritter & Wiltschko 2014 (3.2.1). This will serve as the basis for the analysis of fake indexicals (3.2.2).

3.2.1 Fake past The starting point for the analysis is the translation of Iatridou’s 2000 insight into an analysis within the universal spine hypothesis (Ritter & Wiltschko 2014). The intrinsic [coin(cidence)] feature of functional categories plays the role of Iatridou’s exclusion feature. By hypothesis, past morphology values the unvalued coincidence feature [ucoin] associated with the anchoring category (INFL) as [−coin]. Furthermore, since the content of the head is temporal, it picks out the temporal component of the situation argument (indicated as Utt-time). (10) An analysis of past marking⁶ κP: anchoring

Utt-time

κ κ [−coin]

Ev-time

past Furthermore Ritter & Wiltschko 2014 argue that the utterance situation associated with the anchoring category is an instance of default interpretation. In particular, they claim that it is in fact a pronominal (i.e., anaphoric) argument, but that in the absence of an antecedent (i.e., in matrix clauses) it is interpreted as the utterance situation.

6 The representation in (10) is simplified for expository reasons. Valued κ corresponds to the traditional functional category tense. However, the abstract arguments it relates to each other do not correspond to the traditional specifier or complement of tense. That is, the lower argument (Ev-time) corresponds to the specifier of the VP-complement; the higher argument must be viewed as a second specifier, as in Demirdache & Uribe Etxebarria 1997.

32 | Martina Wiltschko

(11) The external argument is pronominal κP: anchoring

Pro-sit

κ κ [ucoin]

Ev-sit

Since past marking values κ in (10), the situation arguments are restricted to times. To account for fake past marking in counterfactual contexts, Ritter & Wiltschko (2014) argue that in this case, counterfactuality – associated with the linking layer of the spine (i.e., CP) – values the coincidence feature in the anchoring category. In this case, past marking is an instance of past agreement. In particular, fake past in counterfactuals is not the only instance of past tense morphology that lacks past temporal force. The same can be observed in English Sequence of Tense (SOT) effects as well as in Romance subjunctives. SOT effects arise when an embedded clause is selected by a [+past] matrix verb. In this case the embedded INFL must be inflected as past even though its temporal reference need not be past relative to the matrix situation. Thus it allows for both a simultaneous reading as well as a shifted reading. It is only in the shifted reading that past morphology is in fact interpreted as past. The absence of a past interpretation in the simultaneous reading can be viewed as a form of tense agreement with the matrix verb (see Giorgi 2009 for a recent overview). (12) Mary said that she was tired. (i) Simultaneous reading: Time of being tired is at time of saying. (ii) Shifted reading: Time of being tired is before time of saying (Enç 1987: 350 ex. 18) Ritter & Wiltschko 2014 suggest that a similar mechanism is at play in counterfactuals, but here we are not dealing with past agreement but instead it is an instance of agreement between two [−coin] features: the one in INFL must be of the same value as the one in C. Since now past no longer values INFL as [−coin] it is no longer interpreted as past but rather it spells out the independently valued [−coin] feature. This is reminiscent of Iatridou’s 2000 insight that past morphology expresses an exclusion feature. And the intuition that in counterfactuals it is worlds rather than times that are excluded is captured by the assumption that the trigger for [−coin] agreement in INFL is counterfactuality in C, which in turn orders world arguments (Mezhevich 2008a, b).

Fake form

|

33

(13) Counterfactuals ordering worlds κP

Eval-world

κ κ [−coin]

argsit ⟨t, loc, part⟩

irrealis Ritter & Wiltschko 2014 present independent evidence that fake past marking in INFL is dependent on the counterfactual content in C. In particular, it is consistent with the well-known fact that in counterfactuals the inflected auxiliary may move to C, as in (14). This differs crucially from realis conditionals where past marking has temporal force (15a, b) but INFL- to-COMP is ruled out (15c). (14) Had she arrived, I would not have left. (15) a. If she really arrived last night she will be here today. b. * If she really arrived right now, she will be here today. c. * Has she really arrived, she will be here. Ritter & Wiltschko’s 2014 analysis thus captures the intuition that indexical past is somehow more basic than fake past by claiming that in that context it doesn’t serve an interpretive function, namely that of valuation. In such contexts, the anchoring category is necessarily embedded (either under a matrix predicate or under a higher functional head C) and thus the syntactic context is essentially more complex.

3.2.2 Fake indexicals As mentioned above, Wiltschko 2014 assumes that nominal categories are similar to verbal categories. However they differ in that the intrinsic feature is not coincidence but instead identity and moreover that we are dealing with individual rather than situation arguments. Thus, the nominal equivalent of the anchoring category (INFL) – generally assumed to be D – is shown in (16).

34 | Martina Wiltschko

(16) Nominal anchoring κP: anchoring Pro-ind

κ κ [uident]

Ref-ind

In the absence of an antecedent, the pronominal individual argument Pro-ind will, by default, be interpreted as an Utterance individual, i.e., a speech act participant. Thus, both 1st and 2nd person pronouns are analyzed as instances of [+ident] such that it is asserted that the referent is identical to an utterance individual, as in (17a).⁷ In contrast, 3rd person pronouns are analyzed as instances of [−ident] such that it is asserted that the referent is not identical to an utterance individual, as in (17b). (17) a. Local pronouns κP: anchoring Utt-ind

κ κ [+ident]

b. Non-local pronouns κP: anchoring Utt-ind

Ref-ind

κ κ [−ident]

Ref-ind

What is crucial in this analysis, is that 1st and 2nd person pronouns are not intrinsically indexical. Instead, indexicality is supplied by abstract arguments, which come with the anchoring category. In the absence of an antecedent, this pronominal argument corresponds to the utterance argument. Accordingly, fake indexicals must be analyzed as structures in which the abstract argument is bound to an appropriate antecedent, i.e., in the presence of an antecedent it is interpreted as a pronominal argument (Pro-ind), as in (18).

7 I address the question concerning the difference between 1st and 2nd person pronoun in Section 4.2.

Fake form

|

35

(18) “fake indexicals” κP: anchoring Pro-ind

κ κ [+ident]

Ref-ind

4 Consequences In this section, I discuss three consequences of the analysis just introduced. First, I discuss the impersonal use of 1st person pronouns (4.1); second I show independent evidence that the basis for local pronouns is the identity predicate (4.2). And finally, I discuss the properties of 3rd person pronouns in more detail (4.3).

4.1 Impersonal uses of 1st person pronouns Fake indexicals of the type discussed in Section 2.1, are not the only instances where 1st person pronouns are used in a non-indexical way. In particular, they can also be used as impersonal generic pronouns (Gruber 2013, Malamud 2012, Zobel 2010). Illustrative examples are given in (19) and (20). (19) Context: Edith und Brigitte are talking about their sister Bettina who claims not to be able to come home for Christmas because she has to work on her dissertation. Edith says: Wenn ich zu Weihnachten wirklich zuhause sein will, dann schaff ich … If I at christmas really at.home be want then manage I … das auch irgendwie. that too somehow ‘If I really want to spend christmas at home, then I will somehow manage.’ Gruber 2013: 160 (76) (20) Wenn ich als Mannschaft gewinnen will, dann muss ich motiviert auf … If I as team win want then must I motivated on … den Platz gehen the field go.inf

36 | Martina Wiltschko

‘If (one as) a team wants to win, then one/they has/have to enter the field motivated.’ Zobel 2010: 293 Note that the example in (20) is particularly revealing as it adds the modifier ‘als Mannschaft’ (‘as a team’) to the 1st person pronoun. Since one person cannot make up a team, it clearly shows that in this case the 1st person pronoun has a different referent: it expresses a general statement about how teams that want to win have to act. The impersonal or generic use of local pronouns is thus another instance of a fake indexical, i.e., a non-indexical use of an otherwise indexical form, as illustrated in Figure 22.

Indexical

1st person pronoun

bound variable anaphor “fake indexicals” impersonal/ generic

Figure 22: Another fake use of 1st person pronouns.

The logic of the analysis developed in Section 3.2.2 leads us to identify the source of the impersonal generic interpretation in the abstract argument associated with the anchoring argument. In particular, suppose that this abstract pronominal argument can be interpreted generically. In this way, the abstract argument is reminiscent of arbitrary PRO and pro (Williams 1980, Suñer 1983; see Sigurdsson and Egerland 2009 for a recent overview). Hence I have labelled the pronominal argument as Pro-arb in (21).⁸

8 On the affinity between 1st person, arbitrary pro and genericity see Moltmann 2006.

Fake form

|

37

(21) The impersonal generic use of 1st person pronouns κP: anchoring Pro-arb

κ κ [+ident]

Ref-ind

Both, arbitrary pro and the impersonal use of 1st person pronouns have in common that they are restricted to pick out human referents. Consider the examples in (22) which involve impersonal passives and thus no overt subject. In place of the subject, this construction has a silent pronominal element with arbitrary reference (proarb ) Crucially, this arbitrary pronoun is restricted to [+human] individuals. The impersonal passive is well-formed with a verb that is restricted to human agents such as reden (22a). In contrast, verbs denoting an activity that can only be performed by non-humans – horses in the case of wiehern (‘neigh’) – are infelicitous (22b). This is due to the incompatibility between the verb’s selectional restriction and the requirement for the arbitrary pronoun to be restricted to humans. (22) a.

In der Schule wird proarb leise geredet in the school aux proarb silent speak.part ‘People talk silently in school.’ b. # Im Stall wird proarb leise gewiehert in.the stable aux proarb silent neigh.part ‘People neigh silently in the stable.’

Similarly, the first person pronoun when used as an impersonal must refer to a human individual (Gruber 2013). Hence, we find the same contrast as with silent arbitrary pronouns: it cannot function as the agent of a verb that requires a nonhuman agent. This is illustrated in (23). (23) a.

Wenn ich in der Schule reden will, dann muss ich das leise tun. when I in the school speak want then must I that silent do ‘If I want to talk at school I have to do that silently.’ b. # Wenn ich im Stall wiehern will, dann muss ich das leise tun. when I in.the stable neigh want then must I that silent do ‘If I want to neigh in the stable I have to do that silently.’

38 | Martina Wiltschko

As with local pronouns used as bound variable anaphors, we also have the impression that the impersonal use of these pronouns is somehow less basic – hence fake. In case of the impersonal use this is so because this interpretation requires the entire utterance to be a generic statement. That is, the impersonal use of the local pronoun is not available in episodic statements, as shown in (24). (24) # Wenn ich als Mannschaft gestern gewinnen wollte, dann musste … If I as team yesterday win wanted then must.past … ich motiviert auf den Platz gehen. I motivated on the field go.inf ‘If (one as) a team wanted to win yesterday, then one had to enter the field motivated.’ Thus, under this analysis, an impersonal pronoun is constructed with an arbitrary pronominal argument serving as the abstract argument with which the referent is identified by the identity feature. This predicts that an argument of a similar type is also available in the verbal domain, as in (25). (25) generic anchoring κP: anchoring Pro-arb

κ κ [+coin]

Ev-sit

This is indeed so, as evidenced by the well-known fact that present tense does double duty as a marker of genericity, i.e., it does not always express present time. This is shown (26) where present tense on the verb does not indicate that the time of spoiling coincides with the time of utterance. Instead, it is a general statement about what happens when too many cooks work on the broth. (26) Viele Köche verderben den Brei. Many cooks spoil the mash ‘Too many cooks spoil the broth.’ Note that it is unclear how this impersonal use of the 1st person pronoun can be accounted for under Kratzer’s 2009 analysis of fake indexicals, though it too is an instance of a non-indexical, thus fake use. In particular, since it is not interpreted

Fake form

|

39

indexically, it would have to be born as a minimal pronoun acquiring its 1st person feature in the course of the post-semantic syntactic derivation. However, unlike what Kratzer 2009 argues to be the case for the bound variable use, there is no obvious trigger as to why a first person feature would have to be added in this context.

4.2 ich as the spell out of the identity predicate The core of the analysis developed in Section 3.2.2 has it that a local pronoun is not intrinsically indexical but instead that it acquires its indexicality by the syntactic context. The intrinsic contribution of local pronouns is to value the identity feature associated with the anchoring category as [+ident]. This amounts to saying that local pronouns serve as identity predicates. In this subsection, I address the question as to how 1st and 2nd person pronouns differ from each other. And I present some independent evidence that German ich is indeed an identity predicate. Everything else being equal, the analysis presented thus far cannot distinguish between 1st and 2nd person pronouns: they both value the identity feature as [+ident] asserting that the referent is identical to an utterance participant. I follow Wiltschko 2014 where this proposal is modified as follows. Suppose that ich is the identity predicate and the difference between 1st and 2nd person is encoded by means of agreement. While in their nominative instantiations this is not immediately obvious, it is in the accusative, as shown in (27). Both 1st and 2nd person pronouns are composed of an invariant stem ich and a prefix m- for 1st person and d- for 2nd person. (27) a. m-ich 1st person accusative b. d-ich 2nd person accusative Independent evidence for the decomposition of these accusative pronouns comes from a number of considerations (see Leiss 2004 and Wiltschko 2014 for a more detailed discussion). Consider the paradigm of German personal pronouns in Table 1. A closer look at the shaded cells reveals that in the 1st person all but the nominative pronoun start with m-. In contrast, all 2nd person pronouns start with d-. This appears to be more than accidental, as what remains of the pronominal forms

40 | Martina Wiltschko

Table 1: German personal pronouns. 1

2

3 MASC

FEM

NEUT

er

sie

es

ihn

sie

es

ich

du

mich

dich

DAT

mir

dir

ihm

ihr

ihm

POSS

mein

dein

sein

ihr

sein

NOM ACC REFL

sich

is – except for the nominative form – identical: -ich is the accusative stem, -ein is the possessive stem⁹ , and -ir is the dative stem. This is illustrated in (28). (28) a. Accusative 1st [[m]1 -[ich]1 ]1 nd 2 [[d]2 - [ich]1 ]2 3rd [[s]3 - [ich]1 ]3 b. Possessive [[m]1 -[ein]3 ]1 1st 2nd [[d]2 - [ein]3 ]2 3rd [[s]3 - [ein]3 ]3 c. Dative [[m]1 -[ir]3 ]1 1st 2nd [[d]2 - [ir]3 ]2 3rd [ihr]3 Incidentally note that the same decomposition is possible in the French pronominal paradigm (Kayne 2000), where the cognate prefixes m-, t-, and s- serve the same function as in German. This is illustrated in Table 2. Thus, I analyze German m-, d-, and s- as agreement prefixes that attach to the pronominal stem ich, which in turn serves as the identity predicate. This is illustrated in (29).

9 At this point, I have no analysis for the basis for dative and possessive pronouns. See Brandner 2014 for a detailed discussion regarding the decomposition of s-ein.

Fake form

|

41

Table 2: French personal pronouns. 1

2

3 MASC

FEM

il

elle

li

la

me

tu

me

te

DAT

me

te

lui

elle

POSS

mon

ton

son

leur

NOM ACC REFL

se

(29) Decomposing accusative pronouns κP: anchoring

Pro-ind

κ Ref-ind

κ 1: m2: d3: s-

κ [+ident] ich

If this analysis is on the right track, we have independent evidence for the analysis of ich as an instantiation of the identity predicate as opposed to being intrinsically a 1st person indexical pronoun. If it was intrinsically 1st person, then the fact that the same form also occurs in 2nd and 3rd person pronouns would be unexpected. In fact, if that were the case it would seem to make the decomposition in (28) impossible. Further evidence that ich is not intrinsically an indexical first person pronoun stems from the fact that it may be used inside of compounds, without referring to the speaker, i.e., another fake use of indexical pronouns. This is illustrated in (30). (30) a. Ich-sucht me-addiction ‘self-absorption’ b. Ich-gesellschaft me-society ‘self-absorbed society’

42 | Martina Wiltschko Note that on this analysis, the 3rd person accusative pronoun (sich) too contains the identity predicate ich. At first sight, this seems to contradict the analysis of 3rd person I have introduced in Section 3.2.2. In particular, there I assumed that 3rd person pronouns are best analyzed as valuing the identity feature as [−ident] thereby asserting that the referent does not coincide with the utterance participant. However, note that the 3rd person pronoun containing what I take to be the identity predicate (sich) is in fact not an ordinary 3rd person pronoun, but it is instead a reflexive. And crucially, it is the hallmark of reflexives that they encode an obligatory local dependency. Thus, while reflexive 3rd person pronouns do not assert an identity with the utterance participant, they do assert identity with an appropriate antecedent (see Cresswell 1973, Keenan 1988, Gast 2006 for analyses of reflexives as identity predicates). Hence we only need to assume that s-ich is always associated with a syntactic category where the abstract argument is pronominal and thus requires an antecedent. In this way, the reflexive pronoun receives a similar analysis as fake indexicals (see also Kratzer 2009 for arriving at this conclusion albeit for different reasons). Further support for the assumption that the identity predicate may serve as the basis for both 1st person pronouns as well as reflexives stems from the fact that in many languages the former is contained in the latter (Leiss 2004). This is shown in Table 3. Table 3: “First person” contained in reflexives.

Russian German Old Icelandic Gothic Modern Icelandic Norwegian

REFL

1ST NOM

sebja sich sik sik sig seg

ja ich ik/ek ik eg/ég jeg

I take this to establish the proposed analysis of German ich as instantiating an identity predicate, thus valuing the identity feature associated with the nominal anchoring category as [+ident]. Its indexical reference to the speaker comes about by being associated with the syntactic structure, which serves as an indexical shell. Thus, the relation between the form ich and its interpretation as an indexical 1st person is mediated by the syntactic structure. This is summarized in (31).

Fake form

|

43

(31) Inserting the identity predicate into the indexical shell

κP: anchoring Utt-ind

1st person indexical pronoun

κ κ [+ident] ich

Ref-ind identity predicate

Note in passing that this pattern cannot be straightforwardly accounted for under Kratzer’s 2009 analysis of fake indexicals. In particular, in this case ich cannot be born as a fully specified form, because if it were it would have to be interpreted indexically. But at the same time there is no source for 1st person agreement that would trigger the occurrence of the 1st person form in this context. What remains to be accounted for under the analysis presented here however, is the forms of the nominative pronouns: (32) a. b. c. (i) (ii) (iii)

ich du er sie es

1st person nominative 2nd person nominative 3rd person masculine nominative 3rd person feminine nominative 3rd person neuter nominative

Among these pronouns only the 1st person pronoun contains the identity predicate, and in fact there is no marker of 1st person to be found (i.e., the expected 1st person prefix m- is missing). But neither the second nor the third person nominative pronouns are composed with ich. I suggest, that this pattern be analysed as involving zero marking. Hence, the first person pronoun is analyzed as containing a silent 1st person agreement marker (otherwise spelled out as m-), as in (33a) while the second person pronoun is analyzed as containing a silent (possibly elided) version of the identity predicate but an overt form of the pronominal agreement, in the form of the d- prefix, as in (33b).¹⁰

10 I have nothing to say about the source of the vowel -u which accompanies the 2nd person prefix in du.

44 | Martina Wiltschko (33) a. 1st nominative κP: anchoring

Utt-ind

Utt-ind

κ

κ κ

Ref-ind

κ 1: m-

b. 2nd nominative κP: anchoring

κ [+ident] ich

2: d-

Ref-ind

κ [+ident] ich

For 3rd person non-reflexive pronouns the absence of the identity predicate is unsurprising: they do not express identity with an utterance participant. However, 3rd person pronouns still present us with an analytical challenge to which I turn in the next section.

4.3 A closer look at 3rd person pronouns Recall the analysis of fake indexicals repeated below as (34). There are two ingredients to capture the bound variable interpretation: the abstract argument is interpreted as anaphoric (Pro-ind) and the identity feature is valued as [+ident]. This derives the result that the referent is identical to the pronominal argument. (34) Local pronouns as bound variables κP: anchoring Pro-ind

κ κ [+ident]

Ref-ind

This presents us with a dilemma in dealing with 3rd person pronouns. On the one hand 3rd person pronouns differ from local pronouns in that they value the identity feature as [−ident] (see Ritter & Wiltschko 2014) but on the other hand, they too can function as bound variables. Thus, everything else being equal we would expect then that they too should be valued as [+ident] to assert identity with the pronominal argument. And indeed that is the analysis for 3rd person reflexives. So how can we reconcile this dilemma?

Fake form

|

45

While I will not present an analysis here, I will present some data which suggest that this conundrum may in fact be a virtue of the analysis. In particular, in Upper Austrian German, 3rd person pronouns come in two forms: in addition to the regular pronouns there are also phonetically reduced ones. The relevant paradigm is given in Table 4.¹¹ Table 4: Upper Austrian pronouns: full and reduced forms.¹²

sg

pl

nominative

accusative

full

reduced

full

reduced

masc

ea

a

eam

n

fem

sie

s

sie

s

neut

es

s

n/a

s

sie

s

sie

s

What is interesting from the present perspective is the fact that these two paradigms differ in the way they may be interpreted. First, as shown in (35), only full but not reduced forms can be used deictically (35) As an answer to the question: Who sang yesterday? and uttered while pointing at an individual in sight of the utterance context Gestan hot ea/*a gsunga. Yesterday has hef /her sing.part ‘Yesterday he sang.’ Second, both full and reduced forms can be use to refer to a salient discourse referent (i.e., endophorically), as show in (36). (36) Wasst eh, da Peta is gestan auf Besuch kumma Zeast homa … know-2nd prt det Peter is yesterday to visit come.part First have.we … olle gessn, und donn hot ea/a gsunga. all eaten and then has hef /her sing.part ‘You know, Peter came for a visit yesterday. First we all ate and then he sang.’

11 The regular forms do not necessarily bear stress, though they can if they are focussed. 12 The Upper Austrian pronouns are almost identical to those found in Bavarian. See Weiss 1998 for discussion.

46 | Martina Wiltschko

In this respect, it appears that both forms can be used anaphorically. However, when we turn to instances of bound variable anaphors the situation is different. In that case, only the reduced form is well formed; the full form cannot be interpreted as a bound variable, as shown in (37). (37) Neta do Hons hot a Frog kriagt wos *ea/a vastondn hot. Only det Hans has a question gotten which hef /her understood has ‘Only Hans got a question which he understood.’ Finally, neither full nor reduced forms can be used for generic impersonal uses (38a). Instead a dedicated impersonal pronoun man has to be used (38b). (38) a. Des tuat ea/a ned. that does hef /her not ‘He doesn’t do that.’ (i) = talking about a specific individual (ii) ≠ talking about people in general b. Des tuat ma ned. that does one not ‘One doesn’t do that.’ (i) ≠ talking about a specific individual (ii) = talking about people in general The range of meanings associated with the full and reduced 3rd person pronominal forms are summarized in Table 5. Table 5: The range of interpretation associated with full and reduced pronouns.

deictic endophoric BVA generic

full

reduced

   

   

This distribution suggests that endophoric coreference is not achieved in the same manner as bound variable anaphora. There are a few conclusions we can draw from this pattern. First, it appears that weak pronouns do not serve to value the identity feature. This is so because as a 3rd person pronoun the identity feature would be valued as [−ident] but as a bound variable it would have to be [+ident].

Fake form

|

47

This may suggest that reduced pronouns serve as modifiers of a silent [+ident] head as in (39). (39) 3rd person bound variables κP: anchoring

Pro-ind

κ κ

Ref-ind

masc: a/n κ fem : s [+ident] neut : s 0 Note that both ingredients of this analysis are independently needed: modificational prefixes were necessary to analyse reflexives, and a silent identity predicate was needed to account for 2nd person pronouns. The assumption that 3rd person gender pronouns are qualitatively different than local pronouns is not unique to this analysis. It is also what’s assumed in Kratzer 2009. In particular, Kratzer assumes that 1st and 2nd person features are irreducibly referential in that they refer to speakers and hearers as in (40). (40) Participant features [| [1st ]N |]g,c = the speaker(s) in c [| [2nd ]N |]g,c = the hearer(s) in c [| [1st + 2nd ]N |]g,c = the sum of speaker(s) and hearer(s) in c Kratzer 2009: 220 (70) In contrast, 3rd person features are assumed to be the absence of person features but are instead made up of descriptive features such as gender features, as in (41). (41) “3rd person” descriptive features [| [male]N |]g,c = λx. x is one or more males. [| [female]N |]g,c = λx. x is one or more females. [| [thing]N |]g,c = λx. x is one or more things Kratzer 2009: 221 (73)

48 | Martina Wiltschko In contrast, full 3rd person pronouns may be analysed as valuing the identity feature as [−ident] deriving the fact that they neither refer to the utterance participant nor do they allow for a bound variable interpretation.¹³

5 Conclusion In this chapter, I have explored the formal underpinnings of fake forms. This term was taken to be a descriptive label for linguistic forms that are associated with more than one interpretation. There is a sense in which one of the interpretations is considered real (basic) and the other one fake (derived). We started the discussion with two prominent instances of fake forms: fake indexicals and fake past. In reviewing the analyses developed in the papers that introduced the terms fake indexical (Iatridou 2000) and fake past (Kratzer 2009) we have seen that (i) they don’t in fact reflect a sense of fakeness and (ii) they develop fundamentally different ways of viewing fakeness. Iatridou 2000 argues that there is an abstract underlying form, which serves as the basis for both the ‘real’, and the ‘fake’ use of so called past tense morphemes. In contrast, Kratzer 2009 argues that there are two ways pronouns may start out as: (i) as minimal pronouns without feature specifications yielding the fake (bound variable) use and (ii) with fully specified features yielding the real (indexical) use. Crucially, in both analyses, syntax plays an important role in the mediation between form and interpretation, albeit in different ways. For Iatridou 2000, the syntactic context affects the interpretation of linguistic form while for Kratzer 2009 the syntactic context affects the form of a linguistic form. These analyses are thus in line with two very different principles, both of which are typically ascribed to Frege (but see Pelletier 2001 for a critical review). In particular, Iatridou’s 2000 approach adheres to the principle of contextuality (42) while Kratzer’s 2009 analysis adheres to the principle of compositionality (43). (42) Principle of contextuality A word has a meaning only in the context of a sentence. Baker & Hacker 1980: 258

13 A detailed analysis of difference between pronouns that may or may not be bound variables is developed in Baltin, Déchaine & Wiltschko, 2015.

Fake form

|

49

(43) Principle of compositionality The meaning of any complex expression is a function of the meanings of its parts and of the way they are syntactically combined. Partee 2014: 9 (10) I have adopted an alternative that uses key ingredients of both approaches and which allows us to reconcile the tension between the two – apparently contradictory – principles. In particular, I developed an analysis of German 1st person pronouns within the framework of the universal spine hypothesis (Wiltschko 2014). According to this analysis a local pronoun is born as an identity predicate. Evidence for this basic interpretation comes from the use of ich inside of compounds as well as from the fact that ich is a core ingredient in the construction of reflexives. Its indexical use comes about by virtue of associating with the spine, which in turn comes equipped with two sources that can add meaning to a linguistic form: syntactic heads are intrinsically endowed with an identity feature (or a coincidence feature in the case of verbal heads). Moreover, these heads relate abstract arguments whose interpretation is contextually determined. Specifically, the anchoring category, with which the local pronouns associate comes with a pronominal argument. In the absence of an antecedent (in indicative matrix clauses) this argument is (by default) interpreted as the utterance argument yielding the indexical interpretation. In the presence of an appropriate antecedent, it is interpreted anaphorically yielding the bound variable interpretation. And in the context of a generic statement it may be interpreted arbitrarily yielding the impersonal interpretation. In this way the universal spine hypothesis allows us to reconcile the two opposing principles: on the one hand certain linguistic forms (namely those that associate with the spine early) receive different interpretations depending on the syntactic context, in accordance with the principle of contextuality. Since patterns of multi-functionality, including fake forms of the type discussed here are ubiquitous in natural languages, the principle of contextuality is empirically supported. On the other hand however, the principle of compositionality is also obeyed. In particular, syntactic computation is based on the syntactic spine, which does itself add meaningful parts (such as the abstract arguments). In as much as the analysis presented here is successful it supports the view that the relation between form and interpretation is mediated by syntax, not only in the domain of complex expressions, but also in the domain of seemingly simplex forms.

50 | Martina Wiltschko

References Abney, Steven Paul. (1987). The English noun phrase in its sentential aspect. Unpublished Phd Dissertation. Cambridge, MA: MIT. Baker, G.P. & P.M.S. Hacker (1980) Wittgenstein: Understanding and Meaning. Oxford: Blackwell. Baltin, Mark; Rose-Marie Déchaine & Martina Wiltschko, 2015. Two types of variables. Unpublished ms. NYU & UBC Beard, Robert. (1995). Lexeme-Morpheme Base Morphology: A General Theory of Inflection and Word Formation. Albany, NY: SUNY Press. Bernstein, Judy. (2001). The DP Hypothesis: Identifying clausal properties in the nominal domain. In Mark Baltin and Chris Collins (eds.), The Handbook of Contemporary Syntactic Theory. 536–561. Oxford: Blackwell. Bernstein, Judy. (2008). Reformulating the determiner phrase analysis. Language and Linguistics Compass, 2(6), 1246–1270. Borer, Hagit (2005a) In Name only. (Vol. 1). Oxford: Oxford University Press. Borer, Hagit. (2005b). The normal course of events (Vol. 2). Oxford: Oxford University Press. Brandner, Ellen. (2014). On possessive (reflexive) pronouns, equatives, and the structural basis for Condition A. Paper presented at GLOW, Understanding Possession workshop. Brussels, April 2014. Cable, Seth. 2005. Binding local person pronouns without semantically empty features. Ms., MIT, Cambridge, MA Cardinaletti, Anna, and Michal Starke. (1999). The typology of structural deficiencies: A case study of the three classes of pronouns. In Henk van Riemsdijk (ed.), Clitics in the languages of Europe, 145–233, New York: Mouton de Gruyter. Chomsky, Noam. (1970). Remarks on Nominalization. In: Roderick Jacobs and Peter Rosenbaum (eds.), Readings in English Transformational Grammar, 184–221, Ginn, Waltham, MA. Chomsky, Noam. (1981) Lectures on Government and Binding. Dordrecht: Foris. Chomsky, Noam. (1995). The minimalist program. Cambridge, Mass.: MIT Press. Cresswell, Maxwell J. (1978). Logics and Languages. London: Methuen. Déchaine, Rose-Marie, and Martina Wiltschko. (2015). When and why can 1st and 2nd person pronouns be bound variables? In Patrick Grosz, Pritty Patel-Grosz, and Igor Yanovich (eds.), NELS 40: Proceedings of the Semantics Workshop on Pronouns, 1–50. Enç, Myrvet. (1987). Anchoring conditions for tense. Linguistic Inquiry, 18, 633–657. Gast, Volker. (2006). Grammar of Identity: Intensifiers and reflexives in Germanic Languages. London, New York: Routledge (Taylor and Francis). Giorgi, Alessandra. (2009). A grammar of Italian Sequence of Tense. University of Venice Working Papers in Linguistics. 19 Grimshaw, Jane B. (2005). Words and Structure. Stanford, CA: Center for the Study of Language and Information, Stanford University. Gruber, Bettina. (2013). The Spatiotemporal Dimensions of Person: A Morphosyntactic Account of Indexical Pronouns. LOT dissertation series. Netherlands Graduate School of Linguistics. Hale, Kenneth. (1983). Warlpiri and the Grammar of Non-Configurational Languages. Natural Language and Linguistic Theory 1: 5–47.

Fake form

|

51

Halle, Morris, and Marantz, Alec. (1993). Distributed Morphology and the Pieces of Inflection. In Kenneth Hale and Samuel Keyser (eds.), The View from Building 20, 111–176. Cambridge: MIT Press. Iatridou, Sabine. (2000). The Grammatical Ingredients of Counterfactuality. Linguistic Inquiry, 31, 231–270. Jackendoff, Ray. (1997). The Architecture of the Language Faculty. Cambridge: MIT Press. Kaplan, David. (1989). Demonstratives: an essay on the semantics, logic, metaphysics, and epistemology of demonstratives and other indexicals. In: Joseph Almog, John Perry, and Howard Wettstein (eds.) Themes from Kaplan. 481–564. Oxford, UK: Oxford University Press. Kayne, Richard S. (2000). Parameters and Universals. Oxford: Oxford University Press Keenan, Edward. (1988). On Semantics and Binding Theory. In John Hawkins (ed.), Explaining Language Universals, 105–144. Oxford: Blackwell. Koopman, Hilda. (2005). On the parallelism of DPs and clauses in Kisongo Maasai. In Andrew Carnie, Sheila Dooley and Heidi Harley (eds.), Verb First: On the syntax of verb-intial languages. 281–302, Philadelphia: John Benjamins. Kratzer, Angelika. (1998). More Structural Analogies between Pronouns and Tenses. Proceedings from Semantics and Linguistic Theory, 8, 92–110. Kratzer, Angelika. (2009). Making a Pronoun: Fake Indexicals as Windows into the Properties of Pronouns. Linguistic Inquiry, 40(2), 187–237. Leiss, Elisabeth. (2004). Submorphematische Motiviertheit als Grammatikalisierungsergebnis: zur Grammatikalisierung von Reflexivpronomen. Zeitschrift für germanistische Linguistik, 32(2), 233–244. Malamud, Sophia. (2012). Impersonal indexicals: one, you, man, and du … Journal of Comparative Germanic Linguistics, 15(1), 1–48. Mezhevich, Ilana. (2008a). A time-relational approach to tense and mood. In Proceedings of the 27th West Coast Conference on Formal Linguistics, ed. by Natasha Abner & Jason Bishop, 326–334. Somerville, MA: Cascadilla Proceedings Project. Mezhevich, Ilana. (2008b). A feature-theoretic account of tense and aspect in Russian. Natural Language and Linguistic Theory, 26 (359–401). Moltmann, Frederike. 2006. Generic one, arbitrary PRO, and the first person. Natural Language Semantics 14:257–281 Partee, Barbara Hall. (1973). Some Structural Analogies between Tenses and Pronouns in English. The Journal of Philosophy, 70(18), 601–609. Partee, Barbara Hall. (1989). Binding implicit variables in quantified contexts. In Caroline Wiltshire, Randolph Graczyk, and Bradley Music (eds.), Papers from CLS 25, 342–356, Chicago: Chicago Linguistic Society. Partee, Barbara. (2014). A Brief History of the Syntax-Semantics Interface in Western Formal Linguistics. Semantics Syntax Interface. 1: 1–21. Pelletier, Jeff. (2001). Did Frege believe Frege’s principle? Journal of Logic, Language, and Information 10: 87–114. Reinhart, T. (1986) Center and periphery in the grammar of anaphora. In: Barbara Lust (ed.) Studies in the Acquisition of Anaphora, Vol. 1, 123–150. D. Reidel. Rijkhoff, Jan N.M. (2008). Synchronic and diachronic evidence for parallels between noun phrases and sentences. In Folke Josephson and Ingmar Söhrman (eds.), Interdependence of Diachronic and Synchronic Analyses, 13–42, Amsterdam: Benjamins.

52 | Martina Wiltschko

Ritter, Elizabeth, and Wiltschko, Martina. (2009). Varieties of INFL: TENSE, LOCATION, and PERSON, In: Jeroen Craenenbroeck, (ed.) Alternatives to Cartography, 153–201, Berlin, New York: Mouton de Gruyter. Ritter, Elizabeth and Martina Wiltschko (2014). The composition of INFL. An exploration of tense, tenseless languages and tenseless constructions. Natural Language and Linguistic Theory. Online first. http://link.springer.com/article/10.1007/s11049-014-9248-6 Rullmann, Hotze. (2004). Bound-variable pronouns and the status of person/number agreement. Linguistic Inquiry 35.1: 141–178. Schlenker, Philippe. (1999). Propositional Attitudes and Indexicality. Ph.D. dissertation, MIT. Schlenker, Philippe. (2003). A plea for monsters. Linguistics and Philosophy 26: 29–12 Sigurðsson Halldór and Verner Egerland (2009). Impersonal null-subjects in Icelandic and elsewhere. Studia Linguistica 63: 158–185. Speas, Peggy. (2010). Evidentials as generalized functional heads. In Anna-Maria Di Sciullo and Virginia Hill (eds.), Edges, Heads and Projections: Interface properties, 127–150. Amsterdam: John Benjamins. Sproat, Richard. (1985). On Deriving the Lexicon. Unpublished PhD dissertation, Cambridge, MIT Press. Stechow, Arnim von. 2003. Feature deletion under semantic binding. In: Makoto Kadowaki and Shigeto Kawahara (eds.) Proceedings of North East Linguistic Society (NELS) 377–403. Amherst: University of Massachusetts, Graduate Linguistic Student Association Suñer, Margarita. (1983). Pro arb. Linguistic Inquiry 14: 188–191. Weiß, Helmut. (1998). Syntax des Bairischen. Studien zur Grammatik einer natürlichen Sprache. Tübingen: Niemeyer, 1998. Williams, Edwin. (1980). Predication. Linguistic Inquiry 14: 287–304. Wiltschko, Martina. (2014). The universal structure of categories. Towards a formal typology. University of Cambridge Press. Zobel, Sarah. (2010). Non-standard uses of German 1st person singular pronouns. In: Kumiyo Nakakoji, Yohei Murakami, and Eric McCready (eds.) New Frontiers in Artificial Intelligence: JSAI -isAI 2009 Workshops Berlin/Heidelberg: Springer Verlag.

Inna Livitz

Referential dependence across phase boundaries in Russian Abstract: This paper addresses an alternation between null and overt subjects in Russian embedded finite clauses. In this syntactic environment, null pronouns must refer to the matrix subject, while overt pronouns are free to refer either to the matrix subject or to another individual in the discourse. I develop a derivational account of the link between pronominal form and interpretation which provides an explanation both for the obligatory referential dependence of null subjects and for the optional referential dependence of the overt subject pronouns in the same environment. I argue that the availability of a bound-variable reading for an overt pronoun in seeming violation of standard economy conditions on binding stems from the variable status of the embedded CP as a phase. The features of the embedded subject determine whether the embedded clause is spelled out once the CP is complete or whether the phase boundary is extended until matrix vP. I propose that null, referentially dependent pronouns occur when the embedded CP phase is extended, while overt pronouns occur inside a phasal CP. Keywords: null pronouns, finite control, Agree, binding theory, phases

1 Introduction The focus of this paper is on an alternation between overt and silent pronominal subjects in Russian. Although Russian is not a null-subject language, overt pronouns freely alternate with null pronouns in the subject position of embedded finite clauses, as illustrated in (1). There are two important observations to be made regarding (1). First, a null pronoun (0) in the embedded clause obligatorily refers to the matrix subject (1a), while an overt pronoun in the same position is free to refer either to the matrix subject or to a different individual in the discourse (1b). Second, there is no preference for the use of a null pronoun to express referential dependence on the matrix subject.¹

1 I use the term referential dependence throughout this paper to refer to the coindexed readings in (1), subsuming both coreferential and bound-variable interpretations. Note, however, that the

54 | Inna Livitz

(1)

a. Petjai skazal čto vesnoj 0i/∗j poedet v Pariž. will-go.3sg in Paris Petja said that spring ‘Petjai said that hei/∗j will go to Paris in the spring.’ b. Petjai skazal čto vesnoj oni/j poedet v Pariž. Petja said that spring he.nom will-go.3sg in Paris ‘Petjai said that hei/j will go to Paris in the spring.’

The possibility of a referentially dependent reading for the overt pronoun in (1b) is surprising given a well-established generalization in the binding literature; a pronoun is known to be prohibited from being referentially dependent on an antecedent if the pronoun alternates with an element (such as an anaphor) that is obligatorily dependent in reference on the same antecedent (Grodzinsky and Reinhart 1993; Safir 2004, a.o.). Given this generalization, we would expect – contrary to fact – that the overt pronoun would be prohibited from referring to the matrix subject in (1b) because the same syntactic context permits the use of a null pronoun, which is obligatorily referentially dependent on the matrix subject. In this paper I propose a unified explanation for the obligatory referential dependence of the null pronoun in (1a) and for the seeming violation of wellestablished economy conditions on binding by the overt pronoun in (1b). I adopt a derivational approach to referential dependence that is also able to derive the silence of a referentially dependent pronoun from the pronoun’s minimal structure and its unvalued φ-features. The availability of a referentially dependent reading for the overt pronoun in (1b) is explained by the variable status of finite clauses as phases in Russian. Whether finite CP is a phase depends on whether the φ-features on C are valued by the embedded subject; if C’s φ-features remain unvalued – as when the embedded subject is referentially dependent – Spell Out of the phase is postponed. By contrast, when C’s φ-features are valued by a referentially independent pronoun, embedded CP behaves as a phase. I argue, based on Reuland’s (2011) theory of binding economy, that a referentially independent pronoun can have a bound reading only if it is separated from its antecedent by a phase boundary. Section 2 addresses the interpretation of null and overt pronouns in Russian embedded clauses and relates the data to the phenomenon of finite control. Section 3 summarizes the proposal in Livitz (2014) for the link between pronominal silence and referential dependence. Section 4 addresses the availability of a referentially dependent interpretation for overt embedded subjects and the challenges

referential dependency in (1b) is not accidental coreference, since a bound-variable interpretation is clearly available, as demonstrated in Section 2.

Referential dependence across phase boundaries in Russian

| 55

it poses for standard approaches to binding theory. Section 5 presents an account of the alternation between overt and silent pronouns that relies on properties of phasal Spell Out. Section 6 concludes the paper.

2 The interpretation of null and overt subject pronouns As mentioned in the introduction, overt pronouns alternate freely with null pronouns in Russian finite embedded clauses (2), repeated from (1). Null and overt pronouns are not identical in interpretation, however. While a silent pronoun is obligatorily dependent in reference on a matrix antecedent (2a), an overt pronoun is free to refer to a different individual in the discourse (2b).² (2) a. Petjai skazal čto vesnoj 0i/∗j poedet v Pariž. will-go.3sg in Paris Petja said that spring ‘Petjai said that hei/∗j will go to Paris in the spring.’ b. Petjai skazal čto vesnoj oni/j poedet v Pariž. Petja said that spring he will-go.3sg in Paris ‘Petjai said that hei/j will go to Paris in the spring.’ This alternation is restricted to embedded clauses. As (3) illustrates, Russian is not a null-subject language and does not freely allow null subjects in matrix clauses. (3) Vesnoj *(on) poedet v Pariž. spring he.nom will-go.3sg in Paris ‘He will go to Paris in the spring.’ The alternation occurs with a range of embedding predicates, including think, promise, forget, and insist, as well as predicates that embed wh-clauses.

2 Note that the embedded verb agrees in person and number with the embedded subject regardless of whether the latter is silent or overt. Subject-verb agreement in Russian targets exclusively nominative nominals; therefore, the presence of morphological agreement on the embedded verb is consistent with the silent subject bearing nominative case like the overt subject.

56 | Inna Livitz (4) a. Petjai dumaet čto 0i/∗j / oni/j begaet bystree vsex. / he runs faster everyone.gen Petja thinks that ‘Petja thinks that he runs faster than everyone.’ b. Mamai obeščala čto 0i/∗j / onai/j kupit mne morožennoe. / she will-buy me.dat ice-cream mother promised that ‘Mother promised that she will buy me ice cream.’ c. Petjai zabyl čto 0i/∗j / oni/j uže včera zapravil mašinu. Petja forgot that / he already yesterday filled-up car ‘Petja forgot that he already got gas yesterday.’ d. Učeniki nastaival čto 0i/∗j / oni/j napisal sočinenie sam. student insisted that self / he write essay ‘The student insisted that he wrote the essay himself.’ e. Mašai ne pomnit kuda 0i/∗j / onai/j položila den’gi. Maša neg remembers where money / she put ‘Maša doesn’t remember where she put the money.’ Let us consider more closely the interpretation of the overt and silent pronouns in the above examples. Overt pronouns are referentially independent; i.e., they are capable referring to a salient individual in the discourse that does not c-command it. We have seen that they are also capable of being referentially dependent on a local c-commanding antecedent. This referential dependence cannot be an instance of accidental coreference, since the overt pronouns are capable of receiving a bound-variable interpretation, as demonstrated in (5) and (6). (5) [Ni odin bol’noj]i ne otrical čto 0i/∗j / oni/j inogda kurit. not one patient neg denied that / he sometimes smokes.3sg ‘No patient denied that he occasionally smokes.’ (6) Ktoi xvastalsja čto 0i/∗j / oni/j pobedit Petju? Who bragged that / he will-defeat.3sg Petja ‘Who bragged that he will defeat Petja?’ In contrast to overt pronouns, null pronouns in finite clauses have a similar interpretation to PRO in obligatory control constructions (7). (7) Anai hated PROi/∗j to spend Christmas alone. This similarity has been repeatedly noted in literature. The construction in (2a) is regularly referred to in the literature as finite control. The term was first introduced

Referential dependence across phase boundaries in Russian

| 57

in Landau (2004) to describe the silent PRO-like pronouns in Hebrew finite clauses and it has since been applied to silent referentially dependent pronouns in Brazilian Portuguese (Holmberg and Sheehan, 2010; Holmberg et al., 2009; Modesto, 2000, 2008, 2011; Rodrigues, 2004; Ferreira, 2009; Nunes, 2008; Boeckx et al., 2010), Finnish (Holmberg and Sheehan, 2010; Holmberg et al., 2009; Rodrigues, 2004), and Russian (Tsedryk, 2012, 2013) As argued in Tsedryk (2012, 2013) and Livitz (2014), null pronouns in Russian embedded finite clauses behave in many respects like instances of obligatory control into infinitives. First of all, the silent pronoun must be c-commanded by its antecedent. (8) [Doč’ prezidentai ]j ob”javila čto 0∗i/j vystupit s daughter president.gen announced that will-perform.3sg with dokladom. speech ‘The president’s daughter announced that she will give a speech.’ Second, the antecedent of the null subject in a finite control construction must be sufficiently local – it must be the subject of the next highest clause. A null subject is ungrammatical in (9) because the verb play has a masculine ending, which is incompatible with the closest possible antecedent, Maša. (9) * Petjai skazal čto Mašaj podumala v čto 0i ploxo igral Petja said that Maša thought.sg.fem that badly played.sg.masc in futbol. soccer ‘Petjai said that Mašaj thinks that hei plays soccer badly.’ Third, similar to subjects of obligatory control clauses, silent subjects in finite control constructions prohibit split antecedents. (10) * Ivani skazal ženej čto 0i+j prigotovjat pirog vmeste. Ivan.nom said.sg wife.dat that will-prepare.3pl pie together ‘Ivan said to his wife that they will prepare a pie together.’ (Tsedryk 2012: ex. 19)

58 | Inna Livitz

Tsedryk (2012, 2013) also argues that finite control, like obligatory control, allows only a sloppy reading for the embedded subject under ellipsis.³ (11) Ivani nadeetsja čto 0i vyigraet poedinok i Miša tože. and Miša also Ivan hopes that will-win duel ‘Ivan hopes to win the duel and so does Miša.’ (Tsedryk 2012: ex. 16) All of the above are well documented properties of PRO in obligatory control constructions. We can, therefore, conclude that null subjects in Russian finite clauses share the following crucial properties with PRO: they are null pronominals that must be referentially dependent on an antecedent in a designated syntactic position (the controller). In Section 3.5 I argue that the similarity in form and interpretation of these null pronouns to PRO is not accidental, but follows from their syntactic structure. To summarize, two types of elements can occur in the subject position of embedded finite clauses. The first type is an overt referential pronoun – it is capable both of independent reference and of referential dependence on the matrix subject. The second type is a null pronoun obligatorily referentially dependent on a c-commanding antecedent. The following section introduces the derivational theory of PRO’s silence developed in Livitz (2014), which extends to the null subjects in Russian finite control constructions.

3 A defective goal account of pronominal silence In Livitz (2014), I develop an account of the link between silence and referential dependence of PRO that extends to the null subject pronouns in finite control constructions. The proposal is built from three main components, which I discuss in turn below. First, I adopt an approach to pronominal syntax according to which pronouns may differ in structure and featural complexity, and I propose that pronouns that are obligatorily referentially dependent are featurally and structurally deficient compared to pronouns capable of independent reference. Second, I appeal to an account of control in which the reference of PRO has its source in the syntactic operation Agree. Third, I expand on Roberts’ (2010) notion of defective

3 The ellipsis data, however, are not unequivocal. See Livitz (2014) for a detailed discussion.

Referential dependence across phase boundaries in Russian

| 59

goal to argue that the operation Agree can result in the deletion of a referentially dependent pronoun at Spell Out. In this paper, I make several standard Minimalist assumptions about the nature of the syntactic derivation. I assume that lexical items consist of three independent sets of features: phonological, semantic, and syntactic. The syntactic, or formal, features drive the syntactic derivation until the point of Spell Out, after which the phonological and semantic information becomes relevant as the output of the narrow syntax is interpreted by PF and LF.

3.1 Background: pronominal structure Pronouns have often been analyzed as a syntactically heterogenous class, and a number of approaches link the interpretation of a pronoun to its underlying structure. Many theories of pronominal structure have proposed the equivalent of an optional D-layer as part of pronominal syntax (Ritter 1995; Cardinaletti and Starke 1999; Déchaine and Wiltschko 2002 a.o.). In (12) I have reproduced the structure from Déchaine and Wiltschko (2002), which I largely adopt, and according to which pronouns can be full DPs or smaller φPs. I depart from Déchaine and Wiltschko, however, in my assumptions about the interpretation of the pronominal D-layer.⁴ (12) [dp D [φP φ [np N ]]] While there is little consensus regarding the precise role of D in the interpretation of a pronoun, the presence of this layer is generally associated with more referential independence. For instance, Déchaine and Wiltschko argue that pronouns with a D-layer behave like R-expressions and cannot be bound. Alternatively, Cardinaletti and Starke (1999) propose that it is the D-layer (C-layer in their formulation) that allows a pronoun to introduce a new referent. Similarly, Holmberg (2005, 2010) argues that the ability to refer is linked to the referential index housed in the D-layer. According to Holmberg, pronouns that lack a D-layer can only be interpreted as impersonal. Building on Holmberg’s proposal, I assume that a D-layer is necessary for a pronoun to be capable of independent reference. I propose, however, that some D-less pronouns do not receive an impersonal interpretation. Specifically, PRO obligatorily receives a referentially dependent reading because it not only lacks

4 In this paper, I set aside the question of NP pronouns.

60 | Inna Livitz

a D-layer (i.e., it is a φP), but it also lacks φ-feature values, as has been proposed in several recent syntactic accounts of control (Landau, 2008; Sigurðsson, 2008). The values for these features (and thus the reference of PRO) are provided by the controller.

3.2 Background: control and the operation Agree Livitz (2014) builds on the proposal in (Landau, 2004, 2008) that obligatory control is derived via a series of Agree relations linking the controller with PRO and mediated by clausal functional heads. Under Landau’s proposal, PRO has unvalued φ-features, which must be valued via Agree over the course of the derivation. Matrix functional heads like T and v also have unvalued and uninterpretable φ-features. The φ-features on matrix T are valued by the controller DP, and T, in turn agrees with PRO and values its φ-features, thereby determining PRO’s reference. Landau’s derivation, which I adopt here for ease of exposition, is schematized in (13), with arrows representing the Agree relations underlying control. In Section 5.3 I revise Landau’s proposal to take into account the phasal status of vP and CP. (13) [T[uφ:_] … DP[iφ:3sg] [CP [TP PRO[iφ:_] T]]] 1 2 I additionally adopt the approach to formal features and the operation Agree from Pesetsky and Torrego (2007). Pesetsky and Torrego propose that formal features can vary along two distinct parameters – they can be interpretable or uninterpretable and they can be valued or unvalued. This creates a four-way distinction between uninterpretable/unvalued features, interpretable/unvalued features, uninterpretable/valued features, and interpretable/valued features. As is standard, I assume that that person, number, gender (φ-features ) are interpretable (though possibly unvalued) on nominal elements but uninterpretable on any clausal functional heads like v, T, and C. Uninterpretable and unvalued features must enter into an Agree relation with interpretable and valued instances of the same feature type in order to avoid a crash at Spell Out. Agree is triggered by an unvalued feature on a Probe, which searches for another instance of the same feature in its c-command domain. I adopt Pesetsky and Torrego’s (2007) feature-sharing formulation of the operation Agree, according to which Agree copies a feature from the Goal onto the Probe, resulting in the same feature being present in two separate locations. The feature can retain interpretable and uninterpretable instances following Agree,

Referential dependence across phase boundaries in Russian

|

61

but in the same way that copies of an element are actually the same element, features that have undergone Agree are the same occurrence of a feature. An important property of this theory of Agree is that two unvalued features can enter into an Agree relation, which is illustrated below in (14a). The more traditional case of an unvalued feature agreeing with a valued feature is schematized in (14b). The numbered indices next to each feature F reflect the basic idea of the feature-sharing approach to Agree; regardless of whether a given Agree relation is able to value F on the Probe, Agree results in the same occurrence of F in two distinct positions. (14) a. [uF:_][] [iF:_][5] → [uF:_][5] [iF:_][5] b. [uF:_][] [iF:α][5] → [uF:α][5] [iF:α][5]

3.3 Background: defective goals The final component of the proposal in Livitz (2014) is the notion of defective goal first introduced in Roberts (2010) and defined as follows: (15) Defective Goal: A Goal G is defective iff G’s formal features are a proper subset of those of G’s Probe. (Roberts 2010: 62) Roberts (2010) argues that what has been traditionally analyzed as head movement should be reanalyzed as a special case of Agree. Specifically, he points out that Agree between a Probe and a defective goal gives the appearance of movement. Consider the following schematized derivation. There is a lexical item F1 that consists exclusively of the feature bundle F1 . This lexical item is the Goal for a Probe X that has features F1 and F2 . When X enters into an Agree relation with F1 , F1 is a defective goal because its features are a subset of the features of X. After the features on the Probe have been valued, the resulting structure contains two copies of F1 : one on F1 itself, and one on the Probe. This configuration is identical to head movement where F1 has adjoined to X. (16) Agree with a defective goal XP X [F1 : _] [F2 :β]

YP Y

F1 [F1 : α]

XP X [F1 : α] [F2 :β]

YP Y

F1 [F1 : α]

62 | Inna Livitz

While Roberts uses the notion of defective goal to account for head movement, in Livitz (2014) I extend the intuition behind Roberts’ proposal and, similar to Holmberg (2010), apply the notion of a defective goal to account for the silence of certain pronouns. In (16), at Spell Out only one copy of the lexical item F1 can be interpreted by PF, following the reasoning in Nunes (2004). Because the entire content of F1 is contained in X, there are, in fact, two copies of F1 in (16). Therefore, the lower copy of F1 must be deleted before PF, in the same way that the lower copy is deleted for an element that has undergone Internal Merge (17). (17) Deletion of lower copy XP X [F1 : α] [F2 :β]

YP Y

F1 [F1 : α]

3.4 PRO as a defective goal Livitz (2014) argues that PRO in obligatory control constructions is null crosslinguistically because it is necessarily a defective goal in an obligatory Agree relation. As discussed in Section 3.1, I posit that since PRO is incapable of independent reference, it must lack a DP layer. Additionally, PRO’s referential dependence reflects a lack of φ-feature values. PRO, therefore, is a bare φP that consists exclusively of unvalued (though interpretable) φ-features. Having unvalued φ-features means that PRO must enter into an Agree relation (or series of Agree relations) that value those features. PRO acquires its reference as a result of an Agree chain that links it with the controller. However, its minimal feature content ensures that PRO is a defective goal with respect to any head that values its φ-features, since the φ-feature bundle that makes up the entire content of PRO will necessarily be a subset of the φ-feature bundle that is contained on the head that probes PRO. Therefore, PRO must be deleted at PF, as schematized below. Note that PRO’s status as a defective goal ensures its deletion at PF only; since its φ-features are interpretable, PRO is interpreted as an argument of the lower clause at LF. (18) 1[T, uφ:_[]] [vP [dp D, iφ:3SG[]] v [CP [TP [iφ:_[]]T…]]]→ 2[T, uφ:3SG[5]] [vP [dp D, iφ:3SG[5]] v [CP [TP [iφ:_[]]T…]]]→ 3[T, uφ:3SG[5]] [vP [dp D, iφ:3SG[5]] v [CP [TP [iφ:3SG[5]]T…]]]→ 4[T, uφ:3SG[5]] [vP [dp D, iφ:3SG[5]] v [CP [TP [iφ:3SG[5]]T…]]]→

Referential dependence across phase boundaries in Russian

| 63

To summarize, PRO is silent because it must enter into an Agree relation to value its φ-features, but it is a defective goal whenever it does so.

3.5 A defective goal analysis of finite control The defective goal proposal extends straightforwardly to Russian finite control constructions. I preview the analysis here and provide more details in Section 5, following a discussion of bound overt pronouns in Section 4. Null referentially dependent pronouns in Russian embedded finite clauses are bare φPs with unvalued φ-features, like PRO in obligatory control constructions. A series of Agree relations links the embedded subject with the matrix subject. Because of its minimal content, the embedded subject will be a defective goal in any Agree relation for φ-features and will, therefore, be deleted at Spell Out. By contrast, pronouns capable of independent reference are full DPs, as discussed in Section 3.1. This means that these pronouns will not be defective goals in an Agree relation for φ-features, since the probing functional heads will lack a D-feature. As a result, pronouns capable of independent reference will surface overtly. To summarize, null and overt subject pronouns in embedded clauses in Russian differ in their featural composition. Subject pronouns that are incapable of independent reference have minimal structure and featural content, and are, therefore, deleted as defective goals. By contrast referential pronouns have more structure and are not defective goals in an Agree relation for φ-features.

4 The referential dependence of overt subjects I have thus far accounted for the link between the form and interpretation of pronouns in finite control constructions; the notion of defective goal explains why PRO-like pronouns are null, while pronouns capable of independent reference are overt. However, a referentially independent interpretation is not the only one available to an overt pronoun. As mentioned earlier, an overt pronominal subject in Russian embedded finite clauses can optionally be interpreted as referentially dependent on the matrix subject in (19) and (20).

64 | Inna Livitz

(19) Petjai skazal čto vesnoj oni/j poedet v Pariž. Petja said that spring he will-go.3sg in Paris ‘Petjai said that hei/j will go to Paris in the spring.’ (20) [Ni odin bol’noj]i ne otrical čto oni/j inogda kurit. not one patient neg denied that he sometimes smokes.3sg ‘[No patient]i denied that hei/j occasionally smokes.’ The availability of a referentially dependent reading for the overt pronouns in (19) and (20) is unexpected. To appreciate why, we need to consider a well-established empirical generalization regarding the interpretation of pronouns and anaphors that occur in the same syntactic context. A pronoun cannot refer to an antecedent if the pronoun alternates with an anaphor obligatorily bound by that antecedent. Grodzinsky and Reinhart (1993) offer one formalization of this generalization. (21) Rule I: NP A cannot corefer with NP B if replacing A with X, X a variable A-bound by B, yields an indistinguishable interpretation. (Grodzinsky and Reinhart 1993: 79) Safir (2004) proposes a more general principle governing referential dependence that subsumes Rule I in (21). According to Safir, in any given context, referential dependence must be expressed with the most referentially dependent element. (22) Form to Interpretation Principle: If x c-commands y, and z is not the most dependent form available in position y with respect to x, then y cannot be directly dependent on x. (Safir 2004: 50) Safir considers the relative referential dependence of a number of different elements, but in the present discussion we need to compare the null and overt pronouns that occur in embedded finite clauses in Russian. The null pronoun is more referentially dependent than the overt pronoun because it is obligatorily bound by the matrix subject, whereas the overt pronoun is freer in reference with its ability to refer to a salient individual in the discourse. Therefore, we would predict – contrary to fact – that the overt pronoun should not be able to refer to the matrix subject, since the same syntactic environment allows a pronominal form dedicated to expressing referential dependence, namely 0. Note, by contrast, that the referential dependence of an overt embedded subject in the parallel English construction in (23) is completely expected given

Referential dependence across phase boundaries in Russian

| 65

the economy conditions in (21) and (22) because English lacks a more dependent pronominal form that could occur in subject position of a finite clause. Since he is the only pronoun possible in embedded subject position in (23), it can refer to the matrix subject. (23) Johni /every boyi said [that hei/j will leave for Paris on Sunday].

4.1 A different approach to pronoun-anaphor alternations A recent reformulation of binding theory in derivational terms (Reuland, 2001, 2006, 2011) is able to capture the crosslinguistic parallelism in the interpretation of overt pronouns in Russian and English, namely that in both languages overt subject pronouns in embedded finite clauses can be referentially dependent on the matrix subject, in addition to being capable of selecting a different referent from the discourse. Reuland proposes a novel theory of economy conditions on binding that compares not types of elements but ways of establishing reference in the grammar. One way – tangential to the present discussion – is via the discourse. Another way is via variable binding at LF; pronouns can be interpreted as variables in the semantic component of the grammar. As is standardly assumed in the literature, this is precisely how referential dependence is achieved in (23) above. Finally, reference can be established in the narrow syntax via the operation Agree. Similar to Landau’s (2004, 2008) Agree-based approach to control, Reuland proposes that referential dependence is established by feature-checking/ Agree between a pronoun/anaphor deficient in its φ-feature specifications and its antecedent, mediated by functional heads (24).⁵ (24) DP T v pronoun Reuland’s economy conditions ensure competition between variable binding at LF and deriving binding via Agree. Reuland argues that referential dependence must be derived syntactically if such a derivation is possible. If an Agree chain connecting the antecedent and the referentially dependent element is impossi-

5 Note that binding at LF is constrained by the syntax (for instance, the antecedent of a bound pronoun must c-command it). However, interpreting a pronoun as a variable at LF is distinct from the derivation of binding in the narrow syntax – in the latter case, it is the Agree operations themselves that link a pronoun to its antecedent via a syntactic chain, and, therefore, the antecedent-pronoun relationship is subject to stricter locality conditions.

66 | Inna Livitz

ble – as is presumably the case for (23) – then, and only then, can the pronoun be interpreted as a variable at LF. In certain cases, however, the derivation in (24) fails not because of restrictions on Agree, but because of the properties of the referentially dependent element; in these instances economy conditions will prohibit variable binding at LF. Reuland proposes an additional restriction on elements that acquire their reference via Agree – they have to satisfy the Chain Condition (25). (25) Chain Condition: the tail of an Agree-chain cannot be fully specified for φfeatures. (Reuland, 2011) Pronouns (as opposed to anaphors), according to Reuland, are fully specified for φ-features and thus cannot occur in the tail of an Agree chain that connects the pronoun to its antecedent. This means, for instance, that the pronoun him in the ECM construction in (26) cannot be bound by the matrix subject because it is not a suitable tail for an Agree chain. (26) Johni [vP expects [him∗i/j to win]]. Since an Agree chain linking the antecedent and embedded subject position is technically possible in (26) (i.e., nothing blocks the Agree relation), referential dependence via variable binding is blocked by Reuland’s economy conditions. Therefore, the embedded subject cannot refer to the matrix subject – there is no way to derive this referential dependency.

4.2 The role of phases While Reuland does not specify what prevents a syntactic derivation of binding in a sentence like (23) – thereby enabling binding at LF – he does suggest that locality is a key component. I propose that phases play a crucial role in blocking a syntactic binding relation in the case of finite embedded clauses. I make the standard assumption that a syntactic derivation proceeds in cycles (phases). At the end of each phase the complement of the phase head is transferred to semantics and phonology for interpretation (Chomsky, 2000, 2001, 2004, 2008). Material that has been transferred is inaccessible to subsequent computations in the narrow syntax, as per the Phase Impenetrability Condition (PIC).

Referential dependence across phase boundaries in Russian

|

67

(27) Phase Impenetrability Condition: after a phase has been spelled out it is no longer accessible to the computation The effect of the PIC is that only a phase head and its specifier are accessible to higher Probes once a phase is complete. I assume that finite CP and transitive v*P are phases. The PIC makes the subject in Spec,TP of an embedded finite clause inaccessible to Probes from outside the clause, which means that an Agree relation between a matrix functional head and the embedded subject is impossible. Returning to Russian embedded subjects, under Reuland’s formulation, a silent, referentially dependent pronoun is a possible tail for an Agree-chain because it lacks φ-feature values. By contrast, an overt, referential pronoun cannot occur as the tail of an Agree chain. Since a bound-variable interpretation is available to the overt pronoun in Russian embedded clauses, it must be the case that no Agree chain can link the embedded subject position and the matrix subject. This follows straightforwardly from the status of finite CP as a phase. Reuland’s economy conditions on binding capture the similarity between the Russian (20) and the English (23); in both cases, the overt pronoun is separated from its antecedent by a phase boundary, making a referential dependency possible.⁶ We have now arrived at a contradiction, however. If no Agree-chain is possible linking the matrix and embedded subjects in Russian, how can we derive control into finite clauses which crucially relies on the existence of such an Agree chain? In the following section I appeal to a theory of phase extension to account for the seemingly contradictory behavior of the embedded CP.

5 Accounting for the alternation between null and overt pronouns Recall the original puzzle presented in the introduction. Subjects of embedded finite clauses can be null or overt, but null pronouns are more restricted in their reference than overt pronouns. Null pronouns can refer only to the matrix subject, while overt pronouns can refer to the matrix subject or to another individual in the discourse.

6 By contrast, English ECM clauses are not phases.

68 | Inna Livitz (28) a. Petjai skazal čto vesnoj 0i/∗j poedet v Pariž. will-go.3sg in Paris Petja said that spring ‘Petjai said that hei/∗j will go to Paris in the spring.’ b. Petjai skazal čto vesnoj oni/j poedet v Pariž. Petja said that spring he.nom will-go.3sg in Paris ‘Petjai said that hei/j will go to Paris in the spring.’ So far I have provided an account of the restriction on the reference of the null pronoun. Additionally, I have proposed an explanation for the surprising availability of a referentially dependent reading for the overt pronoun by appealing to properties of phasal Spell Out. However, the two explanations appear to be in conflict: on the one hand, a finite CP in Russian cannot be a phase given the existence of finite control, but, on the other hand, it must be a phase given the possibility of a bound-variable reading for an overt embedded subject. In this section I appeal to a theory of flexible phase boundaries in order to account for this seemingly contradictory conclusion.

5.1 Flexible phase boundaries It has been proposed in the literature that phase boundaries can be shifted either through movement of the phase head (den Dikken, 2007) or through agreement of the phase head with a higher head (Rackowski and Richards, 2005). Recent work by Wurmbrand (2013) and Wurmbrand and Bobaljik (2013) argues that a phase is voided or suspended if the phase head depends on a higher head for a feature value. Wurmbrand is aiming to account for an asymmetry in the acceptability of Quantifier Raising (QR) that depends on clause type. QR in English is possible out of subjunctive clauses (29b) and control clauses (29c), but not out of raising clauses (29d) or finite clauses (29a). (29) (Bobaljik and Wurmbrand 2013: 189) a. # Someone said that Sue is married to every man. *∀ ≫ ∃ (finite) b. Sue has requested that they read only Aspects. only ≫ request (subjunctive) c. A different student decided to report on every article. ∀ ≫ ∃ (control) d. # This soldier seems to someone to be likely to die in every battle. *∀ ≫ ∃ (raising)

Referential dependence across phase boundaries in Russian

|

69

Wurmbrand argues that the asymmetry stems from the properties of the phase head associated with the embedded clause. She posits that all embedded clauses are, in principle, phases, but features on the highest head of the clause determine whether the phase will be spelled out once the clause is complete or whether Spell Out will be postponed until the next phase. She argues that subjunctive and control clauses are not treated like phases English, but finite clauses and raising clauses are.⁷ Wurmbrand proposes that the difference between control clauses and finite clauses is that the highest head in a control clause has an unvalued tense or mood feature. That value is later provided by the selecting matrix head, which has an uninterpretable, but valued occurrence of the same feature. It is this featural dependency of the phase head on a higher head that has the effect of extending the phase associated with the embedded clause until the next phase is complete. I propose that finite control across the CP phase boundary in Russian is made possible by a similar featural dependency between finite C and a matrix functional head, though the dependency is for φ-features rather than for tense or mood features.

5.2 The roles of C and the embedded subject I propose that the alternation between overt and silent subjects in finite embedded clauses is explained by the variable status of finite CP as a phase in Russian. In finite clauses with overt subjects, the embedded CP is a phase; in instances of finite control, embedded CP is not a phase. It is the interaction of features on C with features of the embedded subject that determine whether or not the embedded clause is a phase. I propose that finite C enters the derivation with uninterpretable, unvalued φ-features, which must be valued and deleted before the phase can be spelled out. It is precisely the presence of unvalued features on C that provides the possibility of extending the phase in finite control constructions. If the features on C are not valued within the CP phase, C will be required to enter into an Agree relation with a higher functional head to have those features valued. This dependency on a head outside the CP phase for feature values will postpone Spell Out of the CP phase, in the same way that a phase is extended in the examples discussed by Wurmbrand. The difference from Wurmbrand’s examples is that the unvalued features on finite C in Russian are φ-features, not tense or mood features. Since the 7 Combined with the notion of Scope Economy (Fox, 2000), the phasal status of finite and raising clauses explains why the quantifiers cannot raise out of the embedded clause.

70 | Inna Livitz

unvalued features on C are φ-features, it is the embedded subject that ultimately determines whether or not the embedded CP is a phase. Recall from Section 3.1 that while obligatorily referentially dependent pronouns have unvalued φ-features, the φ-features on pronouns capable of independent reference are valued. An Agree relation between finite C and a referentially independent embedded subject results in the valuation of C’s φ-features. Therefore, when the CP phase is complete, there are no unvalued φ-features on C that would necessitate the extension of the CP phase. Instead, the complement of C will be spelled out, and the elements inside it, including the embedded subject in Spec, TP, will be inaccessible to further probes from the matrix clause. That embedded C enters the derivation with uninterpretable features has similarly been proposed by Landau (2004, 2008) for certain types of infinitival clauses. Landau argues that uninterpretable φ-features on infinitival C occur in those infinitival clauses that have dative subjects in Russian (and which form a subset of all Russian infinitival clauses). According to Landau, when infinitival C enters the derivation with uninterpretable φ-features, it acts as a link in the Agree chain underlying control.⁸ Landau posits that one instance of control that involves φ-features on infinitival C is control into infinitival questions. It turns out that infinitival questions also exhibit an alternation between overt and null pronouns identical to the one found in finite clauses. In (30) a silent embedded subject must refer to the matrix subject Petja, while an overt embedded subject can additionally refer to a different salient individual in the discourse.⁹ (30) a. Petjai ne znaet čto 0i/∗j skazat’ na rabote. say.inf on work Petja neg know what ‘Petjai doesn’t know what hei/∗j should say at work.’ b. Petjai ne znaet čto emui/j skazat’ na rabote. Petja neg know what he.dat say.inf on work ‘Petjai doesn’t know what hei/j should say at work.’

8 I am not proposing that all clauses have uninterpretable features on C; for instance, I follow Landau in assuming that many infinitival clauses lack an active C. It is telling that many of the infinitival clauses for which Landau proposes the involvement of C have an active left periphery – they contain wh-words or complementizers, as in (30). 9 Note that dative case on the overt subject is standard for infinitival questions in Russian – the same subject would surface in nominative case in a finite clause. The case distinction between finite and infinitival clauses is largely irrelevant for the purposes of this article.

Referential dependence across phase boundaries in Russian

|

71

5.3 Deriving the alternation between null and overt pronouns in Russian finite clauses This section details the derivation of the alternation between overt and silent pronouns in Russian embedded finite clauses. Before turning to the alternation itself, it is important to introduce a slight modification to the previously assumed Agree relations underlying control. I assume that finite C and v* (a v that introduces an external argument) are phase heads. I propose, therefore, that v, like embedded C, has uninterpretable φ-features. If v’s φ-features remain unvalued after agreeing with C, it must probe upwards into its specifier to agree with the external argument (along the lines of Rezac (2004)). As a result, when the vP is complete, the features on embedded C, the embedded pronoun, and matrix v are valued by the controller in Spec, vP (31). (31) [vP DP[iφ:3SG] [v v[uφ:_] [CP C[uφ:_] [TP PRO[iφ:_] T]] 3

2

1

The derivation that results in a null referentially dependent embedded subject proceeds as follows. The embedded subject is a bare φP with unvalued, but interpretable, φ-features, which must be valued over the course of the derivation. The embedded subject agrees with embedded T and raises to Spec, TP, like any other subject. There, it is probed by embedded C which has uninterpretable, unvalued φ-features (33). (32) Petjai skazal čto 0i/∗j poedet v Pariž. will-go.3sg in Paris Petja said that ‘Petja said that he will go to Paris in the spring.’ (33)

CP

C čto that [uφ:_][ ]

TP

φ i [iφ:_][5] T [fut] [uφ:_][5]

vP ti poedet v Pariž t i will go to Paris

72 | Inna Livitz

In this instance, φP cannot value C’s uninterpretable φ-features, which means that when CP is complete, it still has unvalued φ-features, extending the phase. TP is not spelled out when the CP is complete because C depends on a higher head for its φ-feature values. When embedded C cannot value v’s φ-features, v will Probe its specifier once the external argument is merged there. When the complement of v is subsequently spelled out, the embedded subject will be deleted because it is a defective goal (34). Since the φ-features on the embedded subject are interpretable, it will be transferred to LF and interpreted as an argument of the embedded predicate. (34)

vP

DP Petja [iφ:3sg][ ]

v skazalj said [uφ:3sg][5]

VP

CP tj C čto that [uφ:3sg][5] φ i [iφ:3sg][5]

TP

T [fut] [uφ:3sg][5]

vP ti poedet v Pariž t i will go to Paris

By contrast, if the embedded subject is a referentially independent pronoun, it will be a full DP with valued φ-features. An Agree relation with the embedded subject will value the φ-features on C, and therefore the complement of C will be spelled out when CP is complete (35). Moreover, since the subject pronoun in (35) is a full DP, it will not be a defective goal relative to embedded C and will surface overtly. Since the PIC precludes the possibility of an Agree chain between v and the embedded subject, Reuland’s theory of binding economy predicts that the embedded subject should be able to receive a bound-variable reading in the semantics.

Referential dependence across phase boundaries in Russian

| 73

(35) v

VP

CP

V skazal said

C čto that [uφ:3SG][5]

TP

DPi on he [iφ:3SG][5] [D]

T [fut] [uφ:3sg][5]

vP ti poedet v Pariž t i will go to Paris

There is some tentative independent evidence that the embedded CP is not a phase precisely when the embedded subject is silent. Scrambling and extraction out of finite clauses is generally difficult in Russian (36). Using a null subject in the embedded clauses makes extraction possible for a set of speakers. (36) Q:

A počemu Peti net? and why Petja.gen neg ‘And why is Petja not here?’ A: ?* Petjai ZAVTRA skazal čto oni/j zajdet. Petja tomorrow said that he.nom drop-by.3sg ’Petjai said that hei/j will drop by TOMORROW.’ A: ? Petjai ZAVTRA skazal čto 0i zajdet. Petja tomorrow said that drop-by.3sg ’Petjai said that hei will drop by TOMORROW.’

Similarly, negative concord does not apply across a finite clause boundary, and scrambling nothing into the matrix clause is impossible (37). When the embedded subject is null, it becomes marginally possible to scramble the negative embedded object into the matrix clause and to license negative concord there.

74 | Inna Livitz

(37) A

Ja dumal Petja skazal čto on prineset xotja-by vino. I thought Petja said that he will-bring at least wine ‘I thought Petja said that he would at least bring the wine.’ B ?* Net, Petjai ničegok ne skazal čto oni/j prineset tk . no, Petja nothing neg said that he.nom will-bring.3sg ‘No, Petjai didn’t say that hei/j would bring anything.’ B ? Net, Petjai ničegok ne skazal čto 0i prineset tk . no, Petja nothing neg said that will-bring.3sg ‘No, Petjai didn’t say that hei would bring anything.’

6 Conclusion This paper dealt with an alternation between null and overt pronouns in Russian embedded clauses, where null pronouns are obligatorily referentially dependent on the matrix subject, while overt pronouns are free in reference. The aim of the paper was to account both for the link between the form and interpretation of these pronouns and for the unexpected availability of a bound-variable reading in an environment where a more referentially dependent form is available. I proposed that the distinct interpretations of overt and null pronouns stem from differences in their syntax; null pronouns are bare φPs with unvalued φ-features, while overt pronouns are full DPs with valued φ-features. The silence of a referentially dependent pronoun follows from its status as a defective goal in any Agree relation for φ-features, as argued in Livitz (2014). I argued that Russian finite clauses allow the alternation between these two types of pronouns because of CP’s flexible status as a phase. More specifically, I proposed that embedded C enters the derivation with uninterpretable φ-features, and the Agree relations that it enters into determine whether or not the CP is a phase. An embedded subject with unvalued φ-features cannot value C’s φ-features, which delays Spell Out of the embedded clause until the vP phase and allows a control relation to span the finite clause boundary. By contrast, if the embedded subject is referentially independent, it will value the φ-features on C, which results in the embedded clause being spelled out once CP is complete. Following the reasoning in Reuland (2011), it is precisely this phase boundary that makes a bound-variable reading possible for the overt pronoun. Finally, we are lead to an intriguing conclusion about phasal Spell Out – whether or not a particular phrase is a phase depends not only on elements outside the phase, but also on elements within the phase itself.

Referential dependence across phase boundaries in Russian

| 75

References Boeckx, C., Hornstein, N., and Nunes, J. 2010. Control as Movement. Cambridge: Cambridge University Press. Cardinaletti, A. and Starke, M. 1999. The typology of structural deficiency. In van Riemsdijk, H., editor, Clitics in the Languages of Europe, pages 145–233. Berlin: Mouton de Gruyter. Chomsky, N. 2000. Minimalist inquiries: The framework. In Martin, R., Michaels, D., and Uriagereka, J., editors, Step by Step: Essays in Honor of Howard Lasnik, pages 89–155. Cambridge, MA: MIT Press. Chomsky, N. 2001. Derivation by phase. In Ken Hale: A Life in Language, pages 1–52. Cambridge, MA: MIT Press. Chomsky, N. 2004. Beyond explanatory adequacy. In Belletti, A., editor, Structures and Beyond: The Cartography of Syntactic Structures, vol.3, pages 104–131. Oxford: Oxford University Press. Chomsky, N. 2008. On phases. In Freidin, R., Otero, C. P., and Zubizarreta, M. L., editors, Foundational Issues in Linguistic Theory: Essays in Honor of Jean-Roger Vergnaud, pages 133–166. Cambridge, MA: MIT Press. Déchaine, R.-M. and Wiltschko, M. 2002. Decomposing pronouns. Linguistic Inquiry, 33(3):409–442. den Dikken, M. 2007. Phase extension: Contours of a theory of the role of head movement in phrasal extraction. Theoretical Linguistics, 33(1):1–41. Ferreira, M. 2009. Null subjects and finite control in Brazilian Portuguese. In Nunes, J., editor, Minimalist Essays on Brazilian Portuguese Syntax. Amsterdam: John Benjamins Publishing Company. Fox, D. 2000. Economy and Semantic Interpretation. Cambridge, MA: MIT Press. Grodzinsky, Y. and Reinhart, T. 1993. The innateness of binding and coreference. Linguistic Inquiry, 24(1):69–101. Holmberg, A. 2005. Is there a little pro? Evidence from Finnish. Linguistic Inquiry, 36(4):533– 564. Holmberg, A. 2010. Null subject parameters. In Biberauer, T., Holmberg, A., Roberts, I., and Sheehan, M., editors, Parametric Variation: Null Subjects in Minimalist Theory, pages 88–124. Cambridge: Cambridge University Press. Holmberg, A., Nayudu, A., and Sheehan, M. 2009. Three partial null-subject languages: a comparison of Brazilian Portuguese, Finnish and Marathi. Studia Linguistica, 63(1):59–97. Holmberg, A. and Sheehan, M. 2010. Control into finite clauses in partial null-subject languages. In Biberauer, T., Holmberg, A., Roberts, I., and Sheehan, M., editors, Parametric variation: Null Subjects in Minimalist Theory, pages 125–152. Cambridge University Press. Landau, I. 2004. The scale of finiteness and the calculus of control. Natural Language & Linguistic Theory, 22(4):811–877. Landau, I. 2008. Two routes of control: Evidence from case transmission in Russian. Natural Language & Linguistic Theory, 26(4):877–924. Livitz, I. 2014. Deriving Silence through Dependent Reference: Focus on Pronouns. PhD thesis, New York University. Modesto, M. 2000. On the Identification of Null Arguments. PhD thesis, USC, Los Angeles. Modesto, M. 2008. Topic prominence and null subjects. In Biberauer, T., editor, The Limits of Syntactic Variation, pages 375–410. Amsterdam: John Benjamins Publishing Company.

76 | Inna Livitz

Modesto, M. 2011. Finite control: where movement goes wrong in Brazilian Portuguese. Journal of Portuguese Linguistics, 10(2):3–30. Nunes, J. 2008. Inherent case as a licensing condition for A-movement: the case of hyperraising constructions in Brazilian Portuguese. Journal of Portuguese Linguistics, 7:83–108. Pesetsky, D. and Torrego, E. 2007. The syntax of valuation and the interpretability of features. In Karimi, S., Samiian, V., and Wilkins, W. K., editors, Phrasal and Clausal Architecture: Syntactic Derivation and Interpretation, pages 262–294. Amsterdam: John Benjamins. Rackowski, A. and Richards, N. 2005. Phase edge and extraction: a Tagalog case study. Linguistic Inquiry, 36:565–599. Reuland, E. 2001. Primitives of binding. Linguistic Inquiry, 32:439–492. Reuland, E. 2006. Agreeing to bind. In Broekhuis, H., Corver, N., Huybregts, R., Kleinhenz, U., and Koster, J., editors, Organizing Grammar: Linguistic Studies in Honor of Henk van Riemsdijk, pages 505–513. Berlin: Mouton de Gruyter. Reuland, E. 2011. Anaphora and Language Design. Cambridge, MA: MIT Press. Rezac, M. 2004. Elements of Cyclic Syntax: Agree and Merge. PhD thesis, University of Toronto. Ritter, E. 1995. On the syntactic category of pronouns and agreement. Natural Language & Linguistic Theory, 13:405–443. Rodrigues, C. 2004. Impoverished Morphology and A-Movement out of Case Domains. PhD thesis, University of Maryland, College Park. Safir, K. 2004. The Syntax of Anaphora. Oxford: Oxford University Press. Sigurðsson, H. Á. 2008. The case of PRO. Natural Language & Linguistic Theory, 26:403–450. Tsedryk, E. 2012. Finite control in Russian as movement: Subject extraction, feature deficiency, and parametric variation. Ms. Saint Mary’s University, Halifax. Tsedryk, E. 2013. Finite control in Brazilian Portuguese and Russian: A’-binding or A-movement. Ms. Saint Mary’s University, Halifax. Wurmbrand, S. 2013. QR and selection: Covert evidence for phasehood. In Keine, S. and Slogget, S., editors, North-Eastern Linguistic Society (NELS) 42, pages 277–280. Amherst: University of Massachusetts, Graduate Linguistic Student Association. Wurmbrand, S. and Bobaljik, J. 2013. Suspension across domains. In Matushansky, O. and Marantz, A., editors, Distributed Morphology Today: Morphemes for Morris, pages 185– 198. Cambridge, MA: MIT Press.

Sandhya Sundaresan (University of Leipzig)

Anaphora vs. agreement A new kind of Anaphor Agreement Effect in Tamil Abstract: This paper presents new data pertaining to the Anaphor Agreement Effect (the observation, originally noted in Rizzi, 1990, that anaphors in many languages seem incapable of triggering ϕ-covarying agreement) from the Dravidian language, Tamil. On the one hand, this data will be seen to further support the AAE as a robust crosslinguistic generalization. On the other hand, it will be shown to yield new insight on the theoretical principles underlying this descriptive one, and to question the possible loci for parametric variation – by virtue of employing a hitherto unreported strategy to obey the AAE. Specifically, it will be argued that the verbal agreement triggered in the scope of the anaphor is triggered, not by the anaphor itself, but by a different DP in the local phase. Keywords: Anaphora, agreement, Anaphor Agreement Effect, ϕ-features, reference, Dravidian, Tamil, syntax, semantics, morphology.

1 Introduction It has long been noted in the literature (Borer, 1989, a.o.) that, despite their obvious categorial differences, pro-forms (i.e. pronouns and anaphors) and ϕagreement have much in common: at their core, they both denote referentially deficient grammatical objects. Interestingly (though perhaps unsurprisingly), interactions between the two are often anomalous in some way. The relationship between pronouns and agreement has been well studied with respect to the phenomenon of pro-drop, the original observation, due to Taraldsen (1978) and simplifying greatly, being that languages with rich agreement allow pro-drop to a greater degree than do ones with poor agreement. The interaction between anaphora and agreement has merited less attention in the literature. The data that we do have amassed suggests, however, that there is something irregular happening here as well. This irregularity, termed the “Anaphor Agreement Effect” in Rizzi (1990) and developed since (Woolford, 1999; Tucker, 2011, among others), is the observation that, overwhelmingly across languages, anaphors cannot trigger “regular” (i.e. ϕ-covarying) agreement. Languages have been observed to display a range of interesting strategies to avoid a violation of the AAE. When an anaphor

78 | Sandhya Sundaresan

does occur in agreement-position, the sentence either crashes, or the triggered agreement takes on one of two non-covarying forms: a morphophonological default or “special” anaphoric form. The various parametrized instantiations of the AAE are significant and useful, not only because they stand testimony to the presence of a highly irregular relationship between two linguistic objects, but also because they serve as a potential window into the ϕ-featural make-up of anaphors and into the mechanism of agreement. In this paper, I report on a new type of AAE in the Dravidian language, Tamil, a language that has, in fact, been singled out in the literature (Kayne, 1994; Woolford, 1999; Selvanathan and Kim, 2008) as being a possible counterexample to the AAE. The theoretical import of the evidence presented in this paper is multifold. First, it shows that Tamil does indeed obey the AAE, contra previous claims. Second, and relatedly, it bolsters the robustness of the AAE as a potentially universal principle of language. Third, it adds to the typological variety of parametrized strategies languages have been observed to employ to avoid a violation of the AAE – since it will be shown that Tamil employs a hitherto unattested method to do so. Toward the end of the paper, I speculate on the possible theoretical underpinnings of the AAE as a potential principle of UG.

2 A brief history of the AAE The original observation that led to the formulation of the Anaphor Agreement Effect, as well as the moniker itself, go back to Rizzi (1990). Rizzi’s observation was motivated by minimal pairs like (1) and (2) in Italian, and analogous sentences in Icelandic (not listed here), where the anaphor occurs in a position typically associated with agreement in that language: (1)

A loro import-a solo di se-stessi. to them matters-3sg only of them-selves “Theyi only matter to themselvesi .”

(2) * A loro interess-ano solo se-stessi. to them interest-3pl only them-selves.nom “Theyi only interest themselvesi .” (Intended) The theme object of importa in (1) is in the genitive case, which in turn results in the verb’s surfacing with default 3sg agreement. In (2), the object is in the nominative case and does trigger ϕ-covarying (3pl) agreement on the verb. What is signif-

A new kind of Anaphor Agreement Effect |

79

icant is that this distinction seems to directly regulate the grammaticality of these sentences. The genitive-marked anaphoric object in (1) is licit, whereas the nominative marked one in (2) is not. However, sentences like (2) become marginally acceptable if the agreement on the verb is replaced with default (non-)agreement: (3) ? A loro interess-a solo se-stessi. to them interest-3sg only them-selves.nom “Theyi only interest themselvesi .” Additionally, the same patterns as in (1) and (2) obtain if the 3rd-person se is replaced with 2nd-person voi, yielding a bound 2nd-person form: (4)

A voi import-a solo di voi-stessi. to you matters-3sg only of you-selves “Youi only matter to yourselvesi .”

(5) * A voi interess-ate solo voi-stessi. to you interest-2pl only you-selves.nom “Only yourselves interest you.” (Intended) In addition to corroborating the patterns in (1) and (2), the sentences in (3)–(5) show that the problem with the ungrammatical sentences above has to to with agreement, not with some paradigmatic gap having to do with the absence of nominative anaphors in the morphology (as has been claimed for Icelandic, for instance, by Maling, 1984, among others). On the strength of such data, Rizzi postulated the existence of an Anaphor Agreement Effect, formulated as below: (6) “[T]here is a fundamental incompatibility between the property of being an anaphor and the property of being construed with agreement.” (Rizzi, 1990, 28)

2.1 Parametric strategies to avoid an AAE-violation Subsequent analyses (e.g. Woolford, 1999; Haegeman, 2004; Deal, 2010; Tucker, 2011) have tested the validity of the AAE against a wider range of languages. Woolford (1999) extends the investigation to languages with object agreement, predicting that the existence of the AAE, defined as in (6), in these languages should rule out anaphors in object position – a prediction that is indeed borne

80 | Sandhya Sundaresan

out, for the languages tested. In addition to confirming the systematicity of the AAE, this showed that the occurrence of the AAE has nothing to do with properties that distinguished subjects from objects: e.g. due to the absence of nominativemarked anaphors in subject position, the EPP, or licensing by the ECP. In other words, what matters for the AAE in subject-agreement languages is not that the agreement-controller in such cases is a subject, but that it is an anaphor. Tucker (2011) expands on Woolford’s work, discussing in particular the various parametric strategies employed by the world’s languages to avoid a violation of the AAE. What emerges is an interesting typology that reflects both the robustness of the AAE as a general crosslinguistic phenomenon and the scope of variation in the language-specific strategies used to avoid its violation. I discuss these findings in brief below.

2.1.1 Detransitivization The predicate detransitivization strategy is attested in some languages with object agreement. In these languages, an AAE violation in a clause containing an object anaphor is avoided by detransitivizing the main predicate, thereby obviating its need to agree with the object altogether. Inuit is a language that employs this strategy. In Inuit, the verb is (portmanteau-)marked for both subject and object agreement (7). But when the direct object is an anaphor, object marking on the verb is no longer licit (8) (examples taken from Bok-Bennema, 1991, 51): (7)

Angutip arnaq taku-vaa. man.erg woman.abs see-ind.3sgS.3sgO “The man sees the woman.”

(8) * Hansi-upi immi{i,∗j} asap-paa. Hansi-erg himself.abs wash-ind.3sgS.3sgO “Hansii washed himself{i,∗j} .” (Intended) The structure can be redeemed by suppressing the anaphoric object entirely: in this case, the verb behaves like an intransitive, marking subject agreement alone, as illustrated in (9):¹

1 The pronominal subject in (9) is presumably pro-dropped.

A new kind of Anaphor Agreement Effect |

81

(9) Asap-puq. wash-ind.3sgS “Hei washed himself{i,∗j} .” A variant of the detransitivizing strategy, used in certain other languages, involves cliticizing the anaphor to the clausal predicate such that the anaphor no longer counts as an argument (thus no longer counts as a potential agreement controller): Tucker (2011) proposes that French is such a language.

2.1.2 Default agreement In languages that employ this strategy, the anaphor shows up in agreement triggering position (often though not always marked with some oblique case), but the verb shows up with non-ϕ-covarying, default agreement. This version of affairs is also attested in Inuit. In (10), the anaphoric object is marked with oblique case, and the verb surfaces with default agreement: (10) Anguti immi-nut{i,∗j} taku-vuq man himself-dat see.ind-3sg “The mani sees himself{i,∗j} .” Notice that this is essentially the same strategy that was observed at the outset with Italian (1), repeated below: (11) A loro import-a solo di se-stessi. to them matters-3sg only of them-selves “Theyi only matter to themselvesi .” In (11), the anaphoric subject is marked with oblique case and thus triggers default agreement on the verb. Regardless of this difference between the two languages, the effect is the same: the presence of oblique case on the anaphor (either in subject or object position) ensures that it triggers no ϕ-covarying agreement on the verb, thereby preserving the AAE. In other languages, the anaphor surfaces with nominative case-marking but nevertheless doesn’t trigger ϕ-covarying agreement on the verb; i.e. the verb still shows invariant agreement. Albanian seems to be such a language (the example below reformatted from Massey, 1990, 135):

82 | Sandhya Sundaresan

(12) Drites ii dhimset vetja{i,∗j} . Drita.dat=3sg.dat pity.3sg.past.nact anaph.nom “Dritai pities herself{i,∗j} .” Given that Albanian has a nominative-accusative case system, one might think that the agreement marking on the verb in (12) is triggered by the nominativemarked anaphor. However, as also observed in Woolford (1999), the agreement marking on the verb remains invariant at 3sg, even when the nominative object is in the first-person. The sentence below essentially constitutes a (scrambled) minimal variant to that in (12) (Hubbard, 1985, 191): (13) Vetja{i,∗j} mei dhimset. anaph.nom=1sg.dat pity.3sg.prs.nact “Ii pity myself{i,∗j} .” In effect, then, the Albanian strategy is also a default agreement strategy, similar to those attested in Inuit and Italian.

2.1.3 Protected anaphora This is a sub-variant of the default agreement strategy. In some languages, the anaphor is embedded inside another DP. The agreement on the verb is triggered by the complex DP and not directly by the anaphor: thus a violation of the AAE is avoided (Tucker, 2011, dubs this “the protected anaphora strategy”). Since the structural conditions required for the anaphor to trigger agreement on the verb aren’t met, the verb surfaces with default agreement instead. The protected anaphora strategy has been observed for the Malayo-Polynesian language Selayerese (Woolford, 1999). Selayarese examples like that in (14) initially seem to represent a counter-example to the AAE, since the anaphor seems to be triggering verbal agreement: (14) La-jaŋjang-i kalen-na. 3.erg-see-3.abs anaph-3 “Hei saw himself{i,∗j} .” However, Woolford (1999) shows that the verbal agreement remains invariant at 3rd-person even if the anaphoric object is 1st- or 2nd-person. Woolford presents independent evidence to show that reflexives in Selayerese look superficially sim-

A new kind of Anaphor Agreement Effect |

83

ilar to possessives, arguing on the strength of this, that Selayarese utilizes a protected anaphora strategy, wherein the anaphor is prevented from triggering agreement by itself being a part of a larger DP structure: it is this complex DP that triggers agreement on the verb. Similar protected anaphora strategies have been observed for the Indo-Aryan language Hindi, Modern Greek (Woolford, 1999) and West Flemish Haegeman (2004).

2.1.4 “Anaphoric” agreement In yet other languages, the presence of an anaphor in agreement-position doesn’t lead to ungrammaticality: rather, the anaphor triggers a special form of agreement on the verb. Since this agreement form is not a part of the ϕ-agreement paradigm for the language in question, it cannot be characterized as ϕ-covarying agreement. This strategy is demonstrated for Swahili below (Woolford, 1999): (15) Ahmed a-na-ji/*m-penda mwenyewe. Ahmed IIIsbj-prs-refl/*IIIobj-love himself “Ahmedi loves himselfi .” (emphatic) (16) Ahmed a-na-m/*ji-penda Halima Ahmed IIIsbj-prs-IIIobj-love Halima. “Ahmed loves Halima.” Crucially, the special ji marking on the verb in (15) does not ϕ-covary, nor is it attested elsewhere in the agreement paradigm of the language. Baker (2008, ˆ pp. 150–151) provides parallel examples from the Bantu language Chichewa, adapted below (formatting mine): (17) Ndi-na-i/*dzi-khal-its-a pro[−anaph] y-a-i-kali. 1sS-past-cliv.O-become-caus-fv (them) cliv-assoc-cliv-fierce “I made them (e.g. lions) fierce.” (18) Ndi-na-dzi/*i-khal-its-a pro[+anaph] w-a-m-kali. 1sS-past-refl-become-caus-fv (myself) clii-assoc-cli-fierce “I made myself fierce.”

84 | Sandhya Sundaresan

In (17), the causativized ‘become’ verb shows overt agreement both with the subject and the non-coreferent pro object. In the minimally varying (18), the verb again agrees with the subject, but the usual object agreement marking is replaced by a special reflexive form, namely the infix -dzi-.

2.1.5 Summary To sum up, the following crosslinguistic strategies are attested, which avoid a violation of the AAE: Detransitivization: The reflexive in agreeing position is deleted altogether, yielding an intransitive with inherently reflexive interpretation. Inuit is a language that supposedly exhibits this strategy. Default agreement: The verb surfaces with default agreement, either because the anaphoric subject or object is marked with oblique case (Italian and Inuit) or due to some other mechanism (Albanian). Protected anaphora: The anaphor appears inside a PP or possessor DP and is thus unable to trigger agreement: Greek, West Flemish, and the MalayoPolynesian language Selayarese supposedly employ this “protected anaphora” strategy (see also Haegeman, 2004). Anaphoric agreement: The verb is marked with a special, “anaphoric” agreement which is different from the regular agreement marking within the ϕparadigm of the language in question (e.g. Swahili, Hindi, Modern Greek, Selayarese, West Flemish). It is valid to ask how languages that lack agreement-marking altogether fare with respect to the AAE. Based on Woolford (1999)’s investigation of a sample of such languages, anaphors seem to be freely allowed in all argument positions in these languages. Languages with nominative-accusative case systems lacking in overt agreement – like Khmer, Vietnamese, Thai, Chinese and Malayalam – allow nominative anaphors in subject (as well as object) position. The following Khmer example from Huffman (1970) via Woolford, 1999, (formatting mine) illustrates this: (19) Mɨt [teˇɘŋ-pii neˇɘq]i kɨt thaa kluən{i,∗j} ciə kounsəh. friend both person think that self be student “[The two friends]i resonated that they(self){i,∗j} are students.”

A new kind of Anaphor Agreement Effect |

85

Similar behavior is also observed in languages with ergative-absolutive case systems that lack overt marking for object agreement. In such languages, the anaphor may licitly occur in object position without incurring a violation of the AAE. The following example from the Papua New Guinea language, Enga, illustrates this (Lang, 1973; Woolford, 1999): (20) Baa-mé tánge pi-ly-á-mo. he-erg self hit-pres-3sg.subj-augment “Hei is hitting himself{i,∗j} .” It is tempting to conclude from this data that the AAE is really about the overt marking of agreement, and not about its underlying presence. However, the range of agreement-less languages tested is still rather sparse, so such a conclusion would be premature: i.e. it may just as well be the case that the languages considered here also happen to lack agreement underlyingly (or, put another way, that these languages show no agreement marking because they lack agreement underlyingly). Given these considerations, the following updated generalization of the AAE emerges (Tucker, 2011, p. 30, ex. 40): (21) “Anaphors do not occur in syntactic positions construed with covarying ϕ-morphology.”

3 Enter Tamil The Dravidian languages have been singled out in the literature for their recalcitrant behavior with respect to the AAE. Kayne (1994, 54) first observed that Dravidian languages are potentially problematic for Rizzi’s AAE, noting that in Tamil, a language where nominatives typically trigger ϕ-agreement, a nominative anaphor may licitly occur with a ϕ-covarying clausemate verb. Kayne’s claim has been contested in Woolford (1999) on the grounds that the agreement triggered under the nominative anaphor is in fact not ϕ-covarying: it is either invariant default agreement marking a gerundival clause (22) or is a clearly mismatched (thus also not co-varying) 1sg agreement, as in (23):²

2 Notably, the Dravidian anaphor ta(a)n cannot take 1st- or 2nd-person antecedents, which makes it difficult to argue that this agreement is triggered by the anaphor directly.

86 | Sandhya Sundaresan

˘ (22) [Seetha var-r-ad-aagæ] Murugeesan so-nn-aaru. anaph.nom.sg come-pres-3nsg-nmlz Murugeesan say-pst-3msg “Murugeesan spoke [of Seetha’s coming].” ˘ ˘ (23) Murukeesani [CP taan{i,∗j} var-r-een-nnu] so-nn-aaru. Murugesan.nom anaph.nom come.prs-1sg-comp say-pst-3msg “Murugesani said [that he{i,∗j} would come].” Woolford’s account has since been contested by (Selvanathan and Kim, 2008), who point out that structures like (22) and (23) don’t exhaust the possibilities for agreement under ta(a)n in Tamil. For a non-trivial range of native speakers, sentences like that in (24) are licit as well – the verbal agreement under ta(a)n is 3msg, makes it harder to dismiss the idea that the agreement is triggered directly by ta(a)n, in apparent violation of the AAE: ˘ u] ˘ ˘ (24) Murukeesani [taan{i,∗j} varu-gir-aar-unn so-nn-aaru. Murugesan.nom anaph.nom come-pres-3msg-comp say-pst-3msg “Murugesani said [that he{i,∗j} would come].” Below, I take a closer look at these kinds of Tamil sentences to see whether they really are problematic for the AAE. On the strength of this investigation, I will conclude that such structures do not, in fact, constitute counter-examples to the AAE. However, Tamil employs a new logical possibility to avoid a violation of the AAE. In particular, although the agreement triggered in the scope of the (nominative) anaphor is ϕ-covarying, it is not triggered directly by the anaphor itself. Rather, it is triggered by a different nominal in the local phase. Tamil (and potentially other languages, as I briefly discuss below) thus instantiates a different strategy for preserving the AAE from those discussed earlier. This in turn warrants a revision of our understanding of this generalization.

4 “Anaphoric agreement” in Tamil In Tamil, ϕ-agreement on the verb is typically triggered by a local nominal in the nominative: ˘ (25) [Nii paris-æ tookkapoo- r-æ-nnu] Raman namb-in-aan. you[nom] prize-acc lose.goprs-2sg-comp Raman believe-pst-3msg “Ramanj believed [CP that you would lose the prize].”

A new kind of Anaphor Agreement Effect |

87

The Tamil anaphor ta(a)n may occur with various case-markings depending on its syntactic position (subject vs. direct or oblique object). When ta(a)n occurs with a non-nominative case, covarying ϕ-agreement on the verb is triggered by the (nonanaphoric) clausemate nominative DP in the local domain, just as expected. But when ta(a)n occurs in the nominative, the nature of the covarying ϕ-agreement triggered on its clausemate verb is less straightforward. What is striking is that this agreement seems to covary, not with ta(a)n itself, but with the antecedent of ta(a)n. In the examples below, the ϕ-agreement triggered on the clausemate verb of ta(a)n is marked in boldface: ˘ u] ˘ (26) Miai [Srij [taan{i,∗j} too-pp-aaɭ-unn Mia.nom Sri.nom anaph.sg.nom lose-fut-3fsg-comp ˘ nene-tt-aan-nu] paar-tt-aaɭ. think-pst-3msg-comp see-pst-3fsg “Miai saw [CP that Srij thought [CP that shei /*hej would lose]].” ˘ u] ˘ (27) Miai [Srij [taan{j,∗i} too-pp-aan-unn Mia.nom Sri.nom anaph.sg.nom lose-fut-3msg-comp ˘ nene-tt-aan-nu] paar-tt-aaɭ. think-pst-3msg-comp see-pst-3fsg “Miai saw [CP that Srij thought [CP that hej /*shei would lose]].” ˘ (28) Koɻændæi naɖandadæ-patti joosi-čč-adu. Taani een child[sg.nom] happening-acc-about reflect-pst-3nsg. anaph[nom] why ˘ kašʈappaʈʈ-iru-kk-adu? suffer-prf-prs-3nsg “[The child]i reflected about what had happened. Why had it{i,∗j} suffered so?” When the intended antecedent is 3fsg Mia (26), the agreement under ta(a)n is also 3fsg. But in the minimally varying (27), the agreement under ta(a)n is 3msg, with the only possible antecedent being Sri. Finally, in (28), ta(a)n refers “logophorically” to the extra-sentential attitude-holder Seetha, but the agreement under ta(a)n must still reflect the ϕ-features of this antecedent: if 3nsg koɻændæi (‘the child’) were replaced by masculine or feminine DPs (like Sri or Mia, respectively), the agreement-marking would be correspondingly 3msg -aan or 3fsg -aaɭ instead. The following descriptive generalization thus emerges: (29) The verbal agreement tracks the antecedent of the anaphor ta(a)n.

88 | Sandhya Sundaresan

4.1 Unviable analytic options There are (at least) three possible ways to interpret the generalization in (29) above. Here, I show why two of these three options are unviable. The first option, given that Tamil is elsewhere a uniformly nominativeagreement language (see again (25)), would be to propose that the source of agreement under nominative ta(a)n is ta(a)n itself. In this case, structures like (26)–(28) would necessarily constitute an exception to the AAE. Thus, such an analytic option, is to be dispreferred on grounds of theoretical economy (pending independent empirical evidence to the contrary). Furthermore, since the agreement triggered under ta(a)n may vary, this would be tantamount to proposing, with no independent evidence to support it, that ta(a)n has three different sets of ϕ-features in each of the examples above: i.e. that there are three underyingly distinct anaphors that all happen to be pronounced “ta(a)n”. Additionally, as noted earlier with respect to structures like (23), the verbal agreement triggered under nominative ta(a)n may be 1st-person. This is illustrated below: ˘ so-nn-aan-nnu] ˘ (30) [Krishnani [taan{i,∗j} ʤej-pp-een-nnu] Srij Krishnan anaph[nom]i win-fut-1sg-comp say-pst-3msg-comp Sri nene-čč-aan. thought-pst-3msg “Srij thought [CP that Krishnani said [CP that he{i,∗j} would win]” Under the assumption that the source of agreement is ta(a)n, we would be forced to posit a fourth variant of ta(a)n to account for such sentences – one in which ta(a)n a 1st-person indexical.³ Finally, under such an approach, the fact that the features on the verb track those of ta(a)n’s antecedent would still have to be explained separately. The second analytical option would be to claim that the agreement on the verb under ta(a)n is triggered by ta(a)n’s antecedent – e.g. via some form of longdistance agreement. Such agreement may be transmitted to the verb via ta(a)n, but ta(a)n itself wouldn’t be the actual source of agreement, so the AAE wouldn’t be violated. Such an approach would also automatically capture the antecedent tracking effect outlined in (29). But there are (at least) two independent reasons to reject this option. The first piece of evidence against this view comes from structures involving logophoric de-

3 See Anand (2006) for an analysis along these lines for Malayalam ta(a)n, and Sundaresan (2012) for detailed discussion of why this won’t work for the Tamil counterpart.

A new kind of Anaphor Agreement Effect |

89

pendencies, like that in (28) and from “backward binding” structures (not shown here, but see Sundaresan, 2012) where the antecedent doesn’t c-command the anaphor or its clausemate verb: it is difficult to see how feature-transmission from the antecedent could be made to work inter-sententially or from a non-c-commanding position, respectively. The second piece of counter-evidence comes from structures like (23) (discussed earlier) and (30) below. These are special structures involving the clausal complement of a speech predicate. The anaphor ta(a)n is the nominative subject of this complement; but the ϕ-agreement triggered under it is 1sg (see again the example in (30) above). The agreement pattern in these sentences seems superficially dissimilar to those seen in (26)–(28), where the verbal agreement simply matches the ϕ-features of the antecedent of ta(a)n. But if we look closer, we see that the sentences in (30) and (23) are actually parallel to these others and, in fact, also obey the antecedent tracking generalization described in (29). This is because the 1sg agreement only obtains when the antecedent is the agent of a speech-predicate; if the antecedent were Krishnan, antecedentmatching 3msg agreement would obtain instead. Additional evidence supporting this conclusion comes from number marking on the verb. When the agent of the speech predicate (which also serves as the antecedent of the anaphor) is marked plural, the agreement on the verb under ta(a)n is 1pl not 1sg: ˘ u] ˘ so-nn-aaŋ-gaɭ. (31) Pasaŋ-gaɭi [taaŋ-gaɭ{i,∗j} ʤej-pp-oom/*aangaɭ-unn boy-pl.nom [anaph-pl.nomi win-fut-1pl/*3mpl-comp] say-pst-3m-pl “The boys said [CP that they{i,∗j} would win]” Such considerations show that the agreement “tracks” the anaphor’s antecedent even in cases where its ϕ-features don’t match those of the antecedent. A simple long-distance feature-transmission account of agreement triggered by the antecedent would be hard put to explain such a mismatch. Sundaresan (2012) argues that the 1st-person agreement under ta(a)n instantiates Kaplanian indexical shift for 1st-person (Kaplan, 1989; Schlenker, 2003) – where the 1st-person refers to the Speaker of the context introduced by the speech predicate in the sentence, and not to the Speaker of the utterance context. In other words, the lack of antecedent ϕ-matching in structures like (30) and (31) is not because the agreement doesn’t track the antecedent – but because the evaluation context against which ϕ-features are evaluated is different in the embedded and matrix clauses in these sentences. To sum up, the data and discussion in this section show that the source of agreement on the verb under ta(a)n must be a nominal that is local to it in the syntactic structure – one that is, furthermore, distinct from ta(a)n itself.

90 | Sandhya Sundaresan

4.2 A viable option: a mediating pro Under a standard Minimalist understanding of agreement, our current analytic state-of-affairs may be summarized as follows: Assumption: ϕ-feature agreement is locally implemented as an Agree operation. I.e. verbal agreement (realized on the T head) is triggered by an element that is (phase-)local to T.⁴ Observation I: ϕ-feature agreement on T under nominative ta(a)n is not directly triggered by ta(a)n. Observation II: ϕ-feature agreement on T under nominative subject ta(a)n is not directly triggered by the antecedent of ta(a)n (which is not local to the T head, in any case). Observation III: But ϕ-feature agreement on T nevertheless tracks the antecedent of ta(a)n. This in turn leads us to the following conclusions. There must be a third element (≠ antecedent, and ≠ the anaphor), local to both ta(a)n and the T head, which triggers ϕ-agreement on T. This element must, of course, have valued ϕ-features at the point at which it checks those on T: we might thus envision it as a kind of (null) pronoun, i.e. as a pro. The antecedent-tracking effect of agreement would follow naturally from the assumption that this pro and the antecedent corefer. If the ϕ-features of the antecedent and of pro are computed against the same evaluation context (the default scenario), they would necessarily match. The verbal agreement triggered by pro would thus also match the ϕ-features of the antecedent by transitivity, as observed for (26)–(28). But in cases where these evaluation contexts differ – as in the examples in (23), (30), and (31) – the ϕ-features would not match, although the pro and the antecedent would crucially continue to corefer. Rather, the pro in such cases would be an obligatorily shifted 1st-person indexical denoting the same entity as the antecedent; it would thus trigger 1st-person agreement on the T head. The antecedent tracking effect on verbal agreement, given in (29), would thus hold in different ways in both types of structures: in one instance, yielding ϕ-matching between the antecedent and the agreement, and in the other, yielding 1st-person agreement under indexical shift (see Sundaresan, 2012, for detailed further motivation and discussion of these points, which lie beyond the scope of this paper).

4 It is less pertinent to this account whether Agree for ϕ-features happens in the narrow syntax, as is more standardly assumed, or in the post-syntactic morphology, as has been argued by Bobaljik (2008). I assume that it happens in the narrow syntax, but nothing crucial rests on this choice, at least for the cases of ϕ-agreement discussed here.

A new kind of Anaphor Agreement Effect |

91

An important question that this raises is why the pro is present in the first place (after all, triggering agreement could not be its sole reason for being). What I propose, again in line with prior work in Sundaresan (2012), is that this pro plays a central role in mediating long-distance anaphoric dependencies in languages with perspectival anaphoric systems like Tamil (and also others, like Icelandic and Italian). In such languages, the antecedent of the anaphor always denotes an individual who holds a mental and/or spatio-temporal perspective toward some extended projection containing the anaphor. As such, I contend that this null pronoun is also associated with a perspectival feature which allows it to pick out a perspective-holder individual in the context of evaluation. The perspectival pronoun mediates the relationship between the anaphor and its antecedent: it binds ta(a)n at LF and is non-obligatorily controlled by its antecedent. All three nominals thus corefer in different ways and end up denoting the same perspectiveholder individual. Finally, following analogous data and discussion in Koopman and Sportiche (1989); Bianchi (2003); Speas (2004), among others, on similar logophoric operators in the clausal left-periphery in other languages – I propose that this perspectival pronoun occupies the specifier of a perspectival phrase (PerspP) in the left periphery of the local clause containing ta(a)n. Under this account, valuing the ϕ-features of the clausemate verb of nominative ta(a)n is just an incidental by-product of this perspectival pronoun’s presence in ta(a)n clauses. As a pronoun, I assume uncontroversially, that this pro is born with fully valued ϕ-features: it is thus eligible to serve as a potential ϕ-agreement controller for the probing T head in such structures.

4.3 Formally deriving anaphoric agreement in Tamil We can now see how the two types of antecedent-tracking agreement patterns in Tamil – the first involving ϕ-matching and the second involving 1st-person agreement under indexical shift – may be formally derived. Consider again the ϕ-matching sentence in (26), repeated below: ˘ u] ˘ (32) Miai [Srij [taan{i,∗j} too-pp-aaɭ-unn Mia.nom Sri.nom anaph.sg.nom lose-fut-3fsg-comp ˘ nene-tt-aan-nu] paar-tt-aaɭ. think-pst-3msg-comp see-pst-3fsg “Miai saw [CP that Srij thought [CP that shei /*hej would lose]].” In the model developed here, the antecedent and the anaphor are not directly related in the syntax in any way: the antecedent-anaphor relation is mediated by the

92 | Sandhya Sundaresan

perspectival pronoun in [Spec, PerspP], in the manner outlined above. Recall, too, that Agree for ϕ-features is assumed to happen under phase locality in the syntax or post-syntactic morphology. This means that, as far as the syntax of anaphora is concerned, the only relevant piece of structure is the local phase (CP) containing the anaphor and the perspectival pro in [Spec, PerspP]. The derivation is fairly straightforward. Given the current model, the various players in the Agree relationship are born with the following features: in addition to the ϕ-features on the nominals, there is also a Dep-feature, which is the featural marker of perspective. The perspectival pro is born with a valued Dep-feature, but the anaphor is born with an unvalued Dep-feature. The pro values the Dep-feature on ta(a)n which in turn trigger variable binding of ta(a)n by pro at LF:⁵, ⁶ pro in [Spec, PerspP]

Anaphor (ta(a)n)

T

[Dep: x, ϕ: 3fsg]

[Dep: , ϕ: ]

[ϕ: ]

I assume, furthermore, that Agree for both ϕ- and Dep-features happens in an upward fashion in these cases: i.e. that the Goal (perspectival pro) c-commands the Probe (ta(a)n and T) (see e.g. Zeijlstra, 2012; Bjorkman and Zeijlstra, 2014; Wurmbrand, 012b). The derivation for (32) proceeds bottom-up as follows: 1.

Merge(vP, T) → T’

2.

Merge(DPta(a)n , T’) → TP

3.

Agree(T [ϕ: ], DPta(a)n [ϕ: ]) → {T [ϕ: i ], DPta(a)n [ϕ: i ]}

4. Merge(Persp, TP) → Persp’ 5.

Merge(DPpro , Persp’) → PerspP

6.

Agree({T [ϕ: i ], DPta(a)n [ϕ: i ]}, DPpro [ϕ: 3fsg]) → {T [ϕ: 3fsg], DPta(a)n [ϕ: 3fsg]}

7.

Agree(DPta(a)n [Dep: ], DPpro [Dep: x]) → DPta(a)n [Dep: x]

Step 3 involves an Agree relation between two sets of unvalued ϕ-features, on T and the anaphoric subject DP, respectively. Following Pesetsky and Torrego

5 Unlike the Agree relation for ϕ-features between pro, the T head and potentially also ta(a)n, the Agree relation for Dep between pro and ta(a)n must take place in narrow syntax, since its output must feed operations at both LF and PF. 6 In addition to the features listed below, I assume that both ta(a)n and the perspectival pronoun are endowed with a categorial D feature and case features. These are not included here for reasons of perspicuity.

A new kind of Anaphor Agreement Effect |

93

(2007), I assume that this yields feature sharing for ϕ-features on T and DP such that these essentially function as a joint probe to get these ϕ-features valued. I indicate this notationally by the coindexation of the values that ϕ-Agree will result in.⁷ The tree structure for this CP after Agree and before Spell-Out, thus looks like this: note that underlining is just a visual mnemonic to distinguish inherent features from inherited ones. (33)

CP C

…

-nnu˘ … (comp)

PerspP

DP pro [Dep: x, p: 3, g: f, n: sg]

Persp’

TP

Persp

DP

taan [Dep: x, p: 3, g: f , num: sg]

T’

vP

T

would lose

-aaɭ [p: 3, g: f , num: sg]

At LF, the matching of the Dep-values on DPta(a)n and DPpro triggers binding of the former by the latter (the latter is construed as the binder since it asymmetrically c-commands the former): the two DPs are now targeted to refer to the same entity in the evaluation context; thus, the assignment function maps these elements to

7 A different possibility may be that ta(a)n doesn’t lack just the values for ϕ-features, but lacks ϕfeatures altogether. In this case, ta(a)n would simply bear a categorial feature, case feature, and an unvalued Dep-feature, and the T head alone would serve as a ϕ-probe, getting these features valued by the perspectival pro. It is a matter of ongoing research which of these alternatives is the correct one.

94 | Sandhya Sundaresan

a salient perspective-holder in this context, one that is also a female individual, since pro was born with 3fsg ϕ-features, and now ta(a)n does too.⁸ In (32), the chosen referent is thus Mia (not Sri), who is both a female and a perspectiveholder. This individual is represented by the DP Mia in the linguistic structure in (32): this DP corefers with the pro and, by transitivity, with ta(a)n, and thus in effect serves as the antecedent of ta(a)n, as discussed earlier. Since the ϕ-features of the antecedent are also evaluated against the same evaluation context (i.e. the utterance context), such coreference furthermore entails ϕ-matching between these elements, all of which thus bear 3fsg features. The perspectival pro has already triggered 3fsg agreement on the probing T head in its clause, as we’ve seen – explaining the antecedent tracking effect of verbal agreement described in (29). The derivation of the minimally varying sentence in (27), repeated below, is very similar, yielding the following structure post-Agree and pre-SpellOut: ˘ u] ˘ (34) Miai [Srij [taan{j,∗i} too-pp-aan-unn Mia.nom Sri.nom anaph.sg.nom lose-fut-3msg-comp ˘ nene-tt-aan-nu] paar-tt-aaɭ. think-pst-3msg-comp see-pst-3fsg “Miai saw [CP that Srij thought [CP that hej /*shei would lose]].” The only difference between this and the structure in (33) has to do with the ϕ-feature values on pro.⁹ As illustrated below, the perspectival pronoun in (27) is born with 3msg ϕ-values: it thus values the features on T and ta(a)n as 3msg (yielding 3msg rather than 3fsg agreement on its clausemate T), and also values the Dep-feature on ta(a)n. This yields the following structure right before Spell-Out:

8 I.e. the ϕ-features on these nominals set up presuppositional restrictions on their reference assignment at LF, along the lines discussed in Heim and Kratzer (1998). 9 As mentioned earlier, I assume that there are no restrictions on the ϕ-features that pro may be born with. The syntax essentially overgenerates and ill-formed structures – such as, for instance, a 3fsg pro and ta(a)n denoting a male perspective-holder – are filtered out at LF.

A new kind of Anaphor Agreement Effect |

(35)

95

CP C

…

-nnu˘ … (comp)

PerspP

DP

Persp’

pro [Dep: y, p: 3, g: m, n: sg]

Persp

TP

T’

DP taan [Dep: y, p: 3, g: m, n: sg]

vP

T

would lose

-aan [p: 3, g: m, n: sg]

At LF, the masculine gender feature on these nominals ensures that they are both mapped onto a male (rather than female) perspective-holder in the utterance context, namely Sri (rather than Mia). The DP representing this individual in (34) is Sri, which corefers with pro, and serves as ta(a)n’s antecedent by transitivity, in the manner described above. Since, again, the ϕ-features on this DP are also evaluated against the utterance context, this coreference entails ϕ-matching: Sri, perspectival pro, and ta(a)n all bear 3msg features. The 3msg agreement features on the verb, triggered by pro thus seem to track those of the antecedent. To complete the paradigm, let us now look at the special structures involving 1st-person verbal agreement under ta(a)n. The structure below is a simpler version of (30): ˘ so-nn-aan. (36) Krishnani [taan{i,∗j} ʤej-pp-een-nnu] Krishnan anaph[nom]i win-fut-1sg-comp say-pst-3msg-comp “Krishnani said [CP that he{i,∗j} would win]” The syntactic derivation for (36) proceeds exactly the same as before. The only difference between (36) and a sentence like (32) at the level of the syntax is that

96 | Sandhya Sundaresan

the perspectival pronoun in [Spec, PerspP] is born with 1sg features. There is no special rule that ensures this, as mentioned earlier: the syntax simply overgenerates and ill-formed structures are filtered out at the interfaces. DPta(a)n and T are thus both valued as 1sg. 1.

Merge(vP, T) → T’

2.

Merge(DPta(a)n , T’) → TP

3.

Agree(T [ϕ: ], DPta(a)n [ϕ: ]) → {T [ϕ: i ], DPta(a)n [ϕ: i ]}

4. Merge(Persp, TP) → Persp’ 5.

Merge(DPpro , Persp’) → PerspP

6.

Agree({T [ϕ: i ], DPta(a)n [ϕ: i ]}, DPpro [ϕ: 1sg]) → {T [ϕ: 1sg], DPta(a)n [ϕ: 1sg]}

7.

Agree(DPta(a)n [Dep: ], DPpro [Dep: y]) → DPta(a)n [Dep: y]

A further difference in (36) comes in only later, at LF: as noted earlier, the evaluation context of the embedded CP in this sentence is not the utterance-context but a shifted “context” pertaining to the speech-event denoted by the matrix speech verb. Thus, 1st-person on pro doesn’t denote the speaker of the utterance context, but the speaker invoked by the speech-verb, who is moreover male – namely Krishnan. This speaker is also a perspective-holder with respect to the embedded CP, thus qualifies at LF, to serve as the antecedent of ta(a)n.¹⁰ The nominal denoting this perspective-holder Krishnan is the DP Krishnan in the matrix clause in (36). This DP corefers with pro and ta(a)n, thus serving as ta(a)n’s antecedent, as described for the other two sentences discussed here. However, the matrix CP is not evaluated against the shifted speech-context but against the utterance context. This has the significant conseqence that coreference between Krishnan, pro, and ta(a)n does not entail ϕ-matching. The ϕ-features on pro and ta(a)n match, of course – and are both set to 1sg, as we’ve seen. But with respect to the utterance context, Krishnan denotes a nonparticipant (i.e. 3rd-person) male entity: thus, the DP Krishnan is set to 3msg, and triggers 3msg agreement on its clausemate verb. The tree-structure for the syntactically relevant part of the derivation – namely the embedded CP – is depicted below, post-Agree and before SpellOut:¹¹ 10 I have assumed here that pro bears a masculine gender feature, though nothing crucial rests of this choice. pro could also have been born bearing just 1sg features with the gender feature being either underspecified or absent altogether. 11 The SpeechActP, selected by the speech-verb, sets the evaluation context to be that associated with the speech-verb. For further details and discussion of the nature of SpeechActP and the representation of this context, see Sundaresan (2012).

A new kind of Anaphor Agreement Effect |

(37)

97

CP

Speech-ActP

C nnu˘ (comp)

Speech-Act’

Spec DP [d]

… PerspP

Speech-Act λxλc.Speaker(x)(c) …

DP

Persp’

pro [Dep: y, p: 1, g: m, num: sg ]

TP

Persp

T’

DP vP taan [Dep: y, p: 1, g: m, n: sg]

T

-een [p: 1, g: m, n: sg]

If the perspectival pronoun in (37) had been born with 3rd-person features instead (something that cannot be prevented), the syntactic derivation would still proceed unimpeded. The perspectival pro would successfully value both ta(a)n and the probing T head as 3rd-person in the syntax. However, such a derivation would crash at LF, since this 3rd-person feature would be evaluated against a shifted context (where Krishnan denotes the speaker, thus is 1st-person), and not against the utterance context. 3rd-person on pro and ta(a)n would thus ensure that they do not refer to Krishnan. As such, the DP Krishnan (which does refer to Krishnan) would not corefer with pro and would, by extension, also not function as the antecedent of ta(a)n. In the absence of another salient perspective-holder candidate for antecedent in the discourse, the sentence will crash. But if there is another salient perspective-holder in the structure – as in (38) below, the problem will be obviated:

98 | Sandhya Sundaresan

˘ ˘ (38) [Krishnani [taan{j,∗i} ʤej-pp-aan-nnu] so-nn-aan-nnu] Srij Krishnan anaph[nom]i win-fut-3msg-comp say-pst-3msg-comp Sri nene-čč-aan. thought-pst-3msg “Srij thought [CP that Krishnani said [CP that he{j,∗i} would win]” In (30), we assume that pro in the innermost CP is born with 3msg features: it values ta(a)n and T with these same features in the syntax. At LF, pro and ta(a)n cannot refer to Krishnan – for the reasons given above. But there is another salient perspective-holder which is evaluated as 3msg in the context associated with the speech-verb (and incidentally, also relative to the utterance context): this is the entity denoted by the matrix DP Sri. Thus pro and ta(a)n are mapped onto this entity with the result that Sri is construed as the antecedent of ta(a)n in (38).

5 Tamil anaphoric agreement and the AAE The discussion above shows that Tamil does not employ any of the parametric strategies used by other languages to avoid a violation of the AAE – described in Section 2.1. The Tamil anaphor is not “protected” from triggering agreement, as illustrated for Selayarese, the verbal agreement triggered under the nominative anaphor ta(a)n is not a frozen, default form, as observed for Italian, Inuit, and Albanian; nor is it a special morphological form that obtains only in the scope of ˆ anaphors – as observed for Swahili and Chichewa. But Tamil nevertheless does have a strategy to avoid an AAE violation in the structures discussed above. The agreement triggered under nominative ta(a)n is ϕ-covarying. But it is ultimately triggered, not by ta(a)n, even though ta(a)n is the closest nominal to the probing T head and thus is in a position which normally would trigger ϕ-agreement.¹² Rather, ϕ-agreement in such cases is triggered by some other nominal in the local domain of T. This element, I have argued, is the perspectival pro in [Spec, PerspP] in the left periphery ta(a)n’s local phase. In the account developed here, the anaphor also participates in ϕ-agreement; but it is crucially not a source of ϕ-agreement on T: rather T and the anaphor jointly function as probes for ϕ-valuation.

12 Of course, it is also important that this nominal bear nominative case since, in Tamil as in many other languages (Bobaljik 2008), only nominatives can serve as ϕ-agreement triggers. Since case is not a central concern of this paper, I don’t discuss this point any further here. But the restriction to nominatives is presupposed in my discussion of Tamil.

A new kind of Anaphor Agreement Effect |

99

Our original formulation of the AAE, based on the discussion of parametric strategies in Section 2.1 was that in (21), repeated below: (39) “Anaphors do not occur in syntactic positions construed with covarying ϕ-morphology.” The Tamil data investigated here shows, rather, that we should not think in terms of dedicated syntactic positions associated with triggering ϕ-agreement. Rather, as standard understandings of the operation Agree (Chomsky, 2001, et seq.) may lead us to expect, the ϕ-probing head seeks to Agree with the minimally closest nominal in its search space that can value its ϕ-features. While this may, for a probing T, typically be a nominal in [Spec, TP], Tamil shows that, if this position is occupied by a nominal without (fully) valued ϕ-features, as e.g. an anaphor, probing can continue on until the closest ϕ-valued nominal is found. The Tamil AAE strategy is thus the result of two independent grammatical properties: (i) The anaphoric needs of ta(a)n independently ensure that a ϕ-valued nominal will always be present above it in its local domain. (ii) ta(a)n does not count as a(n) (defective) intervener for ϕ-agreement by a probing T with this higher nominal. As such, our understanding of the AAE must be updated as in (40) below: (40) Anaphor Agreement Effect (updated): a. Short version: Anaphors cannot directly trigger covarying ϕ-agreement which results in covarying ϕ-morphology.¹³ b. Long version: Anaphors cannot directly trigger covarying ϕ-agreement which results in covarying ϕ-morphology. If an anaphor occurs as the minimally closest target for such agreement,¹⁴ then one of the following possibilities must obtain: (i) Agreement ignores the anaphor and is thus able to obtain with the next closest target which does have valued ϕ-features. (ii) There is special (i.e. non ϕ-covarying) “anaphoric” agreement. (iii) There is default agreement. (iv) The derivation crashes. 13 This formulation allows anaphors to indirectly participate in triggering covarying ϕagreement resulting in covarying ϕ-morphology. Thus, in the Tamil structures discussed here, the anaphor may still serve to transmit the ϕ-features of the perspectival pronoun to its clausemate verb. 14 This is, of course, assuming that it also satisfies other restrictions on triggering agreement, like ones in terms of case.

100 | Sandhya Sundaresan

Note that the possibilities listed in (40b) do not form a natural class, thus cannot be reduced further. They can be seen as a disparate collection of Elsewhere scenarios, i.e. the various instantiations of ways that the specific banned operation, as described in (40a), could fail to obtain.

5.1 Tamil strategy in other languages The Niger Congo language Donna Sɔ seems to manifest a phenomenon that looks a lot like the special 1st-person agreement seen in Tamil speech complements (the example below is taken and reformatted from Curnow, 2002): (41) Oumar [inyemɛ jɛmbɔ paza bolum] miñ tagi. Oumar [anaph[sbj] sack.def drop left.1sg] 1sg.obj informed “Oumari told me [CP that he{i,∗j} had left without the sack.]” In (41), ϕ-covarying agreement obtains on the clausemate verb of the anaphor, just like in Tamil. The agreement triggered on the verb under this anaphor is, furthermore, 1sg. Notably, the minimal CP containing the anaphor is a speech complement, as in the shifted Tamil cases in (30)/(36) and (31). Under the old formulation of the AAE (see again (39)), the sentence in (41) would constitute a counter-example to the AAE; but it is entirely accountable under the updated version in (40). The agreement pattern seen in (41) might be explained along the same lines as the 1st-person anaphoric agreement in Tamil: i.e. we might say that the agreement is triggered by an obligatorily shifted 1st-person pro in the left periphery of the embedded CP, which binds the anaphor in its local domain and also corefers with the antecedent Oumar. Of course, further research needs to be undertaken into the anaphoric and agreement systems in Donna Sɔ to see to what extent such an analysis would be viable for this language. But the surface resemblance to Tamil structures like that in (30) above is striking. Analogous structures may also be attested in Amharic. Amharic has been discussed in the literature as a language that manifests indexical shift (Schlenker, 1999, among others). But in some of the clauses where indexical shift has been shown to obtain, the putatively shifted indexical is actually a silent pro (as in (42) below, from Delfitto and Fiorin, 2011, but ultimately due to Malamud (2006)): (42) Profäsərui [proi bät’am bəzu səra ə-sär-allähu] alä. very much work 1sg-work.imp-aux.1sg say.prf.3sg.masc professor “The professori said [CP that hei works very hard].”

A new kind of Anaphor Agreement Effect |

101

Since the subject is silent, we have no obvious way of knowing that it really is a 1st-person indexical (as also pointed out in Delfitto and Fiorin, 2011, 219). That it is tacitly treated as such is actually due to the 1st-person agreement marking on its clausemate verb. But structures like Tamil (30) and now (potentially) (41), raise the possibility that it is a null anaphor instead, and that the 1st-person agreement is triggered by a shifted 1st-person pro higher up in the clause. Indeed, there seems to be independent evidence to support the idea that the pro subject is an anaphor and not a 1st-person pronoun: Delfitto and Fiorin further note that the null subject may be construed de re with respect to the matrix subject. If it were really a shifted 1st-person indexical, this would be unexpected; but a de re construal is predicted to be possible if it is a null anaphor, instead (see also Pearson, 2013, for arguments in favor of possible de re construals involving anaphors). Thus, it is possible that Amharic too adopts the Tamil strategy for avoiding a violation of the AAE. We may speculate that, in general, two properties would have to hold, for a language to be able to adopt the AAE strategy proposed here for Tamil. First, it would need to have a (c-commanding) ϕ-valued nominal in the local domain which could trigger ϕ-agreement. In Tamil, the presence of this candidate was seen to be independently motivated by the perspectival nature of anaphora. Thus, we might predict that other languages which have both overt agreement marking and similar perspectival systems, would also have recourse to this option. Second, the agreement mechanism of this language and the structural and featural status of the anaphor has to be set up such that the anaphor doesn’t count as an intervener (defective or otherwise) for agreement between the ϕ-probing head and this other nominal. We might speculate that in languages that use either the default or anaphoric marking strategy, the anaphor does intervene for such ϕ-agreement. Further research into the anaphora and agreement systems in these and other languages must, of course, be undertaken to better understand the locus of parametric variation here.

6 Conclusion and theoretical speculations I have presented novel data and reconsidered old data from the Dravidian language Tamil, pertaining to the nature of verbal agreement triggered under the nominative anaphor ta(a)n. In the process, I have argued that this agreement, which looks like standard ϕ-covarying agreement, is nevertheless special because the nominative element, the standard source of agreement in Tamil, is an anaphor. As per the AAE, the agreement is triggered on the verb in such cases, not by the anaphor, but by another element in the local phase, namely a per-

102 | Sandhya Sundaresan

spectival pro. The anaphor is indeed involved in the agreement mechanism, but crucially not as the source of agreement. The broader crosslinguistic implication of this is that the AAE, namely the generalization that anaphors cannot trigger covaring ϕ-agreement yielding covarying ϕ-morphology, is valid in Tamil as well. Tamil just uses a different strategy to avoid violating the AAE than the languages discussed so far in the literature. I have so far said nothing about the theoretical motivations behind the AAE, as described in (40). A plausible answer may simply be that anaphors don’t have the (valued) ϕ-features required to trigger agreement. Indeed, a popular view in the literature claims that this lack of some or all ϕ-features is the defining property of an anaphor (Kratzer, 2009; Reuland, 2011; Rooryck and vanden Wyngaerd, 2011). A radical implementation of this position would be to say that anaphors are “minimal pronouns” (Kratzer, 2009), nominals that are born with a fully unvalued set of ϕ-features. They would thus be more like the functional heads T/v that probe to get their ϕ-features valued, than like full-fledged nominals (a position reminiscent of earlier proposals like that in Borer, 1989). A more conservative variant of this would be to say that anaphors lack a (potentially proper) subset of ϕ-features. In this case, however, we may additionally have to posit that a violation of the AAE cannot be circumvented by partial valuation of ϕ-features on the probe by the anaphor. Regardless of which position we take within the ϕ-deficiency account, the timing of Agree operations is crucial. I.e. the anaphor should be unable to value the ϕ-features on T/v even after it has potentially inherited these ϕ-features from another DP (either its antecedent or some other local entity, like the perspectival pronoun argued for here). One way to derive this would be to argue that the probing head can distinguish between inherited and inherent ϕ-features and can only be valued by inherent ϕ-features. Alternatively, we might say that the T head gets its ϕ-features valued before the anaphor does, or at the same time as the anaphor, as under the feature-sharing mechanism assumed for Tamil in this paper. However, this ϕ-deficiency account of the AAE is independently problematic, a main concern being that it is too simple. Indeed, it is for this reason that anaphora in Tamil was syntactically encoded in terms of a Dep-feature and not in terms of ϕ-features. The simple fact is that anaphors in the world’s languages do not seem to be created equal. Based on restrictions observed on their potential antecedence, anaphors seem to have different ϕ-featural specifications. Some seem to lack all ϕ-features, thus place no ϕ-featural restrictions on their antecedents: the Chinese anaphor ziji is an example (Huang and Liu, 2001). Yet others seem to bear some ϕ-features but lack others: the Dravidian anaphor ta(a)n, among many others, places no restriction on the gender or number of its antecedent but restricts its choice of person (3rd-person antecedents are allowed, 1st and 2nd person are

A new kind of Anaphor Agreement Effect |

103

not). At the other end of the spectrum, we have anaphors that do not seem to lack any ϕ-features whatsoever. Heinat (2008), for instance, discusses examples from San Lucas Quiaviní Zapotec and Thai, among others, to show that even R-expressions may be anaphorically bound. The following Zapotec example is from Heinat (2008, p. 151):¹⁵ (43) San Lucas Quiaviní Zapotec: R-ralloh Gye’eihllyi [r-yu’lààa’z Lia Paamm Gye’eihllyi ]. hab-think Mike Mike hab-like f Pam “Mikei thinks Pam likes Mike{i,∗j} .” (literal) “Mikei thinks Pam likes him{i,∗j} .” (intended) If we wanted to maintain the ϕ-deficiency account in the face of such data, we would essentially have to say that all anaphors, including those in languages like Zapotec, underlyingly lack ϕ-features, even if they don’t seem to do so on the surface. Alternatively, we might find that such languages in fact do not observe the AAE – this would then count as potentially significant evidence in favor of the ϕ-deficiency account of the AAE. Another source of theoretical interest has to do with the question of what factors condition the choice of parametric strategy that a language adopts in order to avoid a violation of the AAE. I have offered some brief speculative discussion of this in Section 5.1, but a great deal more empirical research obviously needs to be done to address this and other questions with a sufficient level of descriptive and explanatory coverage. This is part of my ongoing research. Acknowledgment: I am grateful to Gillian Ramchand, Hedde Zeijlstra, and especially Tom McFadden for discussing aspects of this work with me on multiple occasions, and to the editors for putting this volume together. Thanks also go to the audiences at the Pronouns at Tübingen Workshop at the University of Tübingen, at the Colloquium Series at the University of Leipzig, at the Agreement Workshop in Recife, Brazil, and at the University of Cambridge, where variants of this paper were presented, for insightful feedback. Any errors are mine alone.

15 Crucially, evidence from sloppy readings under VP ellipsis show that the R-expression does indeed function like a bound-variable and is not merely accidentally coreferent with its antecedent as in the sentence. “Everyone loves Bill. Even Bill loves Bill!”

104 | Sandhya Sundaresan

References Anand, P. 2006. De de se. PhD thesis, Massachusetts Institute of Technology. Baker, M. C. 2008. The syntax of agreement and concord. Cambridge University Press, United Kingdom. Bianchi, V. 2003. On finiteness as logophoric anchoring. In Guéron, J. and Tasmovski, L., editors, Temps et point de vue/Tense and Point of View, pages 213–246. Université Paris X, Nanterre. Bjorkman, B. and Zeijlstra, H. 2014. Upward Agree is superior. http://ling.auf.net/lingbuzz/ 002350. Göttingen/Toronto. Bobaljik, J. D. 2008. Where’s phi? Agreement as a postsyntactic operation. In Harbour, D., Adger, D., and Béjar, S., editors, Phi Theory: phi-features across modules and interfaces, Oxford Studies in Theoretical Linguistics, pages 295–328. Oxford University Press. Bok-Bennema, R. 1991. Case and agreement in Inuit. Foris, Dordrecht. Borer, H. 1989. Anaphoric AGR. In Jaeggli, O. and Safir, K., editors, The null subject parameter, pages 69–109. Kluwer Academic Publishers, Dordrecht. Chomsky, N. 2001. Derivation by phase. In Kenstowicz, M., editor, Ken Hale: A Life in Language. MIT Press, Cambridge, Mass. Curnow, T. J. 2002. Verbal logophoricity in African languages. In Proceedings of the 2002 Conference of the Australian Linguistic Society. Deal, A. R. 2010. Ergative case and the transitive subject: a view from Nez Perce. NLLT, 28:78– 120. Delfitto, D. and Fiorin, G. 2011. Person features and pronominal anaphora. Linguistic Inquiry, 42(2):193–224. Haegeman, L. 2004. A DP-internal Anaphor Agreement Effect. Linguistic Inquiry, 35(4):704–712. Heim, I. and Kratzer, A. 1998. Semantics in Generative Grammar. Oxford: Blackwell Publishers. Heinat, F. 2008. Probes, pronouns and binding in the Minimalist Program. VDM Verlag. Huang, J. and Liu, L. 2001. Logophoricity, attitudes, and ziji at the interface. In Cole, P., Hermon, G., and Huang, J., editors, Long-distance reflexives, pages 141–195. Academic Press, New York. Hubbard, P. 1985. The syntax of the Albanian verb complex. Garland, New York. Huffman, F. E. 1970. Modern spoken Cambodian. Yale University Press. Kaplan, D. 1989. Demonstratives: An essay on the semantics, logic, metaphysics, and epistemology of demonstratives and other indexicals. In Perry, J., Almog, J., and Wettstein, H., editors, Themes from Kaplan, pages 481–563. Oxford University Press, Oxford. Kayne, R. 1994. The antisymmetry of syntax. MIT Press, Cambridge, Mass. Koopman, H. and Sportiche, D. 1989. Pronouns, logical variables, and logophoricity in Abe. Linguistic Inquiry, 20(4):555–588. Kratzer, A. 2009. Making a pronoun: fake indexicals as windows into the properties of pronouns. Linguistic Inquiry, 40(2):187–237. Lang, A. 1973. Enga Dictionary. Australian National University, Canberra. Malamud, S. 2006. Semantics and pragmatics of arbitrariness. PhD thesis, University of Pennsylvania, Philadelphia. Maling, J. 1984. Non-clause bounded reflexives in Modern Icelandic. Linguistics and Philosophy, 7:211–241.

A new kind of Anaphor Agreement Effect |

105

Massey, V. 1990. Experiencers, themes, and c-command in Albanian. In MITWPL 12: Student Conference in Linguistics, pages 128–143, Cambridge, MA. Pearson, H. 2013. The Sense of Self: Topics in the Semantics of De Se Expressions. PhD thesis, Harvard University, Cambridge, MA. Pesetsky, D. and Torrego, E. 2007. The syntax of valuation and the interpretability of features. In Simin Karimi, Vida Samiian, W. K. W., editor, Phrasal and clausal architecture: syntactic derivation an dinterpretation, pages 262–294. Amsterdam: John Benjamins. Reuland, E. 2011. Anaphora and language design. MIT Press, Cambridge, Mass. Rizzi, L. 1990. On the anaphor-agreement effect. Rivista di Linguistica, 2:27–42. Rooryck, J. and vanden Wyngaerd, G. 2011. Dissolving Binding Theory. Oxford Studies in Theoretical Linguistics. Oxford University Press, Oxford. Schlenker, P. 1999. Propositional attitudes and indexicality: a cross-categorial approach. PhD thesis, MIT. Schlenker, P. 2003. A plea for monsters. Linguistics and Philosophy, 26:29–120. Kluwer Academic Publishers. Selvanathan, N. and Kim, C. 2008. The anaphor agreement effect in Tamil. Snippets, 18. Speas, M. 2004. Evidentiality, logophoricity and the syntactic representation of pragmatic features. Lingua, 114(3):255–276. Sundaresan, S. 2012. Context and (Co)reference in the syntax and its interfaces. PhD thesis, University of Tromsø and University of Stuttgart, Tromsø. Taraldsen, K. T. 1978. On the NIC, vacuous application and the That-trace filter. Bloomington, Indiana, University Linguistics Club. Tucker, M. A. 2011. Even more on the anaphor agreement effect: when binding does not agree. Manuscript written at UCSC. Woolford, E. 1999. More on the anaphor agreement effect. Linguistic Inquiry, 30(2):257–287. Wurmbrand, S. 2012b. The syntax of valuation in auxiliary-participle constructions. In et al., J. C., editor, Proceedings of the 29th WCCFL, University of Arizona, Tucson. Zeijlstra, H. 2012. There is only one way to Agree. The Linguistic Review, 29:491–539.

Michelangelo Falco and Roberto Zamparelli (University of Trento)

The only real pro-nouns Comparing English one and Italian ne as Noun Phrase pro-forms Abstract: The study of pronouns has focused on personal pronouns replacing Determiner Phrases (DPs), paying little attention to pro-forms like English one(s) and Italian ne, which replace subparts of DPs, namely Noun Phrases (NPs). Although ne should be functionally equivalent to one, the two forms seem to obey different constraints. In this paper, we argue that the differences are superficial, and can be ascribed to independent properties of Italian and English. We show that the diverging acceptability of one(s) and ne when these forms appear ‘bare’, i.e. without a determiner, confirms a Carlsonian view of bare nouns, but points to a difference between English and Italian existential bare nouns. We study the conditions for bare ne and show that they shed light on verbs and verb uses which take objects only as representatives of their kinds. Keywords: pro-NPs, pronouns, syntax/semantics interface, kinds, ones, ne

1 Introduction In generative grammar the linguistic analysis of pronominal elements has primarily focused on pronouns replacing whole nominals, i.e. Determiner Phrases (DPs), in the terminology of Abney (1987). These forms, e.g. they in (1), are literally speaking ‘pro-DPs’. (1)

You have [DP four small bikes]i . Theyi are all heavy.

With a few early exceptions, much less attention has been payed to pro-forms that replace subparts of DPs, namely functional projections including the noun and possibly some of its modifiers, marked NP in (2). Pro-forms for these projections include English one (2) (Postal, 1966; Ross, 1967; Jackendoff, 1977), Italian ne (Belletti and Rizzi, 1981; Cordin, 1988; Cardinaletti and Giusti, 1991) and French en (Kayne 1975a). (2) You own [DP four small [NP (red) bikes]i ] and I own [DP two big onesi ].

108 | Michelangelo Falco and Roberto Zamparelli

Pro-NPs cannot be easily assimilated to other pro-forms. For instance, Elbourne’s (2005) influential attempt to see personal pronouns as disguised definite descriptions with indexes, (3) has no obvious way to treat the pro-NP one. (3) them = [DP [ the i] [ NP ]]

NP contextually derived

Moreover, to the best of our knowledge, examples of quantificational binding of a pro-NP cannot be constructed¹ (Déchaine and Wiltschko 2002, Sec. 3.2 reach the same conclusions, within an analysis that tries to derive the distribution of Japanese kare from its NP status). Therefore, even if one uses an anaphoric index to obtain coreference in (2), this index must have very different properties than those of pro-DPs.² Since pro-DP theories do not easily apply to these forms, new theories have to be developed. This article will provide an analysis for a specific but representative case, Italian ne, contrasted with English one(s). We will show that these forms have many similarities and that their differences can be derived from independent properties of Italian and English: the fact that ne, unlike one, is a clitic pro-form, plus some more general differences in the meaning and derivation of determinerless nominals in Italian and English. One striking difference between Italian ne and English one is that the former can appear ‘bare’, i.e. without any associated determiner or numeral, but only after certain verbs and with certain types of modifiers. We will link these two facts, characterizing the class of verbs that can select ne in a way which will account for the modification pattern. The article is structured as follows. In Section 2.1 we delimit the analysis by excluding non pro-NP cases of one and ne. Next, Section 2.2 and Section 2.3 introduce the main points of similarity and difference between the two forms. Many

1 Here is our best attempt. Color is sometimes said to be a second-order property, since one can say My phone is white, and white is a color. Thus, we can try to build (i) (the predicative position of one insures that it is interpreted as a property), aiming for the meaning ‘Seeing red makes me wish my phone was (a nice) red, seeing green, that it was (a nice) green, etc.’. But this meaning is impossible (contrast with: My phones is red, and a particularly nice one). (i)

* Every colori I see makes me wish my phone was (a nice) onei .

Note that it is anyway unlikely that color is a 2nd -order property in (i). It might be a normal property ranging over the nominalization of color terms, hence an ⟨et⟩-type object. 2 For reasons of space, we are not able to carry out a detailed analysis of the binding properties of pro-NPs vs. pro-DPs in this paper, nor address possible similarities between one/ne and kare. We hope to address these topics in future work.

The only real pro-nouns

|

109

of these differences are explained in Section 3 in terms of the different clitic status of ne and one. The last difference – the existence of ‘bare ne’ but not ‘bare one’ – is taken up in Section 4, and explained in terms of the different behavior of bare nouns in the two languages. In Section 5, we address some syntactic and semantic constraints for bare ne, while Section 6 is devoted to the structure of ne antecedents.

2 The distribution of Italian “ne” and English “one” 2.1 Delimiting the analysis: non-pro-NP “ne” and “one” Before laying down the data on one and ne, it will be useful to stake out the facts we want to cover, excluding some non-pro-NP uses of these forms. First, the one we address in this article is distinct from the numeral one and from the impersonal one found in sentences such as One always needs to be careful with data. Pro-NP one can pluralize (we will write it as one(s) in what follows), and the numeral and pro-NP forms can coexist within the same DP (4a): the one before the adjective red is the numeral, while the one after red is the pro-NP. In (4b), one is the numeral; these cases (which might involve a ‘silent noun’ after the numeral, see Kayne 2005, ch. 10) will not be discussed here. (4) a. I have two red socks and you have only [one red one]. b. I have two socks and you have [only/exactly one]. Italian ne is ambiguous, too. Beside being a pro-NP, it can be a pro-PP (Cordin 1988); in (5), for instance, ne plays the role of an ‘of’ PP or a possessive (in (5a) ‘the museums of Florence/its museums’, in (5b) ‘Giorgio’s best friend/his best friend’). (5) a. Il sindaco di Firenze ne cura particolarmente i musei the mayor of Florencei NEi cares_for especially the museums “The mayor of Florence takes special care of its museums” b. Quanto a Giorgio, Marco ne è il migliore amico as for Giorgioi , Marco NEi is the best friend “Marco is Giorgio’s best friend”

110 | Michelangelo Falco and Roberto Zamparelli

We believe that ultimately certain types of the pro-PP ne could be unified with the pro-NP ne usage illustrated in detail in the next sections (see also Cardinaletti and Giusti 1990), but since nothing in the present discussion hinges on this, in what follows we focus on the clear pro-NP cases and set (5) aside.

2.2 Similarities and differences between “one(s)” and “ne” Once the non pro-NP uses are spelled out and set aside, we can clearly see that the pro-NP ne as it appears in e.g. (6), is a faithful counterpart of one(s), modulo the fact that ne is a clitic pronoun, which must be realized to the left of a finite verb, starting from within an object position (there are no subject clitics in standard Italian). Here and after, we gloss ne as one(s). (6) Tu hai [DP due bicii rosse] e io nei ho [DP una blu]. you have [DP four bikesi red] and I one(s)i have [DP a blue] “You have two red bikes and I have a blue one” Following a vast literature on clitics (Kayne 1975b, a.o.) we assume that ne has moved onto the V head from a DP-internal position, which is the same position one(s) occupies. Both one(s) and ne can pick up nouns and modifiers, as in (7) (note that adjectives are often post-nominal in Italian; in (7b) ne refers to bici rosse ‘red bikes’). (7) a. You have [DP four nice [NP red bikes]i ] and I have [DP two ugly onesi ] b. Tu hai [DP quattro belle [NP bici rosse]i ] e io nei ho [DP due you have [DP four nice [NP bikes red]i and I one(s)i have [DP two brutte] ugly] In (7) ones and ne have an antecedent in a previous clause. Ne, however, can also be found in the Clitic Left Dislocation Construction (CLLD, Cinque 1990), where its antecedent is a topicalized form: a bare N, optionally preceded by di ‘of’ (8a), or a full definite DP, obligatorily preceded by di (8b). Other cases are out (8c).³

3 The pro-DP clitic li (theyclit. ) is in complementary distribution with ne in the contexts in (8): it is impossible with (di) ragazzi and di quei quattro ragazzi arrivati ieri ‘of those 4 boys who arrived yesterday’, but perfectly grammatical when the antecedent is alcuni/due/quei ragazzi ‘some/two/those boys’. Again, this fact points to different anaphoric properties for pro-NP and pro-DP pronouns.

The only real pro-nouns |

111

(8) a.

(di) ragazzi, ne ho visti (due) (of) boys, one(s) I_have seen (two) b. Di quei quattro ragazzi arrivati ieri, ne ho visti due of those four boys arrived yesterday, one(s) I_have seen two c. * alcuni/due/quei ragazzi, ne ho visti (due) some/two/those boys, one(s) I_have seen (two)

In this paper, we will sometimes use forms like those in (8a) as antecedents. Their syntactic nature will be taken up in Section 6.

2.3 Three differences between “ne” and “one(s)” A first difference is that one(s) is compatible with the presence of a ‘strong determiner’ (in Milsark’s 1974 sense: roughly definites, demonstratives and universals) (9). Pro-NP ne is incompatible with each of them (10). (9) Speaking of your bikes, I still have {those red ones / the red ones / the ones you gave me / every one I bought} (10) Di bici, ne ho ancora {*quelle / ??le due (che mi hai dato) / of bikes, one(s) I_have still { those / the two (that you gave me) / *ciascuna } each-one } (11) Difference 1: Dstrong MODIFIER ONE / *Dstrong (MODIFIER) NE Note, however, that one(s) cannot appear with the or the NUM, unless it is modified (i.e. *the (two) ones vs. (i.e. the (two) red ones). A second difference between one and ne, originally observed in Cardinaletti and Giusti (1990), is that ne can strand post-nominal modifiers and pre-nominal numerals, but not prenominal adjectives. This contrasts of course with the situation of one(s), which is very often accompanied by prenominal adjectives.⁴ Since many adjectives in Italian can appear before or after N, in a structure such as (12) we do not know if the ne trace is to the left or the right of the stranded adjective:

4 One(s) can of course be followed by PPs, relatives and the rare post-N English adjectives (e.g. the ones alive). There is no contrast with ne on this point.

112 | Michelangelo Falco and Roberto Zamparelli

(12) a. NEi +V [DP Num t i ADJ] b. NEi +V [DP Num ADJ t i ] To see which of (12) is correct, we need to make use of one of the Italian adjectives that change meaning in pre- and post-N position, e.g. puro ‘mere/pure’ or vero ‘genuine/truthful’: (13) a. Una pura formalità / una sostanza pura a mere formality / a substance pure b. Un vero tradimento / una dichiarazione vera A genuine betrayal / a declaration truthful Using ne with these adjectives, we see that the only surviving meaning is the postN one (‘pure’ and ‘truthful’ in (14a and b) respectively), hence the oddity of the two sentences. (14) a. ?? Di formalità, ne devo concludere una pura. of formality, NE I_have to_conclude a pure b. ?? Di tradimento, costui ne ha commesso uno vero of betrayal, he NE has committed one genuine We conclude, with Cardinaletti and Giusti (1990), that the only possible structure is (12a). (15) Difference 2: DET MOD ONE / DET (*MOD) NE The last important difference between one(s) and ne is the fact that one(s) must appear with a modifier of some sort: one(s) can have an unmodified bare noun as antecedent (see (16)), but it cannot serve as a verbal argument all by itself (a ‘bare unmodified ones’, parallel to a normal bare plural), witness (17), where the two examples were chosen to represent the two possible meanings of English bare plurals – kinds and individual objects (Carlson, 1977). In contrast, ne can appear without any modifier or numeral (18), always with the existential interpretation (as if preceded by a narrow scope ‘some’) which characterizes Italian bare nouns (see Section 4.2 below). Some additional constraints on these cases will be discussed in Section 5.2. (16) I like watching birdsi , particularly [large onesi ].

The only real pro-nouns

|

113

(17) a. * I like [large birdsi ], in fact I like [onesi ] regardless of size. b. * I shoot [large birdsi ], you shoot [onesi ,] no matter what their size. (18) Leoni, in Africa, ne ho visti Lions, in Africa, one(s) I_have seen ‘As for lions, in Africa I saw some’ (19) Difference 3: *(MODIFIERS) ONE / (MODIFIERS) NE Having laid out the data, we turn to those aspects that depend on a syntactic difference between ne and one(s): the clitic status of ne and non-clitic status of one(s).

3 The clitic status of “ne” and its consequences Some distributional differences between ne and one immediately follow from the fact that ne is a clitic and one(s) is not. While one can appear within subjects (As for Belgian beers, the dark ones are the best), hence pre-verbally, the requirement for ne to C-command its trace forces the base position of this clitic to be post-verbal: the object of a transitive verb (20a), or the post-verbal subject of an unaccusative (20b) or passive verb (20c). The last two cases take the auxiliary essere ‘be’ and the participle agrees in gender and number with ne, which in turns shares ϕ-features with the numeral (here una, ‘1’) and the antecedent. (20) a. Ragazze, nei ho vista una ti girls-fem, one(s)i I_have seen-sing-fem a-sing-fem t i ‘Girls, I saw one’ b. Ragazze, nei è arrivata una ti girls-fem, one(s)i is arrived-sing-fem a-sing-femi t i ‘Girls, one arrived’ c. Ragazze, nei è stata vista una ti girls-fem, one(s) is been seen-sing-fem a-sing-fem t i ‘Girls, one has been seen’ With unergative verbs like cenare ‘dine’ ne is impossible both with pre-verbal and post-verbal subjects (21) (Burzio, 1986). (21) * Ragazzei , { molti nei hanno cenato / nei hanno cenato molte } { many one(s)i have dined / one(s)i have dined many } girlsi ,

114 | Michelangelo Falco and Roberto Zamparelli

To explain this fact we assume, with Longobardi (2000), that post-verbal subjects of unergatives are the result of subject extraposition to a left-peripheral position, followed by movement of the predicate remnant to a specifier position further to the left of the subject: (22) * [[ti nej hanno cenato]k [molti t j ]i t k ] one(s) have dined many After these movements, nej +V no longer C-commands the subject position, nor the ne trace inside it. A last interesting case is the possibility (marginal for some speakers), of ne in with what looks like the pre-verbal subject of an unaccusative or passive verb. We suggest, instead, that tre in (23) is the topicalized remnant of a post-V subject from which ne has been extracted, as shown in the structure in (24). (23) % Tre, ne sono arrivati ieri. Three, one(s) are arrived yesterday. (24)

[tre t i ]j nei sono arrivati t j ieri

3.1 Strong determiners and “ne” The fact that ne needs to move from within the DP also accounts for (11), i.e. the fact that ne, unlike one(s), is incompatible with ‘strong determiners’ (see (10) above). Demonstratives, universal or proportional quantifiers are known to block subextraction (25) (Fiengo and Higginbotham 1981; Enç 1991; Diesing 1992), so the incompatibility with ne is fully expected. We remain agnostic as to the precise analysis of (25), but we believe that it will naturally extend to the pro-NP ne.⁵

5 Interestingly, the extraction of the pro-PP ne is more liberal, since it is allowed after certain definites: (i)

Quanto alla CED, nei conosco personalmente il maggiore esperto t i as for_the CED, of_iti I_know personally the main expert t i

In a barrier-framework, the difference might be derived from the fact that pro-PP ne in (i) is selected by a lexical element, the noun, while the ne in (10) isn’t. In a phase-based framework, the difference could be cast in terms of antilocality (Grohmann, 2003): the NP cannot move to the edge of the nominal phase (Chomsky, 2001) from a position immediately below Num because this movement is ‘too close’ to be visible at the interface.

The only real pro-nouns |

115

(25) [Who did Mary make {a / 0 / ??each / ??this / ??most} film(s) about t] Constraints on extraction are also responsible for the impossibility of moving ne from a predicate nominal, illustrated in (26). Following insights in Moro (1997) and Zamparelli (1996), we assume that copular be selects a predicative subpart of the DP (Zamparelli’s PDP) without lexically marking it. In the ‘Barrier’ framework adopted in these works, this resulted in the predicative phrase being an island for sub-extraction, as shown with a full Wh DP in (27).⁶, ⁷ (26) * [Brave professoresse]i , Anna e Lucia nei sono [due t i ] [good professors]i , Anna and Lucia one(s)i are [two t i ] (27) ?? [Di cosa]i credi che Carlo sia un insegnante ti ? [of what]i you_think that Carlo is a teacher ti ?

3.2 The impossibility of pre-N adjectives The fact that pre-nominal adjectives are impossible with ne (15) is open to various explanations. A first possibility is that the pre-N adjectives block the movement of the ne N/NP to V. This would be expected if the adjectives were hosted in [Spec,NP] under the assumption that only maximal projections move (28a) (multiple pre-N adjectives would require multiple Spec positions), less so if they are in the specifiers of independent functional projections, as proposed in Cinque (1994), Scott (2002) and other ‘cartographic’ approaches (28b). (28) a. [DP 3 … [NP [AdjP pure ][N󸀠 ne ]]] b. [DP 3 … [FP [AdjP pure ] F0 [NP ne ]]] A second possibility is to resort to some version of Relativized Minimality (Rizzi, 2001). Adjectives and nouns are both predicative categories which share agree6 See Chomsky 1986. Again, we leave open how this constraint could be exactly recast in a phasebased approach. The degraded status of (26) compared to (27) could be due to the fact that in the former ne violates anti-locality. 7 As Moro observes, the presence of the clitic ci/ce ‘there’ make the copula behave more like a lexical verb asserting existence, thus making ne extraction possible: (i)

[Brave professoresse]i , ce nei sono [due t i ] [good professors], there one(s) are [two t] “as for good professors, there are two”

116 | Michelangelo Falco and Roberto Zamparelli

ment features, so it is not entirely implausible that attributive adjectives in the specifier of a functional projection might block leftward NP movement in (28b). The problem of this approach is that it is at odds with well-established theories of N-movement inside DPs. According to Cinque’s theory of adjective position, in Romance languages N moves left, past the lowest adjective (as a head in Cinque 1994, as an NP in Cinque 2010). This of course presupposes that adjectives do not block N(P) movement. Fortunately, the distinction drawn in Sproat and Shih (1990) and Cinque (2010) between ‘direct’ and ‘indirect’ modification offers a promising solution for ne. In Cinque’s theory, direct modifiers have an individual level interpretation, are non restrictive and potentially non-intersective; indirect ones are based on reduced relative clauses, can have a stage-level interpretation and are restrictive and intersective. The latter end up being post-nominal in some languages as a result of leftward movement of the complex that includes all direct modifiers and the noun: (29) [DP … [FP direct_modifiers N ]i indirect_modifiers t i ] We can hypothesize, then, that the adjectives stranded by ne are all cases of indirect modification. Evidence comes from the observation that adjectives which form idiosyncratic meaning combinations with a noun and which cannot be used across a copula, such as nucleare or marino in (30), are also deviant when their nouns are picked up by ne (31): (30) a. Due fisici {nucleari / ??che sembravano nucleari} two physicists {nuclear / who seemed_to_be nuclear} b. una tartaruga {marina / ??che era marina} a turtle {marine / which was marine} (31) a. ?? Di fisici, ne conosco due t nucleari. of physicists, I know two t nuclear b. ?? Di tartarughe, lo zoo ne ha due t marine. of turtles, the zoo has two marine t ones The question, now, is why direct modifiers are impossible, that is why (31) could not be derived by replacing fisici with ne and stranding the direct modifier nucleare on the way to V, as shown in (32). (32) nei +V [DP … [Direct_modifiers t nucleare t]]

The only real pro-nouns |

117

A possible answer is simply that ne corresponds to the FP that hosts all of the direct modification adjectives (that is, a pro-FP, not a pro-NP). This, however, would introduce an asymmetry with English ones. Most English speakers we consulted accept cases such as (33), where ones is clearly a pro-NP (it can be part of N-N compounds as in (b), which are certainly within the direct modification section). (33) a. You met two theoretical physicists and I met two nuclear ones. b. Land turtles are much rounder than sea ones. In addition, (32) must be impossible even for regular nouns, not just pro-NPs, or we would generate the unattested order Ni > indirect modifiers > direct modifiers ti . It looks like NP must pied-pipe its FP – a fact that suggests that the structure in (28a) might after all be on the right track, or at least that FP should be seen as a projection whose features are only minimally distinct from those of NP – an ‘extended NP’ of sorts. The fact that ne can pick up a noun with or without its (direct) modifiers points to the same conclusions: ne sees FP and NP as one and the same projection. The need for ne to carry along its extended projection clashes, however, with the requirements of its clitic status. Clitics must be simple, ‘small’, necessarily unstressed elements. From the cliticization standpoint, ne should raise alone. These two conflicting constraints (being phonologically simple, qua clitic, pied-piping direct modifiers), can be both satisfied only if ne does pied-pipe the whole FP, but this FP contains no overt material apart from ne itself.⁸ (34) [NP/FP (* Direct-modifiers) [NP ne ]]i +V …[DP [NumP t i Indirect-Modifiers t i ]] Additional evidence for the need to pied-pipe direct modification comes from a comparison between the three partitive structures in (35). (35a) is a normal partitive and (35b) seems to be derived by raising the FP/NP nuclear physicists across the partitive head of. But (35c), where the NP has moved alone stranding its direct modifier, is severely ill-formed. (35) a. two of those [nuclear physicists] who hate the bomb b. two [nuclear physicists] of those who hate the bomb c. * two [physicists] of those [nuclear] who hate the bomb

8 Alternatively, we could assume that adjective-hosting FPs are simply absent when empty. This is probably a more parsimonious possibility, but one which should be settled on the basis of theoretical considerations broader than the facts considered here.

118 | Michelangelo Falco and Roberto Zamparelli

Returning to (34), it is important to remark that like all clitic movements, the raising of ne to V does not affect its meaning. At LF, the meaning of N must still be combined with its (indirect) modifiers and with the numeral.

4 The (im)possibility of bare pro-forms 4.1 Bare and modified pro-NPs Let’s now address the fact that one must be modified (but not just by a numeral, see the next section), while ne can appear alone (cf. (19) above). The puzzle can be presented in the following form. (36) shows that a modified one can pick up a whole bare plural (dogs), interpreted either as a kind (36a) or as an existential (36b), and can itself be part of a bare plural. (36) a. I love [dogs]i , no matter what their size. You only like [small onesi ]. b. I saw [stray dogs]i as I was jogging – [large and small onesi ]. But ones alone can never function as a bare plural, as shown in (17), repeated below as (37). Why is this so? (37) a. * I like [large birdsi ], in fact I like [onesi ] regardless of size. b. * I shot [large birdsi ], you shot [onesi ], no matter what their size. This fact is particularly puzzling for theories such as Chierchia (1998), which assume that bare plurals are simple NPs (or Ns). In these theories, one would in fact expect that one(s) should be the only way to refer to bare nominals. In actual fact, the only pro-form that picks up a bare plural is they/them: (38) I like dogsi and you like themi /*onesi , too. We can hypothesize that ones in (38) is out precisely because them is possible, but this intuition must be spelled out with care, since sometimes, as in (17)/(37), we want to refer to a subpart of the whole DP antecedent (i.e. an NP/FP), and they is a pro-DP. The explanation that we would like to propose rests on the theory of bare plurals proposed by Longobardi (2001). In this theory (i) bare nominals in argument positions are uniformly DPs with an empty D (D0 ); (ii) D0 needs licensing. In English, one way of being licensed is to attract nouns with certain features (plural

The only real pro-nouns |

119

count or singular mass); (iii) this movement gives nominals the possibility of functioning as ‘proper names of kinds of things’ (a syntactic implementation of the well-known semantic proposal in Carlson 1977); (iv) like other proper names in English, nouns raise to the D position via head-movement at LF (or feature movement, in syntax). One problem with the original implementation of this theory is that bare plurals can contain modifiers and relative clauses whose meaning must be compositionally assembled. But modifiers (semantic type ⟨⟨et⟩⟨et⟩⟩, see e.g. Heim and Kratzer 1998:4.3) can combine with nouns only if nouns denote properties (type ⟨et⟩). If the noun becomes entity-denoting (⟨e⟩) by moving to D, the relation with its modifiers cannot be maintained. To address this problem, we depart from point (iv) above, and assume the following. First, we follow Chierchia (1998) in proposing that those plural properties which are ‘cognitively salient’ and exhibit a sufficiently regular behavior can be turned into kinds by a null operator (Chierchia’s “down”, ∩ ). This operator applies to the semi-lattice structure that comes from the normal composition of a property-denoting plural noun with its modifiers⁹, and returns a kind (type ⟨ek ⟩). Following Borer (2004), we also propose that to realize its ⟨ek ⟩-type kind meaning, the whole phrase containing the noun and its modifiers must move to [Spec,D0 ] across an empty NumP (39), (39) [DP [FP/NP African elephants]i D0 [NumP Num0 … t i ] …] Applied to the two structures in (17)/(37), this model makes the two bare nouns [large ones] and [ones] move to [Spec,DP], giving (40a) and (b), respectively (like above, we write FP/NP for the functional projection containing ‘direct modification’ material, see Section 3.2 above). (40) a.

DP D󸀠

FP/NPi large

NPi ones

b.

D0

DP D󸀠

FP/NPi NumP

Num0

ti

ones

D0

NumP Num0

ti

9 We assume that the denotation of (count) nouns is a set S of atomic individuals, which the (subsective and restrictive) modifier restricts to a subset S󸀠 . The effect of the plural operator (say, the PL head of Heycock and Zamparelli 2005) is to create the power-set of S󸀠 , minus the empty set. This structure (the set of all possible non-empty sets which can be formed with the elements in S󸀠 ) can be algebraically characterized as a semi-lattice (see e.g. Link 1983)

120 | Michelangelo Falco and Roberto Zamparelli

Adopting the so-called ‘Neo-Carlsonian’ approach to the semantics of bare nouns, we also assume that bare nouns in English undergo this movement in all cases, i.e. even when they are interpreted existentially (contra e.g. Diesing 1992). This is because in this approach the existential meaning of English bare nouns is derived from the kind meaning via the ‘Derived Kind Predication’ (DKP) shift in (41) (here ∪ is Chierchia’s “up” operator, which maps kinds to properties). By hypothesis, the kind meaning which fuels this process requires raising to [Spec,DP]. (41) If P applies to objects and k denotes a kind, then DKP(P,k) = ∃x[∪ k(x) ∧ P(x)] (cf. Chierchia 1998, ex. 28c) A persistent idea in linguistics, from Postal (1966) to Distributed Morphology, is that the phonetic form of a pronoun is a direct spell-out of the features the pronoun is endowed with. With this in mind, let’s look at (40b). Here the FP/NP in [Spec,D0 ] receives features from D0 ;¹⁰ but the spell-out of a plural pro-form with D features in English is they/them, not ones. Qua D-level pro-form, they/them picks up DPs, hence the contradictory flavor of (42).¹¹ (42) ?? I like [large birds]i , in fact I like themi regardless of size. What happens when a pre-nominal modifier is present? In this case, too, the filled FP complex formed by one and its modifiers moves to [Spec,D0 ], as shown in the first tree in (40). Now ones is no longer in a direct Spec/Head relation with D0 , which it does not C-command. Therefore, it becomes impossible to spell-out D0 and ones together, generating them, since the semantic material from the adjective intervenes. The pronominal features in the raised NP must then be spelled out as one(s), as desired.

10 Which features are relevant remains unclear: +definiteness, according to Postal (1966), person, according to Longobardi (2008), or a more specific ‘referential’ feature. 11 It is interesting to note that if we associate the pro-DP with a modifier, as in object depictives (see Williams 1980, Rizzi 1986), them seems to be able to pick up a subpart of the antecedent (compare (ia) with … while wolves prefer live ones). (i)

a. Jackals eat dead [sheep]i , while wolves prefer themi alive. b. I like [girls]i with short hair, you prefer themi with long pony tails.

One reviewer reports the intuition that this reading might come more easily if the pronoun is shortened to ’em. In any event, this phenomenon is probably not linked to pronouns in particular. It is indeed possible to say: I remember those dead soldiers alive, where alive is an indirect modifier.

The only real pro-nouns

|

121

4.2 Italian “ne” The structures in (40) explain why in English a modifier must be present with ones, but not why the putative Italian equivalent of one, i.e. ne, can in fact appear without any modifier or numeral (“bare ne”), as in (43). (43) a. Funghi, ne ho visti t ieri. mushrooms, one(s) I_have seen t yesterday ‘As for mushrooms, I saw them/some yesterday’ b. Di bravi linguisti, ne ho incontrati t. of good linguists, one(s) I_have met t ‘Good linguists, I have met’ These examples can be glossed in English with some or with a DP proform, them, which can however receive a non-maximal reading: the word them in the glosses of (43a) does not imply that I have seen the total set of contextually available mushrooms. This is likely to be an application of DKP to the kind meaning picked up by the pronoun from its antecedent (i.e. DKP(see󸀠 ,themk )). We are now in a position to derive the existence of bare ne, but not bare one(s) from independently established differences between bare nominals in English and Italian. As is well-known, argumental bare nominals in Italian (and in fact, most Romance languages, see Contreras 1986 for Spanish) are limited to lexically selected positions (verbal or prepositional objects). Moreover, when possible, bare DPs never refer to kinds (Longobardi 1994, 2001, Zamparelli 2002, Dayal 2004, a.o., pace Chierchia 1998). To obtain a kind meaning, Italian uses definite nominals. (44) io {odio / adoro / stermino} ??(i) ratti. I {hate / adore / exterminate (the) rats possible meaning: “I hate/love/exterminate rats” (as a species) Note that this does not mean that the basic denotation of Italian nouns should be different from that of English nouns, as proposed in Chierchia (1998). Italian and English behave in fact on a par in the other constructions that express nominal genericity, the singular definite generic (The dodo is extinct) and the subkind reading (most mammals live on average less than man = most types of mammal …). Following the spirit of reducing putative ‘semantic parameters’ to interface conditions (Borer 2004, Longobardi 2008), we propose that Italian is just like English in allowing some plural/mass properties to be turned into kinds by the “up” operator ∪ . Crucially, however, Italian does not carry out the NP-to-[Spec,DP] movement

122 | Michelangelo Falco and Roberto Zamparelli

which would allow its bare nouns to express this meaning. As a result, Italian must express the kind meaning of a plural nominal with different means, i.e. the definite article, in ways which will not concern us here. The net result of this situation is that verbs that require kind-denoting objects take definite objects and do not accept bare nouns in Italian (see (45a) and (46a)), whereas with verbs that take existential objects the two languages behave alike ((45b) and (46b)). (45) a. b.

I {hate / adore / wiped out} rats I just saw rats in the yard.

(46) a. * Io {odio / adoro / stermino} ratti. b. Ho appena visto ratti per strada.

‘kind’ ∃ = (45a) = (45b)

The existential meanings of (45b) and (46b) look identical, but they are actually derived by two different routes: in English, the bare FP/NP raises to [Spec,DP] as explained in Section 4.1, a kind meaning is expressed, but the DKP applies and generates an existential meaning over tokens. In Italian, on the other hand, bare nouns do not raise to the D position (which needs to be licensed externally), the existential reading comes directly from the noun’s base position, where NP is interpreted as a property. The rest of the structure (D itself, or a VP-level existential closure à la Diesing 1992) provides the quantificational force. If this proposal is correct, Italian ne, unlike English one(s), never establishes the Spec/Head relation with D that, in the absence of an adjective, would make it surface as a pro-DP (e.g. the clitics lo/la/li/le). As a result, bare ne is possible in Italian, but strictly with an existential meaning. This predicts, correctly, that kind-selecting verbs like those in (46) will be impossible with bare ne. (47) * Ratti, ne {odio / adoro / stermino}.

cf. (45a)

5 Bare “ne”: syntactic and semantic constraints Two interesting additional constraints on bare ne remain to be discussed: one in its modification pattern, the other in the range of verbs which can select it.

The only real pro-nouns |

123

5.1 Modifiers and bare “ne” As discussed in Section 3.2, ne can be followed by ‘indirect modification’ material, including adjectives and relative clauses. (48) Di lucciole, ne ho viste due { verdognole / grosse / che Marco of fireflies, one(s) I_have seen two { greenish / big / which Marco aveva messo in un barattolo } proprio qui had put in a jar } right here “as for fireflies, I saw {two green ones / two big ones / two which Marco had put in a jar} right here” But if we remove the numeral due ‘two’ from (48), turning the sentence into an example of bare ne, the modifiers become much worse:¹² (49) Di lucciole, ne ho viste (?? { verdognole / grosse / che Marco of fireflies, one(s) I_have seen ( { greenish / big / which Marco aveva messo in un barattolo } ) proprio qui. had put in a jar } ) right here However, if we precede the adjective with di ‘of’, the judgment is reversed: now bare ne is possible, numeral + ne is not (50). (50) Di lucciole, ne ho viste (*due) di { verdognole / grosse / più lente } / slower } of fireflies, one(s) I_have seen (two) of { greenish / big proprio qui. right here To make matters worse, certain subject relatives are still possible with bare ne: (51) Di lucciole, ne ho viste che volavano vicino a Marco. of fireflies, one(s) I_have seen which flew near Marco “Fireflies, I saw some flying near Marco”

12 A caveat: it is possible to rescue (49) by making a pause after the verb and placing the adjective sentence-finally. This is a characteristic sign of right topicalization. Using the adverbial proprio qui ‘right here’ makes this reading less likely, but of course with the right intonation the adverbial can also be right topicalized. The judgements refer to a reading without pauses which is perfectly possible in (48).

124 | Michelangelo Falco and Roberto Zamparelli

Note that a regular bare plural has none of these constraints. It can be followed by any of the modifiers above, but not by di (unlike in French, where this construction seems to have a wider distribution, see Kayne 1994). (52) Ho visto lucciole (*di) { verdognole / che volavano / che Marco I_have seen fireflies (of) { greenish / which flew / which Marco vorrebbe avere }. would_like to_have })

5.2 Kind, Token and Token-of-Kinds predicates As we have seen, the range of verbs that can select ne is semantically restricted. Being an existential, like any Italian bare noun, ne is of course incompatible with kind-selecting verbs (47). But even with token-selecting verbs, bare ne is not always equally acceptable. Verbs like vedere ‘see’, fotografare ‘photograph’ or osservare ‘observe’ combine smoothly with bare ne (see b’s reply in (53)). (53) a: Nella foresta purtroppo non ho visto [bradipi]i . In the forest unfortunately I have not seen [sloths]. b: Io nei ho {visti / fotografati / osservati}. I one(s)i have {seen / photographed / observed} Interestingly, in the same context, a reply like (54), containing verbs such as spulciare ‘groom’, ricontare ‘count again’, stuzzicare ‘poke’, sounds stranger with bare ne, particularly when a manner modifier is added. All these cases improve once an overt numeral is added.¹³, ¹⁴ 13 Another caveat is in order. With stress, especially in exclamative contexts, it is possible to furnish a bare ne with an unpronounced amount expression roughy meaning ‘quite a large number of’, as in (i): (i)

Eh, se ne ha, di guai! Eh, whether one(s) he_has, of troubles! cf. ‘Boy, quite some troubles, he has got!’

Modulo its particular discourse function, this construction patterns like a non-bare case of ne. We set it aside in the discussion that follows. 14 We are aware that a full corpus-based analysis of the verbs that occur with pro-NP ne would have been very interesting. The problem is that while bare-ne appears to be very common (as shown by searches carried out in the ITWAC web corpus), the vast majority of the naturally attested cases are pro-PP ne (i.e. cases like (5)).

The only real pro-nouns

|

125

(54) b: ?? Io ne ho {spulciati per bene / ricontati / stuzzicati a lungo}. I one(s) have {groomed with care / re-counted / poked at length} To make sense of this strange difference, observe that the two sets of verbs also behave differently in the presence of a definite object. (55) a. Ieri per la prima volta ho {visto / fotografato / osservato} yesterday for the first time I_have {seen / photographed / observed} [i bradipi]. [the sloths] can mean: ‘Yesterday I have {seen / photographed / observed} sloths for the first time’ b. Ieri per la prima volta ho {spulciato / ricontato / stuzzicato} yesterday for the first time I_have {groomed / re-counted / poked} [i bradipi]. [the sloths] ‘Yesterday I {groomed / re-counted / poked} the sloths …’ (some salient one) Let’s focus on the meaning of the bracketed nominal. Above and beyond the normal maximal-set reading provided by the definite, which requires a salient set of sloths, (55a) has a natural indefinite reading which can be rendered in English with a bare plural. In (55b), instead, the bracketed nominal tends to keep its expected maximal/unique reading. Zamparelli (2002) points out that Italian definite noun phrases can express an indefinite meaning in cases such as (56), and argues that this meaning arises via the kind meaning these definites make available, through an application of the DKP, as shown in (57). (56) a. Ogni settimana, il mio sito web viene attaccato da [gli hacker]. attacked by [the hackers] every week, my web site is ‘Every week my web site is attacked by hackers’ b. La casa è sporchissima. In cantina ci sono [i topi] e the house is filthy. In the basement there are [the mice] and sotto il lavello vivono [gli scarafaggi]. under the sink live [the cockroaches] ‘The house is filthy. There are mice in the basement and cockroaches live under the sink’

126 | Michelangelo Falco and Roberto Zamparelli

c. Nel 1986 [i ladri] hanno svuotato il mio appartamento. In 1986 [the thieves] have emptied my apartment. ‘In 1986 thieves have emptied my apartment’ (57) (56d) = [DKP(empty_my_apt󸀠 ,the_thieves󸀠k )] This reading of definites is limited to nominals which can refer to kinds (inserting a numeral between the article and noun will block it, precisely because it blocks kind-reference), and common of objects where the species is normally more salient that the individual. Wild animals are a case in point: if cockroaches are taking over one’s sink, one normally worries about how many, not which individual ones. Naming them is just superfluous. Just as there are nouns, like animal names, that favour a kind-oriented view of their referents, so there are verbs that do the same with their arguments. If I intend to photograph or document a certain kind of object, it does not matter which specimen I find (good-looking is better, of course, but to a point). If I want to see or observe ‘brown bears’, any token will do, as long as it is a token of that kind. So, these verbs take individual tokens, not kinds, but they take them merely qua representatives of their kind. Seen from the angle of specificity, they have two sides: they can select arguments which are extremely specific at the kind level, but completely non-specific with the respect to tokens. At a cognitive level, this behaviour might simply be a function of the ease with which an action performed on one token can be transferred to another one. If I see one brown bear, I am able to recognize the next. Likewise, if acting on one token gives me a special knowledge which will apply to any token of the same kind, I am in a sense acting on the species as a whole. Let’s then call predicates that take tokens but are kind-oriented in the way described ‘Token-of-Kind’ (ToK) predicates. Most predicates are not, or not primarily, token-of-kind. Contrast ‘pruning roses’ with ‘touching roses’. The first requires special skills which, once mastered can be transferred to other roses. One can say: now I know how to prune this type of plant. ‘Touching roses’, on the other hand, does not seem to require the same level of expertise, just like ‘remembering’ or ‘counting’ them: these predicates are thus less ToK-oriented. Adding certain manner modifiers (‘randomly touching roses’) makes predicates even more tokenoriented: carrying out the action on one individual will not have any import on our ability to interact with the next. Our claim, now, is that token-of-kind predicates (or uses of predicates) will be precisely those that tend to interpret definites in an ‘indefinite’ way, via the DKP, and those which allow bare ne (in ways to be discussed momentarily). For

The only real pro-nouns

|

127

instance, in (58a) it is easier than in (58b) to understand le rose as ‘roses’ (a bare plural).¹⁵ And in (59), bare ne is better with the first verb than with the second. (58) a. Ho già potato [le rose] (dunque so farlo). I_have already pruned [the roses], (so I know how to do it) b. Ho già toccato a caso [le rose] (??dunque so farlo). I_have already touched randomly [the roses] ( so I know how to do it) (59) Rose, ne ho {potate / ??toccate a caso}. roses, one(s) I_have {pruned / touched randomly}

5.3 Deriving bare “ne” How to derive the token-of-kind oriented meaning of bare ne? We believe that the answer comes from the pattern of modification examined in Section 5.1. We can summarize it as follows: while ne with a numeral cannot strand ‘direct modifiers’, in Cinque’s sense, bare ne does not accept any modifier, direct or indirect, with the only exception of those introduced by di and certain relative clause structures. We will deal with these last cases at the end. Our solution is based on the idea that when verbs like vedere ‘see’ are used as token-of-kind verbs, they embed a form of DKP in their lexical entry: they take a kind-denoting object and extract ‘representative’ tokens to serve as arguments. Specifically, we assume that beside its regular, ⟨eo ⟨eo t⟩⟩ meaning, vedere ‘see’ has the lexical entry in (60), a function which applies the lexical relation vedere󸀠 to an existential closure over the tokens extracted from the kind by means of the ∪ operator. (60) vedere = λxk λy∃g[∪ x(g) ∧ vedere󸀠 (y,g)]

15 Note that the same effect is also visible with the singular definite generic, and is linked to the so-called ‘avant-garde reading’ discussed by Carlson (1977) in connection with sentences such as The horse came to America with Columbus, where the horse is a token, but acts as a representative of the kind. Again, the choice of predicate makes a difference, see build vs. repair, spot vs. cure: (i)

(ii)

a. Meucci built [the telephone] in 1849 b. ? Meucci repaired [the telephone] in 1849 a. John spotted [the ivory-billed woodpecker] in 2006. b. ? John cured [the ivory-billed woodpecker] in 2006.

a specific phone

a specific individual

128 | Michelangelo Falco and Roberto Zamparelli

Like other Italian bare nouns, ne is not normally kind-denoting, witness *ratti, ne adoro ‘rats, ONES I adore’ (see (47) above). In Section 4.1, however, we argued that this fact can be reduced to a syntactic difference – in Italian common nouns do not raise to [Spec,D0 ] while in English they do. So, while a ⟨ek ⟩-type denotation is in principle available to both Italian and English nouns, only the latter language manages to express it via normal bare nouns. Suppose now that the kind meaning is derived from a noun’s property denotation via the “down” operator (i.e. ∩ N).¹⁶ We propose that, when ne incorporates into a ToK verb¹⁷, ne plugs its kind meaning (∩ ne) inside a ToK semantics similar to the one in (60) above, yielding (61): (61) VToK +ne = λy∃g[ ∪∩ ne(g) ∧ VToK (y,g)] The form ∪∩ ne might seem strange at first: why to interpret a property as a kind, if ∪ immediately turns it back into a property? Recall, however, that ∪∩ P is not always equivalent to P: kinds can only be formed from certain semantically plural properties (Chierchia 1998, a.o.). This is useful to explain the fact that while singular ne is normally possible when a numeral is present (62), a bare singular ne is ungrammatical, since the “down” operator cannot turn a singular count noun into a kind (i.e. ∩ ne is undefined if ne refers to Nsing ). We take this as evidence for the need for token-of-kind arguments to go ‘to the kind and back’. (62) A proposito di orso/orsi, ieri ne ho visto *(uno). apropos bear/bears, yesterday one I_have seensing (one) One feature of the analysis in (61) is that it is conditioned on ne-to-V movement, so it does not extend to Italian bare nouns in general. Note that after this movement, ne is interpreted at the verb. This offers a plausible solution to our second puzzle, the reason why modifiers are impossible with bare ne: if ne is interpreted at the verb (i.e. it is not ‘reconstructed’), a modifier cannot apply to its base position; semantically, there is nothing to modify. This leaves the problem of those modifiers which are acceptable with bare ne. We want to propose that they all contain a null pro-form which is able to directly pick up the antecedent of ne. Let’s begin with the di modifiers we saw in (50), repeated below as (63).

16 Alternatively, kinds might be the primitive meanings of nouns, as in Krifka (1995), Zamparelli (1996) – the choice is orthogonal to the present point. 17 According to Kayne’s (1975) approach to clitic movement, ne would initially move as a phrase. We leave open at this stage whether this phrase should be the DP subpart [NP ne], or the whole DP containing just ne.

The only real pro-nouns

|

129

(63) Di lucciole, ne ho viste (*due) di {verdognole / grosse / più lente} of fireflies, one(s) I_have seen (two) of {greenish / big / slower} proprio qui. right here The di we see in these cases cannot be the preposition di ‘of’ – prepositions do not take adjectives. Rather, we prefer to regard it as an instance of the ‘copular di’ we see in the Q-of-Adj construction in (64) (Zamparelli 1996, ch.3, den Dikken 1998). (64) a. Gianni berrebbe (un) qualcosa di fresco. Gianni would_drink (a) something of cool ‘Gianni would drink something cool’ b. Cosa vuoi di fresco? what you_want of cool? ‘Would you like something cool?’ Here, qualcosa ‘something’ is of the same category as ne (note that it can be preceded by an indefinite article, in (64a)), and di is obligatory. Both den Dikken (1998) and Zamparelli (1996) (1996) analyze it as a nominal equivalent of the copula, a particle identified with a KI(nd)P head in Zamparelli (1996) and with F in the present work. This particle takes a small-clause predicative structure whose subject, a pronominal empty category, moves to [Spec,di/of ] to be licensed. Di is the FP head. (65) [FP/NP proi di/of [SC t i Predicate]] Building on this analysis, we propose that in the acceptable modification cases the di-modifiers are not a stranded part of the V argument (which has been incorporated), but rather adjuncts. (66) [VP [DP/FP/NP ne]i +V [FP proi [F󸀠 di [SC t i Modifiers]] Turning to the ‘relative clauses’ we see with bare ne, they are actually pseudorelatives, the same structures we see in (67) applied to a proper name (type ⟨e⟩, just like bare ne). (67) Ho visto Dumbo che { volava vicino / *Marco osservava }. I_have seen Dumbo which { flew nearby / Marco observed } ‘I saw Dumbo {fly nearby / *Marco observed}’

130 | Michelangelo Falco and Roberto Zamparelli

As (67) also shows, this class of Italian modifiers must have a gap in subject position, a fact which has been linked to the presence of a pro as RC head (Guasti, 1988). They contrast with the ill-formed relative we saw in (49) above, which are genuine (object) relatives, cannot have pro heads and pattern as any other modifier.

5.4 “Ne” uses: a summary It may be useful, at this point, to schematically lay out the various ne data and point to their analysis. Let’s first consider the well-formed cases: (68) [FP/NP ne]i +V [DP 3 t i Indirect_modifiers t i ] cf. ex. (48) Ne interpreted in the base as a property, modified by indirect modifiers. Normal indefinite DP interpretation. (69) [FP/NP ne]+VToK [DP D0 tk ] Bare ne interpreted as a token-of-kind by a token-of-kind verb.

cf. ex. (60)

(70) [DP/FP/NP nei ]+V [FP proi di [SC t i modifiers] cf. ex. (63) Like above, ne interpreted at the verb, picked up by the empty pro-form in [Spec,di] Ill-formed cases. (71) * [FP/NP ne]i +V [DP D0 [FP Direct_Mods ti ]j Indirect_Mods tj ] Ne cannot strand Direct Modifiers (Section 3.2). (72) * [FP Direct_Mods [NP ne]]i +V [DP D0 tj Indirect_Mods tj ] Ne cannot be cliticized into V with its Direct Modifiers (see (34)). (73) * [FP/NP ne]+VToK [DP 3 tk ] Ne interpreted as a token-of-kind at V. Just like the modifiers, the numeral cannot apply to the trace position (wrong type). A final, interesting case to consider is the possibility for bare ne not to be interpreted by rule (60) (i.e. λy∃g[ ∪∩ ne(g) ∧ VToK (y,g)]), but rather ‘reconstructed’ in the base position, as it happens when an overt numeral is present. This would give rise to the normal existential interpretation of Italian bare plurals, but would also allow indirect modifiers without di.

The only real pro-nouns

|

131

(74) * [FP/NP ne]i +V [DP D0 ti j (Indirect modifiers)] Bare ne interpreted at the base as a property (ne(x)). The fact that this case is not found suggests that the ‘lexical’ interpretation in (61) is preferred. This could be seen as a blocking effect (an ‘elsewhere condition’: specific rules win over general ones), or perhaps as a function of the additional cost of positing a default existential quantifier to bind the variable of which ne is predicated in (74).

6 “Ne” antecedents Let’s now return to the issue of the forms that can be antecedents for ne in the CLLD constructions. Since ne is a pro-NP, the possibility to pick up bare nouns plus modifiers in topic position (as in e.g. [ragazzi con jeans]i , nei ho visti ‘[boys with jeans]i , one(s)i I have seen’), is expected. Less expected is the possibility of cases like (75a) (di+BN) and (b) (di+DP). (75) a. di ragazzo/ragazzi, ne ho visto/i (uno/due) of boy/boys, one(s) I_have seen (one/two) b. Di quei quattro ragazzi arrivati oggi, ne ho visti due of those four boys arrived today, one(s) I_have seen two As above, the di in (75a) cannot be the preposition di ‘of’: ragazzo is a bare singular count and prepositions do not productively select such complements in Italian¹⁸. We propose that these di+N cases are another instance of the ‘copular di’ construction seen above in (66), where (i), di is a functional head F (KI(nd)P in Zamparelli 1996, the edge of direct modification in the present approach), taking a predicational structure whose subject pro is moved and licensed in [Spec,di]; (ii) ragazzo is a property, which is applied to the pro internal subject, giving (76b). (76) a. [FP pro [F󸀠 di [SC t [NP ragazzo/i ]]]] b. λx[ragazzo󸀠 (x)] Since the whole constituent is an FP, ne can take it as its antecedent.

18 The only other productive case of P+N is found in pseudopartitives, but N must be plural count or singular mass in that construction. There is nothing ‘massy’ in ragazzo ‘boy’.

132 | Michelangelo Falco and Roberto Zamparelli

The situation with (75b), where the antecedent is a normal PP embedding a definite DP, is different. For these cases, we simply propose that the topicalized PP is actually picked up by the pro-PP ne (the one in (5) above), yielding a normal partitive (i.e. 2 of those 4 boys …). Indeed, replacing the definite with an indefinite such as alcuni ‘some’ yields ungrammaticality, as expected from a partitive: (77) * Di alcuni ragazzi, ne ho visti (due) of some boys, one(s) I_have seen (2) (cf. *“I saw two of some boys”).

7 Conclusions This work is, to our knowledge, the first detailed comparison of two pro-NP forms which have received comparatively little attention in the recent literature on proforms. Our study points to the conclusion that one and ne are actually quite similar (the only remaining lexical difference being, perhaps, the fact that ne can pick up both mass and count antecedents, while one can only pick up count ones). The differences that emerge, we argued, can be explained as the result of two factors: the fact that ne is a clitic (so, it cannot be extracted from ‘strong DPs’, and it incorporates into V), and the idea that bare nominals in English, but not in Italian, have to raise to [Spec,DP] and establish a relation with D0 which essentially turns them into pro-DPs. Unlike one(s), Italian ne can be bare, but in this case it is stripped of modifiers and restricted to verbs which select objects denoting non-specific ‘tokens of kinds’. We tried to give an informal characterization of this class, and we proposed a explanation which links the modifier restriction to the verb semantics. We have, however, just scratched the surface. Still missing is a precise characterization of the cases where one and ne simply do not appear overtly (as, arguably, in when he mixed the cards, two (*ones) were set aside), as well as a comparative analysis of their binding properties. We will discuss these issues in future work. Acknowledgment: For helpful comments, we are grateful to the audiences of the workshop Pronouns@Tübingen: The Impact of Pronominal Form on Interpretation, and of a seminar given at the Hungarian Academy of Sciences. The research reported in this paper is in part supported by funding from the Autonomous Province of Trento (Provincia Autonoma di Trento), Post-Doc 2011 grant to Michelangelo Falco for the project SPEC. The authors’ names are in alphabetical order.

The only real pro-nouns |

133

References Abney, S. P. (1987). The English Noun Phrase in its Sentential Aspect. Ph. D. thesis, MIT. Belletti, A. and L. Rizzi (1981). The syntax of ne: some theoretical implications. The Linguistic Review 1, 117–154. Borer, H. (2004). Structuring sense: In name only. Oxford: Oxford University Press. Burzio, L. (1986). Italian Syntax. A Government-Binding Approach. Hingham, Mass: Kluwer. Cardinaletti, A. and G. Giusti (1990). Partitive ne and the QP-hypothesis. Ms., University of Venice. Cardinaletti, A. and G. Giusti (1991). Partitive “ne” and the QP-hypothesis: a case study. In University of Venice Working Papers in Linguistics, Volume 1. University of Venice. Carlson, G. (1977). Reference to Kinds in English. Ph. D. thesis, University of Massachusetts at Amherst. Chierchia, G. (1998). Reference to kinds across languages. Natural Language Semantics 6, 339–405. Chomsky, N. (1986). Barriers. Cambridge, Mass: MIT Press. Chomsky, N. (2001). Derivation by phase. In M. Kenstowicz (Ed.), Ken Hale: a life in language, pp. 1–52. MIT Press. Cinque, G. (1990). Types of A-bar Dependencies. Cambridge, Massachusetts: MIT Press. Cinque, G. (1994). Partial N-movement in the Romance DP. In G. Cinque et al. (Eds.), Paths Towards Universal Grammar, pp. 85–110. Washington, D.C.: Georgetown University Press. Cinque, G. (2010). The Syntax of Adjectives. A Comparative Study. MIT Press. Contreras, H. (1986). Spanish bare NPs and the ECP. In I. Bordelois, H. Contreras, and K. Zagona (Eds.), Generative Studies in Spanish Syntax. Dordrecht: Foris. Cordin, P. (1988). Il clitico ‘ne’. In L. Renzi, G. Salvi, and A. Cardinaletti (Eds.), Grande grammatica italiana di consultazione, Volume I, pp. 647–655. Il Mulino. Dayal, V. (2004). Number marking and (in)definiteness in kind terms. Linguistics and Philosophy 27(4), 393–450. Déchaine, R.-M. and M. Wiltschko (2002). Decomposing pronouns. Linguistic Inquiry 33(3), 409–442. den Dikken, M. (1998). Predicate inversion in DP. In A. Alexiadou and C. Wilder (Eds.), Possessors, Predicates and Movement in the Determiner Phrase, Linguistik Aktuell, pp. 177–214. John Benjamins. Diesing, M. (1992). Indefinites. Cambridge, Mass: MIT Press. Elbourne, P. (2005). Situations and Individuals. Cambridge, Massachusetts: MIT Press. Enç, M. (1991). The semantics of specificity. Linguistic Inquiry 22.1, 1–25. Fiengo, R. and J. Higginbotham (1981). Opacity in NP. Linguistic Analysis 7, 395–422. Grohmann, K. K. (2003). Prolific Domains: On the Anti-Locality of Movement Dependencies. (Linguistik Aktuell 66). Amsterdam: John Benjamins Publishing Company. Guasti, T. (1988). La pseudorelative et les phenomenes d’accord. Rivista di grammatica generativa 13, 35–57. Heim, I. and A. Kratzer (1998). Semantics in Generative Grammar. Malden & Oxford: Blackwell. Heycock, C. and R. Zamparelli (2005). Friends and colleagues: Plurality, coordination, and the structure of DP. Natural Language Semantics 13, 201–270. Jackendoff, R. (1977). X-bar syntax: A study of phrase structure. Cambridge, Massachusetts: MIT Press.

134 | Michelangelo Falco and Roberto Zamparelli

Kayne, R. (1975a). French Syntax: The Transformational Cycle. Cambridge, Massachusetts: MIT Press. Kayne, R. (1994). The Antisymmetry of Syntax. Cambridge, Mass: MIT Press. Kayne, R. (2005, June). Several, few and many. New York University. Kayne, R. S. (1975b). French syntax: the Transformational Cycle. Cambridge, MA: MIT Press. Krifka, M. (1995). Common nouns: a contrastive analysis of English and Chinese. In G. Carlson and F. Pelletier (Eds.), The Generic Book, pp. 398–411. Chicago, IL: University of Chicago Press. Link, G. (1983). The logical analysis of plurals and mass terms: a lattice-theoretical approach. In R. Bäuerle, C. Schwarze, and A. von Stechow (Eds.), Meaning, Use, and the Interpretation of Language, pp. 302–323. Berlin/New York: de Gruyter. Longobardi, G. (1994). Reference and proper names: a theory of N-movement in syntax and logical form. Linguistic Inquiry 25, 609–665. Longobardi, G. (2000). “Postverbal” subjects and the mapping hypothesis. Linguistic Inquiry 31(4), 691–702. Longobardi, G. (2001). How comparative is semantics? A unified parametric theory of bare nouns and proper names. Natural Language Semantics 9(4), 335–369. Longobardi, G. (2008). Reference to individuals, person, and the variety of mapping parameters. In A. Klinge and H. Müller (Eds.), Essays on Nominal Determination, pp. 189–211. John Benjamins. Milsark, G. (1974). Existential Sentences in English. Ph. D. thesis, MIT, Cambridge, Mass. Moro, A. (1997). The Raising of Predicates: Predicative Noun Phrases and the Theory of Clause Structure. Cambridge: Cambridge University Press. Postal, P. (1966). On so-called ‘pronouns’ in English. In F. Dinneen (Ed.), Monograph Series on Language and Linguistics, Volume 19, pp. 177–206. Georgetown University Press. Rizzi, L. (1986). Null objects in Italian and the theory of pro. Linguistic Inquiry 17.3, 501–557. Rizzi, L. (2001). Relativized minimality effects. In M. Baltin and C. Collins (Eds.), The Handbook of Contemporary Syntactic Theory, pp. 89–110. Oxford: Blackwell. Ross, J. R. (1967). Constraints on Variables in Syntax. Bloomington: Indiana University Linguistics Club. Scott, G.-J. (2002). Stacked adjectival modification and the structure of nominal phrases. In G. Cinque (Ed.), Functional Structure in DP and IP. The Carthography of Syntactic Structures, Volume 1. Oxford University Press. Sproat, R. and C. Shih (1990). The cross-linguistics distribution of adjective ordering restrictions. In C. Georgopoulos and I. R. (Eds.), Interdisciplinary approaches to language: essays in honor of Yuki Kuroda, pp. 565–593. Dordrecht: Kluver. Williams, E. (1980). Predication. Linguistic Inquiry 11.1, 203–238. Zamparelli, R. (1996). Layers in the Determiner Phrase. Ph. D. thesis, University of Rochester. Zamparelli, R. (2002). Definite and bare kind-denoting noun phrases. In C. Beyssade, R. BokBennema, F. Drijkoningen, and P. Monachesi (Eds.), Romance Languages and Linguistic Theory 2000; selected papers from ‘Going Romance’ 2000, Utrecht, 30 November–2 December, Volume 232, Amsterdam/Philadelphia, pp. 305–342. John Benjamins.

Ewa Trutkowski and Helmut Weiß (Goethe-University of Frankfurt am Main)

When personal pronouns compete with relative pronouns Abstract: Our paper deals with relative clauses (RCs) headed by 1st /2nd person pronouns and it focuses on two mismatches: the first between the pronominal head noun and (the obligatorily 3rd person marked) relative pronoun, and the second – in case a resumptive pronoun is present – between the relative and resumptive pronoun. With respect to the surface it seems that there are three possibilities for the finite verb in an RC to agree with: (i) with the pronominal head noun, (ii) with the resumptive pronoun, or (iii) with the relative pronoun, cf. Ich, der (ich) malen lerne/lernt (I, who (I) to-paint learn-1SG/-3SG). We will present an account for different agreement types in RCs (mainly within the scope of German(ic)), supported by two experimental surveys. Besides, we will show that the different agreement patterns can also be semantically determined. Finally, we will provide a syntactic analysis which explains the syntax of RCs in German dialects in general. Keywords: German, NRRCs, Agreement, Mismatches, Binding, Structure of (Relative) Pronouns, Left Periphery of RCs, Resumption

1 The phenomenon The focus of this paper is on subject relative clauses, in particular on nonrestrictive relative clauses (NRRCs) where the pronominal head noun (HN) and relative pronoun (RP) are both nominative case marked and the pronoun which acts as the head of the RC is marked for 1st or 2nd person (singular/plural).¹ As (1) shows, in such a situation there are three possibilities for the finite verb in an RC

1 A lot of authors assume that relative clauses and clefts are structurally very similar, cf. already Schachter (1973), Chomsky (1977). The main argument that (in our view) speaks against a unification of RCs and clefts, is that clefts (in contrast to RCs) do not allow the insertion of a resumptive pronoun – hence they do not allow HN agreement (possibly, the latter fact is caused by the first), cf. (i) and (ii): (i)

a.

Weil ich es bin, der (*ich) die ganze Arbeit mache. Because I it am, who-MASC-SG (I) the whole work do-1SG

*Cleft

136 | Ewa Trutkowski and Helmut Weiß

to agree with: Either with the ‘RC-internal’ RP as in (1a), with the ‘RC-external’ HN, cf. (1b), or with the ‘RC-internal’ resumptive pronoun (which we abbreviate as ResP), cf. (1c):² (1)

a. ( *) ich, der sechzig ist, … I, who-MASC-SG sixty is sechzig bin, … b. ( *) ich, der I, who-MASC-SG sixty am c. ich, der ich sechzig bin, … I, who-MASC-SG I sixty am

(RP agreement) (HN agreement) (ResP agreement)

Note that the grammaticality judgements in (1) are from Ito & Mester (2000); brackets indicate that they depart from our own judgements. The data in (1) show that independently of the chosen agreement pattern, a mismatch is unavoidable if one assumes that an RP in German has 3rd person features (contra claims of Benveniste 1971 and many others): either we have (i) disagreement between the pronominal 1st /2nd person marked HN and the 3rd person marked RP or (ii) we have disagreement between the 3rd person marked RP and the 1st /2nd person marked ResP (+ HN). According to these mismatches, Ito & Mester (2000), who first described this phenomenon within the generative literature (in an unpaged squib), postulate ungrammaticality when no ResP is present. They say:

b.

(ii)

ich, der ich die ganze Arbeit mache I, who-MASC-SG I the whole work do-1SG

✓RC

*Cleft a. * Weil er es ist, der er Volljurist ist. Because he it is, who-MASC-SG he fully-qualified-lawyer is b. Aber was macht er, der er immerhin Volljurist und seit But what does he, who-MASC-SG he after-all fully-qualified-lawyer is and for ✓RC Jahrzehnten in der Politik aktiv ist? decades in the political-world active is? (Original datum from: Spiegel-Online, 2.2.2012)

In the following, we will only discuss RCs, just to avoid mixing up two constructions which may possibly be unrelated (contrary to Heck & Cuartero 2008). 2 It can be assumed that the HN agreement in (1c) is caused by the ResP and not by the HN – otherwise we would expect a sentence as Ich, der ich sechzig ist (I, who-MASC-SG I sixty is) to be fine, contrary to fact.

When personal pronouns compete with relative pronouns |

137

First and second person pronominal heads demand corresponding agreement morphology on the verb, i.e., default third person agreement is not permitted. At the same time, first and second person agreement morphology on the verb demands the presence, within the same clause, of a corresponding subject to agree with. This problem is resolved by repeating the pronoun that serves as the head of the relative as an internal subject precisely in such cases, i.e., precisely when needed to support non-third person singular agreement on the verb.

According to Ito & Mester (2000) an agreement mismatch can only be avoided when the form of the finite verb in the RC is syncretic (between 1st and 3rd person). Specifically, they state that in such cases the insertion of a ResP is not obligatory,³ cf. “both forms are acceptable (even though the one with the additional wir⁴ sounds somewhat better⁵)”, cf. (2): (2) a. ich, der ich alles weiß/kann, … I, who-MASC-SG I everything know/can-1/3SG b. ich, der alles weiß/kann, … I, who-MASC-SG everything know/can-1/3SG c. wir, die wir sechzig sind, … we, who-PL we sixty are-1/3PL d. wir, die sechzig sind, … we, who-PL sixty are-1/3PL However, as the following (corpus/internet) data⁶ suggest, insertion of a ResP does not seem to be obligatory, even when the form of the finite verb is nonsyncretic:⁷

3 Trutkowski, Grewendorf & Weiß (2013) and Trutkowski (in prep.) found that syncretic forms do not improve “mismatched” structures (contra Ito & Mester 2000). For reasons of space we do not report this experiment. 4 This statement relates to NRRCs where the head noun is a 1st person plural pronoun. 5 If Ito & Mester are right, and the use of syncretic forms neutralises any possible mismatch, it is unclear why an additional ResP should be able to improve the structure in any way. 6 The examples in (3)–(8) were collected on the internet in June 2013. 7 Note that in German the forms of the 1st and the 3rd person plural always coincide, whereas the forms of the 1st and the 3rd person singular are only syncretic in some conjugation classes. Further syncretisms are found in the conjugation class of weak verbs, namely between the 3rd person singular and the 2nd person plural. Besides, in verbs whose stem ends with an /s/, we find syncretisms between the 2nd and the 3rd person singular (and the 2nd person plural), cf. e.g. motzen (to grumble), küssen (to kiss) or rasen (to rush).

138 | Ewa Trutkowski and Helmut Weiß

RP agreement (parallel to (1a)) (3) Ich, die sich ihr Leben allein aufgebaut hat, habe I, who-FEM-SG REFL her life alone arranged have-3SG, have-1SG endlich jemanden⁸ finally someone (4) Das fragst gerade du, der sich nicht einmal traut, That ask especially you-SG, who-MASC-SG REFL not even dare-3SG, mit seinem Kind offene Gespräche über die Gesellschaft zu führen⁹ with his child open conversations about the society to lead (5) Ihr, die sich schon lange auf diesen Zeitpunkt vorbereitet You-PL, who-PL REFL since long for this moment prepared haben, werdet schon im Innern gefühlt haben, was vor sich geht¹⁰ have-3PL, will-2PL already in inside felt have, what PRT REFL goes (= ‘…what is going on’) HN agreement (parallel to (1b)) (6) Und ich, der noch nie etwas über ebay gekauft habe¹¹ And I, who-MASC-SG yet never something via ebay bought have-1SG (7) Du, der heute den noch siehst, der uns’re Wege lenkt¹² You-SG, who-MASC-SG today him still see-2SG, who our ways direct-3SG (8) Ihr, die aus diesen abscheulichen Gemeinden nicht herauskommen You-PL, who-PL of these despicable parishes not to-get-out wollt¹³ want-2PL Thus, (3)–(8) suggest the following: First, the non-insertion of a ResP does not automatically lead to an ungrammatical structure. Secondly, HN/RP-agreement in

8 http://www.superillu.de/zeitvertreib/kinotv/sylvia-leifheit-traumhochzeit-auf-schlossmirabell-die-highligts-auf-superillu 9 http://www.tacheles-sozialhilfe.de/forum/thread.asp?FacId=1839934 10 http://paoweb.org/download/channel/uriel/engel_uriel_13.02.12.pdf 11 http://schmerzwach.blogspot.de/2011/02/freunde-mal-drei.html 12 http://www.reinhard-mey.de/start/texte/alben/schade-da%C3%9F-du-gehen-mu%C3%9Ft 13 http://www.mcreveil.org/Allemand/journaux/german03.htm

When personal pronouns compete with relative pronouns |

139

1st /2nd person NRRCs may be subject to free variation. However, how free is agreement variation in NRRCs headed by 1st /2nd person pronouns? And how obligatory is the insertion of a resumptive pronoun? By conducting a Google-search for Ich, der noch nie (I, who-MASC-SG yet never) we found a strong preference for RP agreement, namely after 100 results: 98 matches which involve RP agreement and only 2 matches involving HN (without ResP) agreement. Moreover, for NRRCs with a 1st person singular HN DUDEN (online)¹⁴ (a German grammar reference book) allows either for HN+ResP agreement as in (9), or for sole RP agreement, cf. (10). DUDEN (online) does not offer ‘pure’ HN agreement, without an additional ResP: (9) ich, der ich mich immer um Ausgleich bemühe I, who-MASC-SG I me-1SG-ACC (= REFL) always for balance strive-1SG (10) ich, der sich immer um Ausgleich bemüht I, who-MASC-SG REFL always for balance strive-3SG Ito & Mester (2000) as well as Vogel (2007/8) maintain that resumptive pronoun insertion is a necessary condition for 1st /2nd person NRRCs (given the fact that the finite verb within the RC (abbreviated as V.fin-Rel) is non-syncretic). Other authors accept 1st /2nd person NRRCs without ResP insertion: E.g. Heck & Cuartero (2008) argue that in the singular 1st /2nd person NRRCs display (what we call) RP agreement, whereas plural 1st /2nd person NRRCs display (what we call) HN agreement. By assuming this claim to be correct (which is coherent with the DUDEN preference rendered in (9) and (10)), we would predict the examples in (5), (6) and (7) to be ungrammatical, or at least degraded.¹⁵ Thus, the questions which arise from the empirical facts that we have presented are: Is there a (singular/plural) preference for one over the other agreement pattern, and – if so – for which one and why? Why are the mismatches described above tolerated, i.e., what can 1st /2nd person NRRCs tell us about the general properties of (relative) pronouns and relative complementisers (as e.g. wo in nonstandard and dialect uses) and what does this mean for a syntactic analysis of RCs/NRRCs? Before discussing these questions, we will present two experiments that address the agreement patterns in NRRCs headed by 1st /2nd person pronouns.

14 Examples (9) and (10) are taken from DUDEN online, cf. http://www.duden.de/rechtschrei bung/ich 15 However, note that the speakers of (6)–(8) cannot be qualified as low performers of German, e.g. (7) is from a famous German singer-songwriter.

140 | Ewa Trutkowski and Helmut Weiß

1.1 Experimental Investigations of the Empirical Data In order to get reliable grammaticality judgements we conducted some Magnitude Estimation experiments in collaboration with Markus Bader (U Frankfurt/Main). Two of these experiments will be presented below. In all experiments we used 24 × 6 test sentences + fillers (from other experiments). The reference sentence in all experiments was (i) Ich glaube, dass den Bericht der Chef in seinem Büro gelesen hat (= I think, that the report-ACC the boss-NOM in his office read has). A positive number indicated “more acceptable than the reference sentence”, a negative number indicated “less acceptable than the reference sentence”.¹⁶ We tested the following hypotheses from the literature (Ito & Mester 2000, Vogel 2007/8, Heck & Cuartero 2008), cf. (11a-c): (11) Hypotheses with respect to Experiment 1 (Cf. agreement patterns in RCs with 1st /2nd person singular pronominal HNs) – Ito & Mester (2000), Vogel (2007/8): (a) NRRCs headed by a 1st /2nd person (singular) pronoun are only wellformed when the RC contains an overt ResP (modulo data with syncretic verbal inflectional endings). – Heck & Cuartero (2008), introspective judgements and reference to corpus/internet data: (b) The insertion of a ResP is not a necessary condition for the well-formedness of NRRCs which are headed by a 1st /2nd person (singular) pronoun. (c) In NRRCs which are headed by a 1st /2nd person singular pronoun, RP agreement is preferred over ‘pure’ HN agreement (without additional ResP). The test items of Experiment 1 are exemplified in (12) and (13), where (a) represents the ResP agreement pattern, (b) the HN agreement pattern, and (c) the RP agreement pattern:

16 Filler sentences were e.g. the following, cf. (i) and (ii), with respective values: (i)

Der Opa hat gesagt, dass das Buch ihn schon erfreut hat. The grandpa-NOM has said that the book-NOM him-ACC indeed delighted has [Value = 0.23]

(ii)

Der Opa hat gesagt, dass das Buch schon ihn erfreut hat. The grandpa-NOM has said that the book-NOM indeed him-ACC delighted has [Value = 0.04]

When personal pronouns compete with relative pronouns |

141

(12) a. Ich, die ich malen lerne, besuche jede Ausstellung. I, who-FEM-SG I to-paint learn-1SG, visit-1SG every exhibition b. Ich, die malen lerne, besuche jede Ausstellung. I, who-FEM-SG to-paint learn-1SG, visit-1SG every exhibition c. Ich, die malen lernt, besuche jede Ausstellung. I, who-FEM-SG to-paint learn-3SG, visit-1SG every exhibition (13) a. Du, die du malen lernst, besuchst jede You-SG, who-FEM-SG you-SG to-paint learn-2SG, visit-2SG every Ausstellung. exhibition b. Du, die malen lernst, besuchst jede Ausstellung. You-SG, who-FEM-SG to-paint learn-2SG, visit-2SG every exhibition c. Du, die malen lernt, besuchst jede Ausstellung. You-SG, who-FEM-SG to-paint learn-3SG, visit-2SG every exhibition Within the first experiment we asked 36 student subjects who were native speakers of German for their judgements. We used a 2 × 3 design where we manipulated two factors: agreement type (ResP vs. HN vs. RP agreement) and person features of the HN in the matrix clause (1st vs. 2nd person). The results of the first experiment were the following: (14) Main clause HN (pronoun) Agreement pattern within the RC

1st person singular

2nd person singular

ResP agreement (‘pure’) HN agreement RP agreement

0.129 (as 12a) 0.928 (as 12b) 0.141 (as 12c)

0.099 (as 13a) 0.295 (as 13b) 0.116 (as 13c)

On the basis of the results in (14) we can state that we do not have a significant difference between ResP agreement and RP agreement. Moreover, HN agreement is the least preferred agreement pattern for NRRCs with 1st and 2nd person singular pronominal head nouns. This is summarised in (15): (15) Results of Experiment 1 (regarding NRRCs with 1st and 2nd person singular HNs) (a) The presence of a ResP does not improve the acceptability of the structure. (b) RP agreement is preferred over ‘sole’ HN agreement.

142 | Ewa Trutkowski and Helmut Weiß

On the basis of our experimental investigation we can thus confirm Heck & Cuartero (2008) and our intuitive judgments formulated in (11b) and (11c), respectively (which are also implicitly formulated by the DUDEN (online) grammar). Besides, we can reject Ito & Mester’s (2000) and Vogel’s (2007/8) hypothesis given in (11a). In the next Magnitude Estimation experiment we tested which agreement patterns are preferred by NRRCs headed by (1st /)2nd person plural pronouns. As the inflectional endings of the 1st and the 3rd person plural are always syncretic in German, it does not make sense to test 1st person plural pronouns (at least not in the experimental design given here). In Experiment 2 we used a 2 × 3 design where we manipulated two factors: agreement type (ResP vs. HN vs. RP agreement) and number features of the pronominal HN in the matrix clause (2nd person singular vs. 2nd person plural). The hypotheses that we wanted to examine are given in (16), the items that we tested in 24 versions are represented in (17) and (18), and the results of this experiment (in which we asked 42 students who were native speakers of German for their judgements) are summarised in (19) and (20): (16) Hypotheses with respect to Experiment 2 (Cf. agreement patterns in RCs with 2nd person singular/plural pronominal HNs) – Heck & Cuartero (2008), introspective judgements and reference to corpus/internet data: (a) The insertion of a ResP is not a necessary condition for the well-formedness of NRRCs which are headed by a 2nd person singular/plural pronoun. (b) In NRRCs which are headed by a 2nd person singular pronoun, RP agreement is preferred over ‘pure’ HN agreement (without additional ResP). (c) In NRRCs which are headed by a 2nd person plural pronoun, ‘pure’ HN agreement (without additional ResP) is preferred over RP agreement. (17) a. Du, die du malen lernst, besuchst jede You-SG, who-FEM-SG you-SG to-paint learn-2SG, visit-2SG every Ausstellung. exhibition b. Du, die malen lernst, besuchst jede Ausstellung. You-SG, who-FEM-SG to-paint learn-2SG, visit-2SG every exhibition c. Du, die malen lernt, besuchst jede Ausstellung. You-SG, who-FEM-SG to-paint learn-3SG, visit-2SG every exhibition (18) a. Ihr, die ihr malen lernt, besucht jede Ausstellung. You-PL, who-PL you-PL to-paint learn-2PL, visit-2PL every exhibition

143

When personal pronouns compete with relative pronouns |

b. Ihr, die malen lernt, besucht jede Ausstellung. You-PL, who-PL to-paint learn-2PL, visit-2PL every exhibition c. Ihr, die malen lernen, besucht jede Ausstellung. You-PL, who-PL to-paint learn-3PL, visit-2PL every exhibition (19) Main clause HN (pronoun) Agreement pattern within the RC

2nd person singular

2nd person plural

ResP agreement (‘pure’) HN agreement RP agreement

0.113 (as 17a) 0.673 (as 17b) 0.124 (as 17c)

0.167 (as 18a) 0.164 (as 18b) 0.032 (as 18c)

The results in (19) and (20) confirm the hypotheses in (16): (20) Results of Experiment 2 (regarding NRRCs with 2nd person singular/plural HNs) (a) The presence of a ResP is not a necessary condition on NRRCs headed by 2nd person singular/plural pronouns. (b) In NRRCs which are headed by a 2nd person singular pronoun, RP agreement is preferred over ‘sole’ HN agreement. (c) In NRRCs which are headed by a 2nd person plural pronoun, ‘sole’ HN agreement is preferred over RP agreement. I.e., the two experiments show that agreement patterns of NRRCs with 1st /2nd person singular/plural HNs are complementarily distributed: whereas the preferred pattern of NRRCs with singular HNs is RP agreement, the preferred pattern of NRRCs with plural HNs is HN agreement. ResP insertion does not improve the structure. The tables in (21) and (22) report the overall results in form of ANOVAs: (21) ANOVA

Main clause Relative clause Main clause : Relative clause Experiment 1 (36 speakers)

Subject Analysis

Item Analysis

F1

Df1

Df2

p

F2

Df1

Df2

p

4.814 3.966 0.675

1 2 2

35 70 70

0.035 0.023 0.512

2.989 3.182 0.411

1 2 2

23 46 46

0.097 0.006 0.665

144 | Ewa Trutkowski and Helmut Weiß

(22) ANOVA

Pronoun Relative clause Pronoun : Relative clause

Subject Analysis

Item Analysis

F1

Df1

Df2

p

F2

Df1

Df2

p

0.015 8.668 21.355

1 2 2

41 82 82

0.903 0.000 0.000

0.022 7.755 25.952

1 2 2

23 46 46

0.883 0.001 0.000

Experiment 2 (42 speakers)

Note that from a theoretical point of view it is questionable whether the finite verb within a non-restrictive relative clause agrees with the plural HN of the matrix clause (see also Heck & Cuartero 2008). Moreover, it is doubtful whether (relative) pronouns bear 3rd person features instead of being underspecified for (1st or 2nd ) person (thus constituting a non-person, as already suggested by Benveniste 1971; however see Trommer 2008 for contradicting evidence). Later, we will discuss how to resolve these theoretical challenges. As for now, we will present some binding data which confirm the findings of our experiments on a semantic basis.

2 Binding in the Context of 1st /2nd Person NRRCs In this chapter we will show that agreement patterns within NRRCs with 1st /2nd person pronominal HNs can alternatively be determined on a semantic basis. As (in German) reflexive and possessive pronouns inflect for person, we can consider them as co-agreeing elements that have to adjust to the dominant agreement patterns found in NRRCs with 1st or 2nd person pronominal heads. Thus, in the following, we will present NRRCs with 1st /2nd person HNs that contain “bindable” elements. The idea is that reflexive/possessive pronouns can only be bound when agreement chains are not interrupted, cf. Kratzer (2009). I.e., we will use binding contexts as ‘lackmus tests’ to see which agreement patterns are preferred. Binding should only be possible under the “correct” pattern. We will first look at NRRCs in which reflexive pronouns are present, and in a further step at NRRCs that contain possessive pronouns.

2.1 Reflexive Pronoun Binding In contrast to languages as e.g. Polish, cf. (23), German has no uniform reflexive pronoun for all persons and numbers (and genders). Instead, object pronouns (with person, number and case specifications) function as 1st and 2nd person re-

When personal pronouns compete with relative pronouns |

145

flexive pronouns, cf. (24). For the 3rd person singular/plural the reflexive sich¹⁷ is used, cf. (25): (23) Ja *mnie / ✓się wstydzę. I me-ACC / REFL shame-1SG (24) Ich schäme ✓mich / *sich. I shame-1SG me-ACC / REFL

Polish

German

(25) a. Hans rasiert sich. Hans shave-3SG REFL b. Die Schüler rasieren sich. The pupils shave-3PL REFL Reflexive pronouns/reflexively used object pronouns can occur with inherently reflexive verbs (where they cannot be substituted by a proper argument), cf. (26), and with non-inherently reflexive verbs (where they can be substituted by a proper argument), cf. (27): (26) Ich schäme ✓mich / *den Hans. I shame-1SG me-ACC / the Hans-ACC (27) Ich rasiere ✓mich / ✓den Hans. I shave-1SG me-ACC / the Hans-ACC Clearly, when the reflexively used object pronoun is an improper argument it must be bound in order to be interpreted (under phi-feature sharing with its binder/the head that hosts its binder, cf. also Kratzer’s 2009 notion of an ‘agreement chain’). However, when the reflexively used pronoun can be substituted by a proper argument, it does not have to be bound in order to be interpreted. I.e., in cases where the finite verb is not inherently reflexive, the relation between the subject pronoun and the (reflexively used) object pronoun is ambiguous between a binding and a coreference relation when the possible binder and bindee have/share identical features – (28) is such an example:¹⁸ (28) Ich rasiere mich. I shave-1SG me-ACC

17 Note that in (25b) sich can also have a reciprocal reading (where the pupils shave each other). 18 See Harley & Ritter (2002: 487) for similar data (contrasts) in English.

146 | Ewa Trutkowski and Helmut Weiß

When we consider an NRRC as in (29), that means: the binder (which may be the pronominal HN), the bindee (= the reflexive/reflexively used object pronoun) and the V.fin-Rel (which is the verbal head that hosts the binder) must share relevant features. Under these conditions the following example involves binding: (29) (?) ich, die mich rasiere¹⁹ I who-FEM-SG me-ACC shave-1SG That examples which are similar to (29) may involve binding was already pointed out by Kratzer (2009). However, the “problem” with such data is that they get very marginal when we use an inherently reflexive verb, cf. (30) (instead of a noninherently reflexive one as in (29)): (30) ?* ich, die mich schäme I who-FEM-SG me-ACC shame-1SG I.e., in contexts where reflexive pronouns (which constitute co-agreeing elements) must be bound in order to be interpreted, “incorrect” agreement patterns yield ungrammatical (or degraded) outputs, cf. also (31):²⁰ (31) * du, die dich schämst you-SG who-FEM-SG you-SG-ACC shame-2SG The above data suggest that we can use binding in inherently reflexive verb contexts to determine the correct agreement patterns for NRRCs that are headed by 1st /2nd person pronouns. That binding is only possible when agreement patterns are the “correct” ones and agreement chains are not interrupted, is further evidenced by verbs that are ambiguous between a +/−inherently reflexive meaning, cf. (sich) verschlucken (to choke/to swallow): when the agreement chain is not interrupted and (the verbal head that hosts) the binder has the same feature specifications as the object pronoun, the latter is obligatorily bound and the verb is interpreted as inherently reflexive, cf. (32a). However, the object pronoun cannot be bound (and has to be interpreted in coreference with the pronominal HN) when it does not share the same feature specifications as the (verbal head that hosts its) binder, cf. (32b). In (32c) the object pronoun is pseudo-bound, because the

19 Note that irrespective of a proper binding configuration, HN agreement is somehow degraded for singular head (pro)nouns, cf. the experimental findings above. 20 Note that we do not yet have an explanation, why in the singular 2nd person HN agreement is more degraded than 1st person HN agreement.

When personal pronouns compete with relative pronouns |

147

agreement pattern is not the intended/correct one. As a consequence, the verb can hardly display an inherently reflexive meaning (actually, (32b) and (32c) mean that ‘I have swallowed a person, who is me’ (comparable to a situation in which a cat is chasing her tail)):²¹ (32) a.

ich, die sich verschluckt hat I, who-FEM-SG REFL choked have-3SG (✓ ‘to choke’/?? ‘to swallow’) b. ich, die mich verschluckt hat I, who-FEM-SG me-ACC swallowed have-3SG (* ‘to choke’/✓ ‘to swallow’) c. (?) ich, die mich verschluckt habe I, who-FEM-SG me-ACC swallowed have-1SG (?? ‘to choke’/✓ ‘to swallow’)

Due to the fact that the inherently reflexive meaning of the verb verschlucken is hardly an option in (32b/c), we can conclude that it is not only (i) a fully intact agreement chain (cf. Kratzer 2009) but also (ii) a syntactically correct agreement pattern which enables binding (that is, as the experiments have shown, RP agreement in the case of NRRCs with 1st /2nd person singular HNs, and HN agreement in the case of NRRCs with 1st /2nd person plural HNs). We will refer to ‘binding’ instances when conditions (i) and (ii) are fulfilled (cf. (32a)) and to ‘pseudo-binding’ instances when condition (i) is fulfilled but condition (ii) is not (cf. (32c)). Note that in sentences that contain reflexive pronouns, the ResP strategy yields structures which are quite difficult to parse (which may be a reason for the fact that it is not preferred over agreement patterns without ResP). Moreover, even when the inherently-reflexive meaning of verschlucken is present in examples as (33a, b), they remain ambiguous between an inherently reflexive and a non-inherently reflexive meaning (ResPs are underlined in (33)): (33) a. ich, die ich mich verschluckt habe I, who-FEM-SG I me-ACC choked/swallowed have-1SG b. ihr, die ihr euch verschluckt habt you-PL, who-PL you-PL you-PL-ACC choked/swallowed have-2PL Thus, the above binding data constitute additional and independent evidence for the (experimentally surveyed) acceptability judgements with respect to de-

21 This minimal pair was brought to us by Andreas Pankau.

148 | Ewa Trutkowski and Helmut Weiß termining the agreement patterns of NRRCs which are headed by 1st /2nd person pronouns. In the next section we will present NRRC data that contain possessive pronouns which (also) act as co-agreeing elements.

2.2 Possessive Pronoun Binding As will be seen, the data below confirm the agreement patterns and the binding conditions formulated above: Again, binding a possessive within an NRRC headed by a 1st /2nd person pronoun is only possible when (i) (the verbal head that hosts) the binder and the possessive have the same feature specifications and (ii) the “correct” agreement pattern is obeyed. In (34) the same factors (with the respective factor levels) that we envisaged in Experiment 1 are compared in the context of NRRCs that contain a possessive pronoun. As the following minimal pairs show, binding is only possible when the bound element (the possessive pronoun) and the verbal head that hosts the binder (the finite verb) share the same feature specifications (cf. Kratzer 2009). According to this, only (34a) and (34d) are possible candidates for structures that involve binding, however, (34a) is degraded. In (34b) and (34c) the possessive and the verbal head have different feature specifications, hence binding is not an option. Analogous statements can be made with respect to the 2nd person data in (35): (34) a. (?) ich, die meine Oma besuche I, who-FEM-SG my grandma visit-1SG b. ich, die meine Oma besucht I, who-FEM-SG my grandma visit-3SG c. (?) ich, die ihre Oma besuche I, who-FEM-SG her grandma visit-1SG d. ich, die ihre Oma besucht I, who-FEM-SG her grandma visit-3SG (35) a. ? du, die deine Oma besuchst you-SG, who-FEM-SG your grandma visit-2SG b. du, die deine Oma besucht you-SG, who-FEM-SG your grandma visit-3SG c. ? du, die ihre Oma besuchst you-SG, who-FEM-SG her grandma visit-2SG d. du, die ihre Oma besucht you-SG, who-FEM-SG her grandma visit-3SG

(pseudo-binding/ coreference) (only coreference) (no binding) (binding possible)

(pseudo-binding/ coreference) (only coreference) (no binding) (binding possible)

When personal pronouns compete with relative pronouns |

149

Interestingly, co-agreeing possessive pronouns open a window for NRRCs headed by 1st person plural pronouns. As verbal inflectional endings of the 1st and the 3rd person plural are always syncretic in German, we could not yet determine whether the 1st person plural displays (3rd person plural) RP agreement or (1st person plural) HN agreement. However, now we can observe that a 3rd person plural possessive pronoun can hardly receive a bound reading in a syntactic configuration as (36a) – instead, to receive a bound reading, we have to use a 1st person plural possessive pronoun, which is a hint at the supposition that 1st person plural HNs have to obey HN agreement in NRRCs, in analogy to 2nd person plural HNs, cf. (36b): (36) a. wir, die ihre Oma besuchen we, who-PL their grandma visit-1/3PL b. wir, die unsere Oma besuchen we, who-PL our grandma visit-1/3PL

(no binding) (binding possible)

(37) summarises the additional evidence which we hoped to give by presenting the above binding data: (37) Binding within NRRCs headed by 1st /2nd person pronouns Reflexively used object pronouns and possessive pronouns can only be properly bound within NRRCs headed by 1st /2nd person pronouns when (i) occurring in a fully intact agreement chain and (ii) under the “syntactically correct” agreement pattern. Otherwise they remain unbound, pseudo-bound or coreferent (cf. above for details). Obviously, we still do not know, by which conditions/feature specifications the “syntactically correct” agreement pattern is determined, and what “correct” means. Moreover, we have to specify the features which determine HN agreement and RP agreement, respectively. This will be done in the next chapter.

3 Why do Singular and Plural HNs Display Different Agreement Patterns? In order to find out more about the singular/plural agreement split in NRRCs headed by 1st /2nd person pronouns, we will look at differences with regard to feature specifications of the pronominal head nouns and the relative pronouns

150 | Ewa Trutkowski and Helmut Weiß

in the singular and plural, respectively. Let us first assume (following Benveniste 1971, Siewierska 2004 and many others) that 3rd person marking indicates the lack of 1st or 2nd person marking, i.e. the 3rd person is used to mark a “non-person”, i.e. a person that is neither speaker nor hearer. In the following, we will further decompose the features of the pronominal HN and the RP, respectively, aiming at diminishing the amount of features to a minimum.²² Firstly, we suggest that Number features are default-like absent/negatively specified in the singular and only existent/positively specified in the plural. Moreover, we assume that 1st /2nd person pronouns referring to singular individuals have (inherent) Gender features (contra e.g. Sternefeld 2006: 151). I.e., we assume that Gender assignment is part of reference assignment to indexical pronouns, preferably in the sense of Kaplan (1989a, b). However, in our case Gender is not marked at the character, but (only) part of the content (As to character and content see Kaplan 1989a, b²³).²⁴ Under this assumption 1st and 2nd person singular/plural HNs and RPs have the following feature specifications, cf. (38a, b): (38) a. Feature specifications of pronominal HN and RP (German) – Singular

HN features

RP features

1-SG

2-SG

+Speaker −Addressee

−Speaker +Addressee

−Number +Gender

−Number +Gender

−Speaker −Addressee

−Speaker −Addressee

−Number +Gender

−Number +Gender

22 Being aware of the literature on feature decomposition, as e.g. Harley & Ritter (2002) or the contributions in Harbour, Adger & Bejar (2008), we will omit a (detailed) discussion of this topic. Instead, we will refer to standardly assumed notions and features, similar to Kratzer (2009), but minimizing her repertoire where possible and enriching it where we find it necessary. 23 One can sloppily say that the content of an indexical refers to the referent itself, cf. Kaplan (1989a: 502). Moreover, “[c]ontent is generated in a context, and each context is associated with a particular possible world. The agent, time, place are all drawn from that world.” (Kaplan 1989b: 591) “The character of an expression is set by linguistic conventions and, in turn, determines the content of the expression in every context.”, e.g. “ ‘I’ refers to the speaker or writer” (Kaplan 1989a: 505). 24 Note that we are aware of the fact that Gender is seldom realised at the character of an expression, cf. Siewierska (2004: 105) for some typological discussion.

When personal pronouns compete with relative pronouns |

151

b. Feature specifications of pronominal HN and RP (German) – Plural

HN features

RP features

1-PL

2-PL

+Speaker −Addressee

−Speaker +Addressee

+Number −Gender

+Number −Gender

−Speaker −Addressee

−Speaker −Addressee

+Number −Gender

+Number −Gender

The tables in (38a, b) show that in German Number and Gender are in complementary distribution: whenever Number features are marked, Gender features are not. Thus, for German, we can reduce the presence of Gender features to the absence of Number features and conversely. Technically, this could be done by imposing a clustering principle that says Cluster (properties, features) if you can. Besides, we will substitute the (absence of) Speaker and Addressee by the more primitive feature +/−Person (which is sufficient for our discussion), cf. (39a, b): (39) a. Minimal feature specifications of pronominal HN and RP (German) – Singular 1-SG

2-SG

HN features

+Person +Gender/−Number

+Person +Gender/−Number

RP features

−Person +Gender/−Number

−Person +Gender/−Number

b. Minimal feature specifications of pronominal HN and RP (German) – Plural 1-PL

2-PL

HN features

+Person −Gender/+Number

+Person −Gender/+Number

RP features

−Person −Gender/+Number

−Person −Gender/+Number

152 | Ewa Trutkowski and Helmut Weiß

Given the above feature specifications and supposing that agreement involves feature sharing, the following questions arise: First, how does agreement in these kinds of NRRCs work? Second, how to account for the singular/plural difference? As to the second question, we will look at Kratzer (2009) who offers an explanation (for similar data), investigating perspectival relative clauses (which are merged into positions where they are “part of a predicative DP whose subject is a 1st or 2nd person pronoun”, cf. Ich bin die einzige, die ihren Sohn versorgt (= ‘I am the only one who takes care of her son’). Kratzer’s (2009) assumptions can be summarised as follows, (i)–(iv): (i) She claims that “3rd person” features might in reality be gender features (evidence comes from Bantu languages, but also from German), (p. 203), (ii) she postulates a gender feature in the agreement chain whose ultimate source is the internally represented head of a relative clause (assuming an RC analysis à la Kayne), (p. 204), (iii) she claims that “3rd person” RPs carry gender features as well as 1st or 2nd person features, (p. 204), (iv) she offers a unification account: All exponents/agreement chain members are specified for the relevant features (after feature transmission has taken place). Besides, Kratzer (2009) postulates the existence of vocabulary insertion rules as those given in (40) and (41). She notes (2009: 205) that in order “to survive morphophonological spell-out, the possessive pronoun, the relative pronoun, and verbal subject morphology must all find exponents that can spell out feature bundles that contain both a 1st or 2nd person feature and a gender feature. […] There is potential trouble, however, when it comes to spelling out singular verbal subject morphology or singular possessive pronouns. In those cases, the simultaneous presence of a 1st or 2nd person feature and a gender feature creates unresolvable spell-out dilemmas in German”. (40) Vocabulary insertion rules for German present tense verbal inflections [1st] [singular] → -e [2nd] [singular] → -st [male] } { { } {[female]} } [singular] → -t { [thing] { } [2nd] [plural] → -t elsewhere → -en

When personal pronouns compete with relative pronouns |

(41) ²⁵ [1st] [singular] [female] [2nd] [singular] [female] [1st] [plural] [female] [2nd] [plural] [female]

* * -en -t

153

[This column: ET&HW] ← Person-Gender clash ← Person-Gender clash In the plural: No Person-Gender clash/ no spell-out dilemma

Kratzer’s system can be made explicit by referring to her explanation for the ungrammaticality of a restrictive relative clause as *Ich bin die einzige, die meinen Sohn versorge (I am the only one who take-1SG care of my son): In this example binding is possible, however, if we look at the above vocabulary insertion rules, we see that in this singular example we have a spell-out dilemna: It is either gender or person features that can be realised at T (besides number features). As the plural is underspecified with respect to Gender (Elsewhere Case), no spell-out dilemma is present – hence a relative clause as Ihr seid die einzigen, die euren Sohn versorgt (You-PL are the only ones who take-2PL care of your-PL son) comes out as grammatical. Unfortunately, Kratzer (2009) skips the problem how feature compatibility between HN and RP is established. Nevertheless, we think that the employment of Gender in order to account for the singular vs. plural agreement difference in NRRCs with 1st /2nd person HNs seems promising. In the following we will examine a way how this could be done. However, some of our and Kratzer’s assumptions can hardly be combined, namely: In contrast to Kratzer (2009) we assume that 1st /2nd person singular pronouns possess inherent Gender features. Besides, contrary to Kratzer (2009:204) we do not assume that “3rd person” RPs carry (3rd person) gender features and 1st or 2nd person features at the same time. Given the ResP strategy, Kratzer’s latter assumption (and the unification account in general) seem implausible to us: Why should the ResP strategy (Ich, der ich …) exist, when RPs already have 1st or 2nd person features? Because of these (irreconcilable) differences we will pursue a different attempt than Kratzer (2009). In particular, we would like to argue for a feature sharing account that is able to get along without unifying the features of all exponents within the agreement chain. As the assumption that the V.fin-Rel agrees with the matrix clause HN in Person (over a (non-restrictive) clause boundary) is doubtful, we will try to explain agreement in NRRCs with 1st /2nd person head nouns by employing Gender.

25 “[(41)] illustrates which combinations with a person feature can or cannot be spelled out according to [(40)], taking [female] as the representative of gender features.” (Kratzer 2009: 205).

154 | Ewa Trutkowski and Helmut Weiß

Our main (empirical) claim is that Gender is a semantic trigger for syntactic agreement. In particular, we think that the crucial difference between singular and plural HNs (as well as their respective agreement patterns) can be traced back to the fact that the former are specified for Gender, whereas the latter are unspecified for Gender, cf. the feature specifications in table (39a, b) and our assumption that Gender assignment is part of reference assignment (in the sense of Kaplan 1989a, b). Due to the fact that in German Gender and Number are in complementary distribution, we have the following situations/scenarios in NRRCs with 1st /2nd person singular and plural HNs, respectively: A singular HN, e.g. Ich (I) or Du (you-SG), starts out with the features [−Num/ +Gen], but due to the fact that [+Gender] can only be spelled out at the 3rd person, which has the feature specification [−Person], agreement becomes parasitic on the feature [−Person] and thus must be spelled out as RP agreement, i.e. as 3rd person agreement as in Du, die vorträgt (You-SG, who is-giving-a-talk-3SG). As to NRRCs with plural HNs our explanation reads as follows: Since in the plural gender agreement cannot be spelled out at all, cf. the feature specification [+Num/−Gen], agreement takes place via the diametrically opposed (negatively specified) feature value, namely ‘minus’ [−Person] as in Ihr, die vortragt (You-PL, who are-giving-a-talk-2PL). Hence, in the plural [−Gender] agreement becomes parasitic on ‘non-3rd person’ agreement – which comes out as person agreement. Both, the [−Person] agreement in the singular as well as the ‘minus’ [−Person] agreement in the plural can be seen as secondary effects, because German has no genuine gender agreement (in contrast to languages as e.g. Polish). I.e., Gender is a (semantic) agreement trigger in NRRCs with 1st /2nd person HNs, but at the surface/at the finite verb gender agreement must be spelled out by another feature, namely by [−Person] in the singular, and by −[−Person] (= [+Person]) in the plural. That is why we call the kind of agreement found in NRRCs with 1st /2nd person HNs parasitic. Note that the singular features [−Num/+Gen] and the plural features [+Num/−Gen], respectively, are shared by the pronominal HN, the RP, as well as by the finite verb²⁶ (such that feature compatibility within the agreement chain is ensured). Independent evidence for our claim comes from coordinated singular pronouns (which we contrast with “real” plural pronouns). In particular, we argue that the crucial difference between plural pronouns (as wir; ihr) and coordinated singular pronouns (as ich und du; du und du) is the fact that the former are under-

26 As Gender cannot be overtly spelled out at the finite verb, the feature specifications [+Gen/−Num] and [−Gen/+Num], respectively, are ‘parasitically’ mapped into a [−Person] vs. −[−Person] (= [+Person]) distinction.

When personal pronouns compete with relative pronouns |

155

specified for Gender, cf. [−Gender], whereas the latter are not specified for Gender, cf. [Gender] – which is due to the fact that it is hard to determine the Gender of coordinated items. Thus, if we look at the a- vs. b-contrasts in (42) and (43) and take it for granted that coordinated singular items are not uniformly but possibly differently gender marked (cf. feminine vs. masculine vs. neuter), the fact that in such a case HN agreement cannot be triggered (semantically, through Gender) comes “for free”. For this reason, the finite verb must display (default-like) 3rd person plural features, cf. (42b) and (43b). In contrast, HN agreement is the preferred option for plural HNs which are underspecified (but not not-specified) for Gender, cf. (42a) and (43a): (42) – 1st person plural agreement in NRRCs a. wir, die ✓uns / ?*sich über gutes Mensaessen freuen we, who-PL us / REFL about good canteen-food glad-are-1/3PL – Coordinated 1st + 2nd person singular agreement in NRRCs (pseudo-plural) b. du und ich, die ?*uns / ✓sich über gutes Mensaessen you-SG and I, who-PL us / REFL about good canteen-food freuen glad-are-1/3PL (43) – 2nd person plural agreement in NRRCs a. ihr, die das natürlich nicht mitgekriegt ✓habt / *haben you-PL, who-PL that of-course not realized have-2PL / have-3PL – Coordinated 2nd + 2nd person singular agreement in NRRCs (pseudo-plural) b. du und du, die das natürlich nicht mitgekriegt *habt / you-SG and you-SG, who-PL that of-course not realized have-2PL / ✓haben have-3PL I.e., our hypothesis/supposition is that it is the lack of a common “semantic roof” (in form of gender consistency) that differentiates 1st /2nd person plural pronouns from coordinated 1st /2nd person singular pronouns with respect to their agreement behaviour in NRRCs.²⁷ A further piece of evidence is given by the fact that in the singular we (marginally) find HN agreement. As Kratzer (2009) postulates a direct Gender-

27 See Kratzer (2009: 210) for a small discussion on agreement patterns of conjoined DPs in main clauses.

156 | Ewa Trutkowski and Helmut Weiß

Person opposition, a datum as (44) should come out as totally ungrammatical – contrary to fact: (44) (?) ich, die malen lerne I, who-FEM-SG to-paint learn-1SG In our account the feature [Person] is not in direct opposition to [Number/Gender]. Instead, the opposition is indirect (because Gender is only parasitic on Person). I.e., it could be that alternatively we have (non-semantically triggered) syntactic person agreement in (44), that is [+Person] instead of [−Num/+Gen] agreement – which is imperfect, because the agreement chain is disconnected/disturbed by the (mismatch inducing) RP. Thus, in the singular, person (= HN) agreement can be seen as a deficient (non-PF) version of ResP agreement, whereas the ResP-inserting variant can be generally seen as a reflex of establishing “pure” syntactic person agreement instead of semantically (Gender) triggered ‘(minus) non-person’ agreement. The illustration in (45) summarises the possible kinds of agreement in NRRCs with 1st or 2nd person head (pro)nouns: (45)

(i)

1st /2nd person HN (ii)

“Pure” syntactic agreement, via Person [Du, die du vorträgst] [Ihr, die ihr vortragt] Semantically triggered agreement, via Gender [Du, die vorträgt] Gender agreement parasitic on [−Person] [Ihr, die vortragt] Gender agreement parasitic on −[−Person] (= [+Person])

Note that the experimental surveys have already shown that the variants (i) and (ii) are equally acceptable (see (14) and (19)). I.e., pure syntactic agreement (ResP agreement) is not favoured over semantically triggered syntactic agreement (RP/HN agreement). We will conclude by summarising our account for NRRCs with 1st /2nd person HNs. In so doing we will compare our suppositions/explanation with Kratzer’s – who, in essence, comes to the same/similar conclusions, however, based on different premises and assumptions. In particular, we have postulated a uniform agreement mechanism for NRRCs with 1st or 2nd person singular/plural HNs, namely syntactic [−Person] agreement triggered by a semantic Gender distinction. Our (parasitic) Gender agreement hypothesis “automatically” accounts for the Number distinction and the other way

When personal pronouns compete with relative pronouns |

157

round (due to the fact that in German Number and Gender appear as an amalgam). With respect to the binding data presented in Section 2, we regard the RP as the relevant binder that shares its features ([−Num/+Gen] in the singular, [+Num/−Gen] in the plural) with the pronominal head noun and the finite verb, where they become parasitically visible through overt person (and number) marking. In contrast to Kratzer (2009) we do not need specific vocabulary insertion rules, since the different agreement patterns can be derived from more general feature specifications (cf. (38) and (39)). Thus our account gets along without controversial extra assumptions (with respect to unification and feature transmission, e.g. Kratzer’s postulations that RPs also possess 1st or 2nd person features (p. 204) or that the V.fin-Rel is marked for Number and Gender (p. 207)).

4 Syntax RCs like (46a, b) pose a challenge for the syntactic analysis, because they contain two seemingly distinct subjects with different phi-features: an RP which is marked for 3rd person and masculine gender, and a personal pronoun in the 1st /2nd person respectively without overt gender specification. (46) a. ich, der ich schon sechzig bin, … I, who-MASC-SG I already sixty am b. du, der du schon sechzig bist, … you-SG, who-MASC-SG you-SG already sixty are-2SG Though both pronouns are marked for NOM, it is always the personal pronoun that agrees with the verb, if both are present, as in (47). However, if the personal pronoun is absent, the RP can trigger agreement with the verb as well, as in (48).²⁸ Therefore the RP must be a real subject (and not a fake subject like the Vorfeld-es in German). (47) a. * ich, der ich schon sechzig ist, … I, who-MASC-SG I already sixty is b. * du, der du schon sechzig ist, … you-SG, who-MASC-SG you-SG already sixty is

28 The same holds for binding, as we have seen in Section 2.2.

158 | Ewa Trutkowski and Helmut Weiß

(48) a. ich, der schon sechzig ist, … I, who-MASC-SG already sixty is b. du, der schon sechzig ist, … you-SG, who-MASC-SG already sixty is Though both pronouns differ in many respects (e.g., d- vs. personal pronouns, person feature), it is intuitively rather likely that both pronouns must form a syntactic unit in some sense. We will present an analysis which can explain why we can have a two-part subject in these cases. There are at least two possibilities for underlying structures: 1. RP and ResP forming a Big-DP (as for examples proposed for Left Dislocation, cf. Grewendorf 2002); 2. RP and ResP as multiple Spell-Out of chain positions. In the following we will argue for the second approach.

4.1 Structure of pronouns Combining several assumptions made by, e.g., Wiltschko (1998), Freidin & Vergnaud (2001), Fuß & Wratil (2013), Weiß (2015), we assume that: – RPs (being d-Pronouns) are of the category D (i.e., containing a bound morpheme in D° and an agreement morpheme in AgrD°, cf. Wiltschko 1998 and others) – weak²⁹ personal pronouns are AgrD elements (expressing phi-features) – strong pronouns are D elements – A DP contains (at least) a D-, an AgrD, and an N-part (49) a. d-er Mann b. [DP d- [AgrD er [NP Mann]]] (Wiltschko 1998: 149)

29 In this paper, we use the traditional terms strong, weak and clitic (in the sense of Cardinaletti & Starke 1999) to distinguish different types of pronouns. As is shown in Weiß (2015), at least for German dialects it is more appropriate to assume that reduced or clitic pronouns can be used weakly and that only clitics are AgrD elements, whereas reduced pronouns are ambiguous in that they have to be DPs in the prefield, but must be AgrDs in the Wackernagel position. Since this further distinction is not essential for the purpose of the present paper, we keep to the tradional terms.

When personal pronouns compete with relative pronouns |

159

In contrast to Wiltschko (1998: 156 f.), however, we assume that the NP projection is also present with personal pronouns (cf. Freidin & Vergnaud 2001). In our view, only weak personal pronouns are AgrD (or φPs, cf. Roberts 2010), whereas strong personal pronouns (like d-pronouns³⁰) are DPs, which can be combined with nouns (50a, b) – or modified by an RC. In dialects like Bavarian which have clearly distinct forms of strong and weak pronouns, it becomes clear that it is the strong form which combines with the noun (cf. 50c–e). We leave open the question whether the NP is completely absent or only silent/empty with weak pronouns. (50) a. b. c. d. e.

wir armen³¹ Linguisten we poor linguists Weg mit ihm Deppen Away with him idiot wai’e des ned kapiert hob because-Iw that not got have wai I Depp des ned kapiert hob have because Ist idiot that not got * wai’e Depp des ned kapiert hob because-Iw idiot that not got have

(internet example)³² (‘because I didn’t get it’)

30 In German dialects, d-pronouns can substitute personal pronouns in the 3rd person showing that both are interchangeable to a high degree. Though the topic has not been systematically investigated so far, it is known that substitution is only impossible for bound variables (i), whereas it is fine even in donkey sentences (ii), (Wiltschko 1998: 172): (i)

(ii)

* Wai a jedai glaubt, dass deai gscheid ist Because an everyone thinks, that this intelligent is Wenn a Bauai an Eselj hod, dann schlogd deai denj aa If a farmer a donkey has, then hits this that too

31 The weak inflection on the adjective shows that we are not dealing with appositive structures (as argued for by Delorme & Dougherty 1972 for English, and Olsen 1991 for German), since then we would expect a strongly inflected adjective (cf. wir, arme Linguisten). The inflectional pattern is that of possessive pronouns (mein/ich armer Idiot – meine/wir armen Idioten), which is an unambiguous evidence for a non-appositive structure (contrary to what is assumed by Olsen 1991: 37). Thanks to P. Grosz for drawing our attention to this aspect. 32 It is often assumed that personal pronouns in the 3rd person cannot take an NP as a complement (cf. Lenerz 1993), but that seems not to be the case. Though very rarely, one can find examples like (50b) on the internet.

160 | Ewa Trutkowski and Helmut Weiß

Thus, strong pronouns show a derivation as given in (51a): they enter the derivation in the AgrD position, raise to D° where they were spelled out as strong pronouns, whereas the copy in the base position remains silent. In the case of weak (and clitic) pronouns, the pronoun is spelled out in AgrD (51b) and no raising to D occurs: (51) a. [DP Pronst [AgrD Pron [NP … b. [DP [AgrD Pronw [NP …

4.2 Syntax of RCs Following Weiß (2013), we will assume that RCs in German are always introduced by a complementizer wo/was (overtly in dialects and covertly in Standard German) (52a). The RC can contain a gap (or silent pronoun) in the position of the relativized argument (cf. 52a), or a lexical item that spells it out. The dialects exhibit two different strategies: (i) RP which spell-out the highest copy in SpecCP (52b), and (ii) ResP spelling out a lower copy, e.g. in the Wackernagel position (52c) or in the base position (52d). (52) a. der Maurermeister, wo/was bei uns gearbeitet hat the bricklayer-MASC-SG, Rcomp with us worked has b. der Maurermeister, der wo/was bei uns gearbeitet hat the bricklayer-MASC-SG, who-MASC-SG Rcomp with us worked has c. der Maurermeister, wos-er bei uns gearbeitet hat the bricklayer-MASC-SG, Rcomp-he with us worked has (Fleischer 2004) d. die Lyt, wo mer iber sy gschwätzt händ the people-PL, Rcomp we about them talked have-1PL (Fleischer 2004) The ResP is the normal weak personal pronoun which behaves like any other weak pronoun, e.g. it raises to the Wackernagel position (WP), if it is the subject or the direct object of the RC. The resumptive strategy may not be very common among German dialects, but it does exist, and there are even some dialects which use ResPs not only for indirect objects or complements of prepositions (as it is known from Swiss German varieties). One of these dialects is the Middle German variety formerly spoken in Leibitz (or Lubica) in Slovakia. There ResPs seem to be (or to have been) obligatory in any case (53a, b) and they interact with other non-ResPs

When personal pronouns compete with relative pronouns |

161

in the normal way, e.g. subject clitics intervene between the complementizer and object resumptives (cf. 54a, b). (53) a. dr maorermeystr, vozər bae uns hat gəorpt the bricklayer-MASC-SG, Rcomp-he with us has worked b. s waep, vosəs uns gəšpon hat the woman-NEUT-SG, Rcomp-it us spun has (54) a. dr man vos yχ nən deu sæ the man-MASC-SG Rcomp I him there see b. dy frao vost zə deu zyst the woman-FEM-SG Rcomp-2SG her there see-2SG However, only a minority of German dialects uses (or used to use) ResPs, whereas the vast majority exhibit the RP-strategy.

4.3 Analysis In RCs under 1st /2nd person pronouns, the underlying derivation for a sentence like (55b) looks like (55c): the pronoun with its articulated structure (= 55a) is merged in its base position, then raised first to WP and finally to SpecCP. The spellout forms in both landing sites differ, because in the WP the AgrD part is spelled out as a weak personal pronoun and in SpecCP the D-part as an RP:³³ (55) a. [DP der [AgrD ich [NP e ]]] who-MASC-SG I b. ich, der (wo) ich malen lerne I, who-MASC-SG (COMP) I to-paint learn-1SG ‘I, who learns to paint’ c. Ich, [CP [DP der [AgrD ich [NP e ]]]i [C󸀠 (wo) [WP [DP der [AgrD ich [NP e ]]]i [TP … [DP der [AgrD ich [NP e ]]]i malen lernt …³⁴

33 Another possibility to derive the string [der wo ich] in (55b) is to assume, that the DP in (55a) splits up in the course of derivation, whereby the Agr part remains in the WP and only the DP part is raised to SpecCP. Note that this assumption (which C. Poletto, p.c., has suggested to us) does not require another structure for d-pronouns. 34 The analysis suggested in (55c) is also in line with the ‘Distributed Deletion’ approach argued for by Fanselow & Cavar (2002). Thanks to an anonymous reviewer for drawing our attention to this aspect.

162 | Ewa Trutkowski and Helmut Weiß

Though we assume that we are always dealing with a fully fledged structure as in (51a) or (55a), a spell-out form as in (55a) is restricted to RCs, because spellout is determined by the syntactic context. This would make sentences like *wai dea i des net kapiat hob lit. ‘because the I that not understood have’ ungrammatical. Note that different spell-out forms are also attested in other cases of pronoun movement, cf. (56a, b), so that is not particular for our case. (56) a. Was glaubst du, wer morgen kommt What think you, who tomorrow comes b. Wer glaubst du, der morgen kommt³⁵ Who think you, the-MASC-SG tomorrow comes The feature mismatch is an inevitable consequence, since on the one hand 1st /2nd person pronouns in German are not overtly specified for gender and on the other hand, RPs show another person specification than 1st /2nd person pronouns do. The ongoing change from HN agreement (plus ResP) (46a, b) to RP agreement (48a, b) as attested in our experiments may be interpreted as due to a tendency to avoid such mismatches within the RC. Our proposal makes the prediction that the ResP is a w Pron, which seems to be borne out by the data. On the other hand, in some cases the ResP can even be a null pronoun (= pro), e.g. in Bavarian in the 2nd person singular and plural (57a, b), where the pronominal complementizer agreement allows for pro-drop (Weiß 2005): (57) a. du, der (wo-)sd pro a so a Depp bisd you-SG, the (Rcomp)-2SG pro a such an idiot are-2SG b. ös, de (wo-)ds pro sechane Deppm sads you-PL, the (Rcomp)-2PL pro such idiots are-2PL Therefore we would like to propose the hypothesis that it is always the weakest form of a weak pronoun (i.e. a null, a clitic, or a reduced pronoun in the sense of Weiß 2015) that is used as a resumptive pronoun in the unmarked case. Future investigations have to show whether this prediction is borne out or not.³⁶

35 The construction in (56b) is called Relativsatzkonstruktion ‘relative clause construction’ and attested since Middle High German times (Axel-Tober 2012: 81). 36 One of the questions to be adressed in future research is whether stProns are excluded at all or can be used as ResPs, e.g. if they are focused. If the latter is the case, this would require a revision of our analysis.

When personal pronouns compete with relative pronouns |

163

As noted above, if a subject ResP is present, it always triggers agreement with the verb. However, that seems not to be the reason for why a ResP must be present, otherwise it would be unexpected that even with objects a ResP is possible (cf. 58a, b). Since objects do not agree with the verb in German, its spellout cannot be motivated by agreement. It may be the case that emphasis or focus play a role for ResP to get spelled out. (58) a. wir, die man (uns) leicht ärgern kann we-NOM, who-PL one (us-ACC) easily tease can b. ihr, die man (euch) leicht ärgern kann you-PL-NOM, who-PL one (you-PL-ACC) easily tease can There is still one issue to explain, namely the question why D is spelled out as dpronoun and not as strong personal pronoun? Why that should be the case is not apparent for two reasons: on the one hand, according to Wiltschko (1998), d- and personal pronouns are “the same lexical item”. And on the other hand, ‘doubling’ of personal pronoun is not excluded in principle, as many Flemish or Dutch dialects show which allow pronoun doubling. In these cases, at least one copy is pronounced as a strong pronoun, whereas the other one could be pronounced as a clitic, weak, or even strong pronoun (van Craenenbroeck & van Koppen 2008, Barbiers et al. 2010):³⁷ (59) a. Ik paus da se zaailn kommen I think that theycl theyst come

(Wambeek Dutch)

‘I think they are coming’ mee te maken b. Ze heeft zij daar niks Shew has shest there nothing with to do

(Flemish)

‘She’s got nothing to do with it.’ mee te maken c. Zij heeft zij daar niets Shest has shest there nothing with to do

(Flemish Brabant)

‘She has got nothing to do with it.’ As the examples (59b, c) show, it is in principle possible to pronounce the copy in SpecCP as a (weak or strong) personal pronoun. Wiltschko (1998)³⁸ assumes that

37 Example (59a) is van Craenenbroeck & van Koppen’s (2008) ex. (1), (59b, c) Barbiers et al.’s (2010) (6) and (2). 38 Wiltschko (1998: 174): “For semantic reasons, a relative clause needs to contain a variable: it is interpreted as a predicate entering a predication relation with the head noun. […] In German,

164 | Ewa Trutkowski and Helmut Weiß

the variable an RC must contain is provided by the trace of the d-pronoun which has undergone A󸀠 -movement to SpecCP, or by a ResP (or a gap) when A󸀠 -movement is not involved (or targets the WP).³⁹ The variable must be bound by an operator within the RC and the operator is either spelled out by a d-pronoun or by an empty operator (in cases like 52a, c, d above). This may be the reason why D is spelled out in SpecCP as a d-pronoun and not as a personal pronoun: because only d-pronouns can function as operators.

5 Conclusion We have presented empirical material which indicates an ongoing change with respect to the (preferred) agreement pattern of 1st /2nd person NRRCs in German. As our experiments showed, the “traditional” ResP agreement pattern is not better rated than sole RP agreement (in the singular) or sole HN-agreement (in the plural). With respect to these “new” agreement patterns, we claim (in partial accordance with Kratzer 2009) that the crucial factor (feature) that drives agreement in NRRCs with 1st /2nd person pronominal HNs is Gender. With respect to an analysis of the data, we presume that RCs in German are always introduced by a complementizer wo/was (overtly in dialects and covertly in Standard German). Moreover, we suspect that the ResP is a weak personal pronoun (which can also be realized as null) and argue that the RP and ResP constitute multiple Spell-Out possibilities of chain positions.

as in many other languages, the variable is provided by the trace left behind by A󸀠 -movement of the d-word.” 39 In OHG, personal pronouns could introduce RC without an additional RC, cf. (i) and (ii): (i)

fater unser, thu in himilom bist father our, you-SG in heaven are

(Pater noster, Weißenburg catechism)

(ii)

ir then christianiun namun intfangan eigut you-PL the christian name received have

(Exhortation ad plebem christianam)

When personal pronouns compete with relative pronouns |

165

References Axel-Tober, Katrin (2012) (Nicht-)Kanonische Nebensätze im Deutschen. Synchrone und diachrone Aspekte. Berlin, Boston. De Gruyter (Linguistische Arbeiten 542). Barbiers, Sjef, Koeneman, Olaf & Maria Lekakou (2010) Syntactic doubling and the structure of wh-chains. Journal of Linguistics 46: 1–46. Benveniste, Émile (1971) Problems in General Linguistics. Translated by Mary Elisabeth Meek. Cora Gables, FA. University of Miami Papers. Cardinaletti, Anna & Starke, Michel (1999) The Typology of Structural Deficiency: On the three Grammatical Classes. In Clitics in the Languages of Europe, ed. Henk van Riemsdijk, 145– 233. Berlin, New York. Mouton de Gruyter (= Empirical Approaches to Language Typology/ Eurotyp 20-5). Chomsky, Noam (1977) On Wh-Movement. In: P. Culicover, T. Wasow & A. Akmajian (eds.) Formal Syntax. New York. Academic Press, 71–132. Craenenbroeck, Jeroen & Koppen, Marjo van (2008) Pronominal doubling in Dutch dialects: big DPs and coordinations. In S. Barbiers, O. Koeneman & M. Lekakou (eds.) Microvariation in syntactic doubling. Syntax and Semantics volume 36. Emerald, 207–239. Delorme, Evelyne & Dougherty, Ray C. (1972) Appositive NP constructions: we, the men; we men; I, a man; ETC. Foundations of Language 81 (1), 1–29. Fanselow, Gisbert & Cavar, Damir (2002) Distributed Deletion. In: Alexiadou, A. (ed.) Theoretical Approaches to Universals. (Linguistik Aktuell/Linguistics Today 49), Amsterdam, Philadelphia. Benjamins, 65–107. Fleischer, Jürg (2004) A Typology of Relative Clauses in German Dialects. In: Kortmann, Bernd (Hrsg.): Dialectology meets Typology. Dialect Grammar from a Cross-Linguistic Perspective. Berlin, New York. Mouton de Gruyter, 211–245. Freidin, Robert & Vergnaud, Jean-Roger (2001) “Exquisite connections: Some remarks on the evolution of linguistic theory.” Lingua 111, 639–666. Fuß, Eric & Wratil, Meláni (2013) Der Nullsubjektzyklus: Etablierung und Verlust von Nullargumenten. In: J. Fleischer & H. Simon (eds.) Comparing Diachronies. Tübingen. Niemeyer. Harbour, Daniel, Adger, David & Susana Béjar (2008) Phi Theory: phi features across interfaces and modules. Oxford, New York. Oxford University Press. Harley, Heidi & Ritter, Elizabeth (2002) Person and Number in Pronouns: A Feature-Geometric Analysis. Language 78(3), 482–526. Heck, Fabian & Cuartero, Juan (2008) Long Distance Agreement in Relative Clauses. In: Heck, F., G. Müller & J. Trommer (eds.) Varieties of Competition, 13–48. Linguistische Arbeitsberichte (Universitär Leipzig), Band 87. Ito, Junko & Mester, Armin (2000) Ich, der ich sechzig bin: An Agreement Puzzle. In: Chung, S., J. McCloskey & N. Sanders (eds.) Jorge Hankamer WebFest. Internet-Publikation. URL: http://ling.ucsc.edu/Jorge/itomester.html Kaplan, David (1989a) Demonstratives. An essay on the semantics, logic, metaphysics and epistemology of demonstratives and other indexicals. In: Almog, J., J. Perry & H. Wettstein (eds.) Themes from Kaplan. Oxford. Oxford University Press, 481–563. Kaplan, David (1989b) Afterthoughts. In: J. Almog, J., J. Perry, & H. Wettstein (eds.) Themes from Kaplan. Oxford. Oxford University Press, 565–614. Kratzer, Angelika (2009) Making a Pronoun: Fake Indexicals as Windows into the Properties of Pronouns. Linguistic Inquiry 40(2), 187–237.

166 | Ewa Trutkowski and Helmut Weiß

Lenerz, Jürgen (1993) Zur Syntax und Semantik deutscher Personalpronomina. In: Reis, Marga (eds.): Wortstellung und Informationsstruktur. Tübingen. Niemeyer, 117–153. Olsen, Susan (1991) Die deutsche Nominalphrase als “Determinansphrase”. In: Olsen, S, & G. Fanselow (eds.) DET, COMP und INFL. Zur Syntax funktionaler Kategorien und grammatischer Funktionen. Tübingen. Niemeyer, 35–56. Rizzi, Luigi (1990) Relativized Minimality. Cambridge. MIT Press. Roberts, Ian (2010) Agreement and head movement: clitics, incorporation, and defective goals. Cambridge, Mass. MIT Press. Schachter, Paul (1973) Focus and Relativization. Language 49, 19–46. Siewierska, Anna (2004) Person. Cambridge. Cambridge University Press. Sternefeld, Wolfgang (2006) Syntax. Eine morphologisch motivierte generative Beschreibung des Deutschen, Band 1. Tübingen. Stauffenburg Verlag. Trommer, Jochen (2008) Third Person Marking in Menominee. In: Harbour, D., D. Adger & S. Bejar (eds.) Phi Theory: phi features across interfaces and modules. Oxford. Oxford University Press, 221–250. Trutkowski, Ewa (in prep.) Zur Frage von Merkmals-Mismatches und Kongruenz in Relativsätzen mit 1./2.-Person-Köpfen. Ms. Goethe-Universität Frankfurt/Main. Trutkowski, Ewa, Grewendorf, Günther & Helmut Weiß (2013) Kongruenzbeziehungen in Relativsätzen mit 1./2. Person-Köpfen. Talk at the ‘Kolloquium der DFG-Forschergruppe Relativsätze’. Frankfurt am Main, 4 June 2013. Vogel, Ralf (2007/8) „Ich, der ich …“ Seminar Syntax und Morphologie, Winter term 2007/ 2008. University Bielefeld. Weiß, Helmut (2005) Inflected complementizers in Continental West Germanic Dialects. Zeitschrift für Dialektologie und Linguistik 72 (2), 148–166. Weiß, Helmut (2013) Satztyp und Dialekt. In: Meibauer J., M. Steinbach & H. Altmann (eds.) Satztypen des Deutschen. Berlin, New York. Walter de Gruyter, 763–784. Weiß, Helmut (2015) When the subject follows the object. On a curiosity in the syntax of personal pronouns in some German dialects. Journal of Comparative Germanic Linguistics 18, 65–92. Wiltschko, Martina (1998) On the Syntax and Semantics of (Relative) Pronouns and Determiners. Journal of Comparative Germanic Linguistics 2, 143–181.

| Part II: Semantics and pronominal interpretation

Uli Sauerland (Zentrum für Allgemeine Sprachwissenschaft)

The form of bound pronouns: Towards a uniform account Abstract: In at least three environments – de se binding, distributive binding, and focus quantification – some presuppositions exhibit unexpectedly weak projection behavior. This holds for the presuppositions of bound pronouns, but also several other cases of presupposition. In this paper, I first describe a general approach to capture the interaction of presuppositions with quantificational operators within a multi-tiered evaluation procedure. Secondly I discuss data from Condition A, in particular non-bound occurrences of reflexives, that motivate a presuppositional account of Condition A and confirm the general approach. Keywords: presupposition, binding, pronouns, projection, quantification, reflexivity, number, gender, agreement, semantics

1 Introduction The paper addresses a class of examples where presuppositions in the scope of a quantifier exhibit an unexpected projection behavior. Several examples of this type have been discussed in the recent literature, and a number of different proposals for their explanation have already been made. However, no prior proposal extends to all examples of this type. My goal in this paper is to provide a general account of presupposition that can extend to all cases. In this section, I illustrate what I see to be the full range of the phenomenon: bound variable pronouns, full singular definites, and factive verbs. I use the descriptive label Weakened Projection for all these cases. The most widely discussed case of weakened projection are bound variable pronouns. Consider briefly the evidence for weakened projection in this case. The argument is based on the presuppositional approach to ϕ-features (i.e. person, number, and gender marking) on pronouns (Cooper, 1979). Specifically, we assume that pronouns always consist of an index and a set of ϕ-features. For example, I is analyzed as an index (for example, j) and the features sing and 1pers; she as another index (e.g., i) and the features sing, fem and 3pers. Since the index fully determines the pronoun’s interpretation, the ϕ-features are interpreted as presuppositions on the value assigned to the index. Assuming the standard truth-value gap account of presuppositions, the ϕ-features are interpreted as re-

170 | Uli Sauerland stricted identity functions of type ⟨e, e⟩. Specifically, sing is restricted to atomic entities, fem to entities consisting only of female atomic parts, 1pers to entities that contain the speaker, and 3pers to entities that contain neither speaker nor addressee. While they can be satisfied for unbound pronouns, the presuppositions of the ϕ-features are too strong at least for the following: bound pronouns in the scope of only as in (1a), de se bound pronouns as in (1b), and distributively bound pronouns as in (1c): (1a) has a bound interpretation that entails that other people didn’t do their homework. But, even when my is bound, the presuppositions 1pers and sing should restrict its interpretation to the current speaker. In (1b), the referent of the de se pronoun he, Brigitte Bardot, is clearly female, so the presupposition of masc should fail. Finally, the pronoun in (1c) must be used even in a scenario where each boy bought a different bike and therefore there is no bike that a plurality of boys bought. In such a scenario, all referents of they considered in the evaluation of the sentence must be singular, but sing cannot be used.¹ (1)

a. Only I did my homework. b. Lakoff dreamt that he was Brigitte Bardot. c. The boys all were riding the bike they had recently bought.

Weakened projection, however, isn’t restricted to bound pronouns. To my knowledge, the first example with a full definite DP in (3) is due to von Heusinger 2007. In the scenario given below, (2) entails that Sam didn’t talk to the Japanese professors, and didn’t talk to the English professors. At the same time, (2) makes no entailment as to whether Sam talked to non-professors. (2) Scenario: One German professor attended the party, three Japanese professors, five English professors, and also two office workers.

Sam only talked to the GermanF professor. Since focus in (2) is on German, alternatives such as John talked to the Japanese professor are expected to be relevant for the interpretation of (2). However, the uniqueness presupposition of the Japanese professor is not satisfied in the scenario under consideration. More recently Sudo (2012) and Walker (2012) suggest that there are further cases of weakened projection with other presupposition triggers. I return to this discussion below. One further clear example of weakened projection is (3), where

1 English speakers often prefer using the plural bikes in (1c). However, the singular bike is fully acceptable at least in German, and still a plural pronoun must be used inside of the relative clause.

The form of bound pronouns: Towards a uniform account |

171

waitress could be analyzed as presupposing that its argument be female. The alternatives to Mary in (3) are both male. So unless projection is weakened, we expect that (3) should be acceptable even if John or Bill were still waiters. (3) Scenario: John, Mary, and Bill all worked as waiters. But John and Bill moved on to different jobs.

Only Mary is still a waitress. As already mentioned, the goal of this paper is to develop and defend a general account of weakened projection. I present the core proposal in the following section. In Section 3, I present data from Condition A that confirm the general account. Section 4 is the conclusion.

2 The proposal: Tiered alternatives and pure presuppositionality My proposal takes some inspiration from existing proposals though none of the existing proposals explains all five examples from the introduction. Briefly, consider the existing approaches to weakened projection. The most popular approach to weakened projection is a syntactic agreement mechanism operating on variable binding relationships (Kratzer, 1998; Heim, 2008; Kratzer, 2009; Collins and Postal, 2012). I believe this idea should be abandoned: I see no way to extend this proposal to cases of weakened projection not involving a bound pronoun such as (2). My proposal is closer to an idea for bound pronouns due to Benjamin Spector (p.c.) cited by Heim (2008).² Namely, Spector suggests that the presuppositions of ϕ-features on bound pronouns don’t contribute to focus interpretation. I will seek to expand this idea to cases not involving a bound pronoun. At least three recent proposals already discuss cases of weakened projection that occur not with bound pronouns. However none of these develops a fully general proposal that covers all the cases of weakened projection I introduced above. von Heusinger (2007) develops a proposal that is specific to singular definite descriptions. Walker (2012) focusses on establishing a generalization – namely that all cases of soft presupposition triggers of Simons (2001) and Abusch (2010) allow weakened projection. However, it seems to me that Walker’s generalization doesn’t cover some bound pronoun cases, e.g. first person marking of my. Finally,

2 See also Jacobson (2012).

172 | Uli Sauerland

Sudo (2012) proposes two different approaches for two classes of weakened projection. For cases such as (2), he argues for a two-dimensional approach to presuppositions and inherits the problems of that approach. Sudo’s second proposal, however, is specifically intended for cases of distributive binding such as (1c). Sudo proposes that distributive quantification ranges not over the atoms that are part of a plurality, but over functions mapping the plurality to an atomic part of it. Plural agreement then is accounted for by reference to the range of this function. This proposal is, as will become evident below, quite close to my proposal. In the following, I focus initially on the cases of focus alternative quantification such as (1a) and (3) with only. The extension to de se-binding and distributive binding is presented at the end of this section. Before presenting the proposal, I argue that weakened projection is indeed limited to quantifiers over focus alternatives. Note first that quantifiers not involving focus alternatives don’t exhibit weakened projection. For example (4) presupposes that all students in the class are female.³ (4) No student in the class except for one criticized herself. Furthermore, let’s consider quantifiers with exactly plus a numeral. Sudo (2012) observes weakened projection in examples such as (5) and analyzes them as not involving quantification over focus alternatives.⁴ (5) Exactly one student criticized herself. However, an account involving focus alternatives is possible. Specifically, I propose the focus semantic entry for exactly given in (6) within the framework of focus semantics of Rooth (1992). (6) exactlyw (C)(p) = 1 iff. C is totally ordered by entailment and p(w) = 1 and for all q ∈ C : (p → q ∨ q(w) = 0) With this lexical entry, (5) entails that the focus alternative (7) must be false. Because weakened projection allows us to ignore both singular number and femi-

3 Example (4) with focus her allows a reading where the feminity presupposition is accommodated. The resulting interpretation entails that the other students are either male or they didn’t criticize themselves or both. It is therefore still different from weakened projection interpretation of Sudo’s Exactly one student criticized herself, which entails that no male student criticized himself. 4 I leave other modified numerals to future work.

The form of bound pronouns: Towards a uniform account |

173

nine gender on her, (7) entails that there can be no second student who criticized him- or herself.⁵ (7) two student λ x x criticized herx self

2.1 Focus alternatives are pairs The basic idea of my proposal is that during the evaluation of a focus alternative, the actual semantic value of a constituent is always available. The idea is compatible with different approaches to alternative semantics as far as I can see. For concreteness, I propose that the meaning of each focus alternative is actually a pair of the actual value and the focus alternative value. So, the set of focus alternatives of a simple sentence such as John loves MaryF is then a set of pairs as shown in (8). Each pair has the same first member – the actual interpretation of the sentence –, but the second member differs with respect to the focussed constituent. (8) FocAlt(John loves MaryF1 ) = {⟨John loves Mary, John loves Mary⟩, ⟨John loves Mary, John loves Sue⟩, ⟨John loves Mary, John loves Anna⟩, …} For the concrete semantics to generate such focus meanings, I follow Beck (2006) and others and assume that foci are indexed and the h is focus assignment that assigns to each focussed constituent a meaning that may differ from its actual meaning in the second member of the pair as stated in (9a). Non-focussed constituents are interpreted only as the actual meaning as in (9b). Furthermore, functional application of the focus alternatives is point-wise as in (9c) (Let a = αg,h and b = βg,h ). (9) a. αFj g,h = ⟨αg , h(j)⟩ b. ωg,h = ωg for any terminal ω ⟨b1 (a1 ), b2 (a2 )⟩ if a, b are both pairs { { { { { {⟨b(a1 ), b(a2 )⟩ if a is a pair and b isn’t c. α βg,h = { { b1 (a), b2 (a)⟩ if b is a pair and a isn’t { { { { otherwise {b(a) Finally, I assume that abstraction forms functions that take pairs as arguments as in (10). Functions formed by λ abstraction can therefore combine with an ar-

5 Note that I assume that the noun student is numberless (see Sauerland 2003 and below).

174 | Uli Sauerland

gument that contains a focus by normal functional application since a function taking a pair as an argument is different from a pair of two functions.⁶ (10) λ i αg,h = λ x⃗ ∈ D e × D e ∪ D e . αg

󸀠

,h

where g󸀠 (i) = x⃗ and g󸀠 (j) = g(j) for i ≠ j

Note that unfocussed pronouns receive a pair interpretation when the argument of their binder is focussed. This predicts bound pronouns to receive a covariant interpretation within the focus alternatives in the way shown in (11):⁷ (11) FocAlt(JohnF1 loves himself.) = {⟨John loves John, John loves John⟩, ⟨John loves John, Mary loves Mary⟩, ⟨John loves John, Bill loves Bill⟩, … }

2.2 Multi-tier and front-tier interpretation The system proposed so far predicts that unfocussed semantic content will always contribute to both elements of the pair-meaning of a focus alternative. At this point, this prediction also holds for presuppositions. However, we saw above that the prediction isn’t empirically correct for at least some cases of presupposition. I therefore propose that presuppositions in some cases need only be verified for the first member of the paired evaluation. But I show below that a second interpretation where the presupposition applies to both members of the paired evaluation is also available. So, I assume an ambiguity. The relevant clause of the focus evaluation is (12):⁸ (12) If the non-focus interpretation of a word ω is the identity function on some subdomain of type τ, i.e. ωg = idS for S ⊂ D τ (or in another way purely presuppositional, see (30) below), then ωg,h can be a or b: a. multi-tier: ωg,h = ⟨ωg , ωg ⟩ b. front-tier: ωg,h = ⟨ωg , idD τ ⟩

6 The case where the argument of a λ-derived function L isn’t a pair, but a single item x, won’t play any role in the following. I think it is safe to assume that in such cases L is applied to the trivial pair ⟨x, x⟩. 7 At this point, I haven’t investigated the predictions of the system for cases of multiple focus and focus on bound pronouns (see Mayr 2012 and references therein.) 8 At this point, nothing in the proposal formally assures that across all alternative meanings that are computed, the choice between (12a) and (12b) is uniform though I suspect uniformity is required.

The form of bound pronouns: Towards a uniform account |

175

Consider for example the evaluation of an alternative for example (1a) in (13) with h(1) = Mary. The argument of the predicate starting with λ x is then the pair ⟨Uli S., Mary⟩ and therefore the presupposition of the first person feature of my is satisfied if the front-tier interpretation is used. (13) IF1 λx x did myx ’s homeworkg,h The multi-tier interpretation of first person doesn’t allow alternatives to Uli S. as arguments. Since only requires alternatives, the multi-tier interpretation isn’t compatible with binding of my.⁹ In general, the single- and multi-tier interpretations are logically related. Specifically, without accommodation, the multi-tier interpretation always predicts a stronger presupposition than the single-tier interpretation and, for all alternatives that satisfy the presuppositions of the multi-tier interpretation, the two interpretations have the same truth value. So, in a situation where the presupposition of the multi-tier interpretation is satisfied, the single-tier interpretation and the multi-tier interpretation predict the same truth value. Hence the multi-tier interpretation can only be detected if accommodation applies. This prediction is shown by (14): (14) Scenario: Mary and three of the guys each criticized themselves. Nobody else engaged in any self-criticism.

Only MaryF1 criticized HERF2 self. For the analysis of (14), I assume that selective, focus-sensitive accommodation is required and underlies the focus on her that (14) requires. Modifying a proposal of Krahmer (1998) for unselective accommodation, I propose the accommodation operator in (15). 1 { { { { { {0 { { { w (15) Ai XP = {0 { { { { { { { { {undefined

if XPw = 1 if XPw = 0 w ∈ ̸ domain(XP) ∧ ∃⟨p, q⟩ ∈ {XP[i󳨃→a] | a ∈ D⟨e,e⟩ } : w ∈ domain(q) otherwise

9 Sentence (1a) also allows an interpretation where my isn’t bound. I put this interpretation aside throughout this paper.

176 | Uli Sauerland

The novelty lies in the third clause in (15), which restricts accommodation to the focussed trigger whose focus is coindexed with the accommodation operator. This is accomplished here by considering all alternative functions of type ⟨e, e⟩ to the function associated with the feature ‘feminine’, especially the function with no domain restriction. This ensures that rather than being undefined for worlds where a feminine pronoun applies to non-feminine individuals, the function resulting from the application of A is defined as false. Like Krahmer’s (1998) accommodation operator, A erases the distinction between presupposition failure and falsehood. The new thing is though that A requires focus on the presupposition trigger whose presupoosition is erased. The account of weakened projection of presuppositions triggered by ϕ-features is now complete. The account rests on the pure presuppositionality of the features in a way I clarify in the next section. I also argue in the next section that weakened projection with singular definites follows from the presupposition of the singular.

2.3 Pure presuppositionality and other triggers The account of presupposition projection based on single-tier interpretation as defined in (12) cannot straightforwardly apply to all presupposition triggers. Consider the case of a function f of type ⟨et, e⟩ with some presupposition. If we apply f on the front tier, but an identity function (necessarily of type ⟨et, et⟩) on the alternative tier, we generate meanings of different types on the two tiers. This will in most cases create problems for the further composition. For this reason, (12) applies only to presupposition triggers that are purely presuppositional.¹⁰ At this point, I mean by purely presuppositional that the meaning of an item is the identity function on some domain that is a subset of the full set of entities of that type. The ϕ-features all fit this description: They are restricted identity functions of type ⟨e, e⟩. However, other presupposition triggers are arguably not purely presuppositional in this sense (and also not in a second sense which I introduce below). For example, the singular definite determiner is often analyzed as a presuppositional function of type ⟨et, e⟩ and therefore is not purely presuppositional. In this section, I show that my approach to weakened projection nevertheless carries over to the singular definite determiner. For the account, I assume a general presuppositional account of ϕ-features (except for those involving grammatical gender). This extends both the account 10 Schlenker (2013) argues for a similar, but not identical restriction for sign languages. At this point, I have not been able to extensively compare the two proposals.

The form of bound pronouns: Towards a uniform account |

177

of Cooper (1979) and Sauerland (2003). Cooper’s account was restricted to pronouns, while my own 2003 account treated unbound pronouns and full DPs uniformly, but made an exception for bound pronouns. Now I claim that this exception for bound pronouns was unnecessary, but otherwise adopt the Sauerland 2003 account: Full DPs always are the complement of a ϕ-head, and DP internal ϕ-features, whether on the noun, adjectives, or the determiner, are always due to agreement with the ϕ-head (see also Scontras 2013). Specifically, this entails that singular and plural noun forms are interpreted the same, while the difference in number leads to different interpretations of the ϕ-head. Now consider again example (2) with the structure in (16a). The meaning of only requires that the two propositions in (16b) and (16c) must be false. Because sing is only interpreted on the front tier, (16b) and (16c) don’t trigger a presupposition failure. In the scenario given for (2) above, (16a) is correctly predicted to be true. (16) a. Sam only talked to sing[the GermanF professor(s)] b. Sam talked to the English professor(s). c. Sam talked to the Japanese professor(s). However consider the scenario described in (17). In this scenario, (16c) is intuitively false. Therefore (2) is incorrectly predicted to be true because the two alternatives (16b) and (16c) are both false. (17) Scenario: One German professor attended the party, three Japanese professors, and five English professors. Sam talked to the German professor and to one Japanese professor.

My solution to this issue is based on the property of (16c) that both it and its negation are judged false in scenario (17) (Fodor, 1970; Gajewski, 2005). I propose an account of this phenomenon assuming a fourth truth value in addition to 0, 1, and undefined, namely 12 as in fuzzy logic (see Alxatib et al. 2013). I assume that a proposition with truth value 12 is intuitively judged false, but its negation will also have truth value 12 and be judged false too. Distributive quantification receives truth value 12 in the non-uniform case (cf. Gajewski 2005; Romoli 2013). The definition of the *-operator in (18) implements this treatment of distributivity: * applies to distributive predicates of individuals and returns a predicate that is also true collectively.

178 | Uli Sauerland

1 { { { (18) *P(X) = {0 { {1 {2

if P(x) = 1 for all atomic x ⊑ X if P(x) = 0 for all atomic x ⊑ X otherwise

Only, however, requires that its focus alternative have truth value 0 as in (19).¹¹ This is needed independently of weakened projection for examples like John only talked to the JapaneseF professors, which entails that John talked to all the Japanese professors, but none of the English ones. (19) only XPg = 1iff.XPg = 1 ∧ ∀h∀⟨p, q⟩ ∈ XPg,h : XPg → q ∨ q(w) = 0 In this way, my approach extends to example (2). Sudo (2012) and Walker (2012) discuss weakened projection with several other presupposition triggers such as factives, change of state verbs, and focus particles. Their empirical results, however, do not converge and at this point I have not done extensive empirical work myself. Therefore, I leave it open whether the approach based on (12) can extend to all cases of weakened projection. In the following section, I present the extension of my approach to distributive and de se binding.

2.4 Distributive and de se-binding For distributive binding, I propose that distributive quantification involves quantification over pairs of the plurality and atomic parts thereof as in (20).¹² Subscript 2 here denotes the second member of a pair. 1 { { { (20) *P(X) = {0 { {1 {2

if P(⟨X, x⟩)2 = 1 for all atomic x ⊑ X if P(⟨X, x⟩)2 = 0 for all atomic x ⊑ X otherwise

Consider the representation in (21) for (1c). The λ x -predicate is evaluated for pairs ⟨B, b⟩, where B is the sum of all boys and b one individual boy. I furthermore

11 I use here what Rooth (1992) calls a direct association with focus on account for ease of exposition. 12 As I noted already, the approach of Sudo (2012) to plurality for (20) is similar. Sudo proposes that distributive quantification quantifies over choice functions with a single mapping from a plural to one atomic part, e.g. the function represented by the unit set {⟨B, b⟩}. Sudo’s approach can now be seen as a special case of the general approach.

The form of bound pronouns: Towards a uniform account |

179

assume that pl is actually also satisfied by singular individuals (Sauerland et al., 2005) as long as in at least one case it appplies to a true plurality. Therefore pl on they is licensed regardless of whether the front-tier or the multi-tier interpretation is chosen because B is a plurality. (21) The boys all *λ x x were riding the bike theyx had recently bought. For attitude verbs like dream, I also propose that de se-quantification involves quantification over pairs. In this case, the pair must contain the de se-center as one element while the other must be the real world counterpart of the de se-center to license the appropriate agreement properties. In (22), I provide a sample lexical entry for dream.¹³ (22) dreamt(P⟨e ,st ⟩ )(y) = 1 iff. for all pairs ⟨w, x⟩ such that ⟨w, x⟩ is consistent with y’s dream that he is x in world w : [P(⟨y, x⟩)(w)]2 = ⊤. 2

2

Given this entry, masc in (23) is licensed by the real world component in the front tier.¹⁴ (23) George dreamed that he was Brigitte Bardot. There’s one potential issue for both extensions mentioned in this section: The two components of the pair introduced have different properties either with respect to number or with respect to which worlds they occupy. Therefore, the composition of the two tiers may not be able to proceed in parallel. But before engaging that issue in the conclusions, I point out one potential area of empirical support for the unifying approach proposed here: namely, evidence from binding theory in the next section.

13 The types e2 and t2 here indicate the types of pairs of individuals and truth values. Since 1 and 2 are used as subscripts for the pairs, I use ⊤ instead of 1 as a truth value. 14 With de se binding, multi-tier interpretation cannot be available. Otherwise a de se interpretation should be possible for he in (i) because the unmarked masc is licensed on the multi-tier interpretation for the pair ⟨BB, George⟩. (i)

* Brigitte dreamed λ x that hex kissed George.

180 | Uli Sauerland

3 English reflexives The evidence in this section comes from the interpretation of reflexive pronouns in English. First, I spell out a presuppositional account of Condition A – the condition that often requires reflexives like himself to be locally bound in English. Then I argue that there is weakened projection of Condition A across the three domains considered above: alternative quantification, de se binding, and distributive quantification. The presuppositional account of Condition A that I propose can be seen as a semantically precise formulation of parts of the approach to binding theory of Pollard and Sag (1992) and Reinhart and Reuland (1993). These authors argue that a self -reflexive in a verbal argument position reflexive-marks a predicate.¹⁵ I propose that the notion of reflexive marking is semantically captured as the following presupposition of self : (24) self(P)(x)(y) presupposes that x = y and denotes P. When it is the object of a transitive verb, self can incorporate into the verb as shown in (25b) and (25c) (for (25a)): (25) a. Mary criticized herself. b. Mary λ x x self-criticized herx c. Mary self-criticized herx

(binding) (coreference)

Note that the presuppositional approach doesn’t require the object pronoun her in (25a) to be bound. Though binding is an option, representation (25c) is also possible. But the reflexivity presupposition of self in (25c) is only satisfied if the variable assignment that context provides assigns Mary to the variable x. The case of reflexive binding most relevant for the following are exceptional case marking (ECM) structures. In example (26) herself is the subject of to win, but not an argument of expect – only the entire infinitival clause is an argument of expect. (26) Mary expected herself to win.

15 Charnavel and Sportiche (2012) argue that in French a similar condition applies but not only to reflexives in argument positions. I believe such variation can be accommodated within my proposal.

The form of bound pronouns: Towards a uniform account |

181

Therefore the reflexive in (26) cannot reflexive-mark a lexical predicate since (26) doesn’t contain any predicate of type ⟨e, et⟩. I follow Lechner (2012) who proposes that a suitable predicate for reflexive marking must be created in the syntax by λ-abstraction. Following Sauerland (1998) and Nissenbaum (2000), Lechner assumes that movement can target the position between an existing λ-operator and its argument, trigger the insertion of a second λ-operator, and thereby create a binary predicate. On the presuppositional approach, this allows the analyses of (26) in (27). Both representations in (27) can be derived by first moving the matrix subject Mary, then the embedded subject herself, and finally moving self to a position where its sister is the binary predicate λ y λ x x expected [ y to win ]. As already noted in (25), (26) also permits both representations (27a), where herz is bound by Mary, and (27b), where coreference of Mary and herz is presupposed, but no binding relation obtains. (27) a. Mary λz z herz self [λ y λ x x expected [ y to win ] ] b. Mary herz self [λ y λ x x expected [ y to win ] ]

(binding) (coreference)

In the ECM case, there is evidence for the ambiguity in (27): Fiengo and May (1994) discuss the strict interpretation in VP-ellipsis as in (28) (see also Oku 1998). (28) Romney expected himself to win and his wife did too. Data from focus alternative interpretation like (29) corroborate this point.¹⁶ (29) Only Romney expected himself to win. a. Nobody other than Romney expected himself to win. b. Nobody other than Romney expected Romney to win. To determine whether weakened projection (i.e. front-tier application) can apply to the meaning of self we need to determine whether self is purely presuppositional. While self doesn’t satisfy the formal definition given above (self cannot be interpreted as a restricted identity function), it intuitively doesn’t do anything but

16 I focus here on the new data in (29) because the account of the ellipsis data like (28) require also an account of Condition B. I believe though that the ellipsis fact in (28) can be fully accounted for within my account. One related fact Hestvik (1995) discusses is the differing availability of a strict reading in (i b). At this point, I have no account for the contrast in (i). (i)

a. John defended himself well, and Bill did too. (defend Bill / *defend John) b. John defended himself better than Bill did. (defend Bill / defend John)

182 | Uli Sauerland

add a presupposition. I therefore define a second notion of pure presuppositionality as follows in addition to the one defined in (12): (30) A functor F applied to function g is purely presuppositional if both of the following hold: a. F doesn’t change the type of g b. wherever F(g) is defined, g is defined too and F(g) and g have the same result In this case of pure presuppositionality too, I propose that the alternative interpretation procedure can apply the presupposition either only to the front tier or to both tiers. For the front-tier interpretation, the identity function of the same type as the purely presuppositional function must be applied on the other tier – i.e. the type τ in (12) needs to be understood as the argument type of F, but otherwise (12) can remain unchanged. Consider the application of weakened projection to (29). (29) allows both the bound and coreferent representations from (27). If the multi-tier interpretation of self is chosen, the reflexivity presupposition can only be satisfied on the bound representation. But, if the front-tier interpretation of self is chosen, both binding and coreference satisfy the reflexivity presupposition on the front tier. The coreferent representation corresponds to the unbound reading noted for (29). In this way, the presence of the coreferent interpretation argues for weakened projection of the reflexive presupposition. The clause-mate reflexive raises an additional issue: For most speakers, examples like (31) only allow the bound interpretation of the reflexive in contrast to the ECM example just discussed. (31) Only Romney praised himself. I propose the following account of the difference between the ECM case (30) and the transitive (31): First of all, note that the transitive case could in principle be structurally ambiguous between a structure like (25) and a structure where both the subject and object undergo movement just like the ECM structure in (27). I assume, however, that syntactic economy considerations block the latter type of structure for (31) (e.g. Chomsky 1995). Then, only the two structures noted in (25) are syntactically available. Both of these structures contain a verbal complex self criticize composed of the reflexive and the verb. Note that while the reflexivizer on its own has a purely presuppositional interpretation, the complex self criticize does not. I suggest therefore that the relevant level for principle (12) may be the maximal syntactic head (i.e. a compound head possibly containing other heads).

The form of bound pronouns: Towards a uniform account |

183

Then the front-tier interpretation is predicted to be unavailable in (31), and therefore the reflexivity presupposition cannot be weakened. Furthermore, only the bound representation of (31) is compatible with a multi-tier reflexivity presupposition. As a result, we predict that (31) only allows a bound interpretation of him.¹⁷ The uniform analysis of weakened projection I propose in combination with the application of weakened projection to reflexive interpretation in alternative quantification makes a prediction: Similar effects of weakened projection should be detectable also under de se binding and distributive binding. I believe that there is considerable evidence that this prediction is borne out, which corroborates the uniform analysis in an interesting way. In the following, I discuss first a case with reflexives and de se binding (data from Sharvit 2011) and then a case with reciprocals and distributivity (data from Higginbotham 1980). Consider example (32) of Sharvit 2011. Sharvit points out that even when a de re interpretation of the object of vote for is targeted, the reflexive himself must be used. This is surprising because the PRO-subject of vote for must receive a de se-interpretation. Therefore, the verb vote for doesn’t seem to receive a reflexive interpretation. (32) Every male politician promised Palin PRO to vote for himself/*him. Sharvit’s approach to (32) essentially adds a complex additional clause of Condition A for cases like (32). Note though that (32) follows from the approach to weakened projection we have developed here. On this approach the arguments of vote for are two pairs: The de se subject is interpreted as pair of a male politician and its de se counterpart. The object, when it is interpreted de re, denotes the pair of the same male politician twice.¹⁸ Therefore the reflexivity presupposition is satisfied on the front tier even when the object is interpreted de re. Furthermore the same considerations as in footnote 17 argue that a pronoun would not be acceptable in (32). Therefore, the obligatory use of the reflexive in (32) is expected.¹⁹

17 With a coreferent interpretation, (31) is also not fully acceptable with a pronoun instead of a reflexive (Heim, 1998). Fox (2003) independently suggests that the analysis of ellipsis requires that pronouns come with a silent counterpart of self that carries an anti-reflexive presupposition (i.e. anti-self(P) = λxλy : x ≠ y . P(y)(x)). If anti-reflexivity must be applied multi-tier just like reflexivity, then it follows that a coreferent interpretation of (31) is unavailable. 18 I assume that de re involves furthermore application of a concept generator as specified in Percus and Sauerland 2003 (see also Pearson 2013), but this is irrelevant for the reflexivity presupposition. 19 One prediction that I hope to still explore is that examples like (25) under an attitude verb should behave differently from (32) as the multi-tier interpretation should be required in this case as well.

184 | Uli Sauerland

Secondly, consider example (33) with a reciprocal under distributive quantification. As Higginbotham (1980) has pointed out already, the salient interpretation (‘John believes that he’s taller than Mary, and Mary believes that she’s taller than John.’) of (33) is one where the predicate are taller than is interpreted neither reflexively nor reciprocally – indeed, they doesn’t receive a plural interpretation in (33). (33) John and Mary believe that they are taller than each other. Fact (33) follows very naturally from my approach. I assume that reciprocals also must reflexive mark a binary predicate, and furthermore assume an approach to reciprocal interpretation similar to Schwarzschild 1996. Specifically, note that the two components introduced by distributive binding correspond to the range and contrast argument of reciprocals in Schwarzschild’s analysis. Therefore I propose the lexical entry in (34) for the reciprocal:²⁰ (34) each otheri g (P)2 = 1 iff. P(g(i)1 ) is defined and ∀y ∈ g(i)1 \ g(i)2 : P(y) The approach predicts that reciprocals must be interpreted bound by a distributive plural since they provide the necessary pair interpretation of the variable.²¹ For (33), the approach predicts the logical form representation (35), where self reflexivizes the derived binary predicate λ y λ z [y are taller than z]: (35) John and Mary *[λ x x believe that each otherx theyx self λ y λ z [y are taller than z]] Because both each other and they are distributively bound by the same binder, the first components of their assignments will both refer to the plural individual John⊕Mary. Therefore the reflexivity presupposition of self is satisfied within the embedded clause.

20 I assume here that the first component each otheri g (P)1 = P(g(i)1 ). 21 A focussed binder or a de se binder, though also providing a pair in the variable assignment, would never necessarily denote a part of a plurality and an atomic part thereof as required by the reciprocal.

The form of bound pronouns: Towards a uniform account |

185

4 Conclusions In this paper, I have explored a unified approach to a set of phenomena that haven’t received a unified treatment before: quantification over focus alternatives, distributive binding, and de se binding. My approach has focussed on a commonality across all these phenomena: weakened projection. Specifically, presuppositions from the scope of the quantifier seem to not be satisfied even for all the instances relevant for the evaluation of the quantifier in all three cases. I believe that the unifying approach has at least two advantages over existing proposals. In particular, I have shown that a bound pronoun agreement mechanism is unnecessary at least for what have been held to be core cases of the phenomenon by Kratzer (2009) and others. Instead, the present approach allows a fully general analysis of ϕ-features building on Cooper 1979 and Sauerland 2003. The second advantage of the unified approach is discussed in Section 3. There I showed that on the basis of the unified approach a new direction to three open problems in binding theory is possible: the problem of the difference between ECM vs. coargument reflexives, the problem of binding reflexives under de se, and the problem of scope with reciprocals. The unified approach also may lead us to discover interesting differences between the phenomena I started out with. Specifically, one such issue that for reasons of space I cannot discuss exhaustively here concerns the licensing of number on full DPs. As I argued above, example (2) shows that number marking on full DPs can be licensed by single tier evaluation with focus alternatives. The facts in (36) and (37) indicate, however, that with de se and distributive binding this option doesn’t exist. Specifically, (36) is set up such that in the actual world the plural children should be licensed. But only the singular is possible. (36) George, who has three children, was dreaming that he was the mother of one child and that he was breastfeeding his child(*-ren). For (37), assume that the men aren’t brothers, and the mothers each only like one bride. Since the plurality of men have a plurality of mothers, plural marking on mothers ought to be licensed. But actually singular is required. (37) These three men each are marrying the woman that their mother likes best. *…that their mothers like best. At this point, I suggest that this difference relates to the issue noted at the end of Section 2.4: With alternative semantics, composition in the front tier evaluation

186 | Uli Sauerland

will always be possible since it is just computing the ordinary meaning. With de se binding and distributive binding, however, composition in the front tier may be impossible. In fact, composition is probably impossible in (36) at a relevant level: in (36), the predicate child in the evaluation world is applied to the real world entity George. The counterpart ontology of Lewis (1968) predicts this to be illicit since George isn’t an individual of the world where child is evaluated. Example (37) might also be impossible to compose if the transitive noun mother cannot be interpreted cumulatively: Then there is no individual or group of individuals that stands in the mother relation to the plurality of men. Therefore both (36) and (37) could be explained using the assumption that if evaluation of the front tier violates a condition of existence in the current world of evaluation, the front tier evaluation is abandoned without triggering a presupposition failure. Then only the second tier evaluation is available to license number marking.

References Abusch, Dorit. 2010. Presupposition triggering from alternatives. Journal of Semantics 27:37– 80. Alxatib, Sam, Peter Pagin, and Uli Sauerland. 2013. Acceptable contradictions: Pragmatics or semantics? a reply to Cobreros et al. Journal of Philosophical Logic 42:619–634. Beck, Sigrid. 2006. Intervention effects follow from focus interpretation. Natural Language Semantics 14:1–56. Charnavel, Isabelle, and Dominique Sportiche. 2012. Anaphor binding domains in French. Ms. Harvard and UCLA. Chomsky, Noam. 1995. The minimalist program. Cambridge, Mass.: MIT Press. Collins, Chris, and Paul M. Postal. 2012. Imposters: A study of pronominal agreement. Cambridge, Mass.: MIT Press. Cooper, Robin. 1979. The interpretation of pronouns. In Selections from the third Groningen round table, syntax and semantics, volume 10, ed. F. Heny and H. Schnelle, 61–92. New York: Academic Press. Fiengo, Robert, and Robert May. 1994. Indices and identity. Cambridge, Mass.: MIT Press. Fodor, Janet Dean. 1970. The linguistic description of opaque contents. Doctoral Dissertation, Massachusetts Institute of Technology, Cambridge, Mass. Fox, Danny. 2003. Condition B: A note on Jacobson’s paper. Handout, MIT (available from the author’s homepage). Gajewski, Jon. 2005. Neg-raising: Polarity and presupposition. Doctoral Dissertation, Massachusetts Institute of Technology, Cambridge, Mass. Heim, Irene. 1998. Anaphora and semantic interpretation: A reinterpretation of Reinhart’s approach. In The interpretive tract, MITWPL 25, ed. Uli Sauerland and Orin Percus, 205– 246. Cambridge, Mass.: MIT Working Papers in Linguistics.

The form of bound pronouns: Towards a uniform account |

187

Heim, Irene. 2008. Features on bound pronouns. In Phi theory: Phi-features across modules and interfaces, ed. David Adger, Daniel Harbour, and Susana Béjar, 35–56. Oxford, UK: Oxford University Press. Hestvik, Arild. 1995. Reflexives and ellipsis. Natural Language Semantics 3:211–237. von Heusinger, Klaus. 2007. Alternative semantics for definite NPs. In On information structure, meaning and form–generalizations across languages, ed. Kerstin Schwabe and Susanne Winkler, 485–508. Amsterdam, The Netherlands: Benjamins. Higginbotham, James. 1980. Reciprocal interpretation. Journal of Linguistics Research . Jacobson, Pauline. 2012. Direct compositionality and ‘uninterpretability’: The case of (sometimes)‘uninterpretable’features on pronouns. Journal of Semantics 29:305–343. Krahmer, Emiel. 1998. Presupposition and anaphora. Stanford, Calif.: CSLI. Kratzer, Angelika. 1998. More structural analogies between pronouns and tenses. In Proceedings of SALT 8, ed. Devon Strolovitch and Aaron Lawson, 92–110. Ithaca, N.Y.: CLC Publications. Kratzer, Angelika. 2009. Making a pronoun: Fake indexicals as windows into the properties of pronouns. Linguistic Inquiry 40:187 – 237. Lechner, Winfried. 2012. Towards a theory of transparent reflexivity. Unpublished Ms., University of Athens. Lewis, David. 1968. Counterpart theory and quantified modal logic. Journal of Philosophy 65:113–126. Reprinted with postscripts in Lewis 1983. Lewis, David K. 1983. Philosophical papers, volume I. Mayr, Clemens. 2012. Focusing bound pronouns. Natural Language Semantics 20:299–348. Nissenbaum, Jon. 2000. Investigations of covert phrase movement. Doctoral Dissertation, Massachusetts Institute of Technology, Cambridge, Mass. Oku, Satoshi. 1998. A theory of selection and reconstruction in the minimalist perspective. Doctoral Dissertation, University of Connecticut, Storrs. Pearson, Hazel A. 2013. The sense of self. Doctoral Dissertation, Harvard University, Cambridge, Mass. Percus, Orin, and Uli Sauerland. 2003. On the LFs of attitude reports. In Proceedings of Sinn und Bedeutung 2002, ed. Matthias Weisgerber. Konstanz, Germany: Universität Konstanz. Pollard, Carl, and Ivan A. Sag. 1992. Anaphors in English and the scope of binding theory. Linguistic Inquiry 23:261–305. Reinhart, Tanya, and Eric Reuland. 1993. Reflexivity. Linguistic Inquiry 24:657–720. Romoli, Jacopo. 2013. A scalar implicature-based approach to Neg-raising. Linguistics and Philosophy (to appear). Rooth, Mats. 1992. A theory of focus interpretation. Natural Language Semantics 1:75–116. Sauerland, Uli. 1998. Plurals, derived predicates and reciprocals. In The interpretive tract, MITWPL 25, ed. Uli Sauerland and Orin Percus, 177–204. Cambridge, Mass.: MIT Working Papers in Linguistics. Sauerland, Uli. 2003. A new semantics for number. In The Proceedings of SALT 13, ed. Robert B. Young and Yuping Zhou, 258–275. Cornell University, Ithaca, N.Y.: CLC-Publications. Sauerland, Uli, Jan Anderssen, and Kazuko Yatsushiro. 2005. The plural is semantically unmarked. In Linguistic evidence – empirical, theoretical, and computational perspectives, ed. Stephan Kepser and Marga Reis. Berlin, Germany: Mouton de Gruyter. Schlenker, Philippe. 2013. Iconic features. Lingbuzz/001808. Schwarzschild, Roger. 1996. Pluralities. Dordrecht, Netherlands: Kluwer.

188 | Uli Sauerland

Scontras, Gregory. 2013. A unified semantics for number marking, numerals, and nominal structure. In Proceedings of sub 17, ed. Emmanuel Chemla, Vincent Homer, and Gregoire Winterstein, 545–562. Paris, France: Semanticsarchive. Sharvit, Yael. 2011. Covaluation and unexpected BT effects. Journal of Semantics 28:55–106. Simons, Mandy. 2001. On the conversational basis of some presuppositions. In Proceedings of SALT 11, ed. R. Hastings et al. Ithaca, N.Y.: CLC Publications, Cornell University. Sudo, Yasutada. 2012. On the semantics of phi features on pronouns. Doctoral Dissertation, Massachusetts Institute of Technology, Cambridge, Mass. Walker, Andreas. 2012. Focus, uniqueness and soft presupposition triggers. In Logic, language and meaning, 460–469. Springer.

Stefan Hinterwimmer (University of Cologne) and Peter Bosch (University of Osnabrück)

Demonstrative pronouns and perspective Abstract: In this paper we argue that the co-referential and binding properties of German demonstrative pronouns cannot be accounted for by assuming that they avoid topics or grammatical subjects, as has been argued for in previous literature. We propose instead that demonstrative pronouns avoid perspectival centres: They cannot be interpreted as being identical to the individual from whose perspective the event or state introduced by the clause containing them is depicted/evaluated. What seems to be avoidance of subject or topic is just an epiphenomenon resulting from the tendency for subjects to function as topics, and topics as perspectival centres. Whenever the notions come apart, though, demonstrative pronouns can be interpreted as referring to or bound by (the denotations of) DPs functioning as topics, but they can never refer to perspectival centres. Finally, we show that in cases where there is a choice among potential antecedents or binders that do not function as perspectival centres, demonstrative pronouns have a weak (i.e. violable) tendency to avoid topics. This we suggest is due to the fact that they belong to the formally more marked of the two German pronoun series. Keywords: Demonstrative Pronouns, Perspective, Free Indirect Discourse, Topicality, Binding, Perspectival Centre, Propositional Attitudes, Subjects.

1 Introduction The topic of this paper are the co-referential and binding options of German demonstrative pronouns (henceforth: DPros) such as der/die/das as opposed to the personal pronouns (henceforth: PPros) er/sie/es (he/she/it). We will show that the distribution of DPros can be explained, to a considerable extent, via a constraint that has not figured prominently in the discussion of DPros so far: A DPro may not be interpreted as referring to the perspectival centre i.e., to the individual from whose perspective the event or state is depicted that is introduced by the clause in which the DPro occurs. The concept of topic avoidance, for which we argued in previous work (Bosch and Umbach 2006, Hinterwimmer 2015, Bosch and Hinterwimmer to appear), is, as we will show below, unable to account for the full range of data, but we believe that it still plays an important role: On the one hand, the aboutness topic of a sentence is often (but crucially not always)

190 | Stefan Hinterwimmer and Peter Bosch

identical to its perspectival centre. On the other hand, in cases where there is still a choice among the individuals not functioning as the perspectival centre, the DPro, being the marked pronoun variant, may not be resolved to the most prominent one among them, which would be the entity corresponding to the aboutness topic of the respective sentence. The paper is structured as follows: In Section 2, we summarize the analyses argued for in Bosch and Umbach (2006) and Hinterwimmer (2015), according to which DPros in their referential uses are subject to a topic avoidance constraint and in their bound uses, as Hinterwimmer (2015) argued, also to a subject avoidance constraint. In Section 3 we present data that are problematic for both accounts, and in Section 4 we sketch the outlines of a novel analysis that is able to account for the full range of facts under consideration, leaving a concrete formal implementation to future work. Section 5 is the conclusion.

2 The Topic/Subject Avoidance Constraint 2.1 Co-referential uses of DPros DPros often have a strong bias against being resolved to the subject of the preceding sentence while PPros often show a (somewhat weaker) preference for subjects. This was shown in Bosch, Rozario & Zhao (2003) and in Bosch, Katz & Umbach (2007) on the basis of corpus studies as well as experimental results. (1)

Pauli wollte mit Peterj laufen gehen. Aber {eri,j /derj } war leider erkältet. Pauli wanted to go running with Peterj . But he {PProi,j /DProj } had a cold unfortunately. (from Bosch et al. 2003)

At the same time, there is clear evidence that subject avoidance cannot be the crucial factor regulating the distribution of co-referential DPros. First of all, whenever the grammatical subject does not occur in its canonical clause-initial position, the anti-subject bias becomes weaker: While in (2a) there is a strong preference to interpret the DPro as picking up the individual denoted by the object DP den Patienten auf Zimmer 3, this preference is less strong in (2b) – in other words, in the case of (2b), where the direct object DP has been moved across the subject to clause-initial position, the individual denoted by the subject DP der neue Assistenzarzt seems to be available, possibly even preferred, as an antecedent for

Demonstrative pronouns and perspective |

191

the DPro as well (see Bosch and Umbach 2006 for details of related experimental results). (2) a. [Der neue Assistenzarzt]i untersuchte [den Patienten auf Zimmer 3]j . {Derj /Eri,j } war sehr geduldig. [The new assistant doctor]i examined [the patient in room 3]j . He {DProj / PProi,j } was very patient. b. [Den Patienten auf Zimmer 3]i [untersuchte der neue Assistenzarzt]j . {Derj /Eri,j }war sehr geduldig. [The patient in room 3]i was examined by [the new assistant doctor]j . He {DProj /PProi,j } was very patient. Second, and more importantly, in the following discourse there is a strong preference for the DPro to pick up Peter, despite the fact that the DP referring to him is not only the grammatical subject of the preceding sentence, but also occurs in canonical clause-initial position. (3) Woher Karli das weiß? Peterj hat es ihmi gesagt. {Derj /Eri,j } war gerade hier. How does Karli know? Peterj told him. He {DProj /PProi,j } has just been here. (from Bosch and Umbach 2006) Based on these and similar observations, Bosch and Umbach (2006) conclude that the behaviour of DPros cannot adequately be captured in terms of grammatical relations, but rather in terms of the information-structural notion of discourse topic – where discourse topics are the entities the whole discourse segment is intuitively felt to convey relevant information about. Following Prince (1992), Bosch and Umbach (2006) assume discourse topics to be discourse-old, i.e., they were introduced into the discourse before, but not as new referents in the immediately preceding sentence. According to Bosch and Umbach (2006), DPros are precluded from picking up individuals functioning as current discourse topics (‘topic avoidance hypothesis’). Since the referents of grammatical subjects are discourse topics by default (Hockett 1958), subject avoidance may simply be taken as the default case of topic avoidance, Observations as in (3) though, make the difference between subject avoidance and topic avoidance clear. In (3), Peter in contrast to Karl is not a potential discourse topic. While Karl is definitely discourse-old by the time the referential value of the DPro has to be determined, Peter has just been introduced into the discourse in the immediately preceding sentence. Accordingly, the DPro must be resolved to Peter, not to Karl, irrespective of the fact that the DP referring to the former is the grammatical subject of its clause.

192 | Stefan Hinterwimmer and Peter Bosch

2.2 DPros as Donkey Pronouns As was already observed by Wiltschko (1999), DPros may also occur as so-called donkey pronouns, i.e., as pronouns that behave as if they were variables bound by a quantifier, although they are outside of the scope of that quantifier. In a sentence such as (4), the DPro den is understood as varying with the donkeys introduced by the indefinite einen Esel (‘a donkey’) in the if -clause, just as the reference of the PPro er is understood as varying with the farmers introduced by the indefinite ein Bauer (‘a farmer’). The resulting interpretation can roughly be paraphrased as For most pairs of farmers x and donkeys y, such that x owns y, it is the case that x beats y. (4) Wenn [ein Bauer]i [einen Esel]j besitzt, schlägt eri denj . If [a farmer]i owns [a donkey]j , hei beats itj [DPro]. Importantly, however, as in the straightforward (non-co-varying) cases of coreference that we saw before, PPros and DPros are not interchangeable in such sentences either. Replacing the PPro in subject position in the main clause in (4) by a DPro and the DPro in object position by a PPro, as shown in (5), leads to an absurd¹ reading according to which it is the donkeys who beat the farmers, not vice versa (cf. Hinterwimmer 2015): (5) Wenn [ein Bauer]i [einen Esel]j besitzt, schlägt deri ihnj . If [a farmer]i owns [a donkey]j , itj [DPro] beats himi . Observe furthermore in a sentence such as (6a), where the subject of the if -clause occurs in canonical position, the DPro has a preference for being interpreted as varying with the individuals introduced by the indirect object indefinite, while in a sentence such as (6b), where the indirect object has been fronted across the subject, it has a preference for being interpreted as varying with individuals introduced by the subject-indefinite (this interpretation is even clearer when (6b) is read with a strong focus accent on the noun Schäferhund). (6) a. Wenn [ein Schäferhund]i [einem Wolf]j begegnet, greift derj meistens sofort an. If [an Alsatian]i encounters [a wolf]j , itj [DPro] usually attacks immediately.

1 Even though the reading is absurd, it is not excluded in the narrow linguistic sense in which topic reference in (1)–(3) was excluded. Its exclusion may well be the business of pragmatics; see below in Section 3.

Demonstrative pronouns and perspective |

193

b. Wenn [einem Wolf]i [ein Schäferhund]j begegnet, greift derj meistens sofort an. If [an Alsatian]j encounters [a wolf]i , itj [DPro] usually attacks immediately. These observations seem highly reminiscent of the observations in the preceding section and a unified explanation is called for. The notion of discourse topic avoidance, however, cannot be applied to the cases with indefinite antecedents, since indefinites introduce new discourse referents by definition (Heim 1982). Hinterwimmer (2015) thus assumes that avoidance of aboutness topic rather than of discourse topic is the relevant notion, since it is well known that not only referential expressions (i.e., proper names, definites, and demonstratives) but also indefinites can establish aboutness topics (Reinhart 1981, Endriss 2009, Endriss and Hinterwimmer 2009). Intuitively, the aboutness topic of a sentence is the entity about which the predicate denoted by the rest of the sentence is understood to provide relevant information (Strawson 1964, Bosch 1980, Reinhart 1981). In slightly more formal terms, aboutness topics are conceived of as the logical subject of the predicate denoted by the rest of the sentence, the comment. This means that the notion of aboutness topic, unlike the notion of discourse topic, is (in principle) also defined for isolated sentences. But within a coherent discourse, discourse topic and aboutness topic would ordinarily coincide. Furthermore, in the default case, the logical subject corresponds to the grammatical subject. The grammatical subject occupying the highest argument position and thus canonically appearing in clause-initial position, the required configuration automatically holds at the surface, since the syntactic sister of the subjectDP is guaranteed to be a one-place predicate that can be applied to the denotation of that DP. As we have already seen in Section 2.1, however, in languages like German it is also possible to put another DP in sentence-initial position (or another XP, for that matter – constituents other than DPs, such as PPs or adverbial clauses can be fronted in German as well). In many cases (but not all cases – focality might also be a reason, as already said above) fronting serves to mark the fronted constituent as the aboutness topic, thus creating the same configuration that holds for subjects and their syntactic sisters in the standard case². For the cases discussed in Section 2.1, the hypothesis that DPros avoid aboutness topics makes the same predictions as the assumption that they avoid discourse topics: In (1) and (2a), where the grammatical subject of the first sentence

2 For English, we assume the required configuration to be created at LF via covert movement operations in cases where DPs other than the grammatical subjects are the aboutness topics of their sentences.

194 | Stefan Hinterwimmer and Peter Bosch

occurs in canonical clause-initial position, the discourse topics are also the aboutness topics, and the respective individuals can thus not be picked up by the DPros in the second sentences. In (2b), in contrast, where the direct object has been fronted, there are two options: The first one is to interpret the fronted object as the aboutness topic of its sentence. In that case, the surface configuration is retained at LF and the object DP is the subject of the logical predicate denoted by its sister constituent, as desired. Accordingly, the individual it denotes cannot be picked up by the DPro in the following sentence. The second option is to assume that the object has been fronted in order to mark it as contrastive focus (in which case it receives the main accent of the sentence). In Hinterwimmer (2015) it is assumed that contrastive foci are optionally reconstructed into their clause-internal base positions at LF (see Hinterwimmer 2006 and 2008, and the references cited there for independent empirical evidence for this assumption and detailed discussion). Concerning the finite verb in second position, we follow standard assumptions that its surface position is the result of PF-movement, i.e., that it is always interpreted in its VP-internal base position. Hence, if the direct object is reconstructed into its VP-internal base position, the grammatical subject is again the logical subject of the predicate denoted by its syntactic sister (at LF), and the DPro cannot pick up the individual it denotes, but rather has to pick up the one denoted by the object DP. Finally, concerning the discourse in (3), note that the main accent of the second sentence has to be placed on the grammatical subject, Peter, the reason being that in the context of the first sentence it provides the only new information. It can be inferred from that sentence that somebody told Karl the fact under discussion. The focal subject DP Peter can thus be reconstructed. Concerning the weak pronoun es, let us assume that it has been moved across the indirect object for purely phonological reasons, i.e., at PF only: It is well known that weak pronouns in German matrix clauses have a very strong tendency to occur directly after the finite verb, which is presumably due to phonological requirements. The indirect object can then function as the logical subject and thus the aboutness topic in cases like these, too: At LF, where the focal subject-DP has been reconstructed into its base VP-internal position and where neither the weak pronoun es nor the finite auxiliary verb have been moved out of their base positions, the indirect object DP ihm occupies the highest position in the clause, its syntactic sister denoting a oneplace predicate. We thus again have the logical subject-predicate configuration that is constitutive of aboutness topicality. Accordingly, the DPro can only pick up the individual denoted by the focal subject-DP, Peter. The predictions made by discourse topic avoidance and aboutness topic avoidance thus are basically the same (apart from an ambiguity in the case of (2b)) for the non-co-varying straightforward co-reference cases in (1)–(3). But what

Demonstrative pronouns and perspective |

195

about cases like the ones in (4)–(6) that were the initial motivation for replacing the notion of discourse topic avoidance by the notion of aboutness topic avoidance? The simple logical subject-predicate relations that can be assumed in (1)–(3) make no sense for (4)–(6). Intuitively, (4)–(6) are not about a particular farmer, donkey, Alsatian, or wolf, but about farmers, donkeys, Alsatians, and wolves in general. This intuition is supported by the common test for aboutness topicality. If a sentence s can figure as a natural answer to the question What about α? or can naturally be preceded by a sentence like I tell you something about α, then α is the aboutness topic of s, otherwise it is not.³ Crucially, while an example such as (6a), repeated here as (8b), sounds natural as a continuation of the sentence in (8a), it is odd as a continuation of (8c), and likewise for (6b), repeated here as (9b). (8) a.

Ich erzähle Dir etwas über Schäferhunde. I tell you something about Alsatians. b. Wenn [ein Schäferhund]i [einem Wolf]j begegnet, greift derj meistens sofort an. If [an Alsatian]i encounters [a wolf]j , itj [DPro] usually attacks immediately. c. Ich erzähle Dir etwas über Wölfe. I tell you something about wolves. d. # Wenn [ein Schäferhund]i [einem Wolf]j begegnet, greift derj meistens sofort an. If [an Alsatian]i encounters [a wolf]j , itj [DPro] usually attacks immediately.

3 Frey (2004:6), for example, uses this test to show that DPs c-commanding sentence adverbs like probably at the surface in German are aboutness topics, while DPs c-commanded by sentence adverbs are not: (7)

Ich erzähle dir etwas über Maria. I tell you something about Maria a. Nächstes Jahr wird Maria wahrscheinlich nach London gehen. [next year will Maria probably to London go] Next year Maria will probably go to London. b. # Nächstes Jahr wird wahrscheinlich Maria nach London gehen.

196 | Stefan Hinterwimmer and Peter Bosch

(9) a.

Ich erzähle Dir etwas über Wölfe. I tell you something about wolves. b. Wenn [einem Wolf]i [ein Schäferhund]j begegnet, greift derj meistens sofort an. If [an Alsatian]j encounters [a wolf]i , itj [DPro] usually attacks immediately. c. Ich erzähle Dir etwas über Schäferhunde. I tell you something about Alsatians. b. # Wenn [einem Wolf]i [ein Schäferhund]j begegnet, greift derj meistens sofort an. If [an Alsatian]j encounters [a wolf]i , itj [DPro] usually attacks immediately.

These intuitive judgements are accounted for by the analysis in Hinterwimmer (2015) in the following way: First of all, quantificational adverbs (Q-adverbs) such as meistens (‘usually’), immer (‘always’) and the covert generic operator are assumed to quantify over minimal situations, i.e. situations that contain nothing more than what is strictly speaking required to make the respective situation predicate true (Berman 1991, von Fintel 1994). Secondly, contra Lewis (1975), Heim (1982) and Kratzer (1986), it is assumed that if -clauses never provide the domains of quantification for Q-adverbs or the generic operator, but rather that they are always interpreted as “real” conditional antecedents, i.e., as the domains of quantification for a covert quantifier over epistemically accessible worlds or situations (independent empirical evidence for this assumption is provided in Hinterwimmer 2015). The difference to standard conditionals is just that in adverbially quantified or generic sentences the worlds or situations quantified over are not accessible from the world of evaluation, but rather from the situations quantified over by the Q-adverb or generic operator. These latter situations, i.e., the domains of quantification for the Q-adverbs or the generic operator, are in turn determined via the interplay of contextual information, world knowledge, and sentence-internal information. Technically, the domain of quantification for a Q-adverb or the generic operator is initially given in the form of a free variable ranging over situation predicates for which a value needs to be found – just as it is assumed by von Fintel (1994) for monoclausal adverbially quantified or generic sentences, i.e., ones that do not contain an if - or when-clause. Concerning sentences like (4)–(6), Hinterwimmer (2015) assumes that the indefinite occupying the highest DP-position within the respective if -clause – be it the grammatical subject in canonical clause-initial position or a fronted direct or indirect object – provides the domain of quantification in the following way: Via a simple type-shifting mechanism the respective indefinite is shifted from its

Demonstrative pronouns and perspective |

197

original denotation as a generalized quantifier with existential force to a (characteristic function of a) set of situations each of which contains an individual of the respective kind, i.e., a farmer, a donkey, an Alsatian, etc. The free domain variable is then resolved to the respective situation predicate. The resulting interpretation for the sentence in (6a) can (roughly) be paraphrased as follows: For most minimal situations s containing an Alsatian it is the case that all epistemic counterparts of s that are extended to a situation of an Alsatian encountering a wolf are also situations of the respective Alsatian attacking. – A very rough but more intuitive paraphrase would be: Most Alsatians are such that if they encounter a wolf, they attack immediately. For our current purposes the indefinite providing the domain of quantification for the respective Q-adverb or the covert generic operator can now be seen as providing the higher-order logical subject of a higher-order predication. It is said about the respective set of situations that most (all/many/…) of them have a certain property – namely the property that if they are extended in the way specified by the if -clause, they satisfy the main clause predicate. Because of the one-toone relation between the sets of individuals and the sets of minimal situations containing them, this is equivalent to a statement about the corresponding sets of individuals: Hence the intuition that (6a), for example, makes a statement about Alsatians in general. Because the subject indefinite in (6a) and the fronted indirect object in (6b) are thus the higher order logical subjects of their sentences, the DPros in the main clause cannot be interpreted as co-varying with the individuals introduced by the respective indefinite without violating the topic avoidance constraint. The only available option is thus to interpret them as co-varying with the individuals introduced by the respective non-topical indefinite. Concerning the question of how such co-varying interpretations come about in principle – i.e. independently of the question of whether the pronoun in question is a PPro or a DPro – two analyses are available that are in principle compatible with the account just sketched: According to the first one, indefinites in contrast to other quantificational DPs are able to extend their scope beyond their c-command domain, i.e., they can effectively bind variables outside of their syntactic scope (see Groenendijk and Stokhof 1991 and much subsequent work for details). Accordingly, both PPros and DPros can be assumed to be variables that are either resolved to some contextually salient individual (this corresponds to the co-referential interpretation), or as bound by a quantificational DP – be it under c-command, i.e., in the standard way available to all quantificational DPs (see Section 2.3 below) or in the special dynamic way that is only available to indefinite DPs. According to the second analysis at least some occurrences of pronouns – those discussed in the current section being among them – actually correspond to

198 | Stefan Hinterwimmer and Peter Bosch

definite descriptions in disguise (see Evans 1980, Heim 1990 and Elbourne 2005 for different technical implementations of this general idea). On Elbourne’s (2005) account, for example, pronouns in general are the spell-out of a definite determiner combined with an empty NP whose denotation needs to be retrieved from the context. In cases like those discussed in the current section, the respective empty NP is thus taken to be identical to the NP contained in the indefinite DP with whose denotation the denotation of the pronoun varies. Hence, a pronoun co-varying with an indefinite DP such as a donkey thus corresponds to the definite description the donkey. Crucially, the uniqueness presupposition associated with the definite determiner is assumed to be relativized to the situations quantified over by the respective Q-adverb or generic operator, i.e. definite descriptions are assumed to contain situation variables that can either be resolved to some contextually salient situation or be bound by a c-commanding quantifier. On this view, co-variation of indefinites and pronouns thus comes about indirectly – not via the binding of individual variables by the respective indefinites, but rather via the binding of situation variables by the Q-adverbs whose domains of quantification are provided by the indefinites. Hinterwimmer (2015) chooses the second option and analyses pronouns as definite descriptions in disguise; the only difference between DPros and PPros being that the former are subject to the topic avoidance constraint (or rather, to a more general constraint that turns out as topic avoidance in the cases discussed so far; see Section 2.3), while the latter are not. Consequently, in order for the coreferential readings of pronouns discussed in Section 2.1 to be captured as well, a descriptive theory of proper names is adopted: Proper names are assumed to be DPs with either an overt (as in der Peter) or a silent definite determiner whose NP complement consists of the content of the respective name. I.e., in the case of a proper name such as (der) Peter, the NP Peter denotes the property of being called Peter, and the DP as a whole denotes the unique individual that has the property of being called Peter (see Elbourne 2005 and the references cited there for discussion). Thus what happens in such cases is that the empty NP contained in the respective pronoun is taken to be identical with the NP contained in the respective proper name. Note that the essentials of the analysis in Hinterwimmer (2015) could be recast in a theory that assumes pronouns to be variables that may either be resolved to some contextually salient individual or be bound (either under c-command by the standard mechanism available to all quantificational DPs or by the special mechanism available only to indefinite DPs).

Demonstrative pronouns and perspective |

199

2.3 Bound Uses of DPros Wiltschko (1999) observed that in sentences like (10a) and (10b) the respective DPro, in contrast to the PPro, cannot be interpreted as a variable that is bound by the subject of the matrix sentence, but only as referring to some other contextually salient male individual. (10) a. Peteri glaubt, dass {eri /derj } klug ist. Peteri believes that he {PProi /DProj } is smart. b. [Jeder Mann]i glaubt, dass {eri /derj } klug ist. [Every man]i believes that he {PProi /DProj } is smart. Simplifying somewhat, Wiltschko explains the contrast between DPros and PPros in such configurations by saying that DPros are full DPs that consist of an overt determiner (recall that in German DPros are for the most part homophonous with the various forms of the definite determiner) and a covert NP. They are thus referential expressions, just like full definite descriptions, and as such they are (a) subject to Principle C of Binding Theory (Chomsky 1981) – i.e. they may not be co-indexed with c-commanding DPs – and (b) they cannot function as simple variables bound by an operator. PPros, in contrast, are not full referential DPs, but only consist of agreement features. They denote free variables that can either be resolved to some contextually salient individual or be bound by a c-commanding DP. Wiltschko (1999) thus makes use of both options for the analysis of pronouns that were discussed in Section 2.2: She assumes that the first one applies to PPros and the second one applies to DPros. While this accounts for the contrasts observed in sentences as (10a, b), it does not account for the contrasts discussed in Section 2.2 and 2.3. Secondly, and more importantly, once a broader range of data is considered, a general ban against bound variable interpretations of DPros turns out to be just as problematic as the assumption that they are subject to Principle C. While Wiltschko’s account seems to cover the observations with respect to (10), where the relation between DPros and c-commanding subjects is at issue, it runs into difficulties, as shown in Hinterwimmer (2015), in cases exemplified below, in (11)–(13). These examples demonstrate that a bound variable interpretation is available for DPros when the binding quantificational DP is not the subject, but (a) the direct object, (b) embedded in a PP, or (c) the indirect object. The sentence in (12), for example, receives an interpretation that can be paraphrased as For no colleague of Claudia x does Claudia believe that x is smarter than Claudia. Similarly for the other examples.

200 | Stefan Hinterwimmer and Peter Bosch

(11) Oskari lädt [jeden Syntaktiker]j zum Abendessen ein, wenn derj ihmi versichert, dass erj Montague gelesen hat. Oskari invites [every syntactician]j for dinner if hej [DPro] assures himi [PPro] that hej [PPro] has read Montague. (12) Claudiai glaubt von [keiner Kollegin]j , dass diej klüger ist als siei . Claudiai believes of [no colleague]j that shej [DPro] is smarter than heri [PPro]. (13) Peter stellte [jedem Studenten]j mindestens eine Frage, die derj nicht beantworten konnte. Peter asked [every student]j at least one question that hej [DPro] could not answer. These examples show that DPros can in principle be bound by quantifiers. In (11) and (12) though, the quantificational DP presumably does not c-command the DPro at the surface. In (11), the quantificational DP is contained in the VP, while the adverbial phrase containing the DPro is right-adjoined either to vP or to TP. C-command therefore only obtains at LF, after Quantifier Raising has applied and the quantificational DP has adjoined to vP or TP. Likewise in (12), where the quantificational DP is contained in the PP von keiner Kollegin (‘of no colleague’) and therefore does not c-command anything outside this PP before Quantifier Raising has applied. Thus one could still argue the interpretation of DPros is constrained by Principle C. Not so for (13). Here the quantificational DP is the indirect object, while the DPro that binds it is contained in a relative clause that is part of the direct object DP. There is empirical evidence showing that the indirect object c-commands the direct object when both are in their respective base positions, i.e., if the latter has not been scrambled across the former (see Grewendorf 2002 for an overview and for further references). In (14a), for example, the reflexive pronoun cannot only be bound by the subject DP, Hans, but also by the indirect object, dem Studenten (‘the student’). Likewise, (14b) is ungrammatical on the intended reading where the PPro in indirect object position is coreferential with either a proper name or a definite description that is embedded in a relative clause modifying the direct object. (14) a. Hansi zeigte [dem Studenten]j [ein Bild von sichi,j ]. Hansi showed [the student]j a picture of himselfi,j . b. * Maria erzählte ihmi eine Geschichte, die {Martini /[den Dekan]i } beunruhigte. * Maria told him a story that upset {Martin /[the dean ]}. i i i

Demonstrative pronouns and perspective |

201

Since the binding of a reflexive pronoun requires c-command, and since the ungrammaticality of (14b) is most straightforwardly explained as a violation of Principle C, these examples would show that indirect objects (in their base position) c-command direct objects (in their base position). We thus not only have evidence against the claim that DPros cannot be interpreted as bound variables, but also against the claim that they are subject to Principle C. What (11)–(13) have in common, however, and what distinguishes them from (10) is that the quantificational DP that binds the DPro is not the grammatical subject of the respective sentence. In Section 2.1 it seemed that DPros, in their coreferential uses, were banned from picking up discourse referents denoted by grammatical subjects. Upon closer inspection though it turned out that not grammatical subjecthood, but aboutness topicality was the crucial parameter. This raises the question of whether the distribution of bound DPros can be explained in terms of aboutness topic avoidance as well. On the one hand, there is of course strong empirical evidence that DPs headed by quantificational determiners such as every, most, no, etc. cannot be aboutness topics. They are unacceptable in topic-marking constructions such as German Left Dislocation, and they cannot be combined with topic markers such as Japanese wa (see Endriss 2009 and Endriss and Hinterwimmer 2009 and the references cited there for discussion). On the other hand, we have seen in Section 2.2 that the sets restricting situation quantifiers such as the generic operator or overt Q-adverbs can function as higher-order subjects of higher-order predicates, and thus as aboutness topics. We could thus assume that what is true of situation quantifiers is true of individual quantifiers too: The respective restrictor sets can in principle function as aboutness topics. The only difference would be that in the case of quantificational DPs this is masked by the fact that the constituent providing the restrictor (namely the NP) always forms a constituent with the determiner. If this idea is on the right track, and given the fact that the DPros in (11)– (13) can be interpreted as bound variables, while the ones in (10a, b) cannot, one would have to assume that DPros are not only prohibited from picking up individuals functioning as aboutness topics, but also from taking individuals as values that belong to sets of individuals that function as aboutness topics. Taking Jeder Mann (‘every man’) not only as the grammatical subject of (10b), but assuming that its restrictor NP Mann also provides the aboutness topic by default, would thus account for the fact that the DPro in (10b) cannot be interpreted as a bound variable. In (10a), where a referential DP is the grammatical subject, the topic avoidance constraint would explain the unavailability of a bound reading straightforwardly. An account along the lines just sketched makes a prediction, however, that is not borne out by the facts. If we construe a sentence where a DP that is not

202 | Stefan Hinterwimmer and Peter Bosch

the grammatical subject functions as the aboutness topic, a DPro contained in that sentence should be interpretable as bound by the grammatical subject. (15b) would be a case in point. But, although the object-DP Peter clearly functions as the aboutness topic of (b), the DPro, in contrast to the PPro, can only be understood as referring to Peter, not as being bound by the quantifier functioning as the grammatical subject. (15) a. A: Erzähl mir mal was über Peter. Tell me something about Peter. b. B: [Den Peter]i hat jederj schon bewundert, als {deri /eri/j } noch ein Kind war. Everyonej admired Peteri already when he {DProi /PProi,j } was still a child. From these and similar observations Hinterwimmer (2015) draws the conclusion that bound occurrences of DPros are not subject to the topic avoidance constraint, but rather to a subject avoidance constraint. Both constraints can be derived from a more general constraint that prevents the interpretation of DPros from depending on the (denotation of the) currently most prominent DP. In non-binding configurations – where the potential antecedent does not c-command the DPro on the surface or at LF – the currently most salient DP is the one whose denotation functions as the current aboutness topic. In binding configurations, on the other hand – where the potential antecedent c-commands the DPro on the surface or at LF – the currently most salient DP is the grammatical subject of the respective sentence. The idea behind this is that while in binding configurations the relation between pronoun and antecedent is a purely structural one, defined in terms of c-command, in non-binding configurations the relation between pronoun and antecedent DP is a more indirect, discourse-related one. In the latter case, there is no particular structural relation that must hold between pronoun and antecedent – it is not even necessary for them to be contained in the same sentence. Of course, structural factors like the hierarchical position occupied by the respective antecedent and its being de-accented relative to other constituents do play a role, but these can be explained as indirect effects insofar as they serve to mark the respective DP as an aboutness topic, with DPs serving as aboutness topics being the most prominent DPs and thus the preferred antecedents for pronouns by default. Given this difference in the way pronoun and antecedent are related in the two cases, it is not surprising that different conditions apply as far as the specification of the way in which pronouns of the marked variety (i.e., DPros) differ

Demonstrative pronouns and perspective |

203

from unmarked ones (i.e., PPros) is concerned. In both cases, a plausible raison d’etre for the marked pronoun series is that it provides speakers with a means to indicate that it is not the default resolution strategy that is meant to apply. Hence, if the default resolution strategies are different in binding and non-binding configurations, the constraint imposed on the marked pronoun series has to be flexible enough to prevent the respective default strategy from applying. Since the default strategy in the non-binding cases is sensitive to structural factors only indirectly, the constraint on DPros applying in these cases must be spelled out accordingly. Concerning potential binding configurations, in contrast, which can be defined as such only in purely structural terms, it is reasonable to assume that the preferred binding option is specified in purely structural terms as well. Since among the DPs, which can in principle (if they are contained in a superordinate clause) bind a pronoun contained in a subordinate clause, the subject DP can reasonably be assumed to be the most prominent one⁴, it is also the obvious choice for a purely structurally defined “preferred binder”. The constraint on DPros applying in these cases must then be spelled out accordingly. Concerning the question of how the assumption that pronouns, PPros and DPros alike, are definite descriptions with an empty NP can be reconciled with the fact that they behave as bound variables in sentences of the type just discussed, Hinterwimmer (2015) follows Elbourne (2005) in assuming that in cases of binding the free predicate variable introduced by the empty NP is resolved to the property of being identical to a variable which is bound by the respective antecedent DP. As already mentioned in Section 2.2, however, this analysis does not hinge on the assumption that pronouns are definite descriptions in disguise – it could easily be recast in terms of an analysis which assumes pronouns to denote variables that can either receive a value from a contextually determined assignment function or be bound, dynamically or non-dynamically.

3 New Data With the analysis sketched in Section 2 in mind, let us now take a look at the examples in (16). In all of these cases the DPro violates the topic avoidance constraint and can clearly refer to the current aboutness topic, i.e., to Otto in (a) and (b), or to Karin in (c).

4 It bears the least marked case, can bind reflexives and control empty pronouns most easily, is associated with the highest argument role by default etc.

204 | Stefan Hinterwimmer and Peter Bosch

(16) a. Lass uns mal über Ottoi reden. Ottoi ist der fähigste Verkäufer, den ich kenne. {Deri /Eri } könnte sogar einem Blinden einen HD-Fernseher verkaufen. Let’s talk abot Ottoi . Ottoi ist he most gifted salesman I know. He {DProi / PProi } could even sell a HD TV-set to a blind man. b. Was Ottoi betrifft, deni mochte Karinj noch nie. {Deri /Eri } hat siej schon als Kind immer geärgert. As for Ottoi , Karinj never liked himi {DProi }. He {DProi /PProi } already always teased her as a child. c. A: Wo ist eigentlich Karini ? Diei habe ich lange nicht gesehen. Where is Karini , anyway? I havn’t seen heri for a while. B: {Diei /Siei } ist schon gegangen. She {DProi /PProi } has already left. In (a) and (b), the status of Otto as the aboutness topic is clearly indicated; either explicitly by the first sentence in (a), or by the left-dislocated adverbial clause in (b). In (c), the status of Karin as aboutness topic is not indicated quite so clearly by linguistic means⁵. But in the absence of another plausible candidate, and in view of the fact that the fronted DPro does not receive a contrastive accent (at least on the most natural reading of the sentence), Karin can definitely be considered the aboutness topic of the second sentence. The cases in (16) thus show that the topic avoidance constraint as it stands makes incorrect predictions. – One might perhaps be tempted to try and save the constraint by re-interpreting it as a constraint of preference. Not as excluding aboutness topics as referents of DPros, but as saying that whenever there is a choice among two or more referents, one of which is the current aboutness topic, then the DPro would prefer a referent that is not the aboutness topic. But this strategy is blocked by examples like the following (from Bosch and Umbach 2006), where only one grammatically possible antecedent is available, which refers to the aboutness topic, but an interpretation on which the DPro picks up that antecedent still remains clearly very odd. (17) Woher Mariai das weiß? [SU Peter]j hat [DO es] [IO ihr]i gesagt. {?? Diei /Siei } war gerade hier. How does Mariai know? Peterj told heri . She {?? DProi /PProi } has just been here.

5 Since a discussion of whether questions have aboutness topics would take us too far afield, let us ignore the question of whether Karin is already the aboutness topic of the first sentence in (16c).

Demonstrative pronouns and perspective |

205

An even stronger argument against such an escape strategy, and against the topic avoidance constraint, comes from the three continuations of the sentence in (18) under (a), (b), and (c): (18) Als Peteri abends nach Hause kam, war die Wohnung wieder in einem fürchterlichen Zustand. When Peteri came home in the evening, the flat was in a terrible state again. a. {* Deri /Eri } musste erst mal drei Stunden putzen. He {* DProi /PProi } first had to clean up for three hours. b. {* Deri /Eri } hatte doch gestern erst aufgeräumt. He {* DProi /PProi } had only tidied up yesterday, after all. c. {Deri /Eri }kann sich einfach nicht gegen seinen Mitbewohner durchsetzen. He {DProi /PProi } is simply unable to stand his ground against his flatmate. In spite of the fact that in all three cases Peter is the only available antecedent, the DPro is perfectly natural only in (c), but neither in (a) nor (b). This shows that the acceptability of the DPros in the sentences in (16) cannot simply be due to the fact that there is only one antecedent available for the DPro. Rather, it would seem to show that the topic avoidance constraint cannot be maintained as it stands, and other factors need to be considered that make for the difference between the continuations in (a) and (b) on the one hand and (c) on the other. For further evidence against the topic avoidance constraint, we may go back to our initial examples (1) and (3), repeated as (19a) and (20a) below, and look at the variants in (19b) and (20b), which differ only in the concluding sentence – and thus remain unchanged with respect to the aboutness topic. (19) (a) Pauli wollte mit Peterj laufen gehen. Aber {eri,j /derj } war leider erkältet. Pauli wanted to go running with Peterj . But he {PProi,j /DProj } had a cold unfortunately. (b) Pauli wollte mit Peterj laufen gehen. {Eri,j /Deri,j } sucht sich immer Leute als Trainingspartner aus, die nicht richtig fit sind. Pauli wanted to go running with Peterj . He {PProi,j /DProi,j } always picks people as training partners who are not really fit. (20) (a) Woher Karli das weiß? Peterj hat es ihmi gesagt. {Derj /Eri,j } war gerade hier. How does Karli know? Peterj told him. He {DProj /PProi,j } has just been here.

206 | Stefan Hinterwimmer and Peter Bosch

(b) Woher Karli das weiß? Peterj hat es ihmi gesagt. {Deri,j /Eri,j } hätte das nie erfahren sollen. How does Karli know? Peterj told him. He {DProi,j /PProi,j } never should never have been told. In (19b) the DPro may refer to either Paul or Peter, but, against the constraint, there is a strong tendency to interpret that DPro as picking up the aboutness topic of the preceding sentence, Paul; and in the case of (20b) it is clearly also possible to interpret the DPro as picking up the aboutness topic of the preceding sentence, Karl. – With the aboutness topic unchanged between the (a) and (b) discourses, it seems inevitable, again, to look for other factors that can explain the difference. Before we do so, let us take another look at donkey sentences. In Section 2.2 we only discussed cases where the respective adverbial clause contained two potential antecedents for the DPro. What about cases where there is just one potential antecedent available? As shown by (21), DPros can in principle be interpreted as varying with the individuals introduced by the respective indefinite in sentences of this kind (although they may sound slightly less natural than PPros). (21) Wenn [ein Bauer]i in Niedersachsen lebt, hält {deri /eri } meistens Schweine. If [a farmer]i lives in Lower Saxony, he {DProi /PProi } usually keeps pigs. Furthermore, in sentences where the adverbial clause contains two indefinites it is possible to pick up either discourse referent by a DPro – even though the variant with two PPros and the one with a PPro picking up the subject indefinite and a DPro picking up the object indefinite may sound more natural than the one with two DPros: (22) Wenn [ein Bauer]i [einen Esel]j besitzt, schlägt {deri /eri } {denj /ihnj }. If [a farmer]i owns [a donkey]j , he {PProi /DProi } beats it {PProj /DProj }. Finally, even in sentences with a subject and an object indefinite in the adverbial clause and one DPro in the main clause, it is not always as clear as in the examples (6a, b), repeated here as (24a, b), that the DPro picks up the discourse referent introduced by the object indefinite. In the case of (23) the DPro can also be interpreted as varying with the individuals introduced by the subject indefinite, even though there may be a slight preference for interpreting it as varying with the individuals introduced by the object indefinite.

Demonstrative pronouns and perspective |

207

(23) Wenn [ein griechischer Bauer]i [einen Esel]j besitzt, heißt das noch lange nicht, dass es {demi,j /ihmi,j } gut geht. If [a Greek farmer]i owns [a donkey]j , this does not mean yet that it/he {PProi,j / DProi,j is doing well }. (24) a. Wenn [ein Schäferhund]i [einem Wolf]j begegnet, greift derj meistens sofort an. If [an Alsatian]i encounters [a wolf]j , it {DProj } usually attacks immediately. b. Wenn [einem Wolf]i [ein Schäferhund]j begegnet, greift derj meistens sofort an. If [an Alsatian]j encounters [a wolf]i , it {DProj } usually attacks immediately. This contrast between (23) and (24a, b) could be accounted for by an appeal to interfering plausibility considerations. In the case of (23) the more plausible interpretation takes the DPro to pick up the discourse referent introduced by the subject indefinite, the farmer, while there is no such plausibility bias in the case of (24a, b) for either the wolf or the Alsatian. On the most plausible interpretation of (23) ownership of a donkey as an indication of wealth for Greek farmers is rejected. In the absence of a special context, however, it is much less plausible that for a donkey being owned by a Greek farmer would be rejected as an indication that a donkey is doing well. That the interpretation on which the DPro picks up the discourse referent introduced by the object indefinite is perhaps slightly preferred, can thus perhaps be taken to show that there is a pragmatic anti-topicality bias for DPros in cases where they have a choice among competing antecedents. In the absence of any plausibility preferences for one interpretation over the other, the DPro would then be interpreted as picking up the non-topical discourse referent, while in cases where plausibility, on the one hand, and the anti-topicality bias, on the other, pull in different directions, both interpretations are in principle available. The idea of taking the topic avoidance constraint rather as a pragmatic antitopicality bias is supported by comparing (24a, b) to their variants in (25a, b). We have left the initial if -clause, and thus the aboutness parameter, unchanged and changed the main clauses to manipulate the plausibility parameter. Here an interpretation of the DPro as picking up the discourse referent introduced by the topical indefinite leads to a more plausible interpretation of the entire sentence and weakens the anti-topicality bias of the DPro considerably: (25) a. Wenn [ein Schäferhund]i [einem Wolf]j begegnet, bekommt deri meistens einen solchen Schreck, dass eri all die tollen Tricks vergisst, die eri in der Hundeschule gelernt hat.

208 | Stefan Hinterwimmer and Peter Bosch

If [an Alsatian]i encounters [a wolf]j , it {DProi } usually gets such a shock that it {PProi } forgets all the great tricks that it {PProi } learnt at the dog training. b. Wenn [einem Wolf]i [ein Schäferhund]j begegnet, braucht deri meistens nur einmal zu knurren, und die Sache ist geklärt. If [an alsatian]j encounters [a wolf]i , it {DProi } usually just needs to growl once and matters are settled. Let us not forget, however, that in the case of (5), repeated here as (26), where a DPro occurs together with a PPro in the main clause, the DPro can only be interpreted as picking up the discourse referent introduced by the object indefinite and the PPro only as picking up the discourse referent introduced by the subject indefinite, in spite of the fact that this leads to an extremely implausible interpretation: donkeys don’t plausibly beat farmers. (26) Wenn [ein Bauer]i [einen Esel]j besitzt, schlägt deri ihnj . If [a farmer]i owns [a donkey]j , hei {DProi } beats itj . This observation can be reconciled with our assumptions regarding the behaviour of DPros in donkey sentences as follows, though: Let us combine the assumption that topical discourse referents are the most prominent ones with the assumption that DPros are the marked pronoun series in German, while PPros are the unmarked one. Let us furthermore assume that there is a pragmatic strategy according to which the marked pronoun is preferably interpreted in such a way that it picks up the less prominent discourse referent, while the unmarked pronoun preferably picks up the most prominent one (23). If there is just one antecedent available (21) or if two pronouns from the same series match with two potential antecedents from the respective adverbial clause, the pragmatic anti-topic bias just sketched simply does not apply (22). If the pragmatic bias and plausibility considerations pull in different directions regarding the interpretation of a single pronoun from the marked series (23, 25a, b), the pronoun can be interpreted either way. If plausibility considerations favour an interpretation that violates the preferences of both a pronoun from the marked series and one from the unmarked series because the marked pronoun would have to pick up the more prominent antecedent, while the unmarked one would have to pick up the less prominent antecedent, avoiding such a double violation of the pragmatic pronoun resolution strategy is more important than avoiding a globally implausible interpretation. The assumption of a weakened, pragmatic version of the topic avoidance constraint argued for in Hinterwimmer (2015) thus seems to make the right predictions as far as DPros in donkey sentences are concerned.

Demonstrative pronouns and perspective |

209

Before we conclude this section, let us take a second look at DPros that are interpreted as bound variables. As the example in (27) shows, it is also possible to violate the subject avoidance constraint that was assumed to regulate the distribution of bound DPros in Hinterwimmer (2015): In spite of the fact that the DP jede Mathematikerin (‘every mathematician [fem.]’) is the grammatical subject of the main clause, the DPro in the embedded clause can be interpreted as a variable that is bound by it. (27) [Jede Mathematikerin]i wirkte auf Paulj , als wäre {siei /diei } klüger als erj . [Every mathematician]i gave Paulj the impression that she {DProi /PProi } was smarter than himj . Similarly, the variant of (28a) with the DPro, while still not entirely perfect, is much better than the variant of (10b), repeated here as (28b), with the DPro: (28) a. [Jeder Politiker]i wurde schon einmal beschuldigt, dass {eri /deri } korrupt sei. [Every politician]i has been accused that he {DProi /PProi } is/was corrupt. b. [Jeder Mann]i glaubt, dass eri /* deri klug ist. [Every man]i believes that he {* DProi /PProi } is smart. It thus neither seems to be the case that co-referentially interpreted DPros have to avoid aboutness topics nor that DPros interpreted as bound variables have to avoid grammatical subjects. Rather, both bound and co-referentially interpreted DPros seem to be subject to a different constraint, with their aversion against aboutness topics or grammatical subjects in some cases perhaps being derivable from that constraint. The distribution of DPros in donkey sentences, in contrast, is captured quite nicely by a weakened pragmatic version of the topic avoidance constraint. In the next section we propose an analysis that tries to make sense of this pattern.

4 The Proposal Consider the examples in (16) again, repeated here as (29): (29) a. Lass uns mal über Ottoi reden. Ottoi ist der fähigste Verkäufer, den ich kenne. {Deri /Eri } könnte sogar einem Blinden einen HD-Fernseher verkaufen.

210 | Stefan Hinterwimmer and Peter Bosch

Let’s talk abot Ottoi . Ottoi ist he most gifted salesman I know. He {DProi / PProi } could even sell a HD TV-set to a blind man. b. Was Ottoi betrifft, deni mochte Karinj noch nie. {Deri /Eri } hat siej schon als Kind immer geärgert. As for Ottoi , Karinj never liked him [DProi ]. He {DProi /PProi } already always teased her as a child. c. A: Wo ist eigentlich Karini ? Diei habe ich lange nicht gesehen. Where is Karini , anyway? I havn’t seen heri for a while. B: {Diei /Siei } ist schon gegangen. She {DProi /PProi } has already left. What all these examples have in common is that they express statements made from the perspective of the respective speaker. This is particularly clear in (29a), where the second and the third sentence express subjective evaluations of Otto’s abilities by the speaker that are made in a conversation where the speaker has first established Otto as the aboutness topic. (29b) is also most plausibly construed as uttered in a conversation where, for example, the question of which people Karin likes is being discussed. The initial sentence first establishes Otto as the aboutness topic and then expresses the speaker’s conviction that Karin never liked Otto, while the following sentence gives an explanation for Karin’s attitude towards Otto. Again, there is an “involved” speaker uttering sentences that express his or her views regarding a particular matter. In the case of (29c), too, the relevant sentence containing the DPro is part of a dialogue in which it provides the answer to a question about the present location of an individual that is provided on the basis of the speaker’s own perceptions. Bearing this common property of all the examples in (29) in mind, let us now turn to the sentences in (18), repeated here as (30): (30) Als Peteri abends nach Hause kam, war die Wohnung wieder in einem fürchterlichen Zustand. When Peteri came home in the evening, the flat was in a terrible state again. a. {* Deri /Eri } musste erst mal drei Stunden putzen. He {* DProi /PProi } first had to clean up for three hours. b. {* Deri /Eri } hatte doch gestern erst aufgeräumt. He {* DProi /PProi } had only tidied up yesterday, after all. c. {Deri /Eri } kann sich einfach nicht gegen seinen Mitbewohner durchsetzen. He {DProi /PProi } is simply unable to stand his ground against his flatmate.

Demonstrative pronouns and perspective |

211

Let us first consider the felicitous continuation in (30c): It clearly expresses an evaluation of Peter’s character that is made from the perspective of the speaker. This is indicated (i) by the switch from past tense in the opening sentence to present tense in the (c) continuation; (ii) by the presence of the speech act particle einfach (‘simply’), which, like all speech act particles makes a contribution at the level of conventional implicatures (hencefort: CIs)⁶. More concretely, einfach (very roughly) expresses the speaker’s view that the asserted proposition corresponds to a state of affairs that has to be accepted since it is beyond the speaker’s power to change it (although s/he may not be happy about it). Let us have a closer look at the switch from past tense to present tense: In (30a), where both the adverbial clause verb and the main clause verb are marked for past tense, a particular situation that presumably includes Peter, but not the speaker, is introduced that is temporally located on the evening of some day preceding the day where the sentence is uttered. The use of present tense in (30c) indicates that the state introduced by the combination of the negated modal verb kann and the infinitival complement clause is not located at or after the situation introduced in the opening sentence, but rather at a larger interval that includes the utterance time. That Peter is in this state (of being unable to stand his ground against his flatmate) is presumably interpreted as a conclusion that is drawn by the speaker on the basis of the information provided by the first sentence in combination with other facts about Peter and his flatmate. By adding the particle einfach the speaker furthermore gives a subjective evaluation of that conclusion (see above). The continuation in (30b), by contrast, is automatically understood as reporting Peter’s own thoughts in view of the flat’s state in the situation introduced in the opening sentence. An indication that what is reported in (30b) is a thought of Peter, not a thought of the speaker, is provided by the following fact: The parenthetical clause dachte er (“he thought”) can be added without any change in truth conditions. (30b) exemplifies a phenomenon that has only recently come into the focus of researchers working in the field of truth conditional semantics (Doron 1991, Schlenker 2004, Sharvit 2008, Eckardt 2014): Free Indirect Discourse (henceforth: FID). FID is a mixture of Direct Discourse (henceforth: DD) and Indirect Dicourse (henceforth: DD): On the one hand, some items such as temporal adverbs and speech act particles that are normally tied to the utterance context (i.e. to the speaker and the time of utterance, respectively) can be interpreted with respect to the context that the individual whose thoughts are reported is in. In this respect, FID behaves like DD. Other elements such as tense marking and personal

6 I.e. it is not asserted and can thus not be negated, it is tied to the speaker (but see below), etc. See Potts 2005 for further criteria for the identification of CI-content.

212 | Stefan Hinterwimmer and Peter Bosch

pronouns are interpreted with respect to the context of utterance. So with respect to them, FID behaves like ID (see the works cited above for details). Both properties of FID just described can be found in (30b): The temporal adverbial gestern (‘yesterday’) is interpreted not with respect to the context of utterance, but rather with respect to the context in which Peter is located, i.e., the day preceding the day that contains the situation introduced in the opening sentence. Similarly, the two speech act particles doch and erst (which do not have straightforward English translations) are most plausibly understood as expressing Peter’s attitude towards the proposition denoted by (30b), not the speaker’s. Doch in combination with a proposition p (very roughly) expresses the surprise of the author of the respective context that p and some contextually salient proposition q are both true at the same time (where in our case q is the proposition denoted by the opening sentence of (30)) while erst signals that the event introduced by p took place more recently than expected in the context – i.e., in this case more recently than expected in view of the flat’s current state. The third person PPro er and the past perfect marking, in contrast, are interpreted not with respect to Peter’s context, but rather with respect to the speaker’s context – otherwise the first person PPro ich and present perfect would have to be used (since the event of tidying up is simply located prior to the time of Peter’s context, not prior to some salient interval that is located prior to the time of Peter’s context). The use of past perfect is particularly instructive here since it fulfils two functions at the same time: On the one hand, it locates the event introduced by (30b) with respect to both the time of the speaker’s context and the time of the situation introduced in the opening sentence, which defines Peter’s context – i.e. it indicates that the event is prior to the time of that situation, which is itself prior to utterance time. At the same time, it is the use of past perfect which clearly indicates that (30b) is a case of FID: If past perfect marking is replaced by present perfect marking, as in the variant in (31b), the sentence can no longer be interpreted as expressing Peter’s views, but only as expressing the speaker’s views. Crucially for our purposes, this change in interpretation goes hand in hand with the DPro being felicitous again (we will come back to this point in a moment). (31) Als Peteri abends nach Hause kam, war die Wohnung wieder in einem fürchterlichen Zustand. When Peteri came home in the evening, the flat was in a terrible state again. b. {Deri /Eri } hat doch gestern erst aufgeräumt. He {DProi /PProi } had only tidied up yesterday, after all. How can it happen that past perfect marking indicates that (30b) expresses Peter’s thoughts and is thus partially evaluated with respect to Peter’s context, while at

Demonstrative pronouns and perspective |

213

the same time it locates the event introduced by the sentence with respect to the time of the speaker’s context? Simplifying somewhat, the past perfect locates an event or state prior to a reference time which is itself located prior to the utterance time. In (30b) the reference time is the time of the situation introduced in the opening sentence. Using the temporal trace of this situation as the reference time is thus required in order to signal that this situation is the context for the evaluation of parts of (30b), i.e., the context at which Peter is located when he thinks the proposition denoted by that sentence. Since the event introduced by (30b) is located prior to the reference time, while the reference time is located prior to the utterance time, past perfect marking has to be employed. The use of past perfect marking thus satisfies the requirement to (i) signal that (30b) expresses a thought of Peter by making explicit reference to the context of Peter and by locating the event introduced by the sentence with respect to the time of Peter’s context and (ii) to temporally locate the event introduced by the sentence with respect to the utterance time. If, by contrast, present perfect is used, as in the variant in (31b), the event of Peter tidying up his flat is only located with respect to the time of the speaker’s context – namely as prior to that time. No reference to Peter’s context (i.e., the situation introduced in the opening sentence of (30)) is made (recall that tense marking in FID is always interpreted with respect to the speaker’s context, i.e., the present perfect cannot be interpreted as locating the event of Peter tidying up prior to Peter’s context) and hence there is no reason not to attribute the thought expressed by the sentence to the speaker. The crucial observation with respect to our present concerns is now that in (30b) the use of the DPro is infelicitous, while in (31b) it is felicitous. This, together with our observations concerning the sentences in (29) suggests the following hypothesis regarding sentences where there is only one potential antecedent: A DPro cannot be used to refer to the individual from whose perspective the event or state introduced by the sentence containing the DPro is described or evaluated. Let us call such an individual the perspectival centre for the respective proposition. In the examples in (29) as well as in in (30c) and (31b) the perspectival centre is always the author of the utterance context (i.e., the speaker). In (30b), by contrast, it is the individual that is at the same time the only potential antecedent for the DPro. With the generalization just stated in mind let us now finally turn to the continuation in (30a). Here the past tense marking also employed in the opening sentence is retained and the event introduced by the sentence is automatically understood as (i) temporally following the event introduced by the adverbial clause in the opening sentence, and (ii) being caused by the state that the main clause in the opening sentence introduces. While there certainly has to be a narrator reporting the events under consideration, this narrator does not make her presence felt

214 | Stefan Hinterwimmer and Peter Bosch

in any way, i.e., there is no indication that those events are perceived or evaluated from the perspective of the narrator. At the same time, Peter is not only the aboutness topic of both the opening sentence of (30) and the continuation in (30a), but the events/states introduced in both sentences can also easily be construed as being perceived from his perspective. In addition to this, although the propositions denoted by the two sentences certainly do not express thoughts of Peter in the sense that (30b) does, they contain elements that can easily be understood as expressing Peter’s views: the adjective fürchterlich (‘terrible’) and the deontically interpreted modal verb müssen (‘must’), i.e., it might well be Peter who considers the state of the flat to be terrible and who is the source of the (thus intrinsic) obligation to clean up in order to make it conform to his standards again. We assume that due to the interplay of the factors mentioned – the absence of any indication of an “involved” speaker or narrator in combination with Peter being construed as the individual from whose perspective the events or states under consideration are perceived and the presence of elements that can easily be understood as expressing Peter’s views – Peter is the perspectival centre in (30a). The fact that the DPro in contrast to the PPro cannot be used to refer to him thus fits our hypothesis stated above. More generally, we assume that in cases where only an abstract, “uninvolved” speaker is present, aboutness topics are the perspectival centres by default. With these assumptions in place, we can also make sense of the contrast between (19a) and (19b) from above, repeated here as (32a, b). (32) a. Pauli wollte mit Peterj laufen gehen. Aber {eri,j /derj } war leider erkältet. Pauli wanted to go running with Peterj . But he {PProi,j /DProj } had a cold unfortunately. b. Pauli wollte mit Peterj laufen gehen. {Eri,j /Deri,j } sucht sich immer Leute als Trainingspartner aus, die nicht richtig fit sind. Pauli wanted to go running with Peterj . He {PProi,j /DProi,j } always picks people as training partners who are not really fit. The example in (32a) is similar to (30a) in the sense that (i) there is only an abstract narrator, but no “involved” speaker; (ii) Paul is the aboutness topic and the events under consideration can naturally be construed as being perceived from his perspective (the modal verb wollte (‘wanted’) explicitly refers to his intentions); and (iii) the second sentence contains an evaluative element that can easily be understood as expressing Paul’s views: leider (‘unfortunately’). Paul is thus presumably not only the aboutness topic, but also the perspectival centre, and the fact that the DPro cannot be used to refer to him is a consequence of that. In the case of (32b), by contrast, the switch to present tense in the second sen-

Demonstrative pronouns and perspective |

215

tence in combination with the content, which corresponds to a general statement about (one aspect of) Paul’s character, signals a break in the narrative structure and the second sentence can only be understood as an evaluation that is made from the speaker’s perspective. The speaker is thus the perspectival centre in that sentence, and the DPro is accordingly in principle free to pick up each of the two potential antecedents introduced by the previous sentence, with plausibility considerations clearly favouring Paul. Similar considerations apply to (20a) vs. (20b) from above. Concerning the examples in (2a, b), repeated here as (33a, b), we would also expect the DPro in the second sentence to be in principle free to pick up either of the potential antecedents. The second sentence is most likely understood as expressing the speaker’s opinion (or can at least easily be understood that way), hence neither the grammatical subject of the first sentence in (33a) nor the fronted object of the first sentence in (33b) refer to the perspectival centres of the respective continuations. At this point, however, the reasoning applied to the donkey sentences with DPros at the end of Section 3 above comes into play again: Apart from the absolute, non-violable constraint against referring to the perspectival centre of the sentence they are contained in, DPros, being the marked pronoun variant in German, are subject to a weak pragmatic anti-topicality bias. Hence the preference of the DPros in (33) for the non-topic. – This pragmatic bias can be overwritten by plausibility considerations though, as is shown by the fact that the DPros in the variants given (34a, b) do quite easily pick up the respective aboutness topics of the preceding sentences⁷. (33) a. [Der neue Assistenzarzt]i untersuchte [den Patienten auf Zimmer 3]j . {Derj /Eri,j } war sehr geduldig. [The new assistant doctor]i examined [the patient in room 3]j . He {DProj / PProi,j } was very patient. b. [Den Patienten auf Zimmer 3]i [untersuchte der neue Assistenzarzt]j . {Derj /Eri,j } war sehr geduldig. [The patient in room 3]i was examined by [the new assistant doctor]j . He {DProj /PProi,j } was very patient.

7 This does not preclude initial processing difficulties when the structurally defined default is in conflict with plausibility considerations which only come into play at later stages. But here, as throughout this paper, we are not looking into the processing theory of pronouns.

216 | Stefan Hinterwimmer and Peter Bosch

(34) a. [Der neue Assistenzarzt]i untersuchte [den Patienten auf Zimmer 3]j . {Deri /Eri,j } sollte erst mal mit den leichten Fällen anfangen. [The new assistant doctor]i examined [the patient in room 3]j . He {DProj / PProi,j } must start with the easy cases first. b. [Den Patienten auf Zimmer 3]i [untersuchte der neue Assistenzarzt]j . {Deri /Eri,j } war Kassenpatient. [The patient in room 3]i was examined by [the new assistant doctor]j . He {DProj /PProi,j } was not a private patient. Concerning donkey sentences like the ones discussed at the end of Section 3 above, it is now easy to see why the distribution of DPros in them is solely regulated by a weak pragmatic anti-topicality bias in combination with plausibility considerations: Since they all express general statements about situations of a particular kind, the individuals introduced by the indefinites they contain do not qualify as perspectival centres for the respective propositions, independently of whether they are aboutness topics or not. This means that the speaker is always the perspectival centre, and the DPros are in principle free to pick up any of the available discourse referents. Let us finally return to the distribution of DPros interpreted as bound variables. The cases we discussed were all cases where the DPro occurs in a propositional attitude context. If we follow Schlenker (2003) in assuming that propositional attitude verbs, such as glauben, quantify not just over worlds, but over contexts compatible with the speaker’s beliefs, then our hypothesis that DPros cannot be resolved to individuals that function as perspectival centres straightforwardly accounts for all the contrasts observed above. The individuals contained in the set of sets of individuals denoted by the quantificational DP are the perspectival centres for the respective contexts. Interpreting them as bound by that quantificational DP would thus violate the anti-perspectival centre constraint. This explains why the DPro in (10b), repeated here as (35a), cannot be interpreted as bound by the quantificational DP, while both the DPro in (12), repeated here as (35b), and the one in (27), repeated here as (35c), can be interpreted as bound by the respective quantificational DP. Only in (35a) the quantificational DP is at the same time the perspectival centre, while in (35b) the perspectival centre is Karin and in (35c) it is Paul. (35) a. [Jeder Mann]i glaubt, dass {eri /derj } klug ist. [Every man]i believes that he {PProi /DProj } is smart. b. Claudiai glaubt von [keiner Kollegin]j , dass diej klüger ist als siei . Claudiai believes of [no colleague]j that shej [DPro] is smarter than heri [PPro].

Demonstrative pronouns and perspective |

217

c. [Jede Mathematikerin]i wirkte auf Paulj , als wäre {siei /diei } klüger als erj . [Every mathematician]i gave Paulj the impression that she {PProi /DProi } was smarter than himj . In all the other examples discussed above, which do not contain propositional attitude verbs anyway, the perspectival centre is either the speaker or the aboutness topic of the sentence, thus enabling the respective quantificational DPs to bind the variables introduced by the DPros. The only case for which our new analysis does not seem to make the right predictions is (15b), repeated here as (36b): Since the main clause verb bewundern (‘admire’) is not a propositional attitude verb and the perspectival centre is presumably the speaker, B, nothing should prevent the quantificational DP jeder (‘everyone’) from binding the DPro. Still, there is a strong preference for interpreting it as co-referential with the aboutness topic, Peter. We can make sense of this observation if we stick (a) to Hinterwimmer’s (2015) assumption that grammatical subjects are the most prominent DPs in binding configurations (while aboutness topics are the most prominent ones in non-binding configurations) and (b) to the assumption argued for above with respect to donkey sentences that DPros in those cases where they have a choice among the DPs not functioning as perspectival centres avoid the most prominent one among them by default. Since this is only a pragmatic principle, though, it should be defeasible if plausibility considerations favour a different interpretation (just as in the case of the donkey sentences discussed in Section 3 topic avoidance was defeasible). This prediction seems to be borne out by the facts: In the variant in (37b), where a bound variable interpretation of the DPro is much more plausible than one where it picks up the aboutness topic, Peter, such an interpretation is indeed available. (36) a. A: Erzähl mir mal was über Peter. Tell me something about Peter. b. B: [Den Peter]i hat jederj schon bewundert, als {deri /eri,j } noch ein Kind war. Everyonej admired Peteri already when he {DProi /PProi,j } was still a child. (37) a. A: Erzähl mir mal was über Peter. Tell me something about Peter. b. B: [Den Peter]i hat [jeder von seinen Schülern]j bewundert, als {derj /eri,j } noch ein Kind war. Each of his pupilsj admired Peteri when he {DProj /PProi,j } was still a child.

218 | Stefan Hinterwimmer and Peter Bosch

5 Conclusion In this paper we have shown that the co-referential and binding options of German DPros can not satisfactorily be accounted for by the combination of the topic- and subject-avoidance constraints that we have proposed in previous work (Bosch and Umbach 2006, Hinterwimmer 2015, Bosch and Hinterwimmer to appear). Rather, we have argued on the basis of new empirical evidence that DPros are prohibited from being resolved to the respective perspectival centres, where the latter are the individuals from whose perspective the event or state is depicted that is introduced by the clause in which the DPro occurs. As we have seen, the concept of topic-avoidance still plays an important role in the following two respects: First, in the absence of an “involved” speaker functioning as the perspectival centre the aboutness topic of a sentence is often its perspectival centre as well. Second, in cases where there is still a choice among the individuals not functioning as the perspectival centre, the DPro, being the marked pronoun-variant, may not be resolved to the most prominent among them, which would be the one corresponding to the aboutness topic of the respective sentence in non-binding configurations (while in binding configurations it is the grammatical subject). In this paper we have not gone into any details concerning the formal implementation of the analysis sketched in Section 4, leaving that to future work. One obvious option that comes to mind would be to adopt and extend the analysis of Schlenker (2004) (see Eckardt 2014 for a similar view), according to which utterances are always interpreted with respect to two contexts: a context of utterance and a context of thought. Schlenker assumes that normally the two contexts coincide, but in free indirect discourse they are different, with the context of thought corresponding to the perspective of some protagonist. The distribution of DPros could thus be accounted for by assuming that they are prohibited from being resolved to the author of the context of thought. Crucially, in order to account for the full range of data considered, we would have to assume that even in the absence of clear indications for free indirect discourse (such as a clash between tense marking and the information provided by temporal adverbials) a separate context of thought (i.e. one that is distinct from the context of utterance) is introduced in cases where there is (a) no „involved“ speaker but (b) some prominent protagonist from whose perspective the event or state introduced by the respective sentence can plausibly be assumed to be depicted. While such an analysis would be easiest to implement under the assumption that DPros are (either free or bound) variables, it is in principle also compatible with a view on which they are definite descriptions in disguise (see above). As already said, we leave a further exploration of the general consequences of such an analysis as well as of its technical details to future work of ours.

Demonstrative pronouns and perspective |

219

References Berman, S. (1987). Situation-Based Semantics for Adverbs of Quantification. Issues in Semantics 12: 45–68. Bosch, P. (1980). The Modes of Pronominal Reference and Their Constraints. In J. Kreiman and A. Ojeda (eds.): Pronouns and Anaphora. Papers from the Parasession on Anaphora of the 16th Regional Meeting of the Chicago Linguistic Society. Chicago Linguistic Society. University of Chicago, 64–78. Bosch, P., T. Rozario and Y. Zhao (2003). Demonstrative Pronouns and Personal Pronouns. German der vs. er. In Proceedings of the EACL 2003. Budapest. Workshop on The Computational Treatment of Anaphora: 61–68. Bosch, P. and C. Umbach (2006). Reference Determination for Demonstrative Pronouns. In Proceedings of the Conference on Intersentential Pronominal Reference in Child and Adult Language: 39–51. Bosch, P. G. Katz and C. Umbach (2007). The Non-Subject Bias of German Demonstrative Pronouns. In M. Schwarz-Friesel, M. Consten and M. Knees (eds.), Anaphors in Text: Cognitive, Formal and Applied Approaches to Anaphoric Reference: 145–164. Bosch, P. and S. Hinterwimmer (to app.). Anaphoric reference by demonstrative pronouns in German. In search of the relevant parameters. In A. Holler, C. Göb and K. Suckow (eds.), Empirical perspectives on anaphora resolution: Information structural evidence in the race for salience. De Gruyter. Chomsky, N. (1981). Lectures on Government and Binding. Mouton de Gruyter. Doron, E. (1991). Point of View as a Factor of Content. In S. Moore and Adam Z. Wyner (eds.) Proceedings of SALT I. Cornell Working Papers in Linguistics: 51–64. Eckardt, R. (2014). The Semantics of Free Indirect Discourse. Brill. Elbourne, P. (2005). Situations and Individuals. MIT Press. Endriss, C. (2009). Quantificational Topics – A Scopal Treatment of Exceptional Wide Scope Phenomena. Studies in Linguistics and Philosophy, vol. 86, Springer. Endriss, C. and S. Hinterwimmer (2009). Indefinites as Direct and Indirect Aboutness Topics. In C. Fery and M. Zimmermann (eds.), Information Structure. Oxford University Press: 89–115. Evans, G. (1980). Pronouns. Linguistic Inquiry 11: 337–362. von Fintel, K. (1994). Restrictions on Quantifier Domains. PhD thesis, University of Massachusetts, Amherst. Frey, W. (2004). Notes on the Syntax and the Pragmatics of German Left Dislocation. In H. Lohnstein and S. Trissler (eds.), The Syntax and Semantics of the Left Periphery, Mouton de Gruyter: 203–233. Grewendorf, G. (2002). Left-Dislocation as Movement. In S. Mauck and J. Mittelstaedt (eds.), Georgetown University Working Papers in Theoretical Linguistics 2: 31–81. Groenendijk, J. and M. Stokhof (1991). Dynamic Predicate Logic. Linguistics and Philosophy 14: 39–100. Heim, I. (1982). The Semantics of Definite and Indefinite Noun Phrases. PhD thesis, University of Massachusetts, Amherst. Heim, I. (1990). E-Type Pronouns and Donkey Anaphora. Linguistics and Philosophy 13: 137– 177.

220 | Stefan Hinterwimmer and Peter Bosch

Hinterwimmer, S. (2006). The Interpretation of Universally Quantified DPs and Singular Definites in Adverbially Quantified Sentences. In. Proceedings of the 25th West Coast Conference on Formal Linguistics (WCCFL 25): 195–203. Hinterwimmer, S (2008a). Q-Adverbs as Selective Binders: The Quantificational Variability of Free Relatives and Definite DPs. Mouton de Gruyter. Hinterwimmer, S. (2015). A Unified Account of the Properties of German Demonstrative Pronouns. In P. Grosz, P. Patel-Grosz and I. Yanovich (eds.), The Proceedings of the Workshop on Pronominal Semantics at NELS 40, GLSA Publications, University of Massachusetts, Amherst: 61–107. Hockett, C. (1958). A Course in Modern Linguistics. New York: Macmillan. Kratzer, A. (1986). Conditionals. In A. v. Stechow and D. Wunderlich (eds.), Handbuch Semantik/Handbook Semantics, Mouton de Gruyter: 651–656. Lewis, D. (1975). Adverbs of Quantification. In E. L. Keenan (ed.), Formal Semantics of Natural Language, Cambridge University Press: 3–15. Potts C. (2005). The Logic of Conventional Implicatures. Oxford: Oxford University Press. Prince, E. F. (1992). The ZPG Letter: Subjects, Definiteness, and Information-Status. In W. Mann and S. Thompson (eds.), Discourse Description: Diverse Analyses of a Fund Raising Text, John Benjamins: 295–325. Reinhart, T. (1981). Pragmatics and Linguistics: An Analysis of Sentence Topics. Philosophica 27: 53–93. Sharvit, Y. (2008). The Puzzle of Free Indirect Discourse. Linguistics and Philosophy 31: 353– 395. Schlenker, P. (2003). A Plea for Monsters. Linguistics and Philosophy 26: 29–120. Schlenker, P. (2004). Context of Thought and Context of Utterance. A Note on Free Indirect Discourse and the Historical Present. Mind and Language 19: 279–304. Strawson, P. F. (1964). Identifying Reference and Truth Values. Theoria 30: 96–118. Wiltschko, M. (1999). On the Syntax and Semantics of (Relative) Pronouns and Determiners. Journal of Comparative Germanic Linguistics 2: 143–181.

Elena Koulidobrova (Central Connecticut State University) and Diane Lillo-Martin (University of Connecticut)

A ‘point’ of inquiry: The case of the (non-)pronominal IX in ASL Abstract: It has been claimed that in American Sign Language (ASL), the sign glossed as IX is used for a variety of functions, including personal pronouns, locatives, and determiners (Meier and Lillo-Martin 2013). We propose to unify these disparate functions by analyzing IX as a demonstrative, appearing with or without an overt complement. Appealingly simple, this move accounts for a wide range of facts regarding the distribution and interpretation of IX. We focus on points to third-person referents. Such points are directed toward loci – real people, objects, or locations in the physical space around the signer, or places (possibly arbitrarily) associated with their referents in the signed discourse (Lillo-Martin and Klima 1990). The loci themselves raise considerable interesting issues (Barberà 2012, Gagne and Davidson 2014); however, their contribution is often conflated with the contribution of IX alone. Thus, we specifically ask: for a sign involving pointing, what is the nature of IX (i.e. the lexical item realized in a pointing hand-shape) when dissociated from the issue of locus (i.e. the space where it points to)? Keywords: sign language, pointing, index, locus, demonstrative, definiteness

1 Introduction Having a pronoun – or an expression that is able to function as one – has become thought of as a primitive of linguistic theory. For instance, in syntax, pronominal expressions serve as diagnostic tools for the size of the c-command domain and ellipsis identification (e.g. Saito 2007, i.a.); in semantics, they are taken to diagnose operator-variable relations (see Heim & Kratzer 1998); in phonology, they are employed in the discussion of prosodic phrasing and stress (e.g. Selkirk 2005); in pragmatics, they are argued to reveal the ‘accessibility hierarchy’ (as in Ariel 1990). Many of these properties are necessarily contingent on the form of the pronoun under consideration. For example, more accessible antecedents have been argued to be picked out by pronouns that are less prominent in form (null ≫ weak ≫ full; Ariel 1990); personal pronouns have been argued to refer to subjects more readily than to objects, while demonstratives tend to refer to objects rather

222 | Elena Koulidobrova and Diane Lillo-Martin

than subjects (Bosch 2006, i.a.); and in certain binding configurations (such as donkey-anaphora), personal pronouns are possible while demonstrative ones are not (Wiltschko 1998). The characteristics just listed gloss over a number of important features of anaphoric expressions, yet they make a point: while pronouns may very well be expected to exist in a language (i.e. the burden of proof to the contrary lies with the alternative view), the form of such expressions matters. In fact, accurate categorization of expressions of this sort may result in a simpler theory. Consider, for example, the following scenario: a researcher mislabels a lexical item ‘X’ as an item of type ‘Y.’ If an independent analysis of elements of type ‘Y’ exists, and the researcher attempts to reconcile her findings with this analysis, she may encounter various problems, precisely due to the fact that the element in question is not ‘Y.’ She then may propose a new account of ‘Y,’ whose sole purpose is to accommodate the findings. Accidentally, she may arrive at an improved analysis of ‘Y’; however, this possibility is a fluke. The fact of the matter is that a mislabeling has occurred. Perhaps the previously advocated view of ‘Y’ suffices, even if it does not apply to ‘X.’ In fact, were the element in question labeled correctly as ‘X,’ the standard account of ‘X’-type expressions would have predicted its distribution. In this paper, we suggest that a mislabeling of this sort has occurred in the study of the so-called ‘personal pronominal’ (‘Y’ in the analogy above) expression in American Sign Language (ASL): much literature in sign linguistics refers to the relevant lexical item in these terms (see Sandler and Lillo-Martin 2006 and references therein). Yet, we demonstrate that on a number of diagnostics, the element in question behaves as something other than what it is typically labeled. We further show that such behaviors do not require any special account but fall out rather nicely from the view that it is actually something else (i.e. our ‘X’). Concretely, in ASL, the form commonly considered to be a personal pronoun is a point to some area of space, which we gloss IX (a.k.a. INDEX and PRO in Sign Language literature).¹ We describe this element in more detail in the next section. An (nearly) identical form is used for a number of other functions – IX has been described 1 We will use the following conventions for notation. Signs are glossed in upper case words that are their closest translation equivalents; ‘#’ indicates that the lexical item that follows is fingerspelled. Lower-case letters appended to sign glosses refer to spatial loci; the letters do not stand for particular locations in space (which are often arbitrarily assigned), except for ‘neu’ (neutral space) – an area in front of the signer with no assigned locus. IX followed by a word in parentheses indicates that the point was directed at the object named, e.g. IX(phone). A line above glosses indicates non-manual marking; ‘t’ stands for topic marking and ‘wh’ for wh-question marking. Somewhat unconventionally (but replicating the relevant source), we use ‘CL IX’ for a two-handed construction involving a ‘1’-classifier signed with a non-dominant hand and IX with the dominant one.

A ‘point’ of inquiry: The case of the (non-)pronominal IX in ASL

|

223

as a definite determiner and a locative adverbial as well as a personal pronoun (MacLaughlin 1997). In much work on pronouns in ASL, authors explicitly state that they are not planning to subsume all instances of IX under the same analysis; rather, they address only a subset of the cases in which IX appears. What we will try to do here instead is open the possibility that a unified account can be offered for all linguistic cases of IX. The essence of the claim is this: even when it seems to behave as such, ASL IX is not a personal pronoun per se (pace what is commonly claimed in the literature); rather, it is a demonstrative expression (Kaplan 1989, i.a.). Such a view, it seems, allows us to capture various behaviors of IX that are ordinarily set aside by the alternative accounts.

2 Description We begin this venture with the intuition that the form of the lexical item under examination here is not accidental – IX is identical to the co-speech pointing gesture employed in many spoken languages and frequently used along with a demonstrative to pick out a particular (typically present) referent in discourse. The photos in Figure 1 below illustrate the basic forms: pointing to the self (Figure 1a; glossed here as 1IX) as a 1st -person, pointing to the addressee (Figure 1b; 2IX) as a 2nd -person (though its status is controversial; see Meier 1990), and pointing to other (Figures 1c and 1d; 3IX) as a 3rd -person forms. In the ‘other’ cases, the sign is directed either toward a referent physically present in the situation (Figure 1c), or toward a location that is established as associated with a particular referent (Figure 1d). Johnston (2013), in his attempt to identify the ‘purely pronominal points,’ notes that “pointing signs in Auslan [Australian Sign Language], and potentially other S[ign] L[anguage]s, may merely be more conventionalized or regularized forms of gestural pointing” (p. 152). To illustrate: we analyzed 80-minute natural language sample of 4 different Deaf adults addressing Deaf children (part of the CLESS corpus, Lillo-Martin and Chen Pichler 2008). We observed 361 instances of IX, used to pick out a variety of referents with the distribution given in Table 1. In this data set, IX primarily refers to an individual either previously set up in the discourse or one that is explicitly present in some form (see Wulf et al. 2002 for a different type of corpus analysis). However, for ASL at least, the gestural view of IX is difficult to maintain; its linguistic nature can be diagnosed in a number of ways. Here we set aside 1IX and 2IX and focus on signs directed at 3rd -person referents (using the gloss IX as a general term). In our sample, 3rd -person IX appears to fulfill various roles in the

224 | Elena Koulidobrova and Diane Lillo-Martin

(a) 1IX (‘I/me’)

(b) 2IX (‘you’)

(c) 3IX (‘she/he/it,’ present referent)

(d) 3IX (‘she/he/it,’ non-present referent)

Figure 1: Pointing signs considered pronominal in ASL. Copyright Diane Lillo-Martin. Reprinted with permission.

Table 1: Distribution of IX in a portion of the CLESS Corpus (Lillo-Martin and Chen Pichler 2008). Direction of point

Proportion

Self Addressee Other, present, +human Other, present, −human Other, non-present, +human Other, non-present, −human Location

.17 .17 .13 .50 .01 .003 .01

sentence, defined here rather coarsely as: subject, object, location, topic, modifier of a noun (phrase) and some other, as shown in Table 2. When IX is used, its referent must be either physically present (as in Figure 1 (c)) or previously established. If the referent has been assigned an area of space prior to uttering of IX, IX must be signed in the associated area. If the refer-

A ‘point’ of inquiry: The case of the (non-)pronominal IX in ASL

|

225

Table 2: 3rd -person IX in a portion of the CLESS Corpus (Lillo-Martin & Chen Pichler 2008). Sentence role

Proportion

Subject Object Location Topic Modifier of N(P) Other

.53 .09 .11 .02 .04 .21

ent has not been assigned such an area, IX may be signed in a neutral location – in front of the signer. The location used for a referent is often called a ‘locus’ (note that the locus is not a geometric point, but an region with some depth; cf. Liddell 2003). The process of associating a referent with a locus is a complex issue to which we will return, although we will not attempt to fully explain it. In some cases, this process involves signing a name or NP in the location to be used; other cases involve signing a name or NP and then pointing to and/or gazing at a locus.² Once a locus has been associated with a referent, informally speaking, pointing to the locus is interpreted in much the same way as a pronoun. For example, in (1), ‘Doreen’ is associated with the locus labeled ‘a’. Subsequently, a-IX (signed as in Figure 1 (c) or (d)) is interpreted as referring to Doreen. (1)

a-DOREEN a-IX TEACH ASL a-IX SKILLED ‘Doreeni teaches ASL. Shei is skilled.’

A related use of the 3rd -person IX is that of ‘deferred ostention’ (Quine 1968; Recanati 1993) – a case when IX refers not to the item pointed at directly but to something related to the item which has been made salient, perhaps through context. For example, in (2), the point to the phone is interpreted as referring to the person on the other end of the line (the doctor). (2) WHAT IX(phone) SAY, IX(phone) DOCTOR DOCTOR IX(phone), WHAT SAY WHAT IX(phone) ‘What’s the he[pointing at the phone] say? That’s the doctor[pointing at the phone]. What does he[pointing at the phone] say?’

2 When a name, NP, or IX is associated with a locus, this will be indicated by the use of a lowercase letter at the beginning of the gloss.

226 | Elena Koulidobrova and Diane Lillo-Martin

The referent of IX need not be human or even concrete. For example, IX can be used to refer to propositions, as in (3). (3) a-GET #a-JOB DISJ/shift b-GO b-GRADUATE-SCHOOL. a-IX I CAN b-IX IMPOSSIBLE ‘Get a job or go to graduate school? The former I can do, but the latter is impossible.’ In (3), disjunction is achieved by shifting the body between two areas of space: one (area a) in which the expression GET #JOB was signed and another (area b) in which GO GRADUATE-SCHOOL is signed (see Davidson 2013 for discussion). From that point on, the area to one side of the signer’s midline can be called locus ‘a’, and this locus can now be successfully pointed at; the other side is locus ‘b.’ The referent of IX must be clearly recoverable from the context. This context may (as in (1)) but does not need to be linguistic, as illustrated in (4). (4) Context: Hankamer attempts to stuff a 9-inch ball through a 6-inch hoop a. I know I can do it/this[stuff a 9-inch ball thought a 6-inch hoop] t

b. a-IX 1IX CAN ‘I can do it/that.’ (adpt. Hankamer & Sag 1976: 392) Among other qualities of IX is an ability to serve as an argument which obeys linguistic restrictions such as number agreement, as in (5). (5) a-c-STUDENTS WORK HARD. a-c-IX GET-A. ‘The students work hard. They got As.’ The aforementioned uses of IX have traditionally been considered to be pronominal (see Meier and Lillo-Martin 2013 for an overview). But one of the chief reasons for viewing IX as a (personal) pronoun is the fact that it obeys the Binding Conditions, as illustrated in (6). (6) a. a-PETER THINK a-IX SMART ‘Peteri thinks hei/∗j is smart.’ b. BOY a-c-ALL THINK a-c-IX SMART ‘All the boysi think theyi/∗j are smart.’

A ‘point’ of inquiry: The case of the (non-)pronominal IX in ASL

|

227

c. a-PETER THINK a-IX SMART, b-JOHN SAME ‘Peter thinks he is smart; John does too.’ i. John thinks John is smart ii. John thinks Peter is smart d. a-MARYi SMART. a-IXi TEACH {OK a-SELFi /∗ a-IXi }. ‘Mary is smart. She taught {OK herselfi /∗ heri }.’ In (6a), IX is coindexed with PETER; in (6b) it appears bound by the quantifier ALL; in (6c), it allows for a sloppy reading (ii.), typical of bound pronouns; in (6d), IX cannot be bound within one clause and SELF is needed instead. To briefly summarize then, IX in ASL is a linguistic expression that exhibits properties of both linguistic and discourse anaphora. In this, it behaves very much pronominally (cf. Johnson 2009). But this, of course, does not mean that in its form it is necessarily a personal pronoun. What we would like to suggest here is that in fact, it isn’t. To make our point more explicit, we will be arguing that ASL IX is better analyzed along the lines of the English demonstratives this/that.³ Why might we want to engage in this enterprise? There are a few reasons, which we will briefly mention here; in Section 4 we will explore this possibility in more detail and provide evidence to support it.

3 Puzzle Recent research has uncovered some properties of IX that are not typical of personal pronouns. Schlenker (2013) has observed that unlike pronouns in English, IX can be used to pick out complement-set reference, even with non-downward entailing quantifiers (cf. Nouwen 2003), as illustrated in (7). This is unexpected on the personal pronominal view of IX. (7) 1-POSS STUDENT IXarc-ab MOST IXarc-a a-CAME. IXarc-b b-STAY HOME ‘Most of my students came. They [= the ones who didn’t come] stayed home.’ (Schlenker 2013: 1)

3 Note that at first glance, such a claim is incompatible with the English in (6a, b): (i)

Peteri thinks that (person)∗i is smart All the boysi think those (people)∗i are smart

We return to this issue in Section 4.2.

228 | Elena Koulidobrova and Diane Lillo-Martin

In both ASL and the English translation of (7), the same lexical item is used for both groups of students – the ones that came and the ones who did not. In English, use of the same pronominal expression to denote both sets is impossible. However, the utterance in ASL is quite natural. Further, Abner and Graf (2012) observe that ASL displays limited obviation with coordinated pronouns, as illustrated in (8). They propose that ASL uses discourse binding rather than syntactic binding in such cases. They also point out other reasons to favor the discourse binding analysis of ASL interpretation of IX, including its resistance to antecedence by quantified DPs with empty domains. (8) ALLi WRESTLERi INFORM SOMEONEj SWIMMERj THAT IXi,j IXj,i WILL RIDEIN-VEHICLE LIMO GO-TO DANCE ‘Every wrestleri told some swimmerj that himi,j and ∗ himj,i would ride in a limo to the dance.’ (Abner & Graf 2012) These oddities in behavior – atypical of pronouns – create a justification for considering an alternative. Of course, what would need to be accounted for are cases in which IX resembles a personal pronoun. We will be arguing here that such resemblance is just that – a resemblance, and the ability of IX to participate in processes characteristic of personal pronouns is only apparent. Instead, we suggest that upon closer examination, IX behaves as a demonstrative. The idea that pronouns are related to demonstratives is not entirely odd. In Eastern Armenian, for instance, demonstratives na, sa, and da are used in lieu of personal pronouns (Kozintseva 1995). The same can be said for the Basque hau’, hori, and hura (Saltarelli et al. 1988) as well as the Korean ku. While it is not the goal of this paper to argue for a particular theory of the aforementioned elements as well as their counterparts in other world languages, a brief survey of languages represented in WALS shows that out of the 225 languages examined, in 125, demonstratives and 3rd person pronouns are morphologically related if not identical, as detailed in Table 3. However, simply finding that a relationship between demonstrative and personal pronouns is a ‘regular occurrence’ is not enough here. The proposal that IX in ASL is a demonstrative has non-trivial consequences. If our proposal is in line with the cross-linguistic data, certain behaviors of IX are necessarily expected – namely the behaviors one observes with demonstrative expressions in natural languages. Let us see what this means, beginning with complex demonstratives (those with an NP complement) and extending the discussion to their simplex (a.k.a. bare) counterparts.

A ‘point’ of inquiry: The case of the (non-)pronominal IX in ASL

|

229

Table 3: Number of languages in which 3rd person pronouns (PPros) and demonstratives show relationships (www.wals.info/chapter43A). 3rd PPros and demonstratives unrelated 3rd PPros and demonstratives to all demonstratives 3rd PPros and demonstratives to remote demonstratives 3rd PPros and to non-remote demonstratives 3rd PPros and demonstratives by gender markers 3rd PPros and demonstratives for non-human reference

= 100 = 52 = 18 = 14 = 24 = 17

Total: 225

4 Proposal: Demonstrative Demonstratives have puzzled semanticists for quite some time. A number of researchers have argued for a generalized quantifier analysis. Consider (9). (9) a. That student who scored 100 % is a genius. b. Most avid skiers remember that first black diamond run they skied. (King 2001) c. Every time I meet a linguist, that linguist talks to me. We understand (9a) as a case where there is exactly one student who received 100 %, and the property of being a genius holds of that student. (9b) records a case of quantification into a demonstrative: there is no particular skier under discussion, but skiers and black diamond mountains covary. The same could be said for (9c): for every encounter x, a different linguist talks to the speaker. In each of the sentences in (9), that can be easily replaced with the without any visible effect on interpretation. Yet, this is not always so: for all their quantificational behavior, demonstratives often behave like rigidly referring expressions. For instance, consider Nowak’s (2013) scenario involving a consistently rigged race such that the person wearing a particular color (say, red) is necessarily the winner because that is who the judges have been bribed to award the win to. In such a context, demonstratives disallow the covarying interpretation while [the NP] allows it (i.e. whoever wears the red shirt always wins): (10) a. That guy in the red shirt always wins = referential/∗ quantificational b. The guy in the red shirt always wins = referential/quantificational (Nowak 2013)

230 | Elena Koulidobrova and Diane Lillo-Martin

The reason for this difference remains somewhat unclear, and at least some of the literature capitalizes on the fact that such rigidly referring cases are typically understood as involving a demonstration. (11) a. [pointing at the man]That red haired man is an Australian. (King 2008: 2010) b. That chair is that chair [with appropriate demonstrations] (Heim 1991, ctd. in Roberts 2003: 99) c. Mary actually pulled a red scarf from the drawer, but that scarf [pointing at the scarf] could have been blue (Wolter 2006: 196) Roberts (2003), following the original intuitions in Kaplan (1989), offers an account of this demonstration, articulated informally below. (12) […] Demonstratives and definite descriptions are both types of definite NPs, and hence both have presuppositions of familiarity and informational uniqueness. The difference between them is this: the demonstrative, but not the definite description, carries a presupposition that the discourse referent which makes it familiar is anchored by information in the common ground to an individual in the world which is directly indicated by the speaker at the time of utterance of the demonstrative NP.⁴ (Roberts 2003: 130; italics are ours) Demonstratives also exhibit another characteristic: they seem to generally prefer different discourse antecedents than other definite descriptions do (see Bosch

4 Formally: Given a context of evaluation C, with common ground CG s.t DomCG ⊆ DomC, and discourse referent S s.t. ∀i ∈ DomCG ∀⟨w, g⟩ ∈ SatCG[speaker(w)(g(i)) ↔ i = S], (i) ∃δ[δ ∈ DomCG & ∀⟨w, g⟩ ∈ SatCG[demonstration(w)(g(δ)) & accompanies(w)(g(δ), utterance(NPi ))] & (ii) ∀j ∈ DomCG[∀⟨w, g⟩ ∈ SatCG[+(−)proximal(w)(g(j), g(S)) & demonstratum (w)(g(j), g(S), δ)] & ∀k ∈ DomCG[∀⟨w, g⟩ ∈ SatCG[+(−)proximal(w)(g(k), g(S)) & demonstratum(w)(g(k), g(S), δ)] → k = j] & Desc(w)(g(j))] & (iii) j = i ]] where Desc(w)(g(i)) is true iff the individual assigned to i by g has the property denoted by Desc in world w; and +(−)proximal(w)(g(j), g(S)) & demonstratum(w)(g(j), g(S), δ) is true iff the individual assigned to j by g is in the set of entities (non-)proximal to the speaker g(S) and is the demonstratum intended by g(S) for the demonstration g(δ). (Roberts 2003: 118)

A ‘point’ of inquiry: The case of the (non-)pronominal IX in ASL

|

231

et al. 2007 and references therein). This is illustrated in (13), where the demonstratives cannot be interpreted as picking out the subject of the previous sentence. (13) a. Pauli wanted to go running with Peterj . But {the guyi,j /hei,j /that guyj,∗i } had a cold. b. The patienti was examined by the doctorj . {The guyi,j /hei,j /that guyj,∗i } was prepared. This property of demonstratives has been described in various ways: as ‘nonsubject orientation’ (cf. Bosch 2006, i.a.), as referring to the ‘anti-aboutness topic’ (Hinterwimmer 2014), or to a ‘non-default situation’ (Wolter 2006), among others. Whatever the final account, decades of formal examination have now yielded a fairly clear empirical picture: among definite descriptions, demonstrative expressions are special. In some contexts they are interchangeable with personal pronouns and overt noun phrases headed by the definite article (as in (15a)). In other contexts demonstrative expressions pattern on par with either pronouns (as in (14a), (15b), and (17)) or with overt [the NP] (as in (14b) and (16b)). (14) a. [pointing at a man] {That man/∗ the man/he} looks friendly. b. {That man/the man/∗ he} by the fireplace looks friendly. (15) a. A man walked in. {That man/the man/he} handed a flyer to another man. b. A man and a woman walked in. {∗ That man/the man/he} was smiling. (16) a. {∗ That center of the universe/the center of the universe/#it} is moving. b. {Those members of Congress who voted against the bill/the members of Congress who voted against the bill/#they} have proposed a new initiative. (adptd. Wolter 2006: 22) (17) A bishop met another bishop on a dusty road. {That bishopj,∗i /the bishopi,j /hei,∗j } sneezed. In order to differentiate between the relevant uses of IX then, it seems that what we need is a context in which demonstratives are licit and pronouns are not (as in (14b) and (16b), for example). Considering that in a number of such contexts [the NP] remains a possibility, this option will need to be considered as well. Incidentally, a number of works in the literature, going back to Evans and more recently

232 | Elena Koulidobrova and Diane Lillo-Martin

articulated by Elbourne (2005), i.a., have argued that semantically, a pronoun and definite article are very much related: pronouns are definite descriptions involving NP ellipsis. Simplistically, this means that in (18), for example, he is really the bishop with the N (bishop) unpronounced and the realized as he. (18) When a bishop meets a bishop, he blesses him In fact, Elbourne (2005) provides the following lexical entry for the and it in English: (19) [[the]]g = [[it]]g = !f⟨⟨s,e⟩,⟨s,t⟩⟩ . λs : ∃! λx f(λs.x)(s) = 1. ix f((λs󸀠 .x)(s) = 1 (Elbourne 2005: 51) Our argumentation then will proceed as follows: we first demonstrate that [IX NP] cannot be equated with [the NP] – i.e. prenominal IX is not a definite article. This move will allow for a cleaner comparison between personal pronouns vs. demonstratives.

4.1 Is IX a definite article? Assuming the view often advocated in the literature since Postal (1966), a pronoun is, in fact, something like a definite article (give or take a few presuppositions, as in Schlenker 2003) with an elided, or unsaturated, NP argument (cf. Elbourne 2005). A two-fold prediction arises: (a) in non-elliptical cases, IX followed by an NP will behave on par with a definite article, and (b) undoing the ellipsis of/saturating the NP argument will yield no visible difference in meaning – i.e., a definite description of the sort [IX ] ought to behave similarly to [IX NP], at least in cases where it is true for similar expressions in other languages. If there is a difference, it is expected to be subsumed by a general account of the difference between [the NP] vs. [he ] on the traditional view of IX and [that NP] vs. [that ] (also known as ‘complex’ vs. ‘bare’ or ‘simplex’ demonstratives) on our. The literature offers no shortage of discussions related to the contribution of the definite article. Much of the field, following Frege (1893), assumes that a definite description headed by a definite article expresses a proposition which is either unique (Russell 1905, i.a.) or familiar (Heim 1982, i.a.; for an extended discussion of the two views, see Elbourne 2005). For the purposes of this discussion, we assume the Fregean view of the expression the P: the uniqueness condition is a presupposition. For instance, the expression the Kind of France is bald receives the interpretation as in (20a). We also assume, with Wolter (2006), for example, that

A ‘point’ of inquiry: The case of the (non-)pronominal IX in ASL

|

233

a definite description is interpreted relative to a situation which may be default or not (although nothing here is contingent on this particular view; it simply allows a convenient demonstration of the facts). (20) a. Bald (ix. King Of France (x)) defned if and only if there is a unique King of France; true if and only if the unique King of France is bald, else, false. (Schwarz 2009: 9) b. A sentence of the form [the A] B presupposes that there exists a unique individual which is A in s n and asserts that the unique individual which is A is also B in s n . c. s n = ‘default’ situation (discourse context) d. Given a sentence S, a situation variable s is a default situation just in case it is bound in S. Otherwise s n is a non-default situation. e. A default situation is a situation relative to which the main predicate of a clause is interpreted and the truth value of the clause is calculated. (Wolter 2006: 64) The expectation then is that if some lexical item is best described as a definite article, it ought to behave in the manner described in (20). MacLaughlin (1997) argues that IX as in (21) – specifically, ‘prenominal IX’ – encodes definiteness in ASL and is best described as a definite article. (21) IXDET WOMAN BORROW VIDEOTAPE ‘The woman borrowed the videotape.’ (adptd.⁵ MacLaughlin 1997: 124) However, as has been argued elsewhere (Koulidobrova 2012, submitted), prenominal IX fails various tests for being a definite article, behaving, instead, as a different type of element. Let us illustrate. The semantics in (20a) predicts the (and, by the temporary hypothesis, IX) to occur in so-called ‘global uniqueness’ cases. We see this in English in (16a): the universe is expected to have only one center; yet, the precedes the NP. Notice that

5 The sentence has been adapted – in the original, a postnominal IX is also present. However, at this point it detracts from the focus of the argument; nothing in our analysis (or in MacLaughlin’s for that matter) hinges on the presence or absence of the postnominal IX in this utterance. We return to the issue below.

234 | Elena Koulidobrova and Diane Lillo-Martin

a demonstrative in this environment is impossible; intuitively, it requires a larger set from which the relevant individual must be picked out. As the universe has only one center, so does France have only one capital and, in a typical scenario, a church only one priest. Yet as (20) shows, IX is impossible in such environments. t

wh?

(22) a. FRANCE (∗ IX) CAPITAL WHAT ‘What is the capital of France?’ wh

b. TODAY SUNDAY. DO-DO. GO CHURCH, SEE (∗ IX) PRIEST. ‘Today is Sunday. What to do? I’ll go to church, see the priest.’ But perhaps the problem of ∗ IX in (22) is that IX must be anaphoric/familiar, and (22) provides no room for such an anaphoric element. This move, coincidentally, is consistent with argumentation in Schwarz (2009), who points out that the typical analyses of the semantics of the definite article (with English as the language of examination) is not fine-grained enough. In fact, it is best to distinguish between its weak uses (encoding uniqueness and part-whole relations) and strong uses (encoding familiarity and anaphoricity). The weak environment we have already examined (cf. (22)) and discovered that the ASL IX does not behave the way a definite article is expected to in such cases. It is still necessary to review the behavior of IX in strong environments. It turns out, however, that this path leads to a dead-end as well: the anaphoric/familiar PRIEST in (23) cannot be preceded by IX either. wh

t

(23) TODAY SUNDAY. DO-DO. GO CHURCH, SEE PRIEST. (∗ IX) PRIEST NICE. ‘Today is Sunday. What to do? I’ll go to church, see the priest. The priest is nice.’ Compare then the use of IX in ASL with the in the translation sentences in English. In both cases necessarily involving the definite article in English (i.e. (22) and (23)), IX is impossible, and PRIEST remains obligatorily bare. In other words, whatever the definite article encodes – uniqueness or familiarity (see Schwarz 2009 for an extensive overview of the discussion) – IX appears unable to do. We thus conclude that the presence of IX next to the NP (or in lieu of it) signals something else. Taking a step back, let us now consider the alternative: what does it mean when we say that IX is not a definite article? Assuming that a definite article necessarily requires a ‘default’ situation as in (20), the prediction arises that in contexts that are distinct from such situations (i.e. ‘non-default’, e.g. where potential alter-

A ‘point’ of inquiry: The case of the (non-)pronominal IX in ASL

|

235

natives must necessarily be considered), a definite article will be impossible but IX might be allowed. The prediction is borne out for [the NP] cases below; this is true for environments consistent with weak as well as strong definite articles. (24) Weak: a. The Pope retired = # if looking at/pointing out the portrait of Benedict among the portraits of other popes = OK out of the blue t

b. a-IX POPE RETIRE = OK if looking at/pointing out the portrait of Benedict among the portraits of other popes = # out of the blue (25) Strong: a. The new priest arrived = # if pointing out Father John among other priests = OK out of the blue if both interlocutors know that a new priest has arrived t

b. a-IX PRIEST NEW ARRIVE = OK if pointing out Father John among other priests = # out of the blue if both interlocutors know that a new priest has arrived The data in (24) and (25) are reminiscent of the ‘bridging inference’ (Clark 1975, i.a.) paradigm – contexts where the definite article alone is licensed and demonstrative ascriptions are impossible (as in (26a)). Yet, as soon as the discourse offers a contrastive interpretation, the paradigm changes and the demonstrative becomes a preferred choice (as in (26b)). (26) a. A car drove by. {OK The/∗ that} horn was honking loudly. b. The girls received individualized reading lists with sections labeled ‘on reserve’ and ‘in bookstore.’ Every girl went to the reserve section of the library and read {∗ the/OK those} books first. (Wolter 2006: 76, 100) To summarize then: in all of the cases where a definite article is licit and a demonstrative is not, [IX NP] is impossible, and, conversely, in the environments clearly calling for a demonstrative, [IX NP] is fine. In light of the data above, we are well justified in arguing that IX is not a definite article (see more discussion on the

236 | Elena Koulidobrova and Diane Lillo-Martin

matter in Koulidobrova, submitted). This is why in the global uniqueness cases, IX is quite odd, as shown earlier in (20). Let us consider further the nature of restrictions on demonstratives by returning to (16a), repeated here as (27). (27)

∗

That center of the universe is moving.

= (16a)

The demonstrative expression in (27) has an odd reading that there is more than one center of the universe, which of course defies common knowledge. Let us assume, however, that a two-centered universe is possible. This improves (27) dramatically, though only if deixis is involved – it is only possible if the immediate physical surroundings offer a chance at such a demonstration, much as in (14a), repeated here as (28). (28) [pointing at a man] {That man/∗ the man/he} looks friendly.

= (14)

However, we now appear to be comparing one of the centers to another, appealing to a contrast of sorts – a reading that [the NP] ordinarily does not invoke. This observation is further exemplified in (29). (29) a. # The dogs are sleeping and the dogs are not. b. Those dogs are sleeping and those dogs are not. (Dayal 2009: 23) The same test can be applied to (26) and (27), and the judgments reverse. (30) a. That Pope retired = OK if looking/pointing out the portrait of Benedict among the portraits of other popes = # out of the blue b. That new priest arrived = OK if pointing out Father John among other priests = # out of the blue if both interlocutor know that a new priest joined the parish We then naturally expect [IX NP] to behave differently from [the NP] in cases like (14a), (15b), (16a), as well as (10). And so it does: in the context involving the rigged race (see (10)), a quantificational reading arises without IX; however, with it, only a referential reading is possible.

A ‘point’ of inquiry: The case of the (non-)pronominal IX in ASL

|

237

(31) a. The guy in the red shirt always wins = referential/quantificational = (10b) rc

b. {a-IX PERSON/a-IX} RED SHIRT SELF TEND WIN ‘IX person/IX in the red shirt tends to win’ = referential/∗ quantificational rc

c. PERSON HAVE RED SHIRT TEND WIN ‘The person in the red shirt tends to win’ = referential/quantificational We thus arrive at the view that IX diagnoses as something other than a definite article (either weak or strong) and is licit precisely in the cases one might ordinarily encounter a demonstrative instead.

4.2 Is IX without an NP a pronoun? So much for the view that IX may (sometimes) be analyzed as a definite article. However, what of the idea that it is sometimes a personal pronoun – the starting point of this discussion? Let us begin with an empirical observation arising from the data documented in the previous section. It turns out that in cases diagnosing for weak and strong articles (as in (22) and (23), IX alone is also impossible. Note that a (personal) pronoun – a bona fide anaphoric expression – should easily be able to refer to the church in (32a, b) and the priest in (32b) as is seen in the English translations. wh

(32) a. TODAY SUNDAY. DO-DO. GO CHURCH, SEE {∗ IX/OK PRIEST} = (22) ‘Today is Sunday. What to do? I’ll go to church, see ∗ it/the priest.’ wh

t

t

b. TODAY SUNDAY. DO-DO. GO CHURCH, SEE PRIEST. { ∗ IX / OK PRIEST } NICE = (23) ∗ ∗ ‘[…] see him/ it/the priest. The priest is nice.’ ASL and English behave in an opposite manner here, corroborating the view that IX does not pattern as a personal pronoun: a personal pronoun (such as it) should have been able to refer to the church – the NP immediately above IX. However, reference to the church is unavailable in (32a). In (32b), another NP enters – PRIEST. Nothing precludes IX from referring to either of the NPs; yet, this remains impossible. It seems reasonable, then, to turn to the alternative – demonstrative expressions. The difference between demonstrative expressions and other definite descriptions is typically argued to be presuppositional in nature: personal pronouns and demonstratives are essentially definite articles but with an additional piece that differentiates between them. As we mentioned earlier, Roberts (2003) casts

238 | Elena Koulidobrova and Diane Lillo-Martin

the presupposition in terms of the presence/absence of the demonstratum. Others approach the difference between the relevant lexical items in terms of distance – both intuitively and formally (cf. Wolter 2006, Elbourne 2008), reliance on the proximal/distal distinction (in terms of space and time) may after all account for the behaviors of demonstratives. The goal of this paper is not to side with any particular account. Instead, we aim to demonstrate that the distribution of IX simply follows the distribution of demonstratives. To that end then, we expect the aforementioned characteristics of demonstratives to surface with IX. On an account of personal pronouns similar in spirit to that in (19) and, by extension, (20), a personal pronoun will pick out a maximally salient referent ‘uniquely relative to the situation corresponding to the discourse context’ – the ‘default situation’ (Wolter 2006). This intuition is confirmed in Bosch and Umbach (2006) – in German, P(ersonal)-pronouns (such as er below) tend to pick out subjects/discourse-old antecedents – those that are maximally salient from discourse and require no additional descriptive content. In contrast, D(emonstrative)-pronouns (i.e. der in (33)) avoid such referents. (33) a. Pauli wollte mit Peterk laufen gehen. Aber {eri /derk } war erkältet. ‘Pauli wanted to go running with Peterk . But {hei /DEMk } had a cold’ b. [DO Den Patienten]i untersucht [SU der Chefarzt]k . Derk ist nämlich Herzspezialist. ‘[The patient]i was examined by [the head doctor]k DEMk is a heart specialist.’ (Bosch et al. 2003: 2) English complex demonstratives follow this pattern neatly, as shown in (17), repeated here as (34). (34) a. A bishopi met another bishopj on a dusty road. {That bishopj,∗i /the bishopi,j /hei,∗j } sneezed.

= (17)

How the reference to the discourse-old information is formalized remains a matter of some controversy. Hinterwimmer (2014), for instance, argues that demonstrative pronouns disallow reference to the aboutness topic (Reinhart 1981). Whatever the eventual solution, let us now summarize the descriptive generalization emerging from the relevant empirical observations and previous accounts: a personal pronoun (exemplified by the P-series in German) is a uniquely referring expression, typically interpreted relative to some standard/default situation and picking out a salient antecedent which also happens to be what the sentence is about. The question here is this: is it the case that ASL IX qualifies as such an element?

A ‘point’ of inquiry: The case of the (non-)pronominal IX in ASL

|

239

While a focused examination of the paradigm must await further study, the preliminary answer to this question seems to be ‘no’. For example, in (35a), we take ‘mother’ to be the topic/subject/discourse-provided individual. Here, IX most naturally refers to either Mary or the sandwich. On the other hand, when SANDWICH is topicalized as in (35b), IX most naturally refers to either the mother or Mary. t

(35) a. a-MOTHERi PERSUADE b-MARYj MAKE SANDWICHk . neu-IXj,k,?i GOOD ‘My mother persuaded Mary to make a sandwich. IX is good.’ t

t

b. SANDWICHk a-MOTHERi PERSUADE b-MARYj MAKE. neu-IXj,k,?i GOOD ‘Mother is persuading Mary to make a sandwich. IX is good’ Confirmation of the observation in (35) with more signers is needed (something that seems to be clearly required, judging by the experimental results in German, Bosch 2013), but this preliminary datum points away from the personal pronoun view of IX. Recall also that unlike personal pronouns, demonstratives presuppose a demonstration (and a demonstratum), which, on all accounts, is at the heart of deixis. It seems that if IX were a personal pronoun semantically, or at least were able to function as one, it should have happily been able to coindex with a topic/subject NP up the tree and do so without any such demonstration. Yet, recall also that this is impossible in ASL. However, if the signer is pointing at some specific object, which is physically present in the environment and is assigned a locus, the use of IX in becomes felicitous. In (36), this object is the church. And while a personal pronoun happily picks out the linguistic object (church), it cannot refer to a part of the physical entity associated with this object (a rosewindow) or the general area in physical space. A demonstrative supported by deixis, on the other hand, can (as indicated in the English translation). Notably, IX in (36) can refer to either a part of or the whole church (e.g. the rose-window or the building itself), or even the general area in which the church is located. wh

(36) TODAY SUNDAY. DO-DO. GO CHURCH. a-IX NEW. SEEM NICE ‘Today is Sunday. What to do? I’ll go to church. IX is new. Seems nice.’ a. … ([pointing]){Itchurch /∗ itwindow /∗ itbuilding /∗ itarea there / ∗ it object in physical space } is new. b. … [pointing] {Thatchurch /thatwindow /thatbuilding /thatarea there / thatobject in physical space } is new.

240 | Elena Koulidobrova and Diane Lillo-Martin

In other words, IX in (36) refers not to the linguistic object but to the entity itself (Roberts 2003).⁶ Incidentally, as Table 1 in Section 2 shows, in the child directed data set we analyzed, the majority of the instantiations of IX refer to individuals (human and non-human) immediately present in the physical context or cases of deferred ostension – i.e. the demonstratum. Yet, the most important observation regarding the (non-)pronominal status of IX is the following. Recall that the original (6), repeated here as (37), showed that IX appears to be bound – i.e. it serves as a variable, which strengthens the original motivation for its account as a pronoun. In these environments, a personal (and not a demonstrative) pronoun is expected: the IX refers to the subject/discourse/ aboutness topic. ≈ (6)

(37) a. a-PETER THINK a-IX SMART ‘Peteri thinks {hei /∗ that[person]i } is smart’ b. BOY a-c-ALL THINK a-c-IX SMART ‘All the boysi think {theyi /∗ those(people)i } are smart’ c. a-PETER THINK a-IX SMART, b-JOHN SAME ‘Peteri thinks hei is smart; Johnj does too’ i. Johnj thinks {Johnj /∗ that(person)j } is smart ii. Johnj thinks Peteri is smart

However, note that in (37), the locus of IX (namely ‘a’) corresponds to the previously established locus of the NP (or the QP) with which the IX is coindexed. Loci themselves raise considerable interesting issues (Barberà 2012, Gagne & Davidson 2014, Schlenker et al. 2013, et seq.), yet their contribution often is but should not be conflated with the contribution of IX. For instance, extraction out of an island is possible only if the locus has not been previously assigned (Koulidobrova 2012). (38) ‘Motherj , I don’t know what (shej ) likes.’ a. a-MOTHERj , 1-IX DON’T-KNOW WHAT b.

∗

MOTHERj , 1-IX DON’T-KNOW WHAT OK

(Koulidobrova 2012) /OK a-IXj LIKE. /

∗ a-IX

j

LIKE.

6 Perhaps this is precisely the reason backwards anaphora is not considered fully grammatical in ASL (Sandler & Lillo-Martin 2006): for a deictic element to be felicitous, the demonstratum first needs to be introduced (and the demonstration occur).

A ‘point’ of inquiry: The case of the (non-)pronominal IX in ASL

|

241

So, in order to bypass the potential issues associated with previous establishment of loci, we examine cases without it by removing the locus for the NP anteceding the IX in (37). Not surprisingly, in light of other arguments in this paper, it turns out that the paradigm in (37) changes dramatically, as shown in (39). Thus, the altered sentences remain grammatical but the readings they induce have changed. Much like (37), (39) records utterances compatible with Condition B – a personal pronoun ought to be allowed here. However, now IX can no longer refer to any of the previously uttered NPs; it is necessarily disjoint in reference from them. (39) a. PETER THINK {a-IX/neu-IX} SMART ‘Peteri thinks hej/∗i is smart’ b. BOY ALL THINK {a-c-IX/neu-IX} SMART ‘All the boysi think theyj/∗i are smart’ c. PETER THINK {a-IX/neu-IX} SMART, b-JOHN SAME ‘Peteri thinks hej/∗i is smart; Johnk does too’ = Peter and John think someone else is smart d. FEW CHILDREN THINK IXarc SILLY ‘Few childreni think theyj/∗i are silly’ Nor, as the data in (39) demonstrate, can the issue be reduced to some type of requirement on locus matching.⁷ If this were true, we would expect the following: the paradigm in (39) would once again become grammatical with IX uttered in the neutral space (lack of locus) – in front of the body of the signer. It turns out, however, that this move does not help: irrespective of whether IX in such a case is uttered at a particular locus (arbitrarily assigned to the right or to the left of the signer) or in a neutral space, reference to the NP (or QP) above is impossible. We take the behavior of neu-IX in such cases to reflect the nature of this element as different from a personal pronoun, and the different behavior of IX with a locus as something attributable to the contribution of the locus. The differences in the referring possibilities of IX also arise in so-called ‘donkey’ anaphora in which farmers and donkeys co-vary, as in (40). (40) If a farmer owns a donkey, he beats it. Such E-type binding is allowed only when the locus has been previously assigned. Consider (41). In (41a), previous establishment of a locus for ONE STUDENT permits apparent binding of a-IX (and the co-referential interpretation is required).

7 We thank Kathryn Davidson for bringing this to our attention.

242 | Elena Koulidobrova and Diane Lillo-Martin

On the other hand, when no locus is established, as in (41b), the bound reading is unavailable, irrespective of whether IX points to a specific locus or to a neutral location. (41) a. WHEN ONE a-CL STUDENTi COME PARTY, a-IXi,∗k HAVE-FUN. ‘When a student comes to the party, he has fun.’ b. WHEN ONE STUDENTi COME PARTY, a-IX∗i,k /neu-[CL IX∗i,k ] HAVE-FUN. ‘When a student comes to the party, he has fun.’ (Schlenker 2011: 18) The non-binding behavior of the ‘pronominal’ IX in (39, 41b) bears an uncanny resemblance to the demonstrative pronouns in German. For instance, Wiltschko (1998) notes that D-pronouns cannot be interpreted as bound variables: (42) a. Peteri hat geglaubt, daß eri /∗ deri dumm ist ‘Peter believed that he/DEM is stupid.’ b. Jeder Manni glaubt, daß eri /∗ deri dumm ist ‘Every man believed that he/DEM is stupid.’ (Wiltschko 1998: 144) Why this might be remains open to discussion; both semantic and syntactic accounts have been proposed (see, e.g., Hinterwimmer 2014 vs. Patel-Grosz and Grosz, t.v., i.a.). The account advocated by Wiltschko is that a D-pronoun is a definite description; as such, it is subject to Condition C. The very same explanation is available for the ASL IX: on the view that IX is a pronoun, the difference between (37) and (39) is not immediately clear; on the approach to IX as a demonstrative, (40) is excluded as a condition C violation, similarly to (42) and (43b). (43) a. Jilli thinks that shei is nice. b. Jilli thinks that {this/the} person∗i is nice. We have now arrived, we think, at a second conclusion: as IX preceding an NP can hardly be considered a definite article, neither can IX occurring alone be viewed as a personal pronoun. Without the previous establishment of a locus, it cannot be bound; instead, it tends to have a deictic use – i.e. it is picking out some physical object in the environment. At the same time, data suggest that IX tends to pick out antecedents that are ‘non-default’ – when the NP it refers to is not a subject/topic and could be contrasted with another NP. Everything works out nicely if, with and without an accompanying NP, IX is a demonstrative. Such an approach to IX explains some otherwise puzzling ob-

A ‘point’ of inquiry: The case of the (non-)pronominal IX in ASL

|

243

servations about ASL ‘pronouns’ previously employed as evidence for modalityrelated differences between sign and spoken languages: as discussed in Section 3, the possibility of reference to the complement set (Schlenker 2013, (7)), and the lack of syntactic (but, rather, discourse only) binding with IX (Abner and Graf 2012, see (8)); and as discussed below, lack of a consistent formational distinction between ‘pronominal’ and ‘locative’ pronouns (Johnston 2013). These characteristics fall out naturally from the view that interpretation of demonstratives recruits discourse binding, as has been argued for cases like (44) (see the discussion in Roberts 2003): (44) That[pointing to a chair1 ] might be that[pointing to a chair2 ] (Roberts 2003: 99) Let us now take this possibility to heart: [IX NP] is the ASL counterpart of complex demonstratives ([this/that NP]) and IX alone is actually something like English simplex demonstratives ([this/that ]). Following previous research, we assume that simplex demonstratives are best viewed as involving some elided/ phonologically null material following a demonstrative head. Note where this takes us. Suppose we follow King (2001) or Elbourne (2005) in approaching this elided/null material as contributing the property of ‘being a thing’ (King 2001: 141–143), an ‘individual of type ⟨e⟩’ (Elbourne 2005).⁸ The unexpressed ‘individual’ could be a person or a place. The ultimately locative (in terms of distance and time) nature of demonstratives now allows for an approach to the ‘adverbial’ (MacLaughlin 1997) postnominal IX in the original version of the example amended in (21), repeated in full here as (45), to be subsumed under the same analysis. This second IX, which may be translated as here/ there, is comparable to “this/that place,” with the phonologically null place. ⁹ (45) IXDET WOMAN IX[variable path length] BORROW VIDEOTAPE ‘The woman (more or less far away) borrowed the videotape.’

≈ (21)

This approach is particularly plausible since ASL has few overt prepositions, therefore allowing for the possibility of analyzing [that person there] as [that person at that place]. This line of reasoning offers a prediction: if the phonolog-

8 In later works (2008, et seq.), the ONE is a predicate of type ⟨et⟩ (more accurately, and reflecting Elbourne’s situation semantics, ⟨se,st⟩). The unexpressed material is thus ignored for the relevant part of the semantic computation due to vacuous quantification (see a direct comparison of the two views in Elbourne 2008). 9 See Dayal (2004) for discussion of postnominal adverbial modifiers and demonstratives.

244 | Elena Koulidobrova and Diane Lillo-Martin

ically null complement of IX can be interpreted as any type of individual (place included), we expect ambiguity. The prediction is borne out in (46). t

t

(46) HEAR-SAY a-IX RESTAURANT a-IX REALLY GOOD. FOOD ITALY. 1-IX NEVER TASTE. WANT EAT a-IX. ‘I heard that that restaurant over there is really good. The food is Italian. I have never tasted it before. I want to eat {that/there}.’ The formalism remains to be worked out, but the path to capturing the various uses of IX has now been cleared.

5 Loose ends We have thus far challenged two views regarding IX – showing that IX behaves as neither a definite article nor a pronoun. We have also argued that it is best approached as a demonstrative. This of course presents its own challenges: we now turn to the question whether the different types of demonstrative expressions (with and without an overt NP) can be unified. Some researchers have argued that ‘bare’/‘simplex’ demonstrative differ from ‘complex’ ones minimally: in the ‘bare’ cases, the NP is simply elided (King 2001, Elbourne 2008, i.a.). An account of this sort would have difficulty explaining why unlike the complex demonstratives, bare demonstratives seem to resist bound variable interpretation: (47) a. Every time [a baby slaps [her hand]i ]j , [that hand]i/∗j hurts, and the baby cries. b. Every time [a baby slaps [her hand]i ]j , [that ]j/∗i hurts, and the baby cries. In (47a), the complex demonstrative refers to the child’s hand; in (47b), the simplex version thereof refers to the event involving hand-slapping and – crucially – not the hand itself. On the view that that in (47b) is [that NP], and ellipsis is either the usual matter of identity (as in King 2001) or an abstract noun ONE (as in Elbourne 2005), it is unclear where the interpretation in (47b) comes from. To elaborate: the NP-ellipsis account predicts several possibilities for the interpretation of the elliptical structure: that that refers either to the child or the hand that has been slapped (both of which are NPs in the discourse). Yet, neither of these readings is available: the only reading of that in (47b) is the event of self-

A ‘point’ of inquiry: The case of the (non-)pronominal IX in ASL

|

245

hand-slapping administered by a child. Incidentally, if that in (47b) were to be substituted by it, the set of available readings would include both the hand and the event of hand-slapping.¹⁰ King (2001) suggests that the deleted NP node might not host any lexical material at all and, rather, contributes the property of ‘being a thing’ (King 2001: 141–143). However, even with this amendment, it remains unclear why the reference to the event is possible here, and no other reference is available. Similarly, Elbourne (2005) argues that the argument of that in the bare demonstrative cases is a phonologically null noun ONE – an individual of type ⟨e⟩. On this view as well, reference to the baby or to the child in (47b) is not excluded. But we set this issue aside for now, unable to offer any evidence from ASL to help solve the problem – as we had noted earlier, without previous introduction of a locus, IX resists binding: when it is uttered, with or without an NP, it has a strong referential flavor and, thus, the asymmetry between a complex and a simplex demonstrative is bleached out. Similarly, if the complement of the IX is something like Elbourne’s ONE (or King’s ‘thing’) – which, for our purposes we informally label INDIVIDUAL – it is somewhat unclear why it appears not to be possible for either that or IX to be bound by a negative quantifier. (48) a. There is no kid that thinks that [∗ this/that INDIVIDUAL] is stupid b. HERE NOT-HAVE KID THINK [∗ neu-IX INDIVIDUAL] STUPID Schlenker (2011) argues that sign language pronouns are special, since their denotation must be non-empty. However, in light of the data offered here, we no longer need to appeal to such a difference between signed and spoken languages. Instead, we might simply claim that what goes wrong in (48b) is exactly what goes wrong in (48a) as well as (47b) – while the reasons for this remain to be explored, the resistance of the simplex demonstrative to being bound by a (negative) quantifier is not a novel observation. Another question concerns the NP ellipsis itself. As has been argued elsewhere (Koulidobrova 2012), NP ellipsis as in (49a) is productive in ASL. However, it is unable to strand a quantifier in the manner we see in English, as shown in (49b).

10 Note also that in a few dialects of English in which it is generally used to refer to a baby (Received Pronunciation, e.g.), the pronoun in (47) can also refer to the baby. This suggests that some feature-matching/-agreement is involved – in the original (47) ‘it’ is able to refer to the hand (ineligible for a feminine or a masculine pronoun in English). Here we might assume a presuppositional approach to such feature matching (Schlenker 2003); however, what is unclear now is what to do with the reference to the event. At any rate, the point here is that the options afforded by the personal pronoun are different from those available with the demonstrative.

246 | Elena Koulidobrova and Diane Lillo-Martin (49) a. BOB BUILD FEW HOUSE FAST, PAUL DESTROY ok HOUSE ‘Bob built a few houses fast; Paul destroyed (a few of {Bob’s/Paul’s} houses)’ b. MARY FINISH {2-POSS/MANY} BOOK. JOE SHOULD READ 1-POSS/FEW ∗ BOOK ‘Mary finished {your book/many books}. Joe should read {my/a few (books)}’ One thing to note here: if a-IX is a demonstrative, and demonstratives are quantifiers, then we might expect the paradigm in (49) to replicate with IX. And in fact it turns out that IX here behaves on par with the quantifiers: as POSS and FEW cannot be left behind, neither can IX, even with a demonstration and contrast – conditions normally licensing a demonstrative. (50) MARY FINISH a-c-IX BOOK. JOE SHOULD READ d-f-IX ∗ BOOK ‘Mary finished these books. Joe should read those (books)’ The question now, of course, is how to analyze IX when it occurs alone: if [IX ] is a case of NP ellipsis which ‘strands’ the demonstrative, then why is it that in regular NP ellipsis cases such stranding is disallowed? One option here is to bite the bullet and admit that the ellipsis of the NP argument of the demonstrative, thereby yielding a simplex demonstrative, is of a different variety than the other NP-ellipsis attested in a number of languages. This path may explain why, for instance, (48a) improves only if the original NP – and not our INDIVIDUAL – is restated. (51) There is no kid here that thinks that [that kid] is stupid This, however, will be a difficult path to take. For one, we have shown elsewhere that nouns in ASL appear to be able to type-shift between kinds and individuals (as in Chierchia 1998; see discussion in Koulidobrova 2012); this predicts the ellipsis of the sort discussed in Elbourne – one that results in a phonologically null element of type ⟨e⟩ or ⟨e,t⟩ – to be just the type of ellipsis that we observe in (49) and (50). The other option to pursue is that the argument of the demonstrative is not phonologically null but, rather, unsaturated. If this is the case, then we expect cases of simplex demonstratives to be allowed while NP ellipsis stranding the complex demonstrative is impossible. This is precisely what the data demonstrate: in ‘regular’ use of IX, the element is licit irrespective of the overt presence of a supporting NP; in clear ellipsis configurations involving stranding of IX, it is disallowed.

A ‘point’ of inquiry: The case of the (non-)pronominal IX in ASL

|

247

6 Conclusion Let us now take stock. Echoing the suggestions in McBurney (2002), who notes that “the class of signs traditionally referred to as personal pronouns may, in fact, be demonstratives” (McBurney 2002: 365), we have offered explicit evidence against the accounts of IX in ASL that analyze it as both a personal pronoun and a definite determiner, and have shown that the lexical item under examination is best viewed as a demonstrative. We have steered clear from engaging in the debate in the literature regarding the semantics of demonstratives; instead, we have focused on the empirical observations, having adopted an account of the facts that most clearly illustrates our point. A number of interesting conclusions have surfaced during our examination of IX. One is that on any account which equates a personal pronoun with an article, IX cannot be a personal pronoun. This is simply because IX cannot serve as a definite article. In fact, in all the cases requiring a definite article and not a demonstrative, IX is illicit (and no other lexical item will do) – driving us to the conclusion that ASL is a language without an overt definite article. Elsewhere (Koulidobrova 2012) we show that this is a plausible observation: on a number of unrelated diagnostics, ASL behaves on par with languages without overt definite articles. Of course, simply because the language does not have an overt morpheme for the definite article does not mean that it should have no overt personal pronoun – the claim in the literature equating the two is being made about the semantics of the lexical items, not necessarily their overt realization. Therefore, independent evidence was required to defend our point of view. In our examination of the ‘pronominal’ use of IX we discovered that it appears to parallel with German Dpronouns – the ones that tend to pick out the types of referents other demonstrative expressions do.¹¹ Furthermore, if our view of IX is on the right track, the data we presented challenge the approach to the covert nominal argument of demonstratives as a case of NP ellipsis. More examination is in order here. Admittedly, much remains to be worked out, and many issues remain undiscussed. For instance, we have not provided any explanation for why exactly previous introduction of loci changes the paradigm. The question also arises whether the analysis advocated here could incorporate other claims in the literature regarding pointing and pronominals in ASL. We have also avoided discussing

11 We make this claim cautiously. Schwarz (2009), for example, suggests that the analysis of the German articles as may be extended to that of the two types of pronouns – the demonstrative pronoun analyzable along the lines of the strong definite article. However, we have demonstrated, this view will not do for the ASL IX.

248 | Elena Koulidobrova and Diane Lillo-Martin

whether any relationship should be established between the third person IX (our a-IX) and its 1st - and (possibly) 2nd -person counterparts (1IX and 2IX respectively). This question is not trivial: previous research has argued that there is no grammatical 2nd person form (i.e., the relevant grammatical distinction is first vs. non-first; Meier 1990, Meier and Lillo-Martin 2013); and the 1st person form behaves logophorically in attitude reports (Lillo-Martin 1995, Schlenker 2014). Additionally, we would like to express caution against making our claim universal for sign languages. It is possible that cross-linguistic differences will be observed here, especially among languages unrelated to one another. Finally, the data reported here should be further compared with other languages (spoken and signed) from Table 3 – languages in which demonstratives exist but personal pronouns do not. More research is in order.

References Abner, Natasha and Graff, Thomas. 2012. Binding complexity and the status of pronouns in English and American Sign Language. Presented at the Formal and Experimental Approaches to the Sign language Theory (FEAST) Colloquium. Warsaw, April 2012. Ariel, Mira. 1990. Accessing Noun-Phrase Antecedents. Routledge. Barberà, Gemma and Martine Zwets. 2013. Pointing and reference in sign language and spoken Language: Anchoring vs. identifying. Sign Language Studies, 13: 4. 491–515 Bhat, D.N.S. 2013. Third person pronouns and demonstratives. In Matthew S. Dryer & Martin Haspelmath (eds.) The World Atlas of Language Structures Online. Leipzig: Max Planck Institute for Evolutionary Anthropology. (Available online at http://wals.info/chapter/43, Accessedon2014-11-05.) Bosch, Peter. 2013. Anaphoric reference by demonstrative pronouns in German. Presented at the Workshop on the Impact of Pronominal Form on Interpretation, Universität Tübingen, Nov. 2013 Bosch, Peter, Graham Katz, and Carla Umbach. 2007. The Non-subject Bias of German Demonstrative Pronouns. In M. Schwarz-Friesel, M. Consten & M. Knees (eds.) Anaphors in Text: Cognitive, Formal and Applied Approaches to Anaphoric Reference. 145–164 Chierchia, Gennaro. 1998. Reference to kinds across languages. Natural Language Semantics 2: 339–405. Davidson, Kathryn. 2013. ‘And’ or ‘Or’: General Use Coordination in ASL. Semantics & Pragmatics, 6(4), 1–44. Davidson, Kathryn and Deanna Gagne. 2014. Vertical representation of quantificational domains. Presented at Sinn und Bedeutung, Basque Country, Spain, Sep 2013. Dayal, Veneeta. 2004. Number marking and (in)definiteness in kind terms. Linguistics and Philosophy 27 (3): 393–450. Elbourne, Paul. 2005. Situations and Individuals. Cambridge, Mass.: MIT Press. Elbourne, Paul. 2008. Demonstratives as individual concepts. Linguistics and Philosophy, 31(4): 409–466

A ‘point’ of inquiry: The case of the (non-)pronominal IX in ASL

|

249

Frege, Gottlob. 1960 [1893]. On Sense and reference. Reprinted in P. Geach and M. Black, eds., Translations from the Philosophical Writings of Gottlob Frege. Oxford: Blackwell. Hankamer, Jorge and Ivan Sag. 1976. Deep and surface anaphora. Linguistic Inquiry 7: 391– 428. Heim, Irene. 1982. The Semantics of Definite and Indefinite Noun Phrases. Doctoral dissertation, University of Massachusetts, Amherst. Heim, Irene. 1985. Direct reference explained away. Ms. University of Texas at Austin. Heim, Irena and Kratzer, Angelika. 1998. Semantics in Generative Grammar. Oxford: Blackwell. Hinterwimmer, Stefan. 2014. A unified account of the properties of German demonstrative pronouns. In P. Grosz, I. Heim, P. Patel & I. Yanovich (eds.). Pronoun Workshop at NELS 40, GLSA Publications, University of Massachusetts, Amherst. Johnson, Kyle. 2009. Ellipsis. Presented at Hogeschool-Universiteit, Brussels, April 2009 Johnston, Trevor. 2013. Formational and functional characteristics of pointing signs in the corpus of Auslan (Australian Sign Language): Are the data sufficient to posit a grammatical class of ‘pronouns’ in Auslan? Corpus Linguistics and Linguistic Theory 9 (1): 109–159 Kaplan, David. 1989a. Demonstratives. In J. Almog, J. Perry, and H. Wettstein, (eds.), Themes from Kaplan, 481–563. Oxford: Oxford University Press. King, Jeffrey. 2001. Complex Demonstratives. A Quantificational Account. Cambridge, MA: MIT Press. Kozintseva, Natalia A. 1995. Modern Eastern Armenian. Languages of the World/Materials 22. München: Lincom Europa. Koulidobrova, Elena. 2012. When the Quiet Surfaces: ‘Transfer’ of Argument Omission in the Speech of ASL-English Bilinguals. Doctoral dissertation. University of Connecticut. Koulidobrova, Elena. submitted. Elide me bare: Null arguments in American Sign Language. Kuhn, Jeremy. 2014. ASL loci: Variables or features? http://semanticsarchive.net/Archive/ jMzZGJhY/kuhn-ASL-loci.pdf Liddell, Scott. (2003). Grammar, Gesture, and Meaning in American Sign Language. Cambridge: Cambridge University Press. Lillo-Martin, Diane and Deborah Chen Picher. 2008. Development of sign language acquisition corpora. In O. Crasborn, E. Efthimiou, T. Hanke, E.. Thoutenhoofd, & I. Zwitserlood, (eds.), Proceedings of the 3rd Workshop on the Representation and Processing of Sign Languages: Construction and Exploitation of Sign Language Corpora; 6th Language Resources and Evaluation Conference, 129–133. http://www.lrec-conf.org/proceedings/lrec2008/ workshops/W25_Proceedings.pdf Lillo-Martin, Diane and Edward Klima. 1990. Pointing out difference: ASL pronouns in syntactic theory. In S. Fischer & P. Siple (eds.) Theoretical Issues in Sign Language Research 1: Linguistics, 191–210. Chicago: University of Chicago Press. Lillo-Martin, Diane and Richard Meier. 2011. On the linguistic status of ‘agreement’ in sign languages. Theoretical Linguistics 37 (3): 91–141. MacLaughlin, Dawn. 1997. The structure of determiner phrases: Evidence from American Sign Language. Doctoral dissertation, Boston University. McBurney, Susan Lloyd. 2002. Pronominal reference in sign and spoken language: are grammatical categories modality-dependent? In R. Meier, K. Cormier and D. Quinto-Pozos (eds.) Modality and Structure in Sign and Spoken Languages, 329–369. Cambridge: Cambridge University Press.

250 | Elena Koulidobrova and Diane Lillo-Martin

Meier, Richard. 1990. Person deixis in American Sign Language. In S. Fischer & P. Siple (eds.) Theoretical Issues in Sign Language Research 1: Linguistics, 175–190. Chicago, IL: University of Chicago Press. Meier, Richard and Lillo-Martin, Diane. 2013. The points of language. Humanamente: Journal of Philosophical Studies 24, 151–176. Nouwen, Rick. 2003. Plural Pronominal Anaphora in Context. Doctoral dissertation. Utrecht Institute for Linguistics. Nowak, Ethan. 2013. Things a semantics for demonstratives should do. Ms. University of California, Berkley. Postal, Paul. 1966. On so-called ‘pronouns’ in English. In F. Dinneen (ed.), Report on the Seventeenth Annual Round Table Meeting on Linguistics and Language Studies, 177–206. Washington, D.C.: Georgetown University Press. Recanati, Francois. 1993. Direct Reference. From Language to Thought. Oxford: Blackwell. Roberts, Craige. 2003. Uniqueness in definite Noun Phrases. Linguistics and Philosophy 26: 287–350. Reinhart, Tanya. 1981. Pragmatics and Linguistics: An Analysis of Sentence Topics. Philosophica 27: 53–93. Russell, Bertrand. 1905. On denoting.Mind 14: 479–493. Saito, Mamoru. 2007. Notes on East Asian argument ellipsis. Language Research 43(2): 203– 227. Saltarelli, Mario, Azkarate, Miren, Farwell, David, de Urbina, Jon Ortiz and Lourdes Oñederra. 1988. Basque. Croom Helm Descriptive Grammars. London: Croom Helm. Sandler, Wendy and Diane Lillo-Martin. 2006. Sign Language and Linguistic Universals. Cambridge, UK: Cambridge University Press. Schlenker, Philippe. 2003. A plea for monsters. Linguistics and Philosophy 26(1), 29–120 Schlenker, Philippe. 2009. Donkey anaphora in Sign Language II: The presuppositions of pronouns. In P. Grosz, I. Heim, P. Patel & I. Yanovich (eds.). Pronoun Workshop at NELS 40, GLSA Publications, University of Massachusetts, Amherst. Schlenker, Philippe. 2011. Quantifiers and variables: Insights from Sign Language (ASL and LSF). In B. Partee, M. Glanzberg and J. Skilters (eds.), Formal Semantics and Pragmatics: Discourse, Context, and Models. The Baltic International Yearbook of Cognition, Logic and Communication 6, 2011. Schlenker, Philippe, Lamberton, John & Santoro, Mirko. in press. Iconic Variables. Linguistics & Philosophy. Schlenker, Philippe. 2014. SuperMonsters I–II: Attude and action role-shift in Sign Language. http://ling.auf.net/lingbuzz/002069 Schwarz, Florian. 2009. Two Types of Definites in Natural Language. Doctoral dissertation. University of Massachusetts, Amherst. Selkirk, Elisabeth. (2005). Comments on intonational phrasing in English. In S. Frota, M.C. Vigário, & M.J. Freitas (eds.), Prosodies, 11–58. Walter de Gruyter Wiltschko, Martina. 1998. The syntax of pronouns and determiners. A cross-linguistic study. In M. Caldecott, S. Gessner, E.-S. Kim (eds.) Current Research on Language and Linguistics: University of British Columbia Working Papers in Linguistics 1: 293–320. Wolter, Lynsely. 2006. That’s that: The Semantics and Pragmatics of Demonstrative Noun Phrases. Doctoral dissertation. University of California, Santa Cruz. Wulf, Alyssa, Paul Dudis, Robert Bayley and Ceil Lucas. 2002. Variable subject presence in ASL narratives. Sign Language Studies 3: 154–7.

Maria Averintseva-Klisch (University of Tübingen)

Definite or still demonstrative? Some ideas on the semantics of the German distal demonstrative jenAbstract: In my paper I offer a unified two-level semantics for the distal demonstrative jen- in modern written German using corpus data with pronominal and adnominal jen- in anaphoric, anamnestic and determinative uses. I propose that jen- combines an at-issue definiteness component – in the sense of denoting a salient referent in the current discourse representation – with a not-at-issue component conveying that the referent is presented by the speaker as not being anchored in the shared mental space of the speaker and the hearer. In contrast to adnominal jen-, this not-at-issue meaning seems to be optional with pronominal jen-, presumably leading to its nascent grammaticalization as a determinativity marker. Keywords: definiteness; deixis; demonstrative; determinative; German; mental distance; at-issue vs. not-at-issue meaning.

1 Introduction This paper focuses on the German demonstrative jen- (‘that (one)’), which is generally considered to be either the distal counterpart to dies- (‘this (one)’), or to have been grammaticalized as a definiteness marker restricted to written upper register German. I challenge both views using data from the IDS DeReKo corpus (cf. Kupiez et al. 2010) and propose an analysis of jen- in terms of multidimensional meaning: jen- has an at-issue definiteness component and a not-at-issue component of ‘mental distance’. To that end, I first briefly review some analyses of demonstrative uses in general, as well as two recent investigations on jen- specifically (Section 2). In Section 3, some claims made in the literature on jen- are checked against corpus data and then modified in accordance with the findings. This investigation reveals some important differences between pronominal and adnominal jen-. In Section 4, I argue that the behaviour of jen- is adequately accounted for by the assumption that its meaning comprises both at-issue and not-at-issue components (in the sense of Simons et al. 2010). Whereas the at-issue meaning contribution is presumably to convey definiteness, the not-at-issue part places the referent of the

252 | Maria Averintseva-Klisch

jen-NP outside of the shared mental space of the speaker and the hearer. I show how this analysis accounts for the data in Section 3 including the observed difference between pronominal and adnominal jen-. Section 5 summarizes the results.

2 German demonstratives In this section, I first give a brief overview of the demonstratives in German (Section 2.1). Then I sum up common assumptions about their uses and properties from the literature, and review two recent investigations on jen- in particular (Section 2.2). Some questions that stay open or arise after this review will serve as the starting point for the explorative corpus study I present in Section 3.

2.1 Demonstrative paradigms German has two sets of demonstrative pronouns for person and object reference, cf. (1): (1)

a. simple/weak demonstratives: der/die/das (‘this one’) b. compound/strong demonstratives: (i) proximal: dieser/diese/dies; dieser/-e/-es N (‘this; this one’) (ii) distal: jener/jene/jenes; jener/-e/-es N (‘that; that one’)

These two sets are usually referred to as ‘simple’ vs. ‘compound’ demonstratives, these terms alluding to the origin of dies- forms as composed forms of the elder simplex der-forms with the deictic particle si. Alternatively they are referred to as ‘weak’ (der/die/das) vs. ‘strong’ (dies- and jen-) demonstratives. These terms go back to Brugmann (1904), and are meant to indicate the grade of the demonstrative force.¹ However, for der/die/das it is not quite clear whether these forms are really demonstrative by themselves: in adnominal use, der/die/das has lexicalized as the definite article, cf. (2a):

1 This terminological use is restricted to German demonstratives, and should not to be confused with Cardinaletti and Starke’s (1999) terms ‘weak’ and ‘strong’ to differentiate cross-linguistically between phonologically, morphologically and syntactically different pronoun classes.

Definite or still demonstrative? | 253

(2) a. Der Arzt untersucht einen Patienten. ‘The doctor is examining a patient.’ b. Der Arzt hat mich nicht verstanden. ‘This.stress doctor hasn’t understood me.’ (Erben 1980: 225) Whilst in (2a) der is a definite article (potentially contrasting with the indefinite ein (‘a’)) marking that the doctor is uniquely identifiable in the current discourse, the stressed article in (2b) clearly has a demonstrative flavour. The stressed forms, both pronominal and adnominal, are often assumed to be more or less equivalent to the strong dies- forms (e.g. Himmelmann 1997: 50; Erben 1980: 225). However, Gunkel (2006) argues that the demonstrative flavour with der/die/das arises solely as an effect of focusing via stress. Rauh (2003: 410) shows that also personal pronouns can be demonstrative and thus comparable with genuine demonstratives if they are stressed, and argues for distinguishing between lexical item and emphatic stress – the latter being “a specific kind of a pointing gesture” responsible for bringing in demonstrative flavour. Irrespective of the lexicalization issue in the case of the adnominal use, an absense of inherent demonstrativity can be observed for pronominal der/die/das: its main function when used without emphatic stress seems to be to serve as an anaphor: der/die/das is used (i) to pick out non-topical referents, while personal pronouns prefer reference to discourse topics² (see e.g. Bosch and Umbach 2007), or (ii) to topicalize syntactically via placing a pronoun with a non-human referent in non-subject function before the finite verb: as unstressed personal pronouns in German are weak in terms of Cardinaletti and Starke (1999), they do not allow prefinite fronting in this case. So, the status of der/die/das requires further investigation. On the contrary, strong demonstratives are uniformly analysed as being truly demonstrative, at least as far as dies- is concerned (cf. Section 2.3). They are traditionally described as ‘proximal’, i.e. referring to a locally or textually near referent (dies-) vs. ‘distal’, i.e. referring to a locally or textually more remote referent (jen-), cf. e.g. Duden (2009; 2011), Zifonun, Hoffmann, and Stecker (1997), as in (3):

2 The concept of discourse topicality is by no means uncontroversial; the literature specializing in demonstratives agrees upon a rather informal understanding of discourse topic in the sense of a discourse-old referent that a current discourse segment “is about”, cf. Bosch and Umbach (2007: 50).

254 | Maria Averintseva-Klisch

(3) a. Die Aussicht von dieser Bank ist schöner als von jener. (Duden 2009: 286) ‘The view from this bench is more beautiful than from that.one.’ b. Mutter1 und Tochter2 kamen näher, diese2 trug ein Sommerkleid, jene1 ein Kostüm. (Duden 2011: 240)³ ‘The mother1 and her daughter2 approached; the latter2 (dies-) was wearing a summer dress, the former1 (jen-) a suit.’ However, it is controversial whether both demonstratives are still used in this fashion with respect to proximity; e.g. Hauenschild (1982: 183) states that “jener practically never occurs in modern spoken or written language”. Also Levinson (2004: 110) assumes that German has only one demonstrative dies-. In the following, I first briefly list the functions that can be identified for German demonstratives, primarily for the proximal strong demonstrative dies-, and then concentrate on jen- and investigate its uses, sometimes comparing jen- to dies-. My goal is first to describe the meaning of jen- and second to try a unified account of strong demonstratives in German.

2.2 General uses of the demonstrative in German Demonstratives are generally used to re-direct the joint attention of the speaker and the hearer by attracting it to a particular referent that has not been in the centre of joint attention so far (e.g. Diessel 2012: 2417). Himmelmann (1996, 1997) and Levinson (2004) descriptively distinguish eight different uses of demonstratives that result out of this primary function; in the following these are briefly presented:⁴ 1. Deixis proper, i.e. referring to a physically perceptible referent in the actual utterance situation accompanied with a pointing gesture or something similar which helps the hearer to resolve the reference unambiguously, cf. (4): (4) I want this cake, please. (pointing to a cake at the baker’s).

3 Here and in the following examples, the numbers indicating coreference are mine. 4 Both Himmelmann and Levinson present cross-linguistically valid demonstrative systems; out of these, I chose only uses that are possible with dies- in German. Only some of the latter are, in turn, available for jen- (see below). For the sake of convenience, however, I use English examples in this section.

Definite or still demonstrative? |

255

Deixis proper is taken to be the primary use of demonstratives (e.g. Diessel 1999: 8), other uses are taken to have developed from this one. 2. Textual deixis in the sense of referring to segments of the text containing the demonstrative: (5) Our aim in this chapter is to summarize briefly the functions of the demonstrative. Here, the referring expression is part of the textual unit (the chapter) it refers to; since the reference object is necessarily physically present in the utterance situation (a reader can only read this sentence in case he has the chapter referred to before his eyes), this is seen as an instance of deixis. 3. Deixis ‘am Phantasma’ in the sense of Bühler (1934); here the referent is not present in the actual discourse situation, but in the situation that is the subject of the ongoing discourse, cf. (6): (6) And he’s heading … you see a scene where he’s … coming on his bicycle this way. (Pear stories, quoted from Himmelmann 1996: 222) In (6), the speaker is treating the narrated situation as if it were taking place directly before his and the hearer’s eyes, using the demonstrative in an as-if-deictic way. It is plausible that he even accompanies his words with a gesture, thus equating the direction he points to in the actual utterance situation with the direction the bicycle takes in the narrated situation. Under this label, Himmelmann also subsumes cases where the speaker introduces a new referent in a narrative with a demonstrative, thus acting as if the corresponding referent were actually present, cf. (7): (7) One day last year on a cold, clear, crisp afternoon, I saw this huge sheet of ice in the street. (Prince 1981: 232) Here the referent of the this-NP is neither prementioned, nor present in the discourse situation, nor uniquely identifiable for the hearer. Accordingly, in cases like this, this is interchangeable with the indefinite determiner a. Besides, such this-NPs are grammatical in the presentational there-is-construction as in (8), which excludes definite NPs:

256 | Maria Averintseva-Klisch

(8) There is this/a/*the man upstairs who is driving me mad because he jumps rope at 2 a.m. every night. (Maclaran 1980: 813) Such cases have been analysed as ‘indefinite this’ (Ionin 2006, Maclaran 1980, Prince 1981, von Heusinger et al. 2010, Ward and Birner 1995). Von Heusinger et al. (2010) argue that German also has indefinite dies- functioning in an analogous manner. 4. Anamnestic or recognitional use in the sense of referring to referents not yet mentioned in the current discourse, but known to both the speaker and the hearer from some previous communication situations, as in (9); cf. Auer (1981, 1984): (9) By the way, where is this old-fashioned telephone you used to have? The anamnestic use differs from the indefinite use as in (7) and (8) in that the referent is uniquely identifiable for both the speaker and the hearer. In choosing a demonstrative for his referring act, the speaker appeals to the hearer to re-activate his knowledge of the referent. In (9), the hearer is expected to be able to resolve the reference, as he surely knows which particular telephone he owned. In this sense anamnestic reference is definite; accordingly, anamnestic this cannot be substituted with an indefinite. 5. Discourse deixis in the sense of reference to propositions, situations or speech acts that can be reconstructed from the previous discourse, cf. (10): (10) I forgot my keys and had to wait for my parents to come home. a. This was terrible, as it was raining. b. This is a strange way of apologizing, I know, but … In (10a), the demonstrative this refers to a “propositionally structured” (Consten et al. 2007: 82) referent, either the whole content of the previous sentence subsuming two propositions (I forgot the keys; I had to wait) or the second proposition alone (the waiting was terrible). In (10b), this takes up the speech act of uttering the previous sentence. 6. Anaphoric use, i.e. referring to a referent that was already textually introduced, cf. (11):

Definite or still demonstrative?

| 257

(11) I had a wonderful friend1 , I still believe I do. I made a mistake, I lied […] but I couldn’t go on forever, so I came clean, and told this friend1 the truth (http://forums.psychcentral.com) The fifth and sixth uses are often both subsumed under the term anaphora, the decisive aspect being that there is a linguistically encoded antecedent in the text before. In (10) similar to (11), the corresponding referent (a proposition, respectively a speech act) is introduced in the previous discourse by linguistic means. Consten et al. (2007) use the term ‘complex anaphor’ to enhance the fact that in these cases the anaphoric expression does not simply re-address an already introduced discourse referent, but triggers a process of constructing a discourse referent out of a complex linguistic and referential unit. 7. Empathetic or emotive use, where demonstratives serve to mark an emotional attitude (positive or negative) towards the referent, at the same time achieving an effect of “emotional closeness between speaker and adressee” (Lakoff 1974: 351), cf. (12): (12) a. Who is this William Young and where has he been? This wonderful work […] b. You don’t deserve to be discouraged and lied to by a con artist. Which is what this Arthur Agatston is. (attested examples from online reviews cited in Potts and Schwarz (2010: 5)) c. How’s that throat? (Lakoff 1974: 351) Here, the speaker explicitly expresses his strong positive (12a) or negative (12b) emotion towards the respective referent of the proper name, which gets enhanced through the (morphosyntactically unnecessary) adnominal demonstrative. In addition, in using the demonstrative the speaker suggests that the hearer might share his knowledge of the referent and his emotion, thus establishing an emotional relation between himself and the hearer; see also Potts and Schwarz (2010), Acton and Potts (2014). For the distal demonstrative that, Lakoff (1974: 351) argues that “while its spatio-temporal uses are very nearly opposite those of this, its emotive uses are surprisingly close”. She gives example (12c) above to illustrate the point: this utterance might be used by a sympathetic nurse towards a patient, while the possessive your throat would be emotionally neutral here.

258 | Maria Averintseva-Klisch

8. For distal demonstratives also the determinative use is described; here the demonstrative functions as an antecedent for a restrictive relative clause as in (13): (13) United Airlines contacting those who flew with Ebola victim (http://news.yahoo.com, 2.10.2014) In these cases the referent has not been introduced before, but the relative clause is intended to supply the information necessary for the unambigous identification. The distal demonstrative does not have any pointing meaning aspect nor that of distance; those could be replaced with the people who without any change in meaning. Himmelmann (1997: 78) talks of “the purely placeholder function” of the demonstrative and assumes that those has been grammaticalized as a definiteness marker in such cases. Recent investigations of German jen- (Gunkel 2007a, 2007b, Molnár 2010) argue that jen- is not used in contemporary German to denote physical distance in contrast to dies-. Gunkel (2007b: 8) proposes that jen- can convey “metaphorical (subjective) distance”, when jen- refers to “far-off objects in the discourse universe”. However, no precise concept of ‘metaphorical distance’ is given. In general, Gunkel (2007a: 223) and Molnár (2010) distinguish two possible uses of jen- in the present-day German, the anamnestic use as in (14) and the determinative like in (15): (14) Es war einer jener bürgerlichen Sonntagnachmittage, an denen einem die Langeweile durch alle Knopflöcher kriecht. (Gunkel 2007a: 223, ex. (30)) ‘It was one of those middle class sunday afternoons when the boredom keeps kreeping through one’s button holes.’ For (14), Gunkel argues that the “meaning effect” of jen- is to “refer to something that it presumed to be known [to the hearer]” (Gunkel 2007a: 223): the speaker in (14) takes it for granted that everybody knows the boring middle-class sunday afternoons from one’s own experience. As actually the case in (14), anamnestic jen-NPs have a strong tendency to serve as antecedents for (restrictive or appositive) relative clauses. As mentioned above, also proximal dies- has anamnestic use. Molnár (2010) notices that in their anamnestic uses both demonstratives are similar. Hence in (14), jen- could be substituted with dies- without any change in meaning. The only difference between the two anamnestic demonstratives is their frequency: due to jen- being in general seldom used and being “practically never used in oral speech” (Molnár 2010: 329), its anamnestic use is also much more rare than the anamnestic use of dies-.

Definite or still demonstrative?

| 259

(15) Für jene Urlauber, die Sri Lanka einst als Fernreiseziel entdeckten, ist in den Plänen kein Platz mehr. (Gunkel 2007a: 223, ex. (20b)) ‘For those holidaymakers who once discovered Sri Lanka as a holiday destination there is no place in the new plans.’ In (15), the restrictive relative clause is necessary since it is the only source of identification for the referent of the jen-NP. Without the relative clause it is not clear which holidaymakers are not considered in the new plans. In its determinative use, jen- is equivalent with derjenig-,⁵ which also requires the presence of a restrictive relative clause, in that both serve as definiteness markers, and can be substituted with the definite determiner der/die/das. However, unlike the prototypical “determinativity marker” derjenig-, jen-NPs allegedly cannot have a non-specific reading, so that (16) can only refer to a specific runner, and not mean “whoever might be characterized as having achieved the goal first”: (16) Jener Läufer, der das Ziel erreicht hat, hat gewonnen. ‘That runner who reached the goal has won the race.’ Gunkel (2007a: 221) argues that jen- in plural is already grammaticalized as a pure “determinativity marker” both in its pronominal and adnominal occurrences, whilst the grammaticalization of the singular form has not reached this stage as yet. In sum, there is a consensus in the relevant literature that jen- is not used contrastively to dies- in modern German. In particular, it is argued that jen- is not used deictically at all. It is on the way of being grammaticalized as a definiteness marker in determinative contexts, i.e. in contexts where restrictive modifiers, preferably restrictive relative clauses, ensure the non-ambiguous identification of the referent. The preferred use of jen- is taken to be the anamnestic one, in which it is usually (or even necessarily) accompanied by a relative clause or at least some attributive expression. The anamnestic use tends to additionally convey ‘metaphorical distance’, i.e. some kind of non-physical, “subjective” (Gunkel 2007b) self-distancing of the speaker from the referent. Turning back to demonstratives in general, two crucial uncontroversial assumptions are more or less explicitly made in the literature. First, cross-linguisti-

5 From an etymological point of view, the German compound form derjenig- consists of the weak demonstrative der + an adjectival derivation from jen-. It has pronominal and determiner forms. It cannot be used demonstratively. Prototypically, it serves as an antecedent for a restrictive relative clause.

260 | Maria Averintseva-Klisch

cally all descriptively distinguishable uses are grounded in the main function of demonstration, i.e. of the speaker’s directing his own and the hearer’s attention to a referent that has not been in the centre of attention before. Second, if there is a proximal-distal contrast like with this/that, then it is relevant only with the deictic and anaphoric uses, especially in cases where there are two referents to be distinguished, one being locally or textually nearer to the speaker than the other. In the anamnestic and emotive uses, the proximal and distal demonstratives seem to function identically (Lakoff 1974; Himmelmann 1997: 73), and the determinative use is only observed with distal demonstratives (Himmelmann 1997: 77). Generalizing over the uses outlined above, we can reduce the eight descriptively distinguished uses to five ones that are differentiated with respect to three parameters: (i) demonstrativity in the sense of attracting joint attention via pointing introduced above: ‘demonstrative proper’ uses 1–7 vs. the determinative use lacking this parameter; (ii) immediate presence of the object of reference: direct reference to objects in the utterance situation (deixis, uses 1–3) vs. reference to textually introduced discourse referents (anaphora, uses 5 and 6) vs. reference to objects in the shared broader knowledge of the speaker and the hearer (anamnestic use); (iii) emotive meaning component of the demonstrative: given (emotive use) vs. not given (others); for this distinction it is important that all ‘demonstrative proper’ cases can have an additional emotive component (cf. Acton and Potts 2014: 5). In Section 3, I report an explorative corpus study which investigates whether the two assumptions above are valid for German distal demonstrative jen- (as compared to proximal dies-), and whether jen- actually disposes of all the five uses (deictic, anaphoric, anamnestic, emotive and determinative). To give a brief glimpse into the results of the study, the data show that it is indeed problematic to talk of demonstrating with jen-; I found no examples of the deictic use in my data. Similarly, whilst dies- clearly has an emotive use (e.g. empirical data in Consten and Averintseva-Klisch (2010) and Potts and Schwarz (2010: 24)), I found no cases of an emotive use with jen-. As for uses that are possible in principle, there is an important difference between pronominal and adnominal occurrences. The former are only used anaphorically or determinatively in my data whereas the latter also show an anamnestic use in addition to the anaphoric and the determinative uses. In Section 4, I present my analysis that is based on this findings and aims to explain and derive them from the core meaning of jen- in a principled way.

Definite or still demonstrative? | 261

3 Corpus study⁶ As a starting point for my proposal of the semantics of jen-, I conducted a small explorative pre-study of 220 instances of jen-. Being interested in the actual usage of jen- nowadays, I took recently produced texts to investigate it, in spite of the claim that jen- is allegedly restricted to written German, presumably even to its lofty or old-fashioned varieties (e.g. Gunkel 2007a). I took approximately 50 % of the cases from the newspaper Mannheimer Morgen (2005; 2011; 2012; = MM) and approximately 50 % from Wikipedia discussions (2011; = WikiD) to see if there are any register differences concerning the use of jen- here. This was not the case. The study shows the following general results: (i) I found no case of a deictic use. One might expect this result for written discourse. However, assuming a broad understanding of deixis comprising textual deixis and deixis ‘am Phantasma’ as discussed above, one could expect to find instances of the deictic use also in written texts. In comparison, out of 30 examples of dies- (randomly chosen from Mannheimer Morgen 2005), 4 were clearly deictic, like vom Dienstag bis Donnerstag dieser Woche (‘from Tuesday till Thursday this week’) referring to the week in which the text appeared in the newspaper, and one further example allowed both for a deictic and an anaphoric reading. (ii) In only one case out of 220 items, jener is used anaphorically in contrast with dieser, cf. (17): (17) Was von dir als “offensichtliche Banalitäten” heruntergespielt wird, sind übrigens gravierende Mängel im Artikel, die dem unbedarften Leser1 so wie auch dem interessierten Autoren2 durchaus mitgeteilt werden sollten – jenem, damit er weiß, daß der Artikel schlecht ist, und diesem, damit er den Artikel evtl. erstmal über (sic!) findet und ihn dann überarbeitet oder sich auf der Disku dazu umguckt. (WikiD_2011) ‘What you are playing down as “apparent banalities” are, by the way, serious shortcomings of the article, which should in any case be disclosed to the undiscerning reader1 , as well as to the interested author2 – to the former? [= jen-], so that he knows that the article is not good, to the latter? [= dies-], so that he can, first of all find over (sic!) the article, and revise it, or browse around on the discu[ssion] site.’

6 I am very grateful to my student assistants Ben Kings and Pirmin Scharer for their help with the data.

262 | Maria Averintseva-Klisch

Whether the anaphoric reference with jen- in (17) is in accord with the prescriptive grammars, such as Duden, or not, is not quite clear: given that Wiktionary per definition does not clearly distinguish between authors and readers in that everybody can contribute as an author, (17) can plausibly either mean (in accordance with the grammars) that the reader should know that the article is bad, and the author should have the possibility of revising the article and conducting further research – or that the author may learn that his article is bad, and the reader may contribute as an author or at least read up on the subject. Especially the last part “or browse around on the discu[ssion] site” renders the second reading plausible as well. (iii) Out of 220 exported examples, 39 are pronominal and 181 are adnominal occurrences. This finding allows us to make some first non-quantitative statements about the possible distribution of the two types of occurrences in the data. Grammis, the internet grammar of the IDS Mannheim, differentiates between jenas a pronoun, functioning “textdeictically and referring in the perception field (‘Wahrnehmungsfeld’)” and adnominal jen-, the latter having only the “determinativity marker” function, whereas the NP itself provides the information necessary for the identification of the referent.⁷ The data, as I show in the following, suggests the opposite picture, though.

3.1 Pronominal occurrences The pronominal occurrences I found in the data can be divided into two groups: in the largest group in my data (29 of 39 cases), jen- has a determinative use; it serves as a definiteness marker, and licenses a restrictive relative clause, cf. (18) and (19): (18) Die Schreibweise […] soll ja nur ein Anhaltspunkt für all jene sein, die keine wissenschaftlich korrekte Notation, sondern einfach eine Idee davon haben wollen, was es mit dem Berlinischen auf sich hat. (WikiD_2011) ‘This spelling should be a clue for all jen.ones who do not want a scientifically correct notation, but just an idea of what the Berlin dialect is like.’

7 The grammar was accessed in November 2013.

Definite or still demonstrative? |

263

(19) Du hast erkannt, dass sich die Mehrheit wohl für das Lager interessiert. […] Jene, die den Lagerteil Birkenau suchen, finden den Link im Artikel KZ Auschwitz. (WikiD_2011) ‘You’ve realized that the majority seems to be interested in the concentration camp. Jen.ones who are looking for its part in Birkenau will find the link inside the article on the KZ Auschwitz.’ In the examples above, the pronoun seems to function purely as a determinativity marker with a restrictive relative clause. Example (18) has – at least preferably – a non-specific reading, and (19) can only be read non-specifically. Both of these examples oppose Grammis in that pronominal jen- can obviously be used determinatively, as well as Gunkel (2007a), who claims that jen-NPs must have specific readings. As for Gunkel’s (2007) generalization that plural jen- has grammaticalized as a “determinativity marker”, and thus allows non-specific reference in plural, but not in singular: in my data (39 items), there were 24 cases of singular pronominal occurrences, 18 of them in determinative use (75 %), and 15 with plural pronouns, 11 of them in determinative use (73,3 %). Of course these frequencies cannot be used as the basis for statistic generalizations, but what can be observed is a certain tendency of pronominal jen- to occur in its determinative use irrespective of morphological number and semantic specificity.⁸ The data also showed that restrictive relative clauses are not necessary for the determinative reading; the information necessary for the identification of the referent may also be given by other kinds of restrictive modifier, as e.g. in (20): (20) a. […] der Untergang Assads wäre auch jener der Offiziere (MM_2011) ‘Assad’s downfall would also be jen.one of the officers’

8 I haven’t found any examples of singular pronominal occurrences with a non-specific reading, but several that are used in definitions like (i): (i)

Die Physik ist unter Naturwissenschaften jene, deren Begriffe am klarsten definiert sind. (WikiD_2011) ‘Among the sciences, physics is jen.one whose terms are most clearly defined.’

Here, the NP that includes the restrictive relative clause forms a part of the predicate, and is thus not referential at all, making it difficult to talk about specific or non-specific reference.

264 | Maria Averintseva-Klisch

b. […] dass die relative Standardabweichung für F halb so klein wird wie jene von F 90 (WikiD_2011) ‘that the relative standard deviation for F is half as big as jen.one for F 90’ In (20a), a genitive NP, and in (20b), a PP serve as restrictive modifiers, and render the determinative use possible. In both cases, a substitution of jen- with the definite article der/die/das or with derjenig- is valid. The condition on the determinative use is thus not a syntactic, but a semantic one: the pronoun has to be accompanied by some kind of restrictive modifier that constraints the denotation enought to ensure the identification of the corresponding referent. The second group of examples for the pronominal occurrences, 10 of 39 items, are anaphoric uses like in (21) and (22):⁹ (21) Im Gebäude, vor dem Raum 111 traf Steinhäuser1 auf den Lehrer Rainer Heise2 , der gerade diesen Raum verlassen wollte. Jener1 hatte seine Gesichtsmaske bereits abgenommen, deshalb konnte Heise2 Robert Steinhäuser1 erkennen. (WikiD_2011) ‘In the building, in front of the room 111, Steinhäuser1 met the teacher Rainer Heise2 just as he2 was leaving this room. Jen.one1 has already pulled off his mask, therefore Heise2 could recognize Robert Steinhäuser1 .’ (22) DFB-Chef Braun1 hat da seinem Angestellten Vogts2 zu Recht die passenden Worte gesagt. Jener2 hat zwar vor dieser WM in Sachen Matthäus eindrucksvoll über den Schatten springen gelernt, ein ganz Großer ist er deshalb aber noch lange nicht. (MM_1998)¹⁰ ‘In this case, DFB chief Braun1 was right to say these suitable words to his employee Vogts2 . Jen.one2 has truly learned to swallow his pride in the case of Matthäus before this World Cup, he is not one of the greats, yet, just because of that.’ In these examples, there is a textually introduced referent in the sentence preceding jen-, the gunman Steinhäuser in (21) and the football coach Vogts in (22). This referent is picked up with the demonstrative in the following sentence. Interest-

9 To operationalize this criterion for my investigation, I took examples where an antecedent for the jen-NP is present in the last 5 sentences preceding the one featuring this NP to be anaphoric. 10 I owe this example, as well as an impulse to its analysis, to an anonymous reviewer.

Definite or still demonstrative?

| 265

ingly, whilst in (21) anaphoric reference to the referent of the suitable antecedent NP that is farther away is in accordance with the reference grammars (cf. 2.1), in (22), it is the referent of the linearly closer antecedent, Vogts, that is chosen by jen-, although as distal demonstrative, it actually should (and morphosyntactically could) choose the remoter one, Braun. Here, dies- could be substituted for jen- without (or nearly without; see Section 4.3 below) any difference. In those examples in which pronominal occurrences were used anaphorically, there was only one case in which just one referent was suitable as an antecedent. In 5 cases jen- choose the more remote out of the two suitable referents like in (21) above, and in 3 cases it picked up the nearer referent like in (22). Again, the observed frequencies do not allow for statistically supported generalizations, but they present a first tendency that fits with the assumption quoted above that there is no clear proximal/distal-contrast with dies- vs. jen- in their anaphoric uses any more. All anaphoric cases in my data were like the both discussed so far in that they were prototypical cases of anaphora with an NP antecedent. None of the 220 items could be classified as ‘complex anaphora’. The tendency shown by my data is that jen- as a pronoun strongly prefers the determinative use; even in its anaphoric use it seems to lose its distal component and is used irrespectively of the antecedent location, so that it often behaves analogously to definite der/die/das. In Section 4.2, I revisit this observation, and account for it with a two-dimensional semantic analysis of jen-.

3.2 Adnominal jenWith adnominal examples, the determinative reading where jen- is interchangeable with derjenig-, and thus presumably grammaticalized as a definite antecedent for a restrictive relative clause is also attested, cf. (23): (23) Die Teilnehmerkarten all jener Kerwe-Besucher, die insgesamt sieben Gaudispiele absolviert hatten, fanden sich am Abend in einer Lostrommel wieder, aus der Bürgermeister Roland Esche die Gewinner zog. (MM_2005) ‘The participant tickets of all jen country fair visitors who had completed seven jamboree games were put into a lottery wheel in the evening, and the mayor Roland Esche drew the winners.’ Here, the demonstrative NP clearly has a non-specific reference with the restrictive relative clause naming the condition on potential referents: whoever completes

266 | Maria Averintseva-Klisch

seven jamboree games takes part in the lottery. In this use – again, as with pronouns, contrary to Gunkel’s (2007a) observations – not only plural, but also singular NPs can be found: (24) Den allerhöchsten Liebes-Wert hingegen wird immer jener Mensch ernten, der, statt das Negative zu verhindern, das Positive vollbringt. (MM_2005) ‘Still, the highest amount of love will get jen person who, instead of preventing the negative, achieves the positive.’ I will return to such examples below, and propose a slightly different account for them. A considerable number of jen-NP uses are like in (25): (25) In seiner Geburtstadt Karlsruhe feierte man ihn […], doch in jener Stadt, in der er seine größte Erfindung auf den Weg brachte, nämlich in Mannheim, da wäre sein Jubeltag fast vergessen worden. (MM_2005) ‘In Karlsruhe, the city where he was born, he was celebrated; but in jen city, where he started with his greatest invention, that is in Mannheim, his jubilee would have nearly been forgotten.’ It is not so easy to describe the use of the demonstrative NP in (25): it is clearly non-anaphoric, as the corresponding referent, the city, has not been introduced before. Similar cases have been subsumed under the anamnestic use by Gunkel (2007a) and Molnar (2010). However, what clearly speaks against “the speaker acting as if the referent were known to the hearer”, as Molnar (2010: 328) describes the anamnestic use, is that the referent is explicitly named following the jener-NP: that is in Mannheim. One could argue that (25) is an example of a determinative reading; but in this case, the referent has to be unambigously identifiable via a restrictive modifier, as shown above. This is, as is again made clear by the city’s name following, not the case here. Moreover, the prototypical determinativity marker derjenig- cannot replace jen- in (25) whilst replacing jen- with a definite determiner would be possible. In Section 4, I propose an account that can cope with such difficult examples by assuming two meaning levels for jen-. The largest group of jen-NPs are those used anaphorically. As with pronominal jen-, reference to persons (26) or objects (27) is possible: (26) Ein privater Sammler aus der Region1 habe den Anstoß gegeben zu dieser ungewöhnlichen Schau […] Jener private Gönner1 , der nicht genannt werden

Definite or still demonstrative? | 267

möchte, habe andere Sammler davon überzeugen können, ihre Schätze […] ebenfalls zu zeigen. (MM_2001) ‘A private art collector1 from the region has initiated this unusual exhibition. Jen private patron1 , who does not wish to be named, was able to persuade other collectors to exhibit their treasures, too.’ (27) […] von “Bent Fabric”1 lese ich heute zum ersten Mal. […] zu erwähnen, dass der gute Mann u.a. auch mit jenem Künstlernamen1 auftritt. (Wiki-Disc_2011) ‘[…] I’ve read about “Bent Fabric”1 for the first time today. […] to mention that the guy also performs unter jen stage name1 .’ In both cases – as in the majority of my anaphoric data – there is no contrast with respect to distance: jen-NP resumes the only referent available. In these cases, jenis interchangeable with proximal dies- or definite der/die/das without any truthconditional changes. In Section 4, I propose that the choice of jen- implies certain non-truth-conditional meaning aspects. A rather prominent group of anaphoric jen-NPs are those referring to time spans situated in the past, like in (28): (28) Die Ruine der vor 104 Jahren1 erbauten prächtigen Mannheimer Synagoge wird abgetragen. Nichts ist geblieben von dem Glanz jener Zeit1 , als 6600 jüdische Mitbürger ihren Mittelpunkt in F2 hatten. (MM_2005) ‘The ruins of the grand synagogue of Mannheim, which was built 104 years ago1 , are being removed. Nothing has remained of the glamour of jen time1 , as 6600 fellow Jewish citizens were living in F2.’ Here, the referent of the NP jene Zeit (‘that time’) is fixed in the previous sentence as being some 100 years ago, the time when the synagogue was built and soon after; i.e., it lies far away in the past, and is said to be gone without a trace: nothing has remained. In my data there were 41 examples out of the 182 anaphoric ones in which jen-NPs referred anaphorically to a time span; all of these were clearly situated in the past. To sum up: Out of the five uses for demonstratives identified above (deictic, anaphoric, anamnestic, emotive, and determinative) I’ve found the anaphoric, the anamnestic, and the determinative use with full jen-NPs and the determinative and the anaphoric uses with the pronominal jen-. For adnominal jen-, two observations could be made: first, time span referring NPs, like jen-time/days/night etc.,

268 | Maria Averintseva-Klisch

seem to be used rather often. Second, in some cases it was not possible to decide whether the jen-NP was used anamnestically or determinatively, since the actual reference act in these cases showed traits of both uses. Neither with pronominal nor with adnominal jen- a deictic use in the broad sense (comprising textual deixis) could be found. Also no examples in which jen- has a clearly emotive function could be found, but see Section 4.1 below for a detailed definition of the emotivity with jen-.

4 jen- revisited: demonstrativity as a not-at-issue component of mental distance I assume that definite descriptions, personal and demonstrative pronouns, and demonstrative NPs are definite in the sense of von Heusinger (2013: 369), who proposes that a definite NP the X denotes the most salient X in the context i. This more or less corresponds to the concept of internal anchoring in Kamp (1990) or Farkas (2002), but restates this concept more decidedly in terms of salience and not uniqueness. Given this assumption, definite determiners like der in German require an anchoring to the current “context” or discourse representation. This anchoring can be achieved via physical presence of the referent (deixis) or mentioning the referent beforehand (anaphora). In these cases, definite reference is expected. If the referent is neither present in the discourse situation nor aforementioned, but his identifiability arises directly from the properties of the current discourse representation, the anchoring is also given. That is why definites allow for an indirect anaphoric use (bridging reference) like in (29a) and a larger-situation use like in (29b): in both cases the corresponding referent is not aforementioned, but is identifiable via principled bridging references (Irmer 2011). In (29a), the mentioning of a lottery allows the inference that there are some winners, in (29b) the general properties of the discourse situation allow the unique reference: (29) a. […] in einer Lostrommel, aus der Bürgermeister Roland Esche die Gewinner zog. (MM_2005) ‘in a lottery wheel from which the mayor Roland Esche drew the winners.’ b. Es ist unübersehbar: Das Weihnachtsfest gehört wieder zum Schnee von gestern. (MM_2011) ‘One cannot ignore it: the Christmas festivities are again old and forgotten.’

Definite or still demonstrative? | 269

As additional features, I propose that demonstratives bring in the concepts of speaker-anchoring (S-anchoring) and hearer-anchoring (H-anchoring).¹¹ This means that with a demonstrative, the corresponding referent has to be not only uniquely identifiable with respect to the current discourse representation, but also has to stand in a particular relation to the speaker (S) and the hearer (H). The specific relation needed for a particular demonstrative depends on the demonstrative in question. In particular, for German demonstratives dies- and jen-, I assume that the referent of the demonstrative has to be mentally near to both S and H in the proximal dies- case, and mentally distant from both S and H in the distal jen- case. To model the concept of mental distance and proximity I use the distinction between ‘speaker’s place’ (SP), ‘denotation space’ (D), and ‘reference space’ (R), proposed in Ehrich (1982) for local deixis.¹² In particular, Ehrich distinguishes between the space a speaker refers to by using a deictic expression (= D) and the space relative to which D is identified (= R). She argues that the two German distal local adverbs, da and dort (both roughly ‘there’), can be distinguished in the following way: for both, SP is not equal to nor included in D (hence both adverbs are distal), but for dort, R is necessarily distinct from D, whilst with da, R might overlap with both SP and D, cf. (30): (30) Ich wohne in Tübingen und will dort/da auch alt werden. I live in Tübingen and wish there also old become A speaker uttering I live in Tübingen and wish to grow old there would choose dort for ‘there’ only if he is not in Tübingen while uttering it; with da, he might be in Tübingen or not. Modifying Ehrich’s analysis, I propose first that not only the speaker, but also the hearer should be taken into account. Hence, I introduce the ‘hearer’s place’. Behind this proposal is the simple intuition that an act of showing necessarily comprises not only somebody who shows (S) and something that is shown (the referent of the demonstrative), but also somebody else to whom it is shown (H). Second, I understand H’s and S’s places in the more abstract sense of mental 11 This analysis builds on proposals like Elbourne (2008), who argues for an ‘index’, i.e. a reference value a as a parameter in demonstrative semantics. The meaning of demonstratives is then “distal/proximal (x,a,t,w)”, i.e. “x is distal/proximal with respect to a at the time t in the world w”, cf. Elbourne (2008: 432). 12 This distinction itself builds on a corresponding trichotomy in the temporal domain, utterance time, event time, and reference time, as originally proposed in Reichenbach (1947) and developed in Klein (1994).

270 | Maria Averintseva-Klisch

space, i.e. roughly as a notion that includes that part of the private world knowledge of the S and H respectively that is relevant for the current discourse. This change is the reason why in the following, I talk of “spaces”, not of places. In particular, I argue that for proximal dies-, the speaker’s space and the hearer’s space necessarily have to overlap, and this overlapping region has to include the reference space; the denotation space is not restricted. For distal jen-, the reference space (and often also the denotation space, although this is not necessarily required) lies outside of the shared region of the speaker’s space and the hearer’s space. Whether the speaker’s and the hearer’s space overlap or not, is not constrained; cf. the schematic representation in (31): jene Katze (‘that cat’)

(31) diese Katze (‘this cat’)

REF REF S

H

S

H

In prose: diese Katze (‘this cat’) denotes a cat that is (i) either in the shared physical space of S and H; this is the deictic use proper, and the reference space is given via non-linguistic means or linguistically like in this cat on the garage roof, or (ii) in the shared mental space, as with the deictic use ‘am Phantasma’ or the anamnestic use: this cat of my neighbour can be used when the neighbour is present in the shared mental space on H and S as a reference space.¹³ In this case, even if the cat referent itself is not present, the reference via a demonstrative is valid. Jene Katze (‘that cat’) denotes a cat that is not anchored in the shared physical or mental space of S and H. Importantly, the reference space is not in the mental space shared by S and H whilst it is left open whether the reference space and the denotation space overlap, and whether the denotation space is also outside the shared mental space. This proposal implies that the physical space S and H share is a proper part of their shared mental space, which, in turn, is a proper part of S’s public world knowledge. Note that it is left open whether (and if, to what extent) S and H share

13 The concept of the reference space resembles the concept of the ‘anchor’ in the sense of von Heusinger (2011a, b).

Definite or still demonstrative? | 271

their public knowledge, and whether the denotation space (the referent of the jen-NP) and the reference space are part of S’s and H’s mutual public knowledge or only part of the speaker’s. That is, using jen- allows the speaker to present the referent of the jen-NP as being neither anchored¹⁴ in the physical space that S and H are sharing, nor in their shared private world knowledge. Example (32) illustrates this point: (32) […] kann es nicht schaden, den Blick auch einmal nach Timbuktu, jener sagenumwobenen Stadt aus Lehm am mächtigen Nigerfluss in Mali, Afrika, zu richten. Der Publizist Roger Willemsen hat […] uns wissen lassen, dass Bundespräsident Heinrich Lübke dort im Jahre 1956 eine Bronzeplakette eingeweiht habe. (MM_2012) ‘It wouldn’t do any harm to have a look at Timbuktu, jen city of clay shrouded in legend lying on the river Niger in Mali, Africa. The publicist Roger Willemsen lets us know that the federal president Heinrich Lübke has inaugurated a bronze tag there in the year 1956.’ In (32), the whole situation of the federal president’s tag inauguration is presented as having taken place far away and long ago. The distance to the situation itself, as well as its location, the city referred to with a jen-NP, is not only temporal and local, but also epistemic: the situation was not witnessed by the writer, but is only reported second-hand. The location, the city Timbuktu, is not only far away, it is also said to be shrouded in legend. That is, S and H not only cannot physically see the city, but also cannot have any real knowledge about it, even less any private shared knowledge – and this is, as I claim, exactly what the use of jen- here is making explicit. I propose to model the expression of mental distance to S and H with jen- as non-truth-conditionally relevant, not-at-issue meaning.

4.1 At-issue vs. not-at-issue meaning I propose that jen- has two meaning aspects at the levels of at-issue and not-atissue meaning:

14 In all the examples discussed below, the denotation space of the jen-NPs is outside of the shared mental space of S and H, too. That is why in the following, I speak of the referent not being in the shared mental space of H and S, although this is a terminological simplification for “not being anchored via R (and also not having D inside)”.

272 | Maria Averintseva-Klisch

(i) at-issue meaning: jen-NP contributes a salient identifiable referent, (i.e. salient referent anchored in the discourse representation), corresponding to definiteness; (ii) not-at-issue meaning: jen- anchors the referent to S and H (demonstrative meaning), i.e. the referent is presented by the S as being mentally distant from S and H in the sense of the reference space not being in the shared mental space of S and H. For this analysis, I adopt the idea to differentiate two levels of meaning from Roberts (1996), which was recently developed by Potts (2005, 2007) and Simons et al. (2010), to name but a few. At-issue content is the descriptive truthconditional content answering the current question under discussion (QUD; Roberts 1996/2012). A prime example for not-at-issue content is contributed by expressive epithets like in (33): (33) That bastard Kresge is famous. at-issue meaning: Kresge is famous. not-at-issue meaning: Kresge is a bastard/bad in S’s opinion. (Potts 2007: 168) It is possible to question or negate the at-issue content (cf. (34a and b i)) contributed by a sentence explicitly, and it does not project out from under negation or modality. Not-at-issue content, i.e. content not directly addressing the QUD, is not truth-conditionally relevant. This content is not directly questionable, and it projects out from under negation or modality (as in (34c)). It is possible to address it, however, with the “hey, wait a minute” protest (cf. (34b ii)). (34) a. What do you know about Kresge? / Who is famous? – Kresge is famous. b. That bastard Kresge is famous. (i) No, he is not famous. (ii) # No, he is a wonderful person. vs. Hey, wait a minute, he is a wonderful person! c. That bastard Kresge is not famous. → Kresge is bad in S’s opinion. Evidence for the two-dimensional meaning analysis of jen- proposed above comes from example (35): here, it is clear that first, distance is a meaning aspect of jenreference, and second, that truth-conditionally dies- and jen- cannot differ: (35) Im Januar dreiundvierzig sprach man viel von der Stadt Stalingrad1 . Da Matzerath jedoch den Namen dieser Stadt1 ähnlich betonte, wie er zuvor Pearl Har-

Definite or still demonstrative? |

273

bour, Tobruk und Dünkirchen betont hatte, schenkte ich den Ereignissen in jener fernen Stadt1 nicht mehr Aufmerksamkeit als anderen Städten, die mir durch Sondermeldungen bekannt wurden. (Günther Grass, Blechtrommel) ‘In January 1943, one spoke a lot about the city called Stalingrad1 . But as Matzerath emphasized the name of this city1 exactly as he had emphasized Pearl Harbour, Tobruk, and Dünkirchen earlier, I paid no more attention to the events in jen distant city1 than I did to other cities that I had come to know from special announcements.’ The city Stalingrad is explicitly presented as very far away, via the adjective fern (‘distant’), for the first person narrator, spatially as well as emotionally. He states that the ongoings in the distant city cannot really emotionally involve him. The choice of adnominal jen- clearly enhances this distance. Still, it is possible to refer to the same referent with diese Stadt (‘this city’), which would not be possible if the meaning aspect of distance were truth-conditionally relevant. The relevant part of the sentence in (35) containing jen- has the following meaning: (35󸀠 ) I paid no more attention to the events in jen-distant city than I did to other cities at-issue: S paid no more attention to the events in Stalingrad than he did to other cities; not-at-issue: S presents Stalingrad city as being distant in the sense of not in the shared mental space of S and H. Applying the projection test (cf. (34c) above) shows that the not-at-issue meaning part projects out from under negation: (36) Ich schenkte den Ereignissen in jener Stadt meine vs. keine Aufmerksamkeit. ‘I gave the events in jen city my vs. no attention’ Here, changing the truth-conditions for the at-issue meaning by adding or taking away sentential negation does not affect the mental distance conveyed by jen-: the city is still presented as far away and not directly involving the narrator emotionally. Note that this effect remains even if the adjective fern explicitly naming the distance is taken away. Spoken drastically, my analysis denies that the distal demonstrative jen- expresses demonstrativity, at least at the truth-conditional level: truth-conditionally jen- is only definite, demonstrativity only comes in at the not-at-issue level. Taking into account that the deictic use, which is the prototypical use of demonstra-

274 | Maria Averintseva-Klisch

tives, is neither assumed for present-day jen- in the relevant literature, nor found in the corpus data, this analysis makes sense. In the deictic use, both the definiteness requirement (the anchoring to the discourse situation), as well as the demonstrativity requirement (S- and H-anchoring) is trivially given if the corresponding referent is actually present in the current discourse situation. However, the requirement on jen- that the corresponding referent is not part of the common mental space of S and H makes a direct deictic use pragmatically odd. Given the assumption made above, i.e. that common mental space includes common physical space, S would be required to draw the attention of H to something that per definition cannot be present. Hence, assuming that the referent of the jen-NP is distant in the sense of ‘not in the shared physical space of the speaker and the hearer’ and not in the sense of ‘far away (or: less near) in the shared physical space’ leads directly to an explanation for the lack of the deictic uses in my data. The observation for the anaphoric use of jen- made above, i.e. that the distal component of jen- is also not observable for the anaphoric use since there is no contrast with the proximal dies-, serves as additional evidence for my analysis. Thus, demonstrativity comes in only at the not-at-issue level: to be felicitously used as a demonstrative, the reference with jen- has to satisfy two conditions: (i) the reference has to be validly definite, i.e. the referent has to be anchored to the discourse representation; (ii) the reference has to be validly demonstrative in the sense that the referent has to be anchored to both S and H as being mentally distant, i.e. not (anchored via R) in the shared mental space of S and H. In the following, I revisit the three uses of jen- that were found in the corpus data – the anaphoric, the anamnestic, and the determinative use – and show how they can be accounted for with my analysis.

(a) the anaphoric use revisited For examples involving anaphoric reference, condition (i) at the at-issue meaning is trivially given since the corresponding referent has already been mentioned in the previous discourse, cf. example (37), repeated from Section 3: (37) DFB-Chef Braun1 hat da seinem Angestellten Vogts2 zu Recht die passenden Worte gesagt. Jener2 hat zwar vor dieser WM in Sachen Matthäus eindrucksvoll über den Schatten springen gelernt, ein ganz Großer ist er deshalb aber noch lange nicht. ‘In this case, DFB chief Braun1 was right to say these suitable words to his employee Vogts2 . Jen.one2 has truly learned to swallow his pride in the case of Matthäus before this World Cup, he is not one of the greats, yet, just because of that.’

Definite or still demonstrative?

| 275

In addition, the context has to ensure that the at-issue condition (ii) is plausibly met. In (37), this is also the case since the corresponding referent is presumably not known to S and H privately, but is part of public world knowledge. Moreover, S does not seem to be sympathetic with the referent, as is made clear by he is not one of the greats, yet. Up until now, I was able to find four possible varieties of mental distance. They often interact in the specific use: 1. Temporal distance: the referent of the jen-NP is presented as distant in time; cf. (38), repeated from Section 3: (38) Die Ruine der vor 104 Jahren1 erbauten prächtigen Mannheimer Synagoge wird abgetragen. Nichts ist geblieben von dem Glanz jener Zeit1 , als 6600 jüdische Mitbürger ihren Mittelpunkt in F2 hatten. ‘The ruins of the grand synagogue of Mannheim, which was built 104 years ago1 , are being removed. Nothing has remained of the glamour of jen time1 , as 6600 fellow Jewish citizens were living in F2.’ Here, the NP jene Zeit (‘that time’) refers to a time span that is around one hundred years before the utterance time, and that is explicitly stated to be gone forever. 2. Spatial distance: S and H do not share the physical space with the referent, as in (39), which has been discussed above in detail: (39) Im Januar dreiundvierzig sprach man viel von der Stadt Stalingrad1 . Da Matzerath jedoch den Namen dieser Stadt1 ähnlich betonte, wie er zuvor Pearl Harbour, Tobruk und Dünkirchen betont hatte, schenkte ich den Ereignissen in jener fernen Stadt1 nicht mehr Aufmerksamkeit als anderen Städten, die mir durch Sondermeldungen bekannt wurden. ‘In January 1943, one spoke a lot about the city called Stalingrad1 . But as Matzerath emphasized the name of this city1 exactly as he had emphasized Pearl Harbour, Tobruk, and Dünkirchen earlier, I paid no more attention to the events in jen distant city1 than I did to other cities that I had come to know from special announcements.’ 3. Modal distance: the referent exists in some fictitious world, and is thus not part of the real world, which S and H inhabit – and in this sense distant; the fictitious world is presented not as part of shared private knowledge of S and H, but as being generally well-known (e.g. the setting of a widely popular book or movie):

276 | Maria Averintseva-Klisch

(40) Selbst das Unmögliche erscheint möglich: Warum sollte eine Figur, die einem Buch entspringt, keine Gefühle haben und kein Heimweh nach seiner fiktiven Welt empfinden wie eben jener Feuerspucker Staubfinger? (MM2012) ‘Even impossible things seem to be possible. Why should a character originating in a book not have feelings and miss his fictitious world, like jen firebreather Dustfinger?’ In this example, the referent Staubfinger (‘Dustfinger’ in the english translation), a protagonist of Cornelia Funke’s novel ‘Inkheart’, gets evoked anamnestically in a short newspaper article reporting a theatre adaptation of the novel. The protagonist has not been mentioned so far, and the author appeals to the general world knowledge of the readers to locate the referent in the fictitious world of the novel. 4. Emotional distance: by choosing jen-, S expresses with his reference that he is not sympathetic to the referent of the jen-NP. (41) Ein gerader Straßenzug führt von Norden nach Süden, nüchterne Hausfassaden beherrschen die Sicht […] Inmitten jener Schmucklosigkeit am Breiten Weg in Magdeburg entsteht “Die Grüne Zitadelle” (MM_2005) ‘A straight street of houses leads from the north to the south, sober house fronts is the only thing one sees. In between, in jen baldness on the BreiteWeg-street in Magdeburg “The Green Citadel” is growing.’ Here, the baldness of the place described in (41) does not get S’s sympathy. In this case, S arguably uses jen- to express both emotional and spatial distance: the described streets are not in Mannheim, where the newspaper is published, and thus also spatially distant from S and H. Usually, different varieties of mental distance interact, as is actually expected from my analysis: not being in the shared mental space of H and S makes not only spatial, or modal, or temporal, but also emotional distance expectable, and is in no way restricted to only one of these aspects of mental distance in isolation. Recall for instance the Timbuktu-example (32) above, where we could distinguish temporal, local, and modal aspects of mental distance for one single occurrence. Emotional distancing is understood in the sense of not having a positive emotion towards the referent. Note that it does not imply that S has strong negative emotions towards the referent of the jen-NP. The data show that jen- can in fact not be used to mark strong negative emotion towards the referent; this, as well

Definite or still demonstrative? | 277

as marking of strong positive emotion, is the domain of dies-, cf. the following examples: (42) Soviel Geld kann der im ganzen Leben nicht verdienen, wie er als Entschädigung zu zahlen hat, dieser Blödmann! (a chat example, after Consten and Averintseva-Klisch 2010) ‘He can never earn that much money in his life as he will have to pay for compensation, this idiot!’ (43) Der Kirchbau, dessen Einweihung wir heute feiern, zählt zu dem Besten, was freie Bürger leisten können. Dieser wunderbare Bau ist mehr als ein Gebäude. Er steht für das Gute, das uns eint. (the federal president’s Horst Köhler speech at the inauguration of the rebuilt Frauenkirche of Dresden 2005) ‘This church building, whose inauguration we are celebrating today, counts among the best things free citizens may achieve. This wonderful building is more than just an edifice. It is a symbol of the good that unites us all.’ In both cases, jen- instead of dies- would sound very odd. Data like those two examples provide evidence against Bisle-Müller (1991: 70), who assumes “emotional distancing” with dies-. As is obvious from (42) and (43), dies- marks emotional proximity – in contrast to Bisle-Müller’s claim – as is expected since strong emotions render the referent mentally near. This emotional proximity has two aspects: First, S states that the referent is mentally near to him as he has a positive emotional stance (cf. (43)) or a negative emotional stance (cf. (42)) towards the referent. Second, S invites H to share his attitude, thus explicitly suggesting mental closeness between S and H. Note that for English, Lakoff (1974: 351) assumes that both proximal and distal demonstratives, i.e. this and that, can mark what she labels “emotional closeness between speaker and addressee”, as in (44); both examples are from Lakoff (1974): (44) a. This Henry Kissinger knows his way around Hollywood! b. How’s that throat? (a sympathetic nurse to a patient) A comparable use is not attested for German jen-. I believe that for English, both this and that require S, H, and the referent of the demonstrative to be in the shared mental space of S and H whilst jen- is used to refer to something outside this shared mental space (see Averintseva-Klisch to appear for more details to this assumption).

278 | Maria Averintseva-Klisch

(b) the anamnestic use revisited In the anamnestic use, the conditions stated for demonstrative jen-, that is the condition on the at-issue meaning, i.e. situational anchoring, as well as the condition on the non-at-issue S- and H-anchoring are at first glance not satisfied. Since the referent has neither been mentioned before nor is physically present, we can say nothing about reference anchoring by examining the context. However, in these cases, it is the use of the demonstrative itself that serves to force such an anchoring. Gunkel (2007a: 223) decribes anamnestic function as referring to “something that is supposed to be known”. In Consten and Averintseva-Klisch (2012: 259), we instead talk of ‘tentative reference’ with anamnestic demonstratives in the sense of referring “with insufficient linguistic means leaving open if the reference is really resolvable for the hearer and appealing to him to accept this deficient reference.” In any case, this appeal towards H seems to apply to all those cases where the form of the NP automatically, by virtue of the reference act itself, induces (or as-if induces) the necessary reference conditions for anchoring the referent both in the discourse representation and relative to S and H. With jen-, the S- and H-anchoring has to be specified as the referent not being in the shared mental space of S and H, hence temporal, modal (cf. ex. (40) above) or emotional distancing are again expected. Note that the anamnestic use is also possible with dies-, cf. (45): (45) Was is’n eigentlich mit diesem Haustelefon, was wir immer g‘habt ham? (Auer 1984: 637) ‘What has actually happened with this internal phone that we used to have?’ The difference between anamnestic dies- and jen- lies in the character of the tentatively evoked mutual knowledge: dies-reference presents this knowledge as belonging to the shared private knowledge of S and H, whilst with jen-, due to its not-at-issue meaning, it is the shared public knowledge that is relevant.

(c) the determinative use revisited For examples with a determinative reading (46) which has already been discussed above, we can observe the following:

Definite or still demonstrative? | 279

(46) Die Teilnehmerkarten all jener Kerwe-Besucher, die insgesamt sieben Gaudispiele absolviert hatten, […] ‘The tickets of all jen country fair visitors who completed seven jamboree games […]’ First, the at-issue condition, i.e. the referent’s discourse-situational anchoring, is satisfied via the NP itself. Thus, the referential means that are used have to be enough by themselves to achieve the anchoring; that is why additional restrictive modifiers are necessary in this case. Second, we can assume that jen- in its determinative use has lost its not-atissue meaning component of marking that the referent is not in the shared mental space of S and H. For the determinative cases, no traces of mental distance can be found, hence we can doubt the demonstrativity of jen- in this use. In fact, in cases like (46), jen- is interchangeable with derjenig- or der/die/das. Thus, we can recast jen-’s “grammaticalization as pure determinativity marker” proposed in Gunkel (2007a) as the loss of jen-’s not-at-issue meaning component. What the exact conditions for this loss are, is still to be investigated in detail. In any case (and as expected with grammaticalization) this loss seems to be a gradual matter. Recall that in Section 3.2, I noted that for some of my corpus examples, it was difficult to define exactly what kind of demonstrative use they instantiate. In particular, the data show that the anamnestic use and the determinative use do not necessarily form two separate cases, but can be two functions of one and the same occurrence, cf. (47): (47) Als Elias Leondaris die Zuschauer auffordert, mitzutanzen, und diese auch dazu bereit sind, da weht ein Hauch von Stoupa durch Ladenburg. Stoupa – Filmfreunde erinnern sich – ist jener Ort auf dem Peleponnes, in dem der Hollywoodstreifen “Alexis Sorbas” spielt. (MM_2005) ‘And when Elias Leondaris prompts the spectators to dance with him, and they are also willing to join in, a hint of Stoupa wafting through Ladenburg. Stoupa, as movie buffs will remember, is jen place at the Peloponnese where the Hollywood film “Alexis Sorbas” is set.’ Here, the anamnestic appeal to people’s general knowledge about well-known movies (explicitly worded: as movie buffs will remember) is combined with explicitly giving the information necessary for the identification of the referent in the restrictive relative clause. If determinative use has lost its demonstrative character in the sense discussed above, then an anamnestic touch, like in example (47), could again bring in demonstrativity. Since jen- is originally a demonstrative, it

280 | Maria Averintseva-Klisch

might be the case that using it in such ambiguous cases, where a demonstrativity flavour is retained, is preferred over clear determinative uses. A detailed investigation of this issue requires further data, and is therefore left to future work. Summing up, I argue that demonstratives are definites that require anchoring with respect to both S and H in addition to situational anchoring. In particular, jen- signals that its referent is ‘mentally distant’, i.e. that the reference space (and often also the denotation space) lies outside of the shared S- and H-space. This meaning aspect can be modeled as non-at-issue meaning in the sense of Potts (2005; 2007) and Simons et al. (2010). It follows from the analysis of jen- as a linguistic means of reference to someting or someone not in the shared mental space of S and H that the deictic use, as well as (strong) emotive use found with dies- are not expected with jen-. At least in my data, these uses are in fact not attested. In the anaphoric and anamnestic uses, reference to someting or someone ‘mentally distant’ is possible. In fact, my data contain anaphoric and anamnestic cases, the latter only occurring as full lexical jen-NPs. I address this discrepancy in the next section. For the determinative use, I have argued that jen- loses its not-at-issue meaning component.

4.2 Pronominal vs. adnominal jenAs discussed in Section 3, I did not find pronominal occurrences of jen- in an anamnestic use in my corpus data. This can be be expected on the basis of the crucial referential difference between full NPs vs. pronouns: the former contribute descriptive meaning which is needed for referent identification, whilst pronouns lack descriptive meaning, and therefore require other means and sources for the identification of the referent. If the corresponding referent has not been mentioned before (anaphoric use), and is not present in the current discourse situation (deictic use), the identification of the referent is not ensured. Still, cataphoric uses are possible for personal pronouns, like in (48), but not for demonstratives: (48a) is only possible with accompanying deictic pointing but not cataphorically, (48b) is completely ill-formed. (48) Er ist da, wenn es keinen Trost gibt. Als Notfallseelsorger hilft Peter Bublitz aus Amberg Menschen im vielleicht schlimmsten Moment ihres Lebens. (www.mittelbayerische.de) ‘(The beginning of a discourse:) He1 is there, when no consolation is possible. As an emergency pastor, [Peter Bublitz from Amberg]1 helps people in maybe the worst moments of their life.’

Definite or still demonstrative? | 281

a. # Dieser ist da, wenn es keinen Trost gibt. (‘This.one is there …’) b. * Jener ist da, wenn es keinen Trost gibt. (‘Jen.one is there …’) This is exactly what we expect given the analysis above. In the case of a personal pronoun er (‘he’) in (48), discourse representation anchoring via accomodation is available: the pronoun supplies the information that a male person referent has to be accomodated in the current discourse representation. The next sentence then adds information about this referent: the accomodated male referent is an emergency pastor named Peter Bublitz. With a demonstrative, S- and H-anchoring is additionally required. If the person referred to is present in the discourse situation, such anchoring is allowed with dies- (deictic reference). As has been shown above, jen- does not have this option. The possibility to search for a suitable referent in world knowledge, as is appealed for with anamnestic demonstratives, is excluded since the pronoun does not supply enough descriptive information that can be used for the search. In anaphoric cases, the at-issue condition requiring discourse representation anchoring is satisfied. If the context supports the not-at-issue meaning of mental distancing, then pronominal jen- occurrence is valid, as both conditions are satisfied. This holds irrespectively of number marking and the relative linear order of any potential antecedents, as we have seen above. The descriptive poverty of pronominal jen- also explains why the loss of the not-at-issue meaning component is predestined in the pronominal occurrences in a determinative use: the meaning component cannot be supported by other lexical material (as it is the case for the head noun and possible attributes in a full lexical NP), and is therefore especially susceptible to loss via grammaticalization.

5 Summing up In modern German, there is no real opposition of dies- vs. jen- as proximal vs. distal demonstratives on the physical dimension, neither in their anaphoric nor their deictic uses. The corresponding meanings are expressed with other lexical items. To account for this descriptive observation, I propose that demonstratives have two meaning components: they are (i) definite, i.e. require an anchoring to the current discourse representation, and they are (ii) demonstrative in that they require an anchoring with respect to the speaker and the hearer. These meaning components are specified differently for proximal dies- and distal jen-: For the distal demonstrative jen-, I assume that it contributes the definiteness meaning component to the at-issue level, whilst the demonstrativity component

282 | Maria Averintseva-Klisch

is not-at-issue. More precisely, jen- has a not-at-issue meaning component that contributes what I label ‘mental distance’: jen-NPs are used by the speaker to refer to objects that are not in the shared mental space of the speaker and the hearer. To be more exact, the reference space and preferably also the denotation space do not belong to the shared S-and-H-space. In this sense – but only in this one – jen- can still be called a distal demonstrative. For proximal dies- I hypothesize that it is a demonstrative proper, so to speak. Its demonstrative meaning component, i.e. the anchoring of the referent to S and H, such that it is in their shared mental space, is part of the at-issue meaning. This explains the asymmetry between dies- and jen- with respect to their deictic and their anaphoric uses. However, to pursue this issue any further goes beyond the scope of this paper, and is left for further research. As proposed in the literature and backed up by the corpus data, jen- also has a determinative use in which the demonstrative meaning component is absent, and jen- serves as a definiteness marker. This use requires the presence of restrictive modifiers in the jen-NP. Gunkel (2007a) and Grammis propose that jencurrently is subject to grammaticalization losing its demonstrativity, with plural forms (Gunkel) and/or adnominal jen- (Grammis) already having completed grammaticalization as a definiteness marker. I recast the observed loss of demonstrativity in the determinative use as a semantic grammaticalization process, i.e. as loss of the not-at-issue demonstrativity meaning component. Contra Gunkel, this loss does not seem to be restricted to a particular parameter value of grammatical number, i.e. it neither affects only singular nor only plural forms. Moreover, this loss is a gradual process, as jen-uses with both determinative and anamnestic features, that is, cases that still have mental distance traits, are attested. My data show that there is an important difference between adnominal and pronominal occurrences of jen-. For pronominal jen-, no clear anamnestic use is attested, whilst the determinative use is the use that is strongly preferred. One could, contra Grammis, assume that grammaticalization occurs in the pronominal use first since pronominal jen- lacks descriptive material that is needed to overtly justify the speaker’s mental distancing. However, this is a semantic reason for a semantic process. Hence, it is doubtful whether it makes sense to assume that a morpho-syntactic process of the emergence of two different lexemes, an adnominal determiner and a pronoun, goes hand in hand with it. An analysis of jen- as contributing an at-issue definiteness component, as well as a not-at-issue component of mental distance correlates naturally with the observed differences between dies- and jen- in style and register: the latter is assumed to occur specifically in written German. If one assumes that in oral communication, one preferably refers to spatially and/or mentally near objects, this difference does not seem all that unexpected. However, as was also shown with

Definite or still demonstrative? |

283

the data from Wikipedia discussions, jen- is not only used in lofty or formal written communication. Hence, rather than saying that jen- is only used in upper-style communication, one should assume more to the point that referential immediacy (in the sense of referring to objects physically or mentally present) and strong emotiveness of ‘conceptual orality’ (cf. Koch and Oesterreicher 1985) cannot be adequately expressed with jen-. Of course there remain lots of questions that still have to be answered. One open issue is an exact definition of the intuitive concept of ‘mental space’ that I use in this article. To that end, a comparison with other indexical linguistic items, as well as with demonstratives in other languages seems promising. Levinson (2004: 109) pvovides a list of languages that encode spatial distance from both speaker and addressee with a special lexeme. It would be interesting to find out whether these lexemes can also be used to express what I described as mental distance, and if that is the case, to compare these uses with German jen-. Acknowledgment: I am very grateful to Pritty Patel-Grosz for her very helpful comments on the first draft of this paper, and to Sarah Zobel (Tübingen) for the proof-reading of the manuscript.

References Acton, Eric K. & Christopher Potts. 2014. That straight talk: Sarah Palin and the sociolinguistics of demonstratives. Journal of Sociolinguistics 18(1). 3–31. Auer, Peter. 1981. Zur Indexikalitätsmarkierenden Funktion der demonstrativen Artikelform in deutschen Konversationen. In Götz Hindelang & Werner Zillig (eds.), Sprache: Verstehen und Handeln (Akten des 15. Linguistischen Kolloquiums. Münster 1980. Vol. 2), 301–310. Tübingen: Niemeyer. Auer, Peter. 1984. Referential problems in conversation. Journal of Pragmatics 8. 627–648. Averintseva-Klisch, Maria (to appear). Pejorative demonstratives. In Rita Finkbeiner, Jörg Meibauer & Heike Wiese (eds.). 2016. Pejoration (Series Linguistics Today). Amsterdam: John Benjamins. Bisle-Müller, Hansjörg. 1991. Artikelwörter im Deutschen. Semantische und pragmatische Aspekte ihrer Verwendung. Tübingen: Niemeyer. Bosch, Peter & Carla Umbach. 2007. Reference determination for demonstrative pronouns. ZAS Papers in Linguistics 48. 39–51. Brugmann, Karl. 1904. Die Demonstrativpronomina der indogermanischen Sprachen. Leipzig: Teubner. Bühler, Karl. 1934. Sprachtheorie. Stuttgart: Fischer. Cardinaletti, Anna & Michal Starke. 1999. The typology of structural deficiency: A case study of the three classes of pronouns. In Henk van Riemsdijk (ed.), Clitics in the languages of Europe, 145–233. Berlin: Mouton de Gruyter.

284 | Maria Averintseva-Klisch

Consten, Manfred, Mareile Knees & Monika Schwarz-Friesel. 2007. The function of complex anaphors in texts. In Monika Schwarz-Friesel, Manfred Consten & Mareile Knees (eds.), Anaphors in Text, 81–102. Amsterdam: John Benjamins. Consten, Manfred & Maria Averintseva-Klisch. 2010. ‘Nahe Referenten’ – ein integrativer Ansatz zur Funktion demonstrativer Referenz. Sprachtheorie und germanistische Linguistik 20(1). 1–34. Consten, Manfred & Maria Averintseva-Klisch. 2012. Tentative reference acts? Recognitional demonstratives as means of suggesting mutual knowledge – or overriding a lack of it. Research in Language 10(3). 257–277. Diessel, Holger. 1999. Demonstratives: Form, function, and grammaticalization. Amsterdam: John Benjamins. Diessel, Holger. 2012. Deixis and demonstratives. In Claudia Maienborn, Klaus von Heusinger & Paul Portner (eds.), Semantics. An international handbook of natural language meaning. Vol. 3, 2407–2431. Berlin: Mouton de Gruyter. Dudenredaktion. 2009. Vol. 4: Die Grammatik. Unentbehrlich für richtiges Deutsch. Mannheim, Wien & Zürich: Dudenverlag. Dudenredaktion. 2011. Vol. 9: Richtiges und gutes Deutsch. Mannheim, Wien & Zürich: Dudenverlag. Ehrich, Veronika. 1982. Da and the system of spatial deixis in German. In Jürgen Weissenborn & Wolfgang Klein (eds.), Here and there. Cross-linguistic studies on deixis and demonstration, 43–63. Amsterdam: John Benjamins. Elbourne, Paul. 2008. Demonstratives as individual concepts. Linguistics and Philosophy 31(4). 409–466. Erben, Johannes. 1980. Deutsche Grammatik. Ein Abriss. München: Max Hueber. Farkas, Donka. 2002. Specificity Distinctions. Journal of Semantics 19. 1–31. Grammis. Systematische Grammatik. Online grammar of IDS Mannheim. http://hypermedia. ids-mannheim.de (accessed 14 October 2014). Gunkel, Lutz. 2006. Betontes der. In Eva Breindl, Lutz Gunkel & Bruno Stecker (eds.), Grammatische Untersuchungen, Analysen und Reflexionen, 79–96. Tübingen: Narr. Gunkel, Lutz. 2007a: Demonstrativa als Antezedentien von Relativsätzen. Deutsche Sprache 35. 213–238. Gunkel, Lutz. 2007b. Nominalphrasen des Typs derjenige N + Relativsatz in den europäischen Sprachen. In Norbert Fries & Christiane Fries (eds.), Deutsche Grammatik im europäischen Dialog. Beiträge zum Kongress Krakau 2006, 10 p. http://krakau2006.anaman.de (accessed 14 October 2014). Hauenschild, Christa. 1982. Demonstrative pronouns in Russian and Czech – deixis and anaphora. In Jürgen Weissenborn & Wolfgang Klein (eds.), Here and there: cross-linguistic studies on deixis and demonstration, 167–186. Amsterdam: Benjamins. Heusinger, Klaus von, Sofiana Chiriacescu & Annika Deichsel. 2010. Two specific indefinite articles in German. Handout, invited lecture at the University of Santa Cruz, California. people.ucsc.edu/~abrsvn/handout_11.pdf (accessed 14 October 2014). Heusinger, Klaus von. 2011a. Specificity, referentiality and discourse prominence: German indefinite demonstratives. In Ingo Reich, Eva Horch & Dennis Pauly (eds.), Sinn & Bedeutung 15. Proceedings of the 2010 annual conference of the Gesellschaft für Semantik, 9–30. Universaar – Saarland University Press: Saarbrücken.

Definite or still demonstrative?

| 285

Heusinger, Klaus von. 2011b. Specificity. In Klaus von Heusinger, Claudia Maienborn & Paul Portner (eds.), Semantics: An international handbook of natural language meaning, 1024–1057. Berlin: de Gruyter. Heusinger, Klaus von. 2013. The salience theory of definiteness. In Alessandro Capone, Franco Lo Piparo & Marco Carapezza (eds.), Perspectives on linguistic pragmatics, 349–374. Berlin: Springer. Himmelmann, Nikolaus P. 1996. Demonstratives in narrative discourse: A taxonomy of universal uses. In Barbara Fox (ed.), Studies in anaphora, 205–254. Amsterdam: John Benjamins. Himmelmann, Nikolaus P. 1997. Deiktikon, Artikel, Nominalphrase. Zur Emergenz syntaktischer Struktur. Tübingen: Niemeyer. Ionin, Tanya. 2006. This is definitely specific: Specificity and definiteness in article systems. Natural Language Semantics 14. 175–234. Irmer, Matthias. 2011. Bridging inferences: Constraining and resolving underspecification in discourse interpretation. Berlin: de Gruyter. Kamp, Hans. 1990. Prolegomena to a structural account of belief and other attitudes. In Anthony C. Anderson & Joseph Owens (eds.), Propositional Attitudes, 27–90. Stanford: CSLI Publications. Klein, Wolfgang. 1994. Time in language. London: Routledge. Koch, Peter & Wulf Oesterreicher. 1985. Sprache der Nähe, Sprache der Distanz: Mündlichkeit und Schriftlichkeit im Spannungsfeld von Sprachtheorie und Sprachgeschichte. Romanistisches Jahrbuch 36. 12–43. Kupietz, Marc, Cyril Belica, Holger Keibel & Andreas Witt. 2010. The German Reference Corpus DeReKo: A primordial sample for linguistic research. In Nicoletta Calzolari et al. (eds.), Proceedings of the 7th conference on International Language Resources and Evaluation, 1848–1854. Valletta, Malta: European Language Resources Association. Lakoff, Robin. 1974. Remarks on ‘this’ and ‘that’. Chicago Linguistic Society 10. 345–356. Levinson, Stephen C. 2004. Deixis. In Laurence Horn & Gregory Ward (eds.), The handbook of pragmatics, 97–121. Oxford: Blackwell. Maclaran, Rose. 1980. On two asymmetrical uses of the demonstrative determiners in English. Linguistics 18. 803–820. Molnár, Krisztina. 2010. Anamnestische Verwendung der Demonstrativa im Deutschen und im Ungarischen. Deutsche Sprache 4(10). 326–344. Potts, Christopher. 2005. The logics of conventional implicatures. Oxford: Oxford University Press. Potts, Christopher. 2007. The expressive dimension. Theoretical Linguistics 33(2). 165–197. Potts, Christopher & Florian Schwarz. 2010. Affective ‘this’. Linguistic issues in language technology 3(5). 1–30. Prince, Ellen. 1981. On the inferencing of indefinite-this NPs. In Aravind Joshi, Bonnie Webber & Ivan Sag (eds.), Elements of discourse understanding, 231–250. Cambridge: Cambridge University Press. Rauh, Gisa. 2003. Warum wir Linguisten “euch Linguisten”, aber nicht “sie Linguisten” akzeptieren können. Eine personendeiktische Erklärung. Linguistische Berichte 196. 389–424. Reichenbach, Hans. 1947. Elements of symbolic logic. New York: Macmillan. Roberts, Craige. [1996]/2012. Information structure in discourse: Towards an integrated formal theory of pragmatics. Semantics and Pragmatics 6. 1–69. Simons, Mandy, Judith Tonhauser, David Beaver & Craige Roberts. 2010. What projects and why? Proceedings of SALT 20. 309–327.

286 | Maria Averintseva-Klisch

Ward, Gregory & Betty Birner. 1995. Definiteness and the English existential. Language 71(4). 722–742. Zifonun, Gisela, Ludger Hoffmann & Bruno Stecker (eds.). 1997. Grammatik der deutschen Sprache. Berlin: de Gruyter.

Thiago N Galery (DBpedia Association)

Deferred pronouns: A sketch based on Relevance Theory and Dynamic Syntax Abstract: This chapter is about the semantics/pragmatics of deferred uses of pronouns (Nunberg, 1993), whose correct understanding depends on the identification of a specific individual that provides a descriptive interpretation. I argue that some of these cases might be seen as an ‘implicit’ form of anaphora, where the antecedent is inferred in the context and yet is able to provide the pronoun with an interpretation that can interact with other structural aspects of the utterance. My proposal employs a combination of relevance-theoretic pragmatics (Sperber and Wilson 1995, Carston 2002), certain motivated assumptions about visual information processing, and the grammar framework of Dynamic Syntax (Kempson et al. 2001, Cann et al. 2005). Keywords: pronouns, descriptions, object-identification, anaphora, quantification, semantics, pragmatics, Relevance Theory, Dynamic Syntax

1 Deferred pronouns: an overview This chapter deals with pronominal uses that depend on some (usually, perceptual) feature of the environment but receive a descriptive interpretation (instead of a deictic or indexical one). They were first discussed in more detail by Nunberg (1993) and are identified there as ‘deferred pronouns’¹. They can be illustrated by the following example (relevant expressions in boldface): (1)

a. Pointing to a footprint on the ground: ‘He must be a giant!’ (Schiffer, 1981, 49) b. The footprint-maker must be a giant.

In this example, the speaker singles out an individual (the footprint), but her utterance conveys information that is ‘indirectly’ related to it (i.e. about whoever

Note: Part of this paper was developed whilst I was working for Idio Research (http://www. idioplatform.com/). I would like to thank them for the support. 1 More recently these have been called ‘descriptive indexicals’ (Elbourne, 2008). I am sticking to the original term because it is less theoretically biased.

288 | Thiago N Galery

made the print). This form of ‘indirectly related meaning’ seems to be picked up by be the pronoun, so that ‘he’ in (1a) is interpreted as the footprint maker and effectively conveys the truth-conditions stated in (1b). As we shall see shortly, there are other cases in which the descriptive interpretation of these pronominal cases interacts with structural aspects of the utterance (e.g. scope of quantifiers). I believe that this fact raises interesting issues in the grammar/pragmatics interface, which have not been addressed properly by the literature. First, linguists from a generativist background, as well as some (minimalist) philosophers consider the output of semantic interpretation to be free of contextual contributions and yet fully truth-conditional. Under this view, it is mysterious how the dependency between structural elements in the utterance and the descriptive pronominal interpretations are established. Elbourne (2005, 2008) proposes an account within this tradition that most closely approximates to an explanation of the data, by appealing to the idea that pronouns are disguised definite descriptions with an elided nominal element. However, since he doesn’t get into the details of how the nominal element is supplied by the context (as well as the principles and constraints on this operation), I take the account to be partial. Second, some linguists, as well as some (contextualist) philosophers, focus on the description of pragmatic principles governing interpretation, thus accounting for the ways in which the context can contribute to utterance truth-conditions. However, they fail to explain how these pragmatically provided constituents interact with linguistic structure. Powell (2003), taking Relevance Theory (Sperber and Wilson, 1995; Carston, 2002) as a pragmatic framework, is an example of this. My aim for this chapter is to address these shortcomings by presenting an account that integrates grammatical processes with pragmatic ones. In order to explain this interaction, one needs to dispense with a model of linguistic processing where grammar simply ‘ships’ outputs to a pragmatics component. We need a theory where pragmatically introduced elements can participate in structural relationships. This view is central to a recent grammatical theory known as Dynamic Syntax (Kempson et al. 2001; Cann et al. 2005, introduced in Section 4) and for this reason I will adopt it as a model of grammar. Furthermore, I will adopt Relevance Theory as a pragmatic framework (briefly presented in Section 3), but for reasons of space I will not be able to motivate it over other accounts (see Galery 2012, ch. 3 for a full elaboration). These elements come together in a final proposal made in Section 5, but for the moment let us turn to the properties of the data that serve as desiderata for any account of the phenomenon.

Deferred pronouns: A sketch based on Relevance Theory and Dynamic Syntax

|

289

2 Deferred pronouns: interpretative properties In order to assess the essential properties of this type of pronominal interpretation, let’s ask the following question. What is necessary for the pronoun in (1) to be correctly interpreted? As argued in (Galery, 2012, ch. 1), if the hearer in (1) fails to identify the demonstrated object as a footprint (or lacks the knowledge that footprints are causally derived from people walking), she would be unable to retrieve the truth-conditions expressed by (1b) (i.e. that the footprint-maker must be a giant). This motivates the following interpretative property: Identification Dependency: Deferred uses of pronouns depend on the identification of a particular entity as belonging to a certain kind. Moreover, these pronominal interpretations suggest that somehow the communicated truth-conditions are descriptive as opposed to singular (to be elaborated below). That is, the speaker of (1a) does not seem to have anyone specific in mind when she communicates (1b): a proposition about whoever happens to be the maker of the footprint. These remarks seem to suggest the following property: Descriptiveness: The information conveyed by the relevant pronominal uses is descriptive or general. As it stands, Descriptiveness is not very helpful. There is no definition for the terms ‘general’ or ‘descriptive’ applied to truth-conditions. Defining such terms, however, touches on a difficult subject within philosophy and linguistics. Some, including Russell (1910) and Evans (1982), have argued that the audience must have the knowledge or ability to identify a specific entity in the context in order to grasp a singular proposition about it. If we adopt this view, cases like (1) would be descriptive in virtue of the fact that the hearer lacks identifying knowledge of the person who made the print. Thus, the truth-conditions would be fixed by the hearer’s ability to identify what the utterance is about in a descriptive manner, as it were. Others, such as Borg (2004), have argued that such epistemic notions should not be part of theories about linguistic meaning (which she identifies as semantics). Based on the idea of an independent language faculty, necessary for explaining human linguistic behaviour (Chomsky, 1986), and the idea of a language module responsible for explaining the structure of thought and action (Fodor, 1975, 1983), Borg argues that there is room for a syntactic notion of singular truthconditions. According to this position, the utterance in (1) would introduce a sin-

290 | Thiago N Galery

gular concept due to the fact that pronouns are linguistically marked as referential expressions². For the purposes of this chapter, I would like to steer clear of this debate, because I believe Descriptiveness might not only have potential epistemic sources, but also structural ones. The key example that shows this, first introduced by Nunberg, is reprinted below. (2) a. Said by Supreme Court Justice O’Connor: If the Democrats had won the last few presidential elections, we might have been liberals (Nunberg, 1993, 14–15). b. If the Democrats had won the last few presidential elections, the US Supreme Court Justices might have been liberals. Like our first example, the paraphrase in (2b) represents the intuitive truthconditions of the utterance in (2a). That is, O’Connor’s use of ‘we’ is interpreted as the description the US Supreme Court Judges. In his paper, Nunberg also points out that although O’Connor is one of the US supreme court justices at the time (or context) of the utterance in (2a), the group of US Supreme Court Judges in (2b) does not need to include her. In short, it seems that the members of the Supreme Court co-vary with the situations introduced by the modals ‘had won’ and ‘might’. This motivates the following interpretative property: Co-variation: The interpretation of deferred pronouns can co-vary with other expressions in the sentence or discourse. To show that the same type of co-variation can occur as intra-sentential dependencies and is not a by-product of plurality, consider the third person singular cases below: (3) a. Pointing to St. Peter’s Basilica: He’s always Italian (discussed in Stokke 2010). b. For all/most situations s, the pope in s is Italian. (4) In Washington, Andy and Martha discuss whether the American government was justified in killing Osama Bin Laden (Galery, 2012): a. Martha: Do you think a democratic government has the right to kill an unarmed man, even if he’s a declared enemy of the state?

2 Note that an explanation of dependencies between pronouns and quantified antecedents (e.g. pronominal binding and anaphora) becomes very mysterious under these assumptions.

Deferred pronouns: A sketch based on Relevance Theory and Dynamic Syntax

|

291

b. Andy: Well … Every time there is a war, he [pointing to the White House] has tough choices to make. c. For every war-time t, the US President in t has tough choices to make in t. In (3) the pronoun ‘he’ is interpreted as the description the pope, which selects different unique individuals according to the situations/eventualities the adverb ‘always’ quantifies over (see Lewis 1975 for the idea of adverbs as quantifiers). In (4), the pronoun is interpreted as the US president which picks out unique individuals according to different war times. Although I assume that Identification Dependency, Descriptiveness and Covariation are core properties of the data, one might argue that there is an important distinction to make with regards to their nature. Patrick Grosz (personal communication) rightly pointed out that although Co-variation might surface in some cases (usually involving third person pronouns), Identification and Descriptiveness emerge in every situation involving deferred pronouns. I acknowledge the point, but I think it does not challenge the claims made here. Following my earlier remarks, Co-variation could be regarded as a structural source for Descriptiveness, that is, the interpretation is descriptive due to some dependency between contextual contributions to the interpretation of the pronoun and other elements in the utterance (or discourse). In some other cases, like the gigantic footprint in (1), Descriptiveness could have an epistemic source; that is, the hearer arrives at some descriptive interpretation due to some lack of knowledge about the specific individual in question. There is a take-home message from this. The importance of Co-variation emerges from the need for an account where pragmatic processes and grammatical ones are intertwined, for it is the identification of a salient object in the context that provides a descriptive interpretation that is able to enter some structural relationship with other truth-conditional elements³. Still, I take it that any theory of deferred pronouns must explain the three properties discussed here.

3 The fact that Identification Dependency is somehow implicated in an explanation of the covarying readings of such pronouns might be regarded as a meta-property of the data (Galery 2012 calls this Connection).

292 | Thiago N Galery

3 Deferred pronouns and pragmatics Assuming that deferred pronouns require a pragmatic component, one might be tempted to explain the data via the Gricean distinction between two distinct levels of meaning.

3.1 Gricean Pragmatics In a nutshell, Grice (1975) proposed a distinction between two levels of meaning: (i) what-is-said by an utterance and (ii) its associated implicatures. The level of what-is-said is such that (i) it is determined by the intentions of the speaker (to mean that p is to intend to make the audience believe that p), (ii) it is largely determined by the conventional meaning of the words uttered, and (iii) it corresponds to the utterance’s intuitive truth-conditions. In addition to these principles, Grice argued that a subset of speaker meant information (i.e. conversational implicatures) are governed by a cooperative principle of communication: ‘Make your conversational contribution such as is required, at the stage at which it occurs, by the accepted purpose or direction of the talk exchange in which you are engaged’ (Grice, 1975, 26). This principle is complemented by a set of norms that assume that, in order to be cooperative, communicators must make their contribution as informative (Quantity), truthful (Quality), relevant (Relation), and perspicuous (Manner) as required by the conversation. With these two levels of meaning, a Gricean could explain examples like (1) along the following lines. By uttering ‘He must be a giant’ whilst pointing to the footprint, the speaker states, at the level of what-is-said, that the footprint must be a giant (nonsensical proposition)⁴. Another contender for Gricean what-is-said would be the proposition that the salient male must be a giant. Although this alternative is more compatible with the linguistic meaning of the pronoun, it doesn’t correspond to the utterance’s intuitive truth-conditions and hence can’t be what the speaker intends to communicate (for a position that endorses this move, see Borg (2004)). The Gricean would then assume that the proposition at the level of what-is-said flouts the cooperative principle or one of its associated maxims (Quality or Quantity in the propositions used as examples here), triggering the in-

4 This would approximate such pronominal uses to non-literal interpretations, like metonymy, which violate the idea that the linguistic (conventional) meaning of words closely determines what-is-said (‘he’ cannot refer to a footprint and preserve its literalness).

Deferred pronouns: A sketch based on Relevance Theory and Dynamic Syntax

|

293

ference of a proposition that satisfies those principles (i.e. that the footprint maker must be a giant). There is a deep problem for this type of account: it is unable to explain Covariation. This was the main reason used by Nunberg (1993) to reject implicaturebased accounts, and has been recognised by the some in the literature (Recanati, 1993; Elbourne, 2008; Stokke, 2010) as a valid argument against such approach. Moreover, it is hard to see how principles governing either (i) the relationship between convention and word meaning or (ii) rational constraints on cooperative behaviour would explain how the singling out of an individual in the context results in a the conceptual representation that makes the speaker’s utterance intelligible (i.e., Identification Dependency, for a fuller argument against such accounts, see Galery 2012, ch. 2).

3.2 Relevance Theory As an alternative to Gricean pragmatics, I would like to argue that these cases are better explained by Relevance Theory (RT, for short) (Sperber and Wilson, 1995; Carston, 2002). In contrast to Grice, RT assumes that pragmatically inferred content can contribute to an utterance’s truth-conditions. The idea is that an utterance’s truth-conditions are constrained, but not fully determined, by the linguistic meaning of the words used. Moreover, instead of having a set of principles governing conventional meaning and another governing rational communicative behaviour, RT assumes a unified principles governing information flow in a cognitive system. The cornerstone of the framework is the notion of relevance itself, which for the present purposes will be presented as a relative property. According to Sperber and Wilson (1995), Relevance is a property of cognitive inputs (both external and internal stimuli): the greater the input’s positive effects on an individual’s cognitive system (true contextual implications, or warranted strengthenings or revisions of existing assumptions), the greater its relevance; the lower the input’s cognitive costs, the greater its relevance. The relevance of a stimulus increases with an increase in the effects/efforts ratio that it brings to the state the cognitive system is in. Additionally, RT assumes two other principles: Cognitive Principle of Relevance: Human cognition is geared towards the maximisation of Relevance, that is, to the achievement of as many positive contextual (cognitive) effects for as little processing effort as possible (Sperber and Wilson, 1995, 260–266).

294 | Thiago N Galery

Communicative Principle of Relevance: Every ostensive act communicates (i.e. makes mutually manifest) the presumption of its own optimal relevance (Sperber and Wilson, 1995, 260–266) The presumption of the optimal relevance of a a stimulus is captured under the idea that the speaker’s ostensive (communicatively overt) act is (i) relevant enough to be worth processing and (ii) the most relevant one compatible with communicators’ abilities and preferences. Of course, the hearer may have presumptions of relevance that go beyond that of mere optimal relevance. For example, if a question is asked, it is presumed that some form of answer will be given, either explicitly or implicitly. From these principles and definitions, a relevance-theoretic comprehension strategy follows: test your hypothesis in terms of accessibility; stop when your expectations of relevance are satisfied. At this point, we have two questions to answer. First, what benefit does a RTbased pragmatic explanation offer (in comparison to Gricean ones)? Second, how does it pave the way for an account for the data highlighted in Section 2? As an answer to the first question, one could argue that RT is better-suited for explaining deferred pronouns because pragmatically inferred meanings can contribute to an utterance’s truth conditions and potentially enter into semantic relations with other truth-conditional elements in the utterance (like antecedentpronoun dependencies). Moreover, moving from a distinct set of principles on conventional meaning and principles of rational conversation to a unified principle regulating the flow of information within a cognitive system, allows us to explain how the identification of an object in the context can give rise to relevant information that is recruited in the construction of the interpretation of the pronoun. In addition to answering our second question, this point can be used to illustrate how this pragmatic framework can explain the property of Identification Dependency⁵. Reconsider the example in (4), reprinted below. (5) In Washington, Andy and Martha are discussing whether the American government was justified in killing Osama Bin Laden (Galery, 2012): a. Martha: Do you think a democratic government has the right to kill an unarmed man, even if he’s a declared enemy of the state? b. Andy: Well … Every time there is a war, he [pointing to the White House] has tough choices to make.

5 I acknowledge that the explanation given here is a bit sketchy. For reasons of space, I cannot elaborate more, see Galery (2012), chapters 2 and 5 for a richer account.

Deferred pronouns: A sketch based on Relevance Theory and Dynamic Syntax

|

295

The idea here is that the gesture towards the White House is a complementary act to the utterance of the sentence. Like any ostensive act, it is governed by the Cognitive and Communicative principles mentioned above and the relevance-theoretic comprehension procedure. In short, I propose that the understanding of (5b) involves inferred representations such as: (6) The pragmatics of object representation a. i (bare object representation, i.e. visual index) [Low-level principles governing object individuation]. b. White House [conceptual representation that the visual index in (a) gives rise to, given the assumption that conceptual representations carry additional cognitive effects (Cognitive Principle of Relevance)]. c. The American President works at the White House [propositions salient by activation-spread from the concept in (b) and top-down influences arising from current expectation of relevance (Cog. and Com. principles of relevance) Identification Dependency is met]. The schema above is just a taster for a more developed explanation of how the identification of an object might give rise to more complex representations (as in (6c)) that can be used for further linguistic processing (to be presented in Section 5). It assumes that at some level of processing there might be bare preconceptual representations of an object; that is, a representation that identifies and individuates an object without any descriptive information about it (see Pylyshyn 1989, 2001 for the notion of visual index or FINST), as indicated by (6a). Conceptual representations would emerge at a later stage (6b). In the same way that the Cognitive Principle of Relevance regulates the derivation of explicit or implicit meaning, it can offer a way in which representations from one level (pre-conceptual) would incrementally evolve into full-blown conceptual representations⁶. The same principle would also underpin the emergence of contextual assumptions associated with that concept, i.e. (6c), in a similar way to the mechanism behind ad hoc concept construction (Wilson and Carston, 2006, 2007). The gist of my proposal uses the information in (6c), namely, the proposition that the American President works at the White House, as a form of implicit antecedent for the pronoun in ‘Every time there is a war, he has tough choices to make.’, uttered by Andy in (5b). Thus, the pronoun would be interpreted as ‘the president’ via some sort of anaphoric process between these two representations.

6 For a full argument for the proposal sketched here, please see (Galery, 2012, ch. 3).

296 | Thiago N Galery

For this approach to get off the ground, we need a model of linguistic meaning that allows such inferred constituents to participate in the building of truthconditions and a mechanism capable of establishing structural dependencies with these constituents. This is the subject of the next section.

4 Pronouns, underdetermination and grammar 4.1 Dynamic Syntax: an introduction Now that we have provided an explanation for Identification Dependency, let’s take a look at the grammar formalism that could explain Co-variation and, as a consequence Descriptiveness. According to the concluding paragraph in the last section, the salient object in the context would provide some form of implicit antecedent for the pronominal occurrence in the utterance. In this section, I will introduce a grammar formalism capable of explaining this relationship, namely, Dynamic Syntax (Kempson et al. 2001; Cann et al. 2005: DS, for short). DS models pronouns as expressions with underdeterminate meaning, that is, expressions whose value is determined by some other available representations. Moreover, the level of representation to which the formalism delivers its outputs corresponds to the level of the language of thought (see Fodor 1975), as opposed to shallow linguistic form. Linguistic meaning corresponds to certain procedures that are capable of building thought constituents in a structured way. As we shall see, this allows grammatical processes, such as anaphora and ellipsis, to operate at levels of representation that have been enriched by pragmatic processes⁷. More precisely, DS assumes that utterance comprehension can be understood as the monotonic construction of a goal tree based on incoming linguistic stimuli. Such modelling requires some basic ingredients. First, the fact that linguistic representations are hierarchically structured is described by the Logic of Finite Trees (LOFT, Blackburn and Meyer-Viol 1994): a modal vocabulary that describes the relations amongst the various nodes in a tree (points in logical space) via an accessibility relation (represented by the symbols , ).

7 Although these assumptions might align DS with some proposals within Discourse Representation Theory (Kamp, 1981; Kamp and Reyle, 1993), DS assumes a unique level of representation, while DRT might be seen as a semantics that is built on top of a theory of syntax (usually defined along Chomskian lines).

Deferred pronouns: A sketch based on Relevance Theory and Dynamic Syntax

|

297

Second, information that holds at a node, such as truth-conditional contribution or syntactic features is captured by declarative units (DU) that ‘annotate’ or ‘decorate’ that node. These ingredients constitute the declarative part of the framework. Third, linguistic understanding is modelled as mappings from partial trees to further developed trees. Such mappings are regulated by three factors: (i) requirements, i.e. the fact that certain information must hold at a node which effectively promotes tree-building (represented by a ‘?’ preceding the required information), (ii) underspecified information (a node having unfixed position or underdeterminate content), and (iii) grammatical rules that allow for transitions between (partial) trees (such as rules that remove a requirement once it is met). These ingredients constitute the procedural part of the language. It is worth noting that all decorations are described by a Labelled Deductive System in the sense of Gabbay (1996). This allows rules to target information about content, syntactic position, and logical type in a unified way. Finally, the combinatorial possibilities of representations in this framework are expressed in terms of the epsilon calculus built on the lambda calculus (necessary to account for some cases of E-type anaphora: a fundamental ingredient in the explanation of Co-variation). In DS, binary trees are rather simple. In addition to the root node, tree-nodes can be either arguments, always represented by the daughter node at the left-hand side, or functors, always represented by the daughter at the right-hand side. The construction of semantic representations starts with the requirement to build a semantic object of type t; formally expressed by a preceding ‘?’. Tn describes the position of the tree node (Tn(0) is the root node, a preceding 0 indicates an argument node below the mother (e.g. Tn(00)), and a preceding 1 indicates a functor node (e.g. Tn(01)). The process is recursive. Ty describes a semantic type; e (individuals), t (truth-values), cn (common nouns) and the functional categories built from these. ⋄ indicates the point of the parse. In this way, DS not only models a structure underpinning our linguistic production and comprehension, but it also models the procedures behind it. Finally, indicate the (LOFT) accessibility relations between tree nodes. The up-arrow symbol describes something that holds or is required at a node above the current one (mother node), whilst the down-arrow symbol does the same for a node below the current one (daughter node). The arrow modalities are usually accompanied by a subscript 0, 1 or *, which respectively mean up or down argument, functor, or any type of node. With the pointer at the root type-t-requiring-node, lexical and computational actions can develop this partially developed tree into a more specified one (note that the application of all computational rules is optional). To begin with, the computational rule Introduction splits the goal to build a proposition of type t into the

298 | Thiago N Galery sub-goals of building a subject node of type e and a predicate node of type e → t. Afterwards, the computational rule of Prediction moves the pointer to the subject node. These operations are schematically represented from (7a) to (7c) below (the definitions of all computational rules are in a appendix): (7) a. Tn(0), ?Ty(t), ⋄ b. Tn(0), ?Ty(t), ⋄

Tn(00), ?Ty(e) c.

Tn(01), ?Ty(e → t)

Tn(0), ?Ty(t)

Tn(00), ?Ty(e), ⋄

Tn(01), ?Ty(e → t)

As an illustration, parsing an utterance of ‘John smokes’ proceeds by building certain formula values of the subject node based on the actions encoded by the proper name (I’m simplifying the treatment of proper names here for explanatory purposes). These are usually stated in terms of if …, then … else … statements and contain (i) a trigger (a condition that must be met), (ii) a set of actions (executed upon the satisfaction of the conditions), and (iii) an else statement (exit condition if (i) isn’t met). For example: (8) Lexical actions for ‘John’ IF ?Ty(e) THEN put Ty(e), Fo(John󸀠 ) ELSE abort

Trigger Actions Else Statement

The entry in (8) makes the transition from (7c) to (I’m ignoring Tn decorations for simplicity’s sake from now on): (9)

?Ty(t)

?Ty(e), Ty(e), Fo(John 󸀠 ), ⋄

?Ty(e → t)

In short, understanding ‘John’ allows the system to decorate the subject node with information about a semantic type Ty(e) and a formula value Fo(John󸀠 ). At this point, the computational rule Thinning applies. It states that when we have a re-

Deferred pronouns: A sketch based on Relevance Theory and Dynamic Syntax

|

299

quirement and an equivalent type statement decorating the same node (i.e. both ?Ty(e) and Ty(e)), the requirement can be discharged. After Thinning, Completion applies. This rule allows the pointer to move from the type-complete daughter node to the mother node and annotate the type information there. Afterwards, the computational rule of Anticipation can apply. This rule allows the pointer to move to daughter nodes with outstanding requirements, in this case, the predicate node annotated with ?Ty(e → t). At this stage, ‘smokes’ can be parsed. Its lexical entry can be cashed out along similar lines as (8), the main difference is that the lexical trigger for the verb is a ?Ty(e → t) requirement and that the contribution it makes to the tree is a λx.Smoke(x)󸀠 concept (I’m abstracting away from tense, see Cann 2011 for a more detailed account). (10)

?Ty(t)

Ty(e), Fo(John󸀠 )

?Ty(e → t), Ty(e → t), Fo(λx.Smoke󸀠 (x)), ⋄

A subsequent use of Thinning removes the requirement at the predicate node. Completion moves the pointer to the mother (root) node, annotating it with the type-complete information from the predicate node. At this point, the rule of Elimination can apply. The rule basically applies modus ponens over the types Ty(e) and Ty(e → t) and annotates the resulting Ty(t) information. Then, Thinning can remove all completed requirements, indicating the successful build-up of the tree. (11)

Ty(t), Fo(Smoke 󸀠 (John󸀠 )), ⋄

Ty(e), Fo(John󸀠 )

Ty(e → t), Fo(λx.Smoke󸀠 (x))

4.2 Pronouns and Underdetermination Tree development can also exploit lexical underdetermination. Pronouns are the paradigmatic case: their contribution varies according to the context in which they are used. DS captures this phenomenon by postulating that these expressions introduce metavariables (represented by bold-face capital letters, e.g., U, which differ from the object-level variables bound by quantifiers). In effect, metavariables provide substitution sites for available formula values from different

300 | Thiago N Galery

cognitive sources (e.g. from a local tree, or a inferred or linguistically derived LINKed tree – to be presented below). Parsing an utterance of ‘He is stressed’ proceeds similarly to the previous case. After the application of Introduction and Prediction (or any other means for introducing the subject node, see Cann 2011 for an alternative), the pronoun ‘He’ can be parsed. Its lexical entry is as follows (abstracting away from case restrictions). (12) Lexical Actions for ‘he’ IF ?Ty(e) THEN put Ty(e), Fo(UMale󸀠 ), ∃x.Fo(x), [↓]⊥ ELSE abort As the lexical rule above indicates, the parse of the pronoun results in a typecomplete annotation with a metavariable which presupposes that its substituens must be a proper formula value (without requirements) that ranges over male entities. Still, with the type Ty(e) information at that node, Completion can apply and Anticipation would move the pointer to the predicate node, at which point the verb can be parsed. Similarly to ‘smoke’, I assume, for the moment, that ‘stressed’ is an intransitive verb that takes only one argument (abstracting away from tense). The resulting representation would look like this. (13)

?Ty(t)⋄

Ty(e), Fo(U), ?∃x.Fo(x), [↓]⊥

Ty(e → t), Fo(λx.Stresses󸀠 (x))

In practical terms, the parse cannot be completed unless we eliminate the requirements in all nodes, and for this to be the case, we need to find a proper value for the metavariable U. This process, which DS identifies as Substitution (similar to the notion of saturation in Recanati 1993), is not governed by grammar and thus becomes a matter of pragmatics. Antecedents and other types of dependencies can provide constraints or helpers for determining a possible range of values. One way to model such constraints in DS is the LINK construction.

4.3 LINK: Inter-semantic relations At this moment, we are in a position to better explain the relationship between pronouns and other formula values. For example, consider the utterance ‘John likes himself’. In this case, we know that the pronoun ‘him’ introduces a metavari-

Deferred pronouns: A sketch based on Relevance Theory and Dynamic Syntax

|

301

able in the structured representation of the utterance. In principle, it could be replaced by any formula value, but the presence of the self anaphor instructs the system to retrieve the formula value Type(e) node (mirroring the notion of local c-command using the or accessibility relations). Now in cases of intersentential anaphora, e.g. ‘John smokes. He is stressed.’, the two sentences generate two distinct tree structures, precluding our modal vocabulary to represent the dependency between ‘John’ and ‘He’. In order to explain these dependencies, DS uses another structural relationship, namely, LINK structures. This notion is especially important for my account of deferred pronouns because it provides a model for cross-sentential anaphora, and as I mentioned before, I take deferred pronouns to be an implicit form of this type of pronominal dependency. Like the arrows, the LINK relation exploits a modality (although a different one) such that the node from which it originates provide structural constraints on further interpretation (tree-building). Intuitively, LINKed trees are structured representations connected by the sharing of terms (either standard argument nodes or situation/eventuality nodes, see Gregoromichelaki 2006). Moreover, a set of LINKed trees can be seen as an inferential unit in the sense of Carston (2002) and thus provides a contextual frame for the processing of further linguistic material and for inference. From the link mother node, Tn(m) expresses a modal accessibility relation over trees (as opposed to tree-nodes) that reads ‘tree node Tn(m) can be found one step across the LINK structure’. In more practical terms, the piece of discourse in (14) has the following tree structure associated with it (abstracting away from tense): (14) John smokes. He is stressed. Tn(m), Ty(t), Fo(Smokes󸀠 (John󸀠 ))

Tn(n), Ty(e), Fo(John󸀠 )

Ty(e → t), Fo(λx.Smokes󸀠 (x))

Ty(t), Stresses󸀠 (John󸀠 ), ? Fo(John)

Tn(m), Ty(e), Fo(John󸀠 )

Ty(e → t), Fo(λy.Stresses󸀠 (y))

LINKs can be created from any Ty(e) node (the LINK ‘mother’) via the LINK Adjunction rule (see appendix). The rule also requires that the formula value at the node from which the LINK originates to be found somewhere in newly created tree structure (the LINK ‘daughter’). Thus, the representation above can be explained

302 | Thiago N Galery

by introducing a LINK relation at the subject of ‘John smokes’, which requires the LINK daughter to have the same formula value. When ‘He is stressed’ is parsed, the pronoun’s metavariable U is replaced by Fo(John)󸀠 , satisfying requirements imposed by the pronoun (a proper formula value must be found) and by the LINK relation (the formula value from mother must be found at a LINK daughter). After the LINK is established, computational rules (LINK evaluation) can combine the formula values of the type t root node of the two trees (conjunction introduction).

4.4 Quantification and Co-variation So far we have only covered cases where the expressions are simple constants referring to particular entities in the world. However, one crucial datum to be explained is the covariation between expressions in a sentence or discourse. So far, the examples which highlighted this property have contained more complex expressions, like quantifiers such as ‘always’ in ‘He [pointing to St. Peter’s basilica] is always Italian’ (e.g. example (3)). I would like to make a few remarks on the picture of quantification adopted by DS, so that when the time comes, we can put all the pieces together. DS adopts a model of quantification based on epsilon calculus (Hilbert and Bernays, 1939), thus assuming that quantifiers are type Ty(e) expressions. This can be understood as a witness of the set relations imposed by the quantifier. Although simple with regards to their logico-syntactic information, quantified terms are complex terms in their semantics. They are structured by an epsilon term representing the quantifier type (τ for universals or ϵ for existentials), a common noun Ty(cn) acting as a restrictor, and a bound variable. For example, ‘Every man’ would correspond to the epsilon term τ, x, Man󸀠 (x). Finally, epsilon terms are associated with evaluation rules that unpack their meanings so that the right set relations are captured. In particular, the epsilon calculus preserves the following equivalences: (15) Epsilon Calculus and Predicate Calculus equivalences: a. ∀xFx ↔ F(τ, x, F(x)) b. ∃xFx ↔ F(ϵ, x, F(x)) Like the linguistic processing of any expression, the parsing of quantifiers, such as, ‘every’, ‘some’, or ‘a’, trigger lexical actions that build a subject node containing a quantifier head (a τ or ϵ term), the quantifier’s bound variable (e.g. x), and

Deferred pronouns: A sketch based on Relevance Theory and Dynamic Syntax

|

303

also a node for the parsing of the restrictor’s lexical material. Here is a tentative representation of the lexical entry for ‘every’. (16) Lexical actions for ‘every’ IF ?Ty(e) THEN put(Indef(−)); make(); go(); put(Fo(λP(τ, P)); Ty(cn → e) go(); make(); go(); put(?Ty(cn)), ELSE abort In sum, the lexical actions of ‘every‘ create two nodes from a type Ty(e) subject node: an argument node of type Ty(cn) (the nominal node) and a functor node of type Ty(cn → e) (the determiner node). With the pointer at the nominal node, the restrictor can be parsed; let us suppose the restrictor is the noun ‘man’. Here are the associated lexical rules: (17) Lexical actions for ‘man’ IF ?Ty(cn) THEN make(); go(); put(Fo(λx.(man, x)); Ty(e → cn), go(); make(); go(); freshput(x), Fo(x); go() put(Sc(x)); go() ELSE abort In short, the restrictor builds two nodes from the nominal type Ty(cn): an argument type Ty(e) node containing a bare variable x (the variable node) and a functor node of type Ty(cn → e) (the restrictor node). Most importantly, these lexical actions introduce a scope statement about the bound variable in the root of the tree (e.g. Sc(x)). This statement means that the scope of the variable must be determined in order for the tree-structure to be complete. If the utterance contained another quantifier, its variable would be added to the statement, for example, Sc(x < y), informing that the first variable (x) has precedence over the second (y). The presence of the scope statement allows a computational rule, namely, Quantifier Evaluation, which removes the variable from the scope statement and introduces the appropriate connective for each quantifier: G(τ, x, F(x)) unpacks as F(a) → G(a) and G(ϵ, x, F(x)) unpacks as F(b) ∧ G(b), where a and b are complex terms that stand for a witness of the relevant set relations. Let us consider the parse of an utterance of ‘A man sleeps’. After all lexical actions take place, the root of the tree is decorated by: Ty(t), Fo(Sleeps󸀠 (ϵ, x, Man󸀠 (x))), Sc(x). The presence of variables in the scope statement (i.e. Sc(x)) trig-

304 | Thiago N Galery

gers the application of the Q-Evaluation, which allows for the following substitutions: (18) Quantifier set relations and witness terms a. f τxψ (ϕ) = ψ[a/x] → ϕ[a/x], where a = τx(ψ → ϕ) b. f ϵxψ (ϕ) = ψ[b/x] ∧ ϕ[b/x], where b = ϵx(ψ ∧ ϕ) c. f Qxψ (ϕ) = (ψ[c/x])(ϕ[c/x]), where c = v Q x((ψ)(ϕ)) d. f w[x1 ] (ϕ) = w[x1 ] : ϕ In practical terms, the unpacking of quantifier terms follows the following recipe: (19) a. First, find the scope statement containing the variable of interest, the restrictor of that variable, i.e. Man󸀠 (x): represented by ψ in the rule, and the nuclear scope of the quantifier, i.e. Sleep󸀠 (x): represented by ϕ in the rule. b. Second, find the corresponding rule for the binder; i.e. (f ϵxψ (ϕ) = ψ[b/x]∧ ϕ[b/x], where b = ϵx(ψ ∧ ϕ)) c. Third, do what the rule tells you: (i) Construe an arbitrary witness for the restrictor: b = ϵ, x, (Man󸀠 (x) ∧ Sleeps󸀠 (x)) (ii) Relate restrictor and nuclear scope using the witness term and the right connective: Fo(Man󸀠 (b) ∧ Sleeps󸀠 (b)) (iii) Remove the variable x from the scope statement (not expressed if it contains 0 variables) (iv) Finally we end up with: Ty(t), Fo(Man󸀠 (b) ∧ Sleeps󸀠 (b)), where b corresponds to the term in (i)

4.5 Situation variables The final ingredient in my proposal is the notion of situation variables. So far, we have given examples of utterances where the verb contains arguments corresponding to subjects and objects in the sentence. In fact, this has been an oversimplification. As indicated in (18d), the quantifier rules are able to manipulate special world/situation/eventuality variables. These are assumed to be of type Ty(e) (represented as Ty(e s ); the subscripts just help the reader to distinguish them from standard individuals) and predicates would be associated with an extra argument slot for such variables (i.e. intransitive verbs would be of type Ty(e s → e → t)). Thus, an utterance of ‘John smokes’ is better represented by:

Deferred pronouns: A sketch based on Relevance Theory and Dynamic Syntax

(20)

|

305

Ty(t), Fo(Smoke󸀠 (John󸀠 )(s)), ⋄

Ty(e s → t), Fo(λy.Smoke󸀠 (John󸀠 )(y))

Ty(e s ), Fo(s)

Ty(e), Fo(John)

Ty(e → e s → t), Fo(λx.λy.Smoke󸀠 (x)(y))

Since situation variables are essentially type Ty(e) expressions, any computational and/or lexical rule that targets Ty(e) nodes would naturally extend to situation nodes (e.g. LINK Adjunction and Evaluation, see Gregoromichelaki 2006 for a semantics of ‘if’ as situation relatives, for node-position underdetermination rules, see Cann 2011).

5 The proposal This is the point at which the elements introduced in previous sections will be combined into a unified account of the phenomenon. I would like to mention that I will focus on third person singular pronouns; for plurals and expressions with other person features see (Galery, 2012, ch. 5). Let us reconsider the example in (4), reprinted below: (21) In Washington, Andy and Martha are discussing whether the American government was justified in killing Osama Bin Laden (Galery, 2012): a. Martha: Do you think a democratic government has the right to kill an unarmed man, even if he’s a declared enemy of the state? b. Andy: Well … Every time there is a war, he [pointing to the White House] has tough choices to make. As suggested in Section 3, I take the utterance above to involve a double ostensive act. There is the gesture towards the building, and there is the utterance itself. On the pragmatic framework I am assuming, the hearer retrieves information from the actions of the speaker on the basis of the cognitive principles of relevance, the definition(s) of (optimal) relevance and the comprehension strategy that follows from them. Thus, the gesture towards the White House would involve incremental processes of representation that look like:

306 | Thiago N Galery

(22) a.

i (bare object representation, i.e. visual index) [Low level principles governing object-individuation]. b. White House [conceptual representation that the visual index in (a) gives rise to, given the assumption that conceptual representations carry additional cognitive effects (Cognitive Principle of Relevance)]. c. The American President works at the White House [propositions salient by activation-spread from the concept in (b) and top-down influences arising from current expectation of relevance (Cog. and Com. principles of relevance) Identification Dependency is met].

Previously, we discussed how the hearer’s expectations of relevance can deviate from those of optimal relevance. In the example above, the interpretation of (21b) must provide an answer to the question in (21a). This expectation is responsible for selecting (22c) as a mutually manifest proposition (i.e. a contextual assumption in a set that might include other propositions). Within the grammar framework that we have described in Section 4, we could assume that the implicit proposition in (22) is structured as (abstracting away from situation variables for the moment): (23) Ty(t), Fo(WorkAt󸀠 (ϵ, x, President󸀠 (x))(ϵ, y, WhiteHouse󸀠 (y))), ⋄

Ty(e), Fo(ϵ, z, President󸀠 (z))

Ty(e → t), Fo(λy.WorkAt󸀠 (ϵ, x, President󸀠 (x))(y))

Ty(e), Fo(ϵ, y, WhiteHouse󸀠 (y))

Fo(λx.λy.WorkAt󸀠 (x)(y))

Above, we are assuming that the equivalent of ‘the president’ in the language of thought is expressed by the epsilon term ϵ, x, President󸀠 (x). Given that this representation is mutually manifest to conversational participants, it would be the antecedent for the interpretation of the pronoun ‘he’ in (21) (a value that replaces the metavariable introduced by the pronoun via Substitution). Let us now consider how the representation above feeds into the interpretation (21b). Like our toy example in Section 4, let’s assume the introduction of subject and predicate nodes is done via the rules of Introduction and Prediction. The only difference this time is that the subject node is of type Ty(e s ) (a situation node). So we end up having a representation similar to:

Deferred pronouns: A sketch based on Relevance Theory and Dynamic Syntax

(24)

|

307

?Ty(t), 0 ?Ty(e s ), 1 ?Ty(e → t)

?Ty(e s ), ⋄

?Ty(e s → t)

Following Gregoromichelaki (2006), I assume a Situation Metavariable Introduction rule, which introduces a situation metavariable whose value should be replaced by an appropriate formula value in the context (akin to the notion of saturation in Recanati 1999). The rule licenses the transition to the following tree: (25)

?Ty(t)

?Ty(e s ), Fo(S), ?Sc(S), ?∃xFo(x), ⋄

?Ty(e s → t)

This is the point at which the adverbial ‘Every time there is a war’ can be parsed. Since, the expression describes temporal aspects of the event denoted by the verb, DS models these expressions as modifiers of the situation argument. Following Kempson (2010) and Gregoromichelaki (2006, 207–221), this is done via the adjunction of LINKed trees. On this basis, Galery (2012, ch. 5) proposes a Situation Adjunction rule that creates a LINKed structure attached to the situation node. This rule enforces identity of formula values between the LINK daughter and the value of the metavariable in the Ty(e s ) node (along the lines of Gregoromichelaki’s treatment of conditionals), that is, what if-clauses do lexically, the Situation Adjunction Rule does generally and optionally (again, for the definition of the rules, see appendix and Galery 2012, ch. 5)⁸. The rule allows for the transition of the tree above to: (26)

?Ty(t)

?Ty(e s ), Fo(S), ?Sc(S)?∃x(Fo(x) ∧ (x))

?Ty(e s → t)

?Ty(e s ), ⋄

8 Ideally, adjuncts like ‘every time’ would introduce the LINK relation and impose the sharing of terms themselves, making the rule proposed here redundant.

308 | Thiago N Galery

At this point, the content encoded by ‘Every time there is a war’ contributes to the LINKed tree. The quantifier head ‘every’ makes the transition from a type e requiring node to a tree structure where this node has a type cn s → e s daughter with formula value λP.τ, P (the determiner node) and type cn s requiring node (the nominal node). For the sake of simplicity, I will omit operations evoked by this complex restrictor and treat the whole expression as deriving Fo(s, Time󸀠 s∧War󸀠 s (i.e. Every war-time), after the application of relevant rules (see Galery 2012, ch. 5 for details). We have: (27)

?Ty(t)

?Ty(e s ), Fo(S), ?Sc(S)?∃x(Fo(x) ∧ (x))

?Ty(e s → t)

Ty(e s ), Fo(τ, s, Time󸀠 ∧ War󸀠 s), ⋄

Ty(cn s ), Fo(s, Time󸀠 ∧ War󸀠 s)

Ty(e s ), Fo(s)

Ty(cn s → e s ), Fo(λP.τ, P)

Ty(e s → cn s ), Fo(λz.z, Time󸀠 ∧ War󸀠 z)

The LINK Evaluation rule then copies the formula value of the LINKed daughter up to the situation node (the LINK mother) and inserts an underdetermined scope statement about the variable of interest (i.e. Sc(s)) at the root of that tree (see Gregoromichelaki 2006, 217). Then, Anticipation moves the pointer to the ?Ty(e s → t) node, where new argument and functor nodes can be created. (28)

?Ty(t), Sc(s)

?Ty(e s ), Fo(τ, s, Time󸀠 ∧ War󸀠 s)

Ty(e s ), Fo(τ, s, Time󸀠 ∧ War󸀠 s), ⋄

Ty(cn s ), Fo(s, Time󸀠 ∧ War󸀠 s)

Ty(e s ), Fo(s)

?Ty(e s → t)

?Ty(e), ⋄

Ty(cn s → e s ), Fo(λP.τ, P)

Ty(e s → cn s ), Fo(λz.z, Time󸀠 ∧ War󸀠 z)

?Ty(e s → e → t)

Deferred pronouns: A sketch based on Relevance Theory and Dynamic Syntax

|

309

With the pointer at the subject type e node, the lexical actions associated with the pronoun ‘he’ can be parsed. The pronoun introduces a fresh metavariable U and a requirement that it should be replaced by an appropriate formula value. At this point, I assume that the contextual assumption (e.g. that the US President works at the White House ), made mutually manifest by the speaker, establishes a LINK relation to the tree. Thus, the implicit antecedent imposes a requirement that the value Fo(ϵ, z, President󸀠 z) must be found in the structure. Since the metavariable introduced by the pronoun is compatible with this requirement, the quantificational antecedent replaces the metavariable U⁹. For space reasons, I will not present this (yet another) LINKed tree and instead represent the term that derives from it in underline. We now have: (29)

?Ty(t), Sc(s)

?Ty(e s ), Fo(τ, s, Time󸀠 ∧ War󸀠 s)

?Ty(e s → t)

Ty(e), Fo(ϵ, z, President󸀠 z)

?Ty(e s → e → t), ⋄

Ty(e s ), Fo(τ, s, Time󸀠 ∧ War󸀠 s)

Ty(cn s ), Fo(s, Time󸀠 ∧ War󸀠 s)

Ty(e s ), Fo(s)

Ty(cn s → e s ), Fo(λP.τ, P)

Ty(e s → cn s ), Fo(λz.z, Time󸀠 ∧ War󸀠 z)

Finally, the pointer can move to the predicate node (via Anticipation), where information about the predicate ‘has tough choices to make’ can be parsed. For the sake of simplicity, I will treat this predicate as a term whose formula value is Fo(λt.λ.vMTC󸀠 tv) (where MTC stands for making tough choices). Although this might be a brute-force simplification, a full account of this predicate is not necessary for the explanation of the data in this paper. After the predicate has been parsed, Thinning, Elimination and Completion compile the remainder of the tree.

9 As an alternative, the epsilon term could be constructed by the re-use of actions that were involved in the build-up of the contextual assumptions. This account would be more in the spirit of Elbourne’s NP deletion account for pronouns (Elbourne, 2005), but I cannot present it here

310 | Thiago N Galery Ty(t), Sc(s), Fo(MTC󸀠 ((ϵ, z, President󸀠 z)(τ, s, Time󸀠 ∧ War󸀠 s))), ⋄

(30)

Ty(e s ), Fo(τ, s, Time󸀠 ∧ War󸀠 s)

Ty(e s → t), Fo(λ.vMTC󸀠 (ϵ, z, President󸀠 z)v)

Fo(ϵ, z, President󸀠 z)

Fo(λt.λ.vMTC󸀠 tv)

Ty(e s ), Fo(τ, s, Time󸀠 ∧ War󸀠 s)

Ty(cn s ), Fo(s, Time󸀠 ∧ War󸀠 s)

Ty(e s ), Fo(s)

Ty(cn s → e s ), Fo(λP.τ, P)

Ty(e s → cn s ), Fo(λz.z, Time󸀠 ∧ War󸀠 z)

The presence of the variable s in the scope statement at the root triggers a Scope Evaluation rule (see appendix) that relates the term’s restrictor to its nuclear scope via a logical connective. Before we move on, let us identify the formula value we need to unpack, which corresponds to the expression below (underline represents term from implicit antecedent): (31) Fo(MTC󸀠 ((ϵ, z, President󸀠 z)(τ, s, Time󸀠 ∧ War󸀠 s))) The formula above states that the predicate making tough choices applies to two quantified arguments. Since the root contains only one variable in the scope statement¹⁰ (Sc(s)) and this variable is bound by a universal τ term, most of the explanation for Co-variation will be achieved by unpacking the universal quantifier over situations. If we take a look at the procedure illustrated in the section on quantification (see (18)), the first step in evaluating the scope statement is to replace the complex quantifier terms by witness terms. The formula in (31) above would then be translated as (underline represents term from implicit antecedent): 10 At this point, we could discuss whether the fact the the implicit antecedent that replaces the metavariable of the pronoun can insert a variable in the scope statement of the tree structure above. Regardless of the position one takes on this issue, the situation variable s would outscope any variable inserted after it (except the ones inserted by indefinites, but this is irrelevant, since the antecedent is implicit and doesn’t carry lexical indefinite features). For this reason, I will omit representing other variables within the scope statement, and explain Co-Variation by the unpacking of the tau term that binds the variable s.

Deferred pronouns: A sketch based on Relevance Theory and Dynamic Syntax

|

311

(32) Fo(MTC󸀠 ((b)(a))) Despite its apparent simplicity, the terms a and b express complex set relations. Let’s take a look at the existential term b, which originates from (ϵ, z, President󸀠 z). Within the context of the original term, we can identify the restrictor as x, President󸀠 (x) and the nuclear scope is MTC󸀠 ((x), (a)). Given that the rule in (18b), relates existential terms by the introduction of a conjunction, the witness term b is equivalent to ϵ, x, [President󸀠 (x) ∧ MTC󸀠 ((x), (a))]. Instead of using this complex expression, I will stick to the simpler term b, but this is the way in which epsilon terms reflect their surrounding environment. Now, let’s move on the universal term a, which originates from the universal (τ, s, Time󸀠 ∧ War󸀠 (s)). It corresponds to a term that binds a variable in the scope statement (Sc(s)). The scope statement is motivated by ordering the variables which are inserted by determiners in a utterance. For example, Sc(x < y) means that x has wide scope over y. In practical terms, this means that the quantifier evaluation rules apply from right to left, so whatever rules that unpack the restrictor and nuclear scope of y (the variable would then be removed from the scope statement Sc(x < y)) will generate complex terms that will then enter into set relations determined by the rules that manipulate the (quite complex) restrictor and nuclear scope of the wide-scoping variable x. Walking though the steps of the quantification rule, let us break down the formula that includes the witness terms, namely, Fo(MTC󸀠 (b))(a)), into sub-parts. We know that a stands for (τ, s, Time󸀠 ∧ War󸀠 s) and it is in this more complex term that we can find the restrictor, namely, s, Time󸀠 ∧ War󸀠 s. The nuclear scope surrounds the witness term and corresponds to MTC󸀠 (b)(s). Since tau terms introduce material implications as part of their definitions, in the application of the rule in (18), a corresponds to τ, s, [WarTime󸀠 (s) → MTC󸀠 (b, s))]. The application of the quantifier evaluation rule removes s from the scope statement and leaves the following decorations at the root node (note that a and b correspond to more complex terms discussed above): (33) Sc(s), Fo(WarTime󸀠 (a) → MTC󸀠 (b, a)) This formula states that for any wartime situation a, that situation is one in which an arbitrary individual b has to make tough choices. Obviously, b itself denotes a set, namely the non-empty set of American Presidents who make tough choices, thus requiring that whoever makes tough choices in situation a must also be a US president in that situation. This explains the relation between wartimes and Presidents at those times, which the utterance is able to convey. Since there is

312 | Thiago N Galery

no other variable in the scope statement, the formula value corresponding to the utterance content cannot be unpacked further. The parse is complete.

6 Conclusion It seems that the account just sketched is able to formally explain the final property of the data mentioned initially: Co-variation, and by consequence one of the forms in which expressions in natural language (or thought) can be said to be descriptive (Descriptiveness). Given that descriptive material that figures as an explanation emerged through a process of inference and the incremental extraction of information from the identification of an individual in the environment, I take it that Identification Dependency is also accounted for. I believe that this explanatory sketch based on Relevance Theory and Dynamic Syntax offers great explanatory gain when it comes to phenomena in the grammar/pragmatics interface and therefore such theories should be considered as real contenders when it comes to explaining phenomena involving relations between grammatical and extra-linguistic processes in language comprehension. Acknowledgment: I am very thankful for comments from Robyn Carston, Ruth Kempson and Eleni Gregoromichelaki, who read many iterations of the chapter of my PhD thesis on which this is based. I’m also very thankful to the audience at the conference ‘The impact of pronominal form on interpretation’, specially Patrick G. Grosz, Pritty Patel-Grosz, and Paul Elbourne. Thanks to Matthew Reeve for suggestions and comments on a draft.

References Blackburn, P. and Meyer-Viol, W. 1994. Linguistics, Logic and Finite Trees. Logic Journal of IGPL, 2(1):3–29. Borg, E. 2004. Minimal Semantics. OUP, Oxford. Cann, R. 2011. Towards an Account of the English Auxiliary System. In Kempson, R., Gregoromichelaki, E., and Howes, C., editors, The Dynamics of Lexical Interfaces, pages 279–318. CSLI. Cann, R., Kempson, R., and Marten, L. 2005. The Dynamics of Language. Elsevier. Carston, R. 2002. Thoughts and Utterances: The Pragmatics of Explicit Communication. Blackwell. Chomsky, N. 1986. Knowledge of language: Its nature, origin, and use. Praeger Publishers. Elbourne, P. 2005. Situations and Individuals. MIT Press.

Deferred pronouns: A sketch based on Relevance Theory and Dynamic Syntax

|

313

Elbourne, P. 2008. Demonstratives as individual concepts. Linguistics and philosophy, 31(4):409–466. Evans, G. 1982. The Varieties of Reference. Oxford University Press, USA. Fodor, J. A. 1975. The Language of Thought. Harvard University Press. Fodor, J. A. 1983. The modularity of mind: An essay on faculty psychology. MIT Press. Gabbay, D. 1996. Labelled Deductive Systems. Oxford University Press. Galery, T. 2012. Descriptive Pronouns Revisited. PhD thesis. Gregoromichelaki, E. 2006. Conditionals in Dynamic Syntax. PhD thesis, King’s College London, London. Grice, H. P. 1975. Logic and Conversation. Syntax and Semantics, 3: Speech Acts:41–58. Hilbert, D. and Bernays, P. 1939. Die Grundlagen der Mathematik II. Springer, New York. Second Edition, 1970 reprint. Kamp, H. 1981. A Theory of Truth and Semantic Representation. In Groenendijk, J. A., Janssen, T., and Stokhof, M., editors, Formal Methods in the Study of Language, pages 277–322. Foris, Dordrecht. Kamp, H. and Reyle, U. 1993. From Discourse to Logic: an Introduction to Modeltheoretic Semantics in Natural Language, Formal Logic and Discourse Representation Theory, volume 1. Kluwer Academic Publishers, Dordrecht. Kempson, R. 2010. Language as Systems for On-line Concept Construction. Talk given at the CSMN workshop on word meaning. Kempson, R., Meyer-Viol, W., and Gabbay, D. 2001. Dynamic Syntax: The Flow of Language Understanding. Wiley-Blackwell. Lewis, D. 1975. Adverbs of quantification. In Portner, P. and Partee, B., editors, Formal semantics: the essential readings, pages 127–147. Blackwell, Oxford. Nunberg, G. 1993. Indexicality and Deixis. Linguistics and Philosophy, 16(1):1–43. Powell, G. 2003. Language, Thought and Reference. PhD thesis, University College London, London. Pylyshyn, Z. 1989. The Role of Location Indexes in Spatial Perception: A Sketch of the FINST Spatial-Index Model. Cognition, 32(1):65–97. Pylyshyn, Z. 2001. Visual indexes, preconceptual objects, and situated vision. Cognition, 80(1-2):127–158. Recanati, F. 1993. Direct Reference: From Language to Thought. Blackwell, Oxford. Recanati, F. 1999. Situations and the Structure of Content. In Murasugi, K. and Stainton, R., editors, Philosophy and Linguistics, pages 113–165. Westview Press, Boulder. Russell, B. 1910. Knowledge by Acquaintance and Knowledge by description. Proceedings of the Aristotelian Society, 11:108–128. Schiffer, S. 1981. Indexicals and the Theory of Reference. Synthese, 49(1):43–100. Sperber, D. and Wilson, D. 1995. Relevance: Communication and Cognition. Blackwell, 2nd edition. Stokke, A. 2010. Indexicality and Presupposition. PhD thesis, University of St. Andrews, St. Andrews. Wilson, D. and Carston, R. 2006. Metaphor, relevance and the ‘emergent property’ issue. Mind & Language, 21(3):404–433. Wilson, D. and Carston, R. 2007. A unitary approach to lexical pragmatics: relevance, inference and ad hoc concepts. In Burton-Roberts, N., editor, Pragmatics, pages 230–259. Palgrave.

314 | Thiago N Galery

7 Appendix: Computational rules Introduction {. . . {. . .?Ty(Y) . . . , ⋄} . . . } {. . . {. . .?Ty(Y),? Ty(X), ? Ty(X → Y) . . . , ⋄}, . . . } Prediction {. . . {Tn(n), . . .? X, ? Y, ⋄} . . . } {{Tn(n), . . .? Ty(X), ? Y}, { Tn(n), ?X, ⋄}{ Tn(n)?Y}} Thinning {. . . X, . . .?X, ⋅ ⋅ ⋅ ⋄} {. . . X, . . . , ⋄} Completion {. . . {Tn(n) . . . }, { Tn(n), . . . Ty(X), ⋅ ⋅ ⋅ ⋄}} {{Tn(n), . . . Ty(X), . . . , ⋄}, { Tn(n), . . . Ty(X), . . .}} where i ∈ {0, 1, ∗} Anticipation {{Tn(n) . . . , ⋄}, { Tn(n), . . .?X}} {{Tn(n) . . . }, { Tn(n), . . .?X ⋅ ⋅ ⋅ ⋄}} Elimination { (Fo(a), Ty(X)), (Fo(b), Ty(X → Y)) . . . , ⋄} {. . . {Ty(Y), Fo(b(a)), (Fo(a), Ty(X)), (Fo(Y), Ty(X → Y)) . . . , ⋄}} Star Adjunction {{Tn(a) . . . , ?Ty(t), ⋄}} {{Tn(a) . . . , ?Ty(t),}{ Tn(a) . . . , ∃x.Tn(x), ?Ty(e), ⋄}} Merge {ND . . . , ND󸀠 . . . , } {ND ⊔ ND󸀠 } where ⋄ ∈ ND󸀠 and ND ∪ ND󸀠 is compatible. Local Star Adjunction {{Tn(n)}, . . . {Tn(m), Tn(n), Ty(X), ⋄}, . . . } {{Tn(n)}, . . . {Tn(m), Tn(n), Ty(X)},{ Tn(m), ?Ty(X), ?∃x.Tn(x), ⋄} . . . }

Deferred pronouns: A sketch based on Relevance Theory and Dynamic Syntax

|

315

Link Adjunction {{Tn(X) . . . , Ty(e), Fo(a) . . . , ⋄} . . . } {{Tn(X) . . . , Ty(e), Fo(a)}{< L −1 > Tn(X), ?Ty(t), ? Fo(a), ⋄} . . . }

Q-Evaluation Rule {. . . {Ty(t), . . . , World(w[x1 ]), Scope(x1 < . . . x n ), Fo(ϕ[vx n ψ n /x n ]), . . . }} {{Ty(t), . . . , World(w[x1 ]), Scope(x1 < . . . x n − 1), . . . , Fo(f v n x n ψ[vx n ψ n /x n ] (ϕ)), . . . }} where for x occurring free in ϕ and w[x 1 ] = a world variable x1 or w[x1 ] = v1 x1 ψ, and the values f v n x n ψ[vx n ψ n /x n ] (ϕ) for v ∈ ϵ, τ, Q and f w [x1](ψ) are defined by: a. b. c. d.

f τxψ (ϕ) = ψ[a/x] → ϕ[a/x], where a = τx(ψ → ϕ) f ϵxψ (ϕ) = ψ[b/x] ∧ ϕ[b/x], where b = ϵx(ψ ∧ ϕ) f Qxψ (ϕ) = (ψ[c/x])(ϕ[c/x]), where c = v Q x((ψ)(ϕ)) f w[x1 ] (ϕ) = w[x1 ] : ϕ

Situation Metavariable Insertion Rule {. . . Tn(n), ?Ty(t), }, { Tn(n), ?Ty(e s ), ⋄}, { Tn(n), ?Ty(e s → t)}} . . . {. . . Tn(n), ?Ty(t), ⋄}, { Tn(n), Ty(e s ), Fo(S), ?Sc(S), ?∃xFo(x)} { Tn(n), ?Ty(e s → t)}} . . .

Situation Adjunction Rule {. . . Tn(n), ?Ty(t)}, { Tn(n), Ty(e s ), Fo(S), ?Sc(S), ?∃xFo(x), ⋄} { Tn(n), ?Ty(e s → t)}} . . . . . . Tn(n), ?Ty(t)}, { Tn(n), Ty(e s ), Fo(S), ?Sc(S), ?∃xFo((x)∧ < L > Fo(x))} { Tn(n), ?Ty(e s → t)}}, {< L−1 > Tn(n), ?Ty(e s ), ?∃xFo(x), ⋄} . . .

LINK Evaluation Rule: Conditionals {{Tn(X), Ty(t), . . . }, { Tn(X), ?Ty(e s ), ?∃x.Fo(x), . . . }, {< L > Tn(X), Ty(e s )Fo(a[x] ), ⋄}} {{Tn(X), Ty(t), Scope(U < x) . . . }, { Tn(X), ?Ty(e s ), ?∃x.Fo(x), ?Sc(x), Fo(a[x] ), ⋄}, {< L > Tn(X), Ty(e s )Fo(a[x] ) . . . }}

| Part III: Experimental insights on the syntax and semantics of pronouns

Francesca Filiaci (University of Edinburgh)

Restrictions on the interpretation of null and overt subjects in Italian and Spanish and the Morphological Ambiguity hypothesis Abstract: This paper outlines a possible cross-linguistic difference between Italian and Spanish regarding the interpretation of overt subject pronouns (Spanish él/ella and Italian lui/lei). More precisely, it presents evidence suggesting that Italian overt pronouns may be more strongly associated with reference to a non prominent antecedent than their Spanish counterparts (Filiaci et al., 2013). I will pursue the hypothesis that such difference may be related to the relative ambiguity of the person agreement morphology on the verbal head. The results of a self-paced reading experiment, suggest that there is no direct link, within each language, between the interpretation of a subject pronoun and the ambiguity, in terms of homophony among persons, of the verb that agrees with it (Filiaci, 2011). Keywords: Anaphora resolution, Italian, Spanish, subject pronouns, pro drop

1 Italian and Spanish as pro drop languages Italian and Spanish are two closely typologically related Romance languages sharing many syntactic and morphological characteristics. One of them is that they both allow the syntactic subject position of a tensed clause to be phonetically empty, a property that is referred to as subject drop, or pro drop. This property, of course, is shared by many other typologically diverse languages, but in both Italian and Spanish it is thought to be associated with the presence of a rich verbal morphology, expressing overtly number and person agreement features. The intuition that these two properties should be related is not recent, but it has generated much controversy whenever researchers have tried to capture the relationship formally (see for example Alexiadou and Anagnostopoulou, 1998; Borer, 1989; Jaeggli and Safir, 1989; Rizzi, 1986b; Rohrbacher, 1999; Taraldsen, 1978, among many others). This issue will be discussed more in detail in the second part of this chapter. In pro drop languages, null and overt subjects are not freely interchangeable, in the sense that, in a given context, their alternation produces subtle interpretational differences. Traditionally, Italian and Spanish have been considered equiv-

320 | Francesca Filiaci

alent both from the point of view of their null subject syntax and morphology, and from that of the distribution and interpretation of null and overt forms. This assumption has had implications for cross-linguistic research. For example, it is more or less explicitly assumed that the acquisition of the properties of null and overt subjects by first and second language learners should follow the same trajectory in the two languages, or that we should find identical patterns of language loss in situations of language contact (a phenomenon known as first language attrition). This assumption is explicit in Sorace et al. (2009). In this study the authors look at the interpretation of Italian null and overt subjects in two groups of bilingual children, English/Italian and Spanish/Italian bilinguals, and compare it to that of a group of monolingual peers. The logic is that, if the intuitions of the Spanish/Italian bilinguals are similar to those of the monolinguals and both these groups are different from the English/Italian bilinguals, then such difference must be due to the cross-linguistic influence of English, a non-pro drop language, over Italian. If, on the other hand, the two bilingual groups patterned together and differently from the Italian monolinguals, that would be an indication that bilingualism itself affects the interpretation of anaphoric forms, independently of the language pairing. The data followed the latter pattern and was interpreted in support of the latter hypothesis, but of course if the interpretation of null and overt pronouns in Italian and Spanish turned out to obey different constraints, then the possibility of cross-linguistic influence could not be ruled out any more. The hypothesis that Italian and Spanish can be considered equivalent from the point of view of the interpretation of their null and overt anaphoric subjects was tested by Filiaci (2011) and Filiaci et al. (2013) using a series of self-paced reading experiments. In the next section, I will briefly compare the pronominal systems of Spanish and Italian (Section 1.1) and introduce the Position of Antecedent Strategy (Carminati, 2002), a processing strategy proposed for the resolution of anaphoric subjects in Italian (Section 1.2). In Section 1.3, I will question whether Italian and Spanish should be considered comparable from the point of view of the assumptions made by the Position of Antecedent Strategy. I will then present the results of two self-paced reading experiments suggesting that, in the context of intrasentential anaphora, the interpretation of overt pronouns differs across the two languages (Section 1.4). The results are then discussed in terms of relative weakness of the Spanish pronominal forms (see Cardinaletti and Starke (1999) and also Bresnan (1997) for an account within the framework of Optimality theory). The second part of the chapter (Section 2) returns to the idea that the availability of null subjects in Italian and Spanish is linked to the presence of a rich verbal agreement morphology, and introduces the Morphological Ambiguity hypothesis. Section 2.2, presents the results of a third experiment, suggesting that there is no

Anaphoric subjects in Italian and Spanish | 321

direct link in either language between the relative ambiguity of a verbal form, in terms of person marking, and the interpretation of its overt pronominal subject.

1.1 The pronominal systems of Italian and Spanish Both Spanish and Italian have a series of nominative personal pronouns that can appear in the preverbal and in the postverbal position. The data presented in this study is limited to the third person singular. One relevant difference between Italian and Spanish is that the former, but not the latter, has two series of third person pronouns that can be used as subjects: the egli/ella, esso/essa series and the lui/lei series. Although prescriptive grammars recommend the use of egli/ella as subject (with esso/essa to refer to objects and animals), in everyday use these forms have become obsolete, and have been replaced by lui/lei, which were originally the oblique forms (Serianni and Castelvecchi, 1991). In their cross-linguistic classification of personal pronouns, Cardinaletti and Starke (1999) analyse Italian lui/lei as strong pronouns and egli/ella as weak. The Italian null subject pro is also analysed as a weak form. These two categories of pronouns, weak and strong, differ form the prosodic, morphological, syntactic and semantic point of view. Weak pronouns can restructure prosodically, tend to be morphologically reduced compared to strong ones, cannot occupy a thematic base position or a dislocated position, and cannot be coordinated or modified (see Cardinaletti and Starke (1999) or Cardinaletti (1997) for examples). More importantly here, weak pronouns need to refer to a prominent antecedent in the discourse, whereas strong pronouns are excluded from reference to a prominent antecedent. This characteristic is illustrated in the example below from Cardinaletti and Starke (1999, p. 198): (1)

a. Giannii partirà quando proi avrà finito il lavoro. b. Giannii partirà quando eglii avrà finito il lavoro. finito il lavoro. c. * Giannii partirà quando luii avrà Giannii will.leave when hei will.have finished the job

Cardinaletti and Starke (1999) do not analyse Spanish pronouns in their work. As for the status of strong and weak forms in a language, it seems that strong forms should be considered more basic, in the sense that weak forms are derived from strong ones by erasure of structure (a structure mirroring the functional structure projected by nominal phrases and verbal clauses), and of the features associated with it. This erasure is the result of an economy operation which applies when the relevant features are not necessary for the required interpretation. Although

322 | Francesca Filiaci

the authors are not explicit about it, it seems that their analysis would predict that strong forms cannot be equivalent to weaker forms, that is to say, it should not be possible to find a language in which strong and weak forms can be used interchangeably. It should also predict that the lack of weak forms in a language would imply the lack of application of a very general economy principle within the language. Within the framework of Optimality Theory (Prince and Smolensky, 2004), Bresnan (1997, 2001) provide an analysis of pronoun typologies across languages that makes largely similar predictions to Cardinaletti and Starke’s. Bresnan identifies five types of pronominal forms: zeros, bound pronouns, clitics, weak pronouns and free pronouns.¹ Each of these forms can be associated with three sets of properties that characterise pronouns: pro (the set of defining semantic properties of pronouns), agr (the optional property of being specified for number, gender and person) and top (the optional property associated with pronouns specialised to refer to topic antecedents). Any from can be associated in principle with any set of properties, but relevant Optimality Theory constraints (namely harmony, struct and faith, see Bresnan (2001) for details) operate on the set of potential associations to produce the inventory that is actually instantiated in human languages and to single out some associations as more marked than others, and therefore cross-linguistically rarer. According to Bresnan, after the application of these universal constraints, the unreduced pronoun (the free pronoun) is the most unmarked form, and should therefore be most common across languages. In other words, Bresnan predicts that there should be languages that only allow free pronouns but not reduced ones in their inventories. In these languages free forms will be unspecified with regard to the property of referring to a highly prominent (topic) antecedent. English, in which overt subject pronouns can freely refer to a highly prominent antecedent, is an example of such language. Other languages will include both free and reduced forms in their inventories. In such languages, the harmony constraint imposes that only the reduced form be morpho-syntactically marked, or specialised, to be bound to a topic antecedent. In contrast, the unreduced form remains unmarked in this sense, and acquires its property of referring to non-topics only because used in opposition to the reduced form.

1 According to Bresnan’s definition, Italian and Spanish should not be considered as languages including a zero anaphor in their inventories, because of the presence of obligatory subject-verb agreement morphology on the verb, which has been sometimes analysed as a bound or clitic pronoun rather than an agreement marker (for in depth analyses, more or less explicit about this point, see for example Alexiadou and Anagnostopoulou, 1998; Borer, 1989; Platzack, 2004; Rizzi 1982, p. 131). Related to this question see also Ariel (2006).

Anaphoric subjects in Italian and Spanish

| 323

As mentioned above, Bresnan’s account makes a more explicit, but similar prediction to Cardinaletti and Starke’s that the unreduced pronominal form should be in some sense more basic, common, or unmarked, than the reduced one. In addition, it makes the prediction that if a language lacks reduced forms, the unreduced ones should be able to refer to prominent discourse antecedents, as they are not specified for topic-anaphoricity. To summarise, we have seen that, unlike Italian, Spanish only has one series of third person subject pronouns which, within both the Generative and the Optimality Theory framework, would be analysed as unreduced, base forms, with the anaphoric characteristics of strong pronouns. Based on the analysis of Cardinaletti and Starke, these forms should be incompatible with reference to a highly prominent antecedent. According to Bresnan’s analysis, on the other hand, these forms should not be inherently marked for topic anaphoricity, except in the contexts where they can alternate with the reduced form pro. In the next section we will look at existing evidence on the interpretation of Italian null and overt subjects in context. In particular I will introduce the Position of Antecedent Strategy.

1.2 The interpretation of null and overt subjects: the Position of Antecedent Strategy As mentioned above, in null subject languages pro and overt personal pronouns are not freely interchangeable, but tend to be interpreted as referring to different antecedents in a given context. This tendency was illustrated by the contrast between Italian examples (1a) and (1c). This behaviour can be captured in more general terms by a framework like Accessibility theory (Ariel, 1990), which claims that there is a systematic relationship between the form of anaphoric expressions and their antecedent preferences in the discourse. More precisely, the intuition is that reduced anaphoric expressions² tend to corefer with highly prominent antecedents, easily accessible in the memory, while informationally/semantically/morpho-phonologically richer expressions tend to skip those antecedents and refer to less prominent ones. This tendency could be related to general cognitive mechanisms, on the basis of which discourse partici-

2 According to Accessibility theory, expressions can be ranked along a continuous scale using three criteria: informativity (the amount of information they contain), rigidity (how uniquely they identify a potentially ambiguous antecedent) and attenuation (roughly their phonological size). Broadly speaking, expressions that are phonetically reduced also tend to contain less information and to be less rigid, and vice versa (but see Ariel, 1990, for the details).

324 | Francesca Filiaci

pants tend to achieve a balance between the processing cost of an expression and its communicative effect, in the spirit of Relevance theory (Sperber and Wilson, 1995). That is, if a referent is somehow prominent and strongly encoded in the memory of speaker/hearer, and therefore readily accessible, an informationally poor, morpho-phonologically reduced expression will be sufficient to retrieve it. In contrast, if a referent is not readily available in the memory, then the use of an informationally rich expression, more costly in terms of processing and memory space, is functionally justified, because it provides a more effective cue for the retrieval of the inaccessible referent (see Almor, 1999, 2000; Almor and Nair, 2007). Prominence of an antecedent is therefore a key concept together with the assumption that prominent antecedents are more strongly encoded in memory and more readily available for retrieval. Gundel et al. (1993) talks about prominence in cognitive terms, and relates it to the focus of attention of speaker/hearer at any given point in the discourse. The focus of attention obviously changes as the discourse unfolds, and can be influenced by both linguistic and non linguistic contextual factors. We could argue that anything that can attract the hearer’s attention on a certain argument may impact its prominence and therefore accessibility in the memory. Some accounts have put the emphasis on structural factors finding that special prominence is associated to the first mentioned argument of an utterance (Gernsbacher, 1989), to the syntactic subject (Gernsbacher, 1989; Grosz et al., 1995; Arnold, 2001), or to certain information primitives like topics or foci, although these latter concepts are often used vaguely in the literature (Gernsbacher, 1989; Gundel et al., 1993; Arnold, 2001). Other accounts have put the emphasis on semantic factors like the implicit causality of the verb or the coherence relations between clauses, which may have the effect of modifying the hearer’s expectations about the upcoming discourse and direct their attention on specific arguments of the verb (see for example Stevenson et al., 1994, 2000; Kehler et al., 2008; Rohde et al., 2011). In a study focusing on intra-sentential anaphora, Carminati formulates the Position of Antecedent Strategy, a processing strategy describing the resolution preferences of null and overt subjects in Italian: (2)

The Position of Antecedent Strategy: The null pronoun prefers an antecedent which is in the SpecIP position (or in the AgrS position), while the overt pronoun prefers an antecedent which is not in the SpecIP position.

(Carminati, 2002, p. 33)

Anaphoric subjects in Italian and Spanish | 325

As can be seen, this strategy assumes that prominence relations within a sentence can be operationalised in terms of syntactic position. The validity of this processing strategy was tested using a series of self-paced reading experiments and questionnaires. Sentences like (3) below were shown to Italian native speakers, clause by clause, in a moving window, and the reading times for the second clause (the one containing the anaphoric null or overt subject) were recorded and analysed. All the sentences begin with a subordinate clause introducing two antecedents, a preverbal subject (‘Maria’) and a postverbal object (‘Vanessa’). The first clause is followed by a main clause starting with an anaphoric subject (an overt pronoun such as ‘lei’ in (3a) and (3c) or the null 0, as in (3b) and (3d)). In sentences (3a) and (3b) the most plausible antecedent for the anaphor is the preceding preverbal subject, whereas in (3c) and (3d) the most plausible antecedent is the postverbal object. (3) a. Quando Mariai è andata a trovare Vanessaj in ospedale, leii le ha portato un mazzo di fiori. b. Quando Mariai è andata a trovare Vanessaj in ospedale, 0i le ha portato un mazzo di fiori. ‘When Mariai went to visit Vanessaj at the hospital, shei /0i brought her a bunch of flowers.’ c. Quando Mariai è andata a trovare Vanessaj in ospedale, leij era già fuori pericolo. d. Quando Mariai è andata a trovare Vanessaj in ospedale, 0j era già fuori pericolo. ‘When Mariai went to visit Vanessaj at the hospital, shej /0j was already out of danger.’ Carminati found that sentences like (3b), with a null (0) anaphoric subject and a plausible preverbal subject antecedent, and (3c), with an overt subject pronoun and a plausible postverbal object antecedent, were read faster and understood more accurately than sentences like (3a) and (3d), where the anaphors need to be resolved against their antecedent biases to yield a plausible interpretation of the sentence. The current study extend the Position of Antecedent strategy testing its validity in Spanish. In order to proceed with the comparison though, there are two aspects that we need to consider: firstly whether the anaphoric forms available in Italian and Spanish are comparable (which I touched upon already in Section 1.1), and secondly whether the potential antecedents (preverbal subjects and postverbal objects) and their relative prominence in terms of syntactic position

326 | Francesca Filiaci

are comparable across the two languages. The latter question will be explored in the next section.

1.3 Are Italian and Spanish comparable? The Position of Antecedent strategy states that the anaphoric preferences of Italian null and overt subject pronouns are determined by the syntactic position of the antecedent. In fact, Carminati assumes that particular prominence should be associated with certain syntactic positions relative to others, and more precisely with the IP Specifier, the sentence-internal position occupied by the preverbal subject of a tensed verb. When it comes to null subject languages though, the literature is split with regard to the exact syntactic position of overt preverbal subjects. The problem arises in relation to the status of the verbal morphology and its ability to license empty subjects. Leaving aside the technical details that differentiate the various proposals, several researchers have suggested that the verbal morphology in pro drop languages has some sort of pronominal content allowing for the dropping of subjects. Rizzi (1982, 1986a,b, 1997) proposes an optionally specified [+pronominal] feature, Platzack (2004) following Borer (1989) talks about pronominal agreement, subject to Binding Principle B, Alexiadou and Anagnostopoulou (1998) suggest that the agreement morpheme may be an independent lexical entry able to satisfy the Extended Projection Principle (Chomsky, 1982). The issue with these proposals is that the verbal agreement morphology in Italian and Spanish is obligatory and does not appear in complementary distribution with overt subjects, leaving us with the problem that sentences with an overt subject should be analysed as having two subjects for one verb. One of the solutions put forward was to suggest that preverbal subjects in pro drop languages occupy a left-dislocated position outside the clause and stand in a relation of clitic doubling with the verbal morphology (see Alexiadou and Anagnostopoulou, 1998, and references therein). In Spanish such solution was explored by Ordóñez and Treviño (1999) and supported with empirical evidence form ellipsis, quantifier extraction and quantifier scope, while Italian syntacticians seem to have warmed less to the proposal (cfr. for example Cardinaletti, 1997) and Carminati herself seems to disregard it. If a cross-linguistic difference exists between the syntactic structure of SVO sentences in Italian and Spanish, this could potentially confound our crosslinguistic comparison. It should be noted though, that the suggestion that overt subjects are always left-dislocated is not uncontroversial even in Spanish. Based on an asymmetry in the distribution of bare nominals shown by the contrast between (4) and (5) below,

Anaphoric subjects in Italian and Spanish | 327

Casielles-Suárez (2001, 2004) and Casielles-Suárez et al. (2006) suggest that two different syntactic positions are available for Spanish preverbal subjects: a leftdislocated, adjoined position, as in (5), which can be occupied by any arguments of the verb and indeed by multiple arguments, adjoined with no restrictions on their relative order, as shown in (6a) through (6d); and a unique, sentence-internal position, where subjects optionally rise by DP-movement to receive a topic interpretation (to check a Topic feature in terms of Chomsky (1995), or to satisfy the Topic criterion in terms of Rizzi (1996)). This would be the structure shown in (4), where the preverbal position can be occupied by the whole DP ‘Los niños’ but not by a bare noun (‘niños’). (4) * (Los) niños juegan en la calle. ‘(The) children play in the street.’ The ungrammaticality the bare noun in example (4) can be contrasted with the grammaticality of (5) below, where the bare noun ‘Niños’, subject of the embedded clause, was dislocated to the left of the main sentence. (5) Niños, no creo que vengan a la fiesta. ‘Children, I don’t think they will come to the party.’ Examples (6a) through (6d) show that this left-dislocated position that is available to bare nouns (here ‘libros’) can be occupied by multiple arguments, not necessarily subjects, with no restrictions on the relative order. (6) a. A Juan yo libros[,] no le dejaría nunca. to Juan I books not cl would.lend-1sg. never ‘I would never lend books to Juan.’ b. Yo libros a Juan … c. Libros yo a Juan … d. Yo a Juan libros … According to Casielles-Suárez these two preverbal positions are different not only from the syntactic point of view, but also from the point of view of their informational characteristics. Dislocated arguments are found in the pragmatically and intonationally marked Focus-Background articulation, which separates the Focus, the informative part of the sentence, from the Presupposition or Background, the part containing all and only the information that is already known to speaker and hearer, which is dislocated to escape the scope of focus (see Casielles-Suárez,

328 | Francesca Filiaci

2001, 2004, and references therein). By contrast, sentence-internal subjects are unique elements that move to initial position to be interpreted as sentence topics, defined as ‘the point of departure for the sentence as a message’ (Casielles-Suárez, 2001, p. 8), and are separated form the rest of the utterance, known as the Comment or Rheme. This articulation is not intonationally nor pragmatically marked, and can be used at the beginning of a discourse and out of the blue. Here, I will adopt Casielles-Suárez’s analysis and assume that pragmatically and intonationally unmarked out of the blue SVO sentences have an equivalent syntactic and informational structure in Italian and Spanish. My position may even depart slightly from Carminati’s in that I think that the particular prominence associated with the IP specifier is not due to its higher syntactic position per se, but to the fact that subjects in this position are interpreted as sentence topics, defined as above. Since multi-sentence discourses are not expected to be formed by disconnected utterances, but to be related to a certain topic, it seems reasonable to assume that arguments introduced as sentence topics should be expected to be continued across a larger stretch of discourse, and therefore entered in the memory store in a way that keeps them particularly accessible. On the basis of this assumption, I would claim that any cross-linguistic difference found between Italian and Spanish should not be attributed to different properties of the antecedents in terms of syntactic position, prominence or accessibility, but to differences at the level of the anaphoric expressions and of the parsing strategies deployed to interpret them. In the next section I will present empirical data comparing the interpretation of null and overt anaphoric subjects (It. lui/lei, Sp. él/ella) in Italian and Spanish. The data show no noticeable cross-linguistics differences in the processing of null subjects, while Italian overt pronouns seem to be significantly more strongly associated than their Spanish counterparts with a shift in reference from the previous preverbal subject to a less prominent antecedent.

1.4 The empirical data The data presented in this section is a selection from a larger study including two different experiments (Filiaci, 2011; Filiaci et al., 2013). The first pilot experiment was a clause by clause, self-paced reading task, replicating Carminati (2002), using the same materials, sixteen experimental sentences of the type illustrated in example (3), and the same method, outlined in Section 1.2 above. In addition the same sentences were translated into Spanish and the experiment run with a group of Spanish native speakers. Table 1 shows the reading times for the second clause, the clause containing the anaphoric subjects.

Anaphoric subjects in Italian and Spanish

| 329

Table 1: Descriptive statistics for the reading times (in milliseconds) of the second clause in Experiment 1. ITALIAN

SPANISH

Antecedent

N

Mean

Std. Deviation

N

Mean

Std. Deviation

NULL ANAPHORA

Object Subject

32 32

2569 1941

798 636

32 32

2319 1998

916 587

OVERT ANAPHORA

Object Subject

32 32

2266 2750

698 777

32 32

2389 2507

784 880

The Italian data replicated Carminati’s results, providing evidence for the double bias described by the Position of Antecedent strategy. More precisely, clauses containing null subjects were read significantly faster when the meaning of the sentence suggested that the antecedent of pro was the previous preverbal (SpecIP) subject, while the opposite was the case when sentences contained an overt pronoun, that is, they were read significantly faster if the pronoun did not refer to the previous preverbal subject. The Spanish data showed the same pattern as Italian in the null subject condition, but the overt pronoun condition did not yield any statistically significant differences in relation to the position of the antecedent. Moreover, Spanish sentences maintaining reference to the preverbal subject antecedent were read faster in the null subject condition than in the overt pronoun condition, while if the subject reference shifted to the less prominent object antecedent, there was no significant difference between the reading times of the clauses containing pro or an overt pronoun. When the residual reading times for both languages were analysed together using Language as a between subjects factor, the results showed a marginally significant interaction between Language and Antecedent only in the overt pronoun condition. This interaction revealed that sentences with an overt subject referring to an object antecedent are read significantly faster in Italian than in Spanish. These results altogether suggest that there may be a difference between Italian and Spanish in the way overt subject pronouns are processed intra-sententially. Crucially, though, the interaction involving the Language variable was only marginally significant. Moreover, this experiment did not yield any information about possible differences in the time course of the anaphora resolution. For these reasons a second experiment was run, using more experimental sentences and presenting the materials phrase by phrase. The next sections describe more in detail the methodology of the second experiment and the analysis of the reading times for the final wrap up region of the experimental sentences, the region that

330 | Francesca Filiaci

yielded significant results for the effects that are of interest here. The results are presented and discussed in their entirety in Filiaci (2011) and Filiaci et al. (2013).

1.4.1 Experiment 2: Method The second pair of experiments consisted in a phrase by phrase self-paced reading task run using Psyscope X software (Cohen et al., 1993). The materials appeared at the press of a button of a USB button box, in a moving window, on the screen of a 13" MacBook. Two variables were manipulated within subjects: the type of anaphora (pro vs. overt pronoun) and its antecedent (the subject vs. the object of the previous clause). The experimental sentences started with a subordinate clause, introducing the two antecedents, followed by a main clause, containing the temporarily ambiguous null or overt anaphoric subject. Like in the previous experiment, the antecedent of the anaphoric subject was disambiguated by the meaning of the second clause. The same items (n = 48) were translated into Italian and Spanish to obtain two equivalent sets of materials. Sentences (7) and (8) show an experimental item in the four experimental conditions in Italian and Spanish respectively: (7) a. Quando Carloi ha chiesto aiuto a Diegoj per preparare l’esame, luii lo ha superato con voti eccellenti. b. Quando Carloi ha chiesto aiuto a Diegoj per preparare l’esame, 0i lo ha superato con voti eccellenti. ‘When Carloi has asked help to Diegoj to prepare for the exam, hei passed it with excellent marks.’ c. Quando Carloi ha prestato aiuto a Diegoj per preparare l’esame, luij lo ha superato con voti eccellenti. d. Quando Carloi ha prestato aiuto a Diegoj per preparare l’esame, 0j lo ha superato con voti eccellenti. ‘When Carloi has given help to Diegoj to prepare for the exam, hej passed it with excellent marks.’ (8) a. Cuando Carlosi pidió ayuda a Diegoj para preparar el examen, éli aprobó con notas excelentes. b. Cuando Carlosi pidió ayuda a Diegoj para preparar el examen, 0i aprobó con notas excelentes. ‘When Carlosi has asked help to Diegoj to prepare for the exam, hei passed it with excellent marks.’

Anaphoric subjects in Italian and Spanish | 331

c. Cuando Carlosi ayudó a Diegoj a preparar el examen, élj aprobó con notas excelentes. d. Cuando Carlosi ayudó a Diegoj a preparar el examen, 0j aprobó con notas excelentes. ‘When Carlosi helped Diegoj to prepare for the exam, hej passed it with excellent marks.’ Four experimental lists were created for each language and 40 filler sentences were included in each list. Half of the experimental items (and of the filler sentences) were followed by a comprehension question, asking to identify the antecedent of the anaphoric subject. The participants were a group of Italian native speakers (n = 32) recruited among undergraduate and postgraduate students at the University of Padua (Italy), and a group of Spanish native speakers (n = 32) recruited at the University of La Laguna (Spain).

1.4.2 Results: the wrap up region In order to take into account the systematic differences in the length of the regions across languages, I calculated for each participant the regression between the raw reading times and the length of each region in terms of number of characters, and analysed the residual reading times for the verb phrase of the main clause (i.e. the verb following pro or the overt pronoun), the region following the verb phrase and the final wrap up region. The data were submitted to a 2 × 2 × 2 ANOVA with Antecedent (subject vs. object) and Anaphora (null vs. overt pronoun) as within subject factors and Language (Spanish vs. Italian) as between subjects factor. The example below shows the three regions analysed (following the double forward slash). This section only includes the results for the final wrap up region.³

3 Reading times tend to slow down at the end of a clause or a sentence and this finding has been traditionally interpreted as evidence that a certain amount of processing is delayed to clause boundaries. According to Just and Carpenter (1980, p. 345) sentence wrap up effects involve ‘a search for referents that have not been assigned […] and an attempt to handle inconsistencies that could not be resolved within the sentence’. More recently this idea has been challenged, although results for alternative explanations have not always been conclusive (see Hirotani et al., 2006; Warren et al., 2009).

50

111

100 150 200 250 300

Object Subject

189

50

20

0

0 -100 -50

-77

-52

-68

-100 -50

100 150 200 250 300

Antecedent

3

Italian

Spanish

Italian Language

Language

(a) pro

Spanish

-300

-200

-200

-121

-300

Residual Reading Times (msec.)

332 | Francesca Filiaci

(b) Overt Pronoun

Figure 1: Residual reading times for the final wrap up region.

(9) Cuando /Antonio /pidió ayuda /a Diego /para preparar /el examen,/ (él) // aprobó / con notas / excelentes. Anaphor Verb Phrase VP+1 Wrap Up Figure 1 shows the residual reading times for the final wrap up region in each condition and the 95 % confidence intervals. The ANOVA shows a main effect for Anaphora that is marginally significant by subjects and fully significant by items (F1 (1, 62) = 3.36; p < .08; F2 (1, 94) = 5.93; p < .05), with overall faster reading times for the overt pronoun condition than the null subject condition (−49 msec. vs. 51 msec.); a highly significant main effect for Antecedent (F1 (1, 62) = 22.62; p < .000; F2 (1, 94) = 27.48; p < .000), with faster reading times for sentences with a subject antecedent (−124 msec. vs. 126 msec.); a significant interaction Anaphora by Antecedent (F1 (1, 62) = 18.89; p < .000; F2 (1, 94) = 50.95; p < .000); and a significant three-way interaction between Anaphora, Antecedent and Language (F1 (1, 62) = 5.41; p < .05; F2 (1, 94) = 14.78; p < .001). In the null subject condition, we find a highly significant effect of Antecedent (F1 (1, 62) = 31.46; p < .000; F2 (1, 94) = 66.80; p < .000), indicating that sentences containing null subjects are wrapped up significantly faster when the anaphora retrieves a prominent (subject) antecedent (−99 msec. vs. 151 msec.); and an interaction Antecedent by Language only significant in the analysis by items (F1 (1, 62) = 1.9; p < .20; F2 (1, 94) = 4.04; p < .05).

Anaphoric subjects in Italian and Spanish

| 333

In contrast, overt pronoun anaphors showed no main effects, but a fully significant interaction Antecedent by Language (F1 (1, 62) = 4.95; p < .05; F2 (1, 94) = 6.46; p < .05). This interaction indicates that, with overt pronouns, the processing penalties encountered at the wrap up region vary between Italian and Spanish depending on the antecedent. More precisely, in Italian the analysis reveals no significant difference between subject or object antecedent condition, whereas in Spanish, there is a significant effect for Antecedent (F1 (1, 31) = 6.85; p < .05; F2 (1, 47) = 4.82; p < .05), with significantly faster reading times for sentences with a subject antecedent. If we compare across languages, when an overt pronoun retrieves an object antecedent the wrap up region is read significantly faster in Italian than in Spanish (F1 (1, 62) = 5.35; p < .05; F2 (1, 94) = 4.12; p < .05), whereas no significant cross-linguistic difference was found for an overt pronoun retrieving a subject antecedent.

1.4.3 Discussion To summarise, Experiment 1 suggested that, in intra-sentential contexts, overt subject pronouns (Italian lui/lei vs. Spanish él/ella) may be resolved differently in Italian and Spanish. More precisely, Spanish overt subjects do not seem to show any significant resolution preference based on the syntactic position of the antecedent. In fact, when subject reference switches to a non prominent antecedent, Spanish sentences with overt subjects are not read significantly faster than sentences containing pro, which is the case in Italian. Additionally, when comparing across languages, overt pronouns referring to a non prominent antecedent tend to be read faster in Italian than in Spanish. Experiment 2 confirmed that the cross-linguistic differences are reliable only in the overt pronoun condition, and that when subject reference across clauses switches to a less prominent object antecedent (the dark grey bars in Figure 1), sentences containing an overt subject are wrapped up significantly faster in Italian than in Spanish. One discrepancy between the two experiments is that, when we look at the wrap up region for the overt subject sentences in each individual language (Figure 1 (b)), in Italian we have gone from the object antecedent preference of Experiment 1 to a non significant difference between antecedents, while in Spanish we have gone from the lack of significance in Experiment 1 to an unexpected subject antecedent preference in Experiment 2. It seems that in both languages subject antecedents have become easier to access for overt pronouns. This could be related to the fact, already noted by Carminati (2002) in Italian, that the overt subject bias seems to be less stable and more dependent on contextual

334 | Francesca Filiaci

factors than the null subject bias, or even to the methodological differences between Experiment 1 and 2, in the sense that the less natural task of Experiment 2 may have interfered with the encoding and retrieval of the antecedents in memory (see Filiaci, 2011, p. 150 for a discussion). As for the null subject bias, it appears to be consistent and reliable across experiments and across languages. This cross-linguistic consistency could be taken as an indication that preverbal subjects in Italian and Spanish can be considered equivalent in terms of prominence, or more precisely, in terms of their informational status within the sentence. If we assume that in both languages the informational status of an argument is encoded syntactically, the data might be taken as indirect evidence that preverbal subjects in the two languages occupy an equivalent syntactic position in the type of sentences used in these experiments. As for the overt subject results, taken together they suggest that Spanish él/ella may not be as incompatible as Italian strong pronouns lui and lei with reference to a highly prominent subject antecedent, and do not seem to be so strongly associated with a shift in subject reference to a non prominent antecedent. In terms of Cardinaletti and Starke (1999), Spanish él/ella here appear to display antecedent preferences that could be associated to those of their weak Italian cognates egli and ella, rather than those of their strong counterparts lui/lei (see example (1)). On the other hand, Cardinaletti and Starke use a whole series of semantic, distributional, and prosodic/phonological diagnostics to classify pronouns as strong or weak, and, likewise, claim that a whole range of characteristics is associated to, and determined by, the structural deficiency of weak pronouns. For example, Cardinaletti (1997, p. 38) shows, with the sentences reported here as (10) that the Italian weak pronoun egli cannot be used postverbally (10b), in a dislocated position (10c), and in isolation (10d). (10) a. Lui/Gianni/Egli ha aderito. b. Ha aderito lui/Gianni/*egli. ‘He/Gianni has adhered.’ c. Lui/Gianni/*Egli Mariai non l’i ha appoggiata. He/Gianni/He Mariai not cli has supported d. ‘Chi è venuto?’ ‘Lui/Gianni/*Egli.’ ‘Who is come?’ ‘He/Gianni/He’ An informal survey of Spanish is sufficient to reveal that native speakers do not rule out él and ella in these contexts, suggesting that they display, in this case, distributional characteristics associated with strong pronouns. A possible explanation for this is that Spanish includes two series of homophonous pronouns, one weak and one strong. If this is the case, the options that readers have when they

Anaphoric subjects in Italian and Spanish | 335

encounter the anaphoric subjects tested here would be different in Spanish and Italian. To begin with, as suggested in Section 1.3, speakers/readers may plausibly assume that reference to the same topic will be maintained across consecutive clauses, and the highly significant main effect for Antecedent in Experiment 2, showing faster wrap up times for sentences maintaining reference to the preverbal subject across clauses, supports this assumption. In Italian, the strong pronoun lui or lei at the beginning of the main clause can only be interpreted as a cue that subject reference has switched from the previous topic to a less prominent antecedent, and facilitate this reading. This interpretation will have to be reanalysed at the end of the sentence only if the semantics impose a different interpretation, yielding a processing penalty when the pronoun is forced to co-refer with the most prominent antecedent. By contrast, Spanish readers assuming a topic continuation, upon encountering él or ella have the possibility to apply the economy operation proposed by Cardinaletti and Starke and erase the structure and features that do not allow coreference with a prominent antecedent. We do not know exactly when the erasing operation occurs, or if it requires additional processing resources compared to the interpretation of the pronoun as strong, but this is likely to depend at least in part on how committed the reader is to a subject antecedent interpretation, or, in other words, on how strong is the expectation for a topic continuation. If the decision to erase structure has been made early, it will need to be reanalysed if the semantics impose an object antecedent interpretation. If the expectation was not so strong, it is possible that the reader may delay erasing the structure, and accept the interpretation referring to a non-prominent antecedent. With reference to Optimality Theory, the Spanish overt pronoun should be analysed as an unmarked free pronoun, not specified for topic-anaphoricity (Bresnan, 2001). On the other hand, Spanish does have a reduced anaphor that can be used in opposition to the free pronoun: the null subject, which could alternate freely with the overt pronoun in the contexts tested. This possibility should in theory have been enough to exclude the overt pronoun from topic anaphoricity. In the context of either analysis, the question that arises is why Spanish should allow an economy operation that has the result of increasing the ambiguity of the overt pronoun, when a weak alternative, pro, is available in the same context. Or, in Optimality Theory terms, why is it that pro does not exclude the overt pronoun from topic anaphoricity when the two can alternate? A possibility is that pro may be unavailable or dispreferred in certain contexts for independent reasons. The next part of the chapter explores the idea that this could be the case and that the reason for it may have to do with the characteristics of the verbal agreement morphology.

336 | Francesca Filiaci

2 Pro-drop and verbal morphology As mentioned in Section 1, the study of the pro drop phenomenon has been accompanied by the intuition that the possibility to leave subjects unexpressed must be related to the presence of a rich verbal agreement morphology, arguably because it spells out explicitly person and number features on the verbal head, allowing for the identification of the subject. A first empirical problem with this intuition is the existence of languages like Chinese, which do not have subject-verb agreement, but still allow argument drop. A further problem arises if we want to define what rich agreement is exactly, as there are languages with a relatively diversified verbal paradigm, like French or German, which still do not allow thematic null subjects, and conversely there are languages with relatively poor verb morphology that allow them. It seems therefore clear that, if there is a relationship between subject drop and verb agreement morphology, this must be more complex and indirect than originally thought. Jaeggli and Safir (1989) attempted to solve this problem with the formulation of the Morphological Uniformity hypothesis, arguing that subject drop is allowed exactly in those languages in which verbs are uniformly inflected (like Italian or Spanish) or uniformly uninflected (like Chinese), and is disallowed in those with mixed paradigms. The hypothesis makes many correct predictions, although from a theoretical point of view, the reasons for the uniformity requirement are not clear, and from the empirical point of view, it still makes wrong predictions when enough languages are considered. Other researchers have suggested that argument drop in languages without agreement morphology (like Chinese) may be altogether a different phenomenon, which should be accounted for in relation to other characteristics of the languages (see for example Givón, 1979; Huang, 1994, 2000; Rohrbacher, 1999, p. 246). Rizzi (1982, 1986a,b, 1997) suggested that, for thematic null subjects to be available in a language, the verb morphology needs to meet two requirements: to be rich, that is to realise overtly the person features that make the identification of pro possible, and to be strong, where strength is an abstract property of the agreement, independent of its morphological characteristics and parameterised across languages. According to Rizzi, Italian strong agreement is characterised by an optional [+ pronominal] feature, that is only specified in sentences without an overt subject. The idea that the verb morphology in null subject languages has pronominal characteristics has been developed by others. For example, Platzack (2004), following a proposal by Borer (1989), suggests that verb agreement across languages can be either anaphoric, and therefore subject to Binding Principle A, requiring

Anaphoric subjects in Italian and Spanish | 337

it to be bound within its governing category, or pronominal, and therefore free in its governing category according to the requirements of Binding Principle B. As we have seen in Section 1.3 the attribution of pronominal characteristics to the verb morphology gives rise to problems regarding the status and position of overt subjects in null subject languages. Also Rizzi’s solution of an optionally specified feature is not without problems, since it amounts to saying that the nature of Italian morphology varies depending on wether a sentence has an overt subject or not. In the context of a study on verb rising in Germanic languages, Rohrbacher (1999) puts forward a different approach, and proposes the concept of referential agreement. According to Rohrbacher, pro drop is licensed by referential agreement, which also triggers verb rising to IP. Agreement inflection is referential exactly in those languages in which both first and second person are distinctively marked through subject-verb agreement in at least one number of at least one tense of the regular verb paradigm. A verbal form is distinctively marked for a certain feature if its phonological matrix is different from that of the forms which carry a different value of that feature. For Rohrbacher the reason why distinctively marked first and second person features have an impact on the syntactic properties of the verb is that they identify speaker and hearer in the conversation, so they have referential abilities that make inflection a referential category, with its own lexical entries, rather than a functional element with grammatical meaning only, produced by post-syntactic spell-out rules. For a language to have thematic null subjects though, the null subject needs not only to be licensed, but also identified. According to Rohrbacher the identification of the content of pro depends on formal aspects of the agreement affixes and is at least in part determined by language-specific requirements. For example, in Yiddish, topic drop is allowed with all persons of the indicative present, while pro drop only occurs with the second person singular, the only one that is uniquely and overtly marked by the agreement morphology (Rohrbacher, 1999, p. 254). Italian, by contrast, does not seem to have an overtness requirement, since pro drop is possible with all persons of the indicative present, including the third person, which only consists of the verb stem and the thematic vowel, as shown in Table 2. On the other hand, the uniqueness requirement may have an impact on the availability of pro in Italian, since in the subjunctive present, in which the three singular persons are homophonous (see Table 2), pro does not seem to be available with the second person, and in the subjunctive past, where first and second person are homophonous, pro is preferred with the first person, according to Cardinaletti (1997). In a similar vein, within the framework of Accessibility Theory, Ariel (1990) explains the different rates of null subject occurrence across persons and tenses in

338 | Francesca Filiaci

Table 2: Verbal agreement morphology for the Italian verb parlare (singular forms).

1st 2nd 3rd

Indicative Present

Subjunctive Present

Subjunctive Past

parlo parli parla

parli parli parli

parlassi parlassi parlasse

modern Hebrew as depending on the amount of informativity of the agreement inflection. Ariel distinguishes in fact between poor agreement markers, only marked for gender and number features, and informative agreement markers, which are overtly marked for person. Because of their reduced informativity, poor agreement markers can only access extremely salient antecedents and therefore tend to occur with an overt subject. On the contrary, more informative agreement can access relatively less salient antecedents, and can therefore be used without an overt subject relatively more often. The Morphological Ambiguity hypothesis investigates the idea that the different properties of subject pronouns in Italian and Spanish may be related to the fact that pro is available to different degrees in each language. More precisely, it explores the possibility that the availability of pro may depend on languagespecific requirements concerning the explicitness and uniqueness of the person agreement morphology on the verb.

2.1 The Morphological Ambiguity hypothesis As seen above, according to the intuitions of Cardinaletti, in Italian there may be a requirement that the verb agreement morphology be unique, for pro to be freely available, although this requirement seems to affect second persons rather than first and third. In Italian the homophony between different persons in the regular verb paradigm is limited to the subjunctive forms shown in Table 2. In Spanish there is a more pervasive ambiguity between first and third person agreement in several tenses, as shown in Table 3. Enríquez (1984) already explored the hypothesis that the ambiguity of the Spanish agreement morphology may have an impact on the availability of pro, or to be exact, on the occurrence of overt pronouns. Enríquez collected a corpus of spontaneous speech from eighty native speakers of Spanish living in Madrid, and looked at the effect of several linguistic variables on the realisation of subject pronouns. One of the variables was the ambiguity of the tense, that is whether

Anaphoric subjects in Italian and Spanish

| 339

Table 3: Ambiguous verbal agreement morphology in Spanish.

1st 2nd 3rd

Indicative Imperfect

Subjunctive Present

Subjunctive Imperfect

Subjunctive Future

Conditional

hablaba hablabas hablaba

hable hables hable

hablara/hablase hablaras/hablases hablara/hablase

hablare hablares hablare

hablaría hablarías hablaría

it had six different endings, or first and third person singular coincided. The data showed a wide range of variation in subject expression across persons and tenses. Significantly more overt pronouns were used with the conditional (an ambiguous tense, see Table 3), but when all the ambiguous tenses were pooled together, the ambiguity turned out not to have a significant impact on subject expression. The last experiment presented in this chapter is an attempt to use a self-paced reading task to investigate whether, within a language, the ambiguity of a verb form, in terms of its person agreement morphology, has an impact on the availability of pro and on the interpretation of overt subject pronouns. Based on the considerations outlined in Section 2 and 2.1, my hypothesis is that in both Italian and Spanish the availability of pro may be somehow limited when the verbal morphology is not unique in terms of person marking. If this is the case, the impact on overt pronouns should be the following: 1. In Italian, the antecedent preference of the overt pronoun may relax, or not apply, if pro is not available due to the tense ambiguity. 2. In Spanish, pronouns should be interpreted as weak mainly in contexts where pro is unavailable, that is with ambiguous tenses, but the trend may be different when these are compared with unambiguous tenses.

2.2 The experiment The data presented in this section is part of a larger experiment, fully described in Filiaci (2011). Like Experiment 2, it consisted of a phrase by phrase self-paced reading task, with the same materials translated in Italian and Spanish. The sentences presented the same type of temporary ambiguity as in Experiment 2, in addition, the verbal tense of the second clause was manipulated using the indicative present as an unambiguous tense, since it has six different endings both in Italian and in Spanish, and the subjunctive present as an ambiguous tense, given its ambiguous third person morphology both in Italian and in Spanish (see Tables 2 and 3).

340 | Francesca Filiaci

2.2.1 Method The participants were 48 adult monolingual speakers of Spanish and 48 adult monolingual speakers of Italian, recruited among undergraduate and postgraduate students at the University of La Laguna (Spain) and at the University of Padua (Italy). The experiment included 80 experimental sentences and 80 fillers. Three variables were manipulated within subjects: Anaphora, Antecedent and verb Tense. The Anaphora could be a null or overt pronoun; the Antecedent of the anaphor could be either the subject or the object of the previous clause; the Tense could be either the Subjunctive Present (an ambiguous tense, as in examples (11a) and (11c) in Italian, and (12a) and (12c) in Spanish), or the Indicative Present (an unambiguous tense, shown in (11b) and (11d), and in (12b) and (12d)). The results presented here will be limited to the overt subject conditions, shown in (11) and (12), and to the final sentence wrap up region. (11) a. Beatricei ha ottenuto una promozione da Carmenj , nonostante leii sia inesperta per il nuovo lavoro. b. Beatricei ha ottenuto una promozione da Carmenj , anche se leii è inesperta per il nuovo lavoro. ‘Beatricei has obtained a promotion from Carmenj , although shei is inexperienced for the new job.’ c. Beatricei ha dato una promozione a Carmenj , nonostante leij sia inesperta per il nuovo lavoro. d. Beatricei ha dato una promozione a Carmenj , anche se leij è inesperta per il nuovo lavoro. ‘Beatricei has given a promotion to Carmenj , although shej is inexperienced for the new job.’ (12) a. Beatrizi ha obtenido un ascenso de Carmenj , aunque ellai sea inexperta para el nuevo trabajo. b. Beatrizi ha obtenido un ascenso de Carmenj , aunque ellai es inexperta para el nuevo trabajo. ‘Beatricei has obtained a promotion from Carmenj , although shei is inexperienced for the new job.’ c. Beatrizi a dado un ascenso a Carmenj , aunque ellaj sea inexperta para el nuevo trabajo.

Anaphoric subjects in Italian and Spanish | 341

d. Beatrizi a dado un ascenso a Carmenj , aunque ellaj es inexperta para el nuevo trabajo. ‘Beatricei has given a promotion to Carmenj , although shej is inexperienced for the new job.’ Based on the hypotheses above, my predictions are that, when the verbal morphology is ambiguous and pro (possibly) unavailable, the subject pronoun should be used to indicate explicitly the person features necessary to the identification of the subject, yielding the following effects: 1. In Italian the structural bias of the overt pronoun should be absent or significantly weaker in sentences (11a) and (11c), compared to sentences (11b) and (11d). 2. In Spanish there should be no overt subject bias in the ambiguous tense (12a) vs. (12c), but at least a tendency towards a preference in the unambiguous tense (12b) vs. (12d).

2.2.2 Results: the wrap up region To test the above predictions the residuals of the reading times were submitted to a 2 × 2 × 2 ANOVA with Antecedent (Subject vs. Object) and Tense (ambiguous Subjunctive vs. unambiguous Indicative) as within subject factors and Language (Spanish or Italian) as a between subject factor. Figure 2 shows the mean residual reading times of the final wrap up region for each group, in each condition, and the 95 % confidence intervals. In this region, the ANOVA shows a main effect for Antecedent (F1 (1, 94) = 14.24; p < .001; F2 (1, 158) = 22.3; p < .000) with significantly faster reading times for object antecedent sentences (−60 msec. vs. 57 msec.), as expected. It also reveals an interaction between Language and Antecedent that is marginally significant by subjects and fully significant by items (F1 (1, 94) = 3.7; p = .057; F2 (1, 158) = 5.78; p < .05). This suggests that, like in the previous experiments, the antecedent preferences of the overt pronoun depend at least in part on the language. No main effects nor interactions involving the Tense variable were found. In order to understand the Language by Antecedent interaction, I analysed separately the relevant conditions. When the antecedent is a preverbal subject, like in Experiment 2, there are no significant differences between Italian and Spanish, but when the antecedent is an object, there is a main effect for Language, showing again that sentences with object antecedents are wrapped up significantly faster in Italian than in Span-

84

68

-34

-120

-87

10

10

11 -7

-80 -50 -20

40

40

63

-80 -50 -20

-160

-112

Italian

Spanish

Italian Language

Language

(a) Indicative (unambiguous)

Spanish

-200

-120 -200

140 70 100

70 100

140

Object Subject

-160

Residual Reading Times (msec.)

180

Antecedent

180

342 | Francesca Filiaci

(b) Subjunctive (ambiguous)

Figure 2: Residual reading times for the final wrap up region.

ish (F1 (1, 94) = 5.42; p < .05; F2 (1, 158) = 4.98; p < .05, −100 msec. vs. −20 msec.). Finally, analysing separately Italian and Spanish, both languages show significant main effects for Antecedent with faster reading times for object antecedents (in Italian: −100 msec. vs. 76 msec.; F1 (1, 47) = 9.92; p < .01; F2 (1, 79) = 17.70; p < .000; in Spanish: −20 msec. vs. 37 msec.; F1 (1, 47) = 4.71; p < .05; F2 (1, 79) = 4.75; p < .05).

2.2.3 Discussion The results of this last experiment confirm the cross-linguistic differences between Italian and Spanish found in Experiment 1 and 2, but do not support either of the predictions made by the Morphological Ambiguity hypothesis, formulated in Section 2.2.1. In fact, no significant effect or interactions were found involving the ambiguity of the tense. As far as Spanish is concerned, these results are in line with those of Enríquez (1984), which did not reveal any significant impact of the ambiguity of the verbal tense on the frequency of subject realisation in Madrid spoken Spanish. In Italian, as noted above, according to Cardinaletti (1997), it is the second person that seems to be most affected by the homophony of the person agreement and lose its null subject properties, so we might still expect different results if we tested different persons, but this question remains open for further investigation.

Anaphoric subjects in Italian and Spanish | 343

One last point should be made about what could appear as a discrepancy between the results of Experiment 1 and 2 and the present results. With Spanish overt pronouns, Experiment 1 did not reveal any significant antecedent preference, while Experiment 2 revealed a significant preference for subject antecedents. By contrast, here the Spanish data show significantly faster reading times for object antecedents similarly to Italian. On the other hand Language still interacted with the antecedent preference, because the object preference was yet again stronger in Italian than in Spanish. This effect can be interpreted as an effect of the changes in the linear order of main and subordinate clause, and in the coherence relation between clauses (see Kehler et al., 2008; Rohde et al., 2011; Hemforth and de la Fuente, 2014). For further discussion of this point see Filiaci (2011) and references therein.

3 Conclusion In this chapter I have compared the anaphoric preferences of null and overt pronominal subjects in Italian and Spanish, testing the assumption that these two languages are equivalent from the point of view of their null subject properties. The data suggest that personal pronouns in these two languages may have different characteristics and follow at least partly different antecedent preferences, with Italian pronouns more strongly associated to a shift in subject reference to a non prominent antecedent than their Spanish counterparts. In addition, the data did not reveal any significant or reliable cross-linguistic differences with regard to the resolution of null subjects, which suggests that their preferred antecedents, preverbal subjects, should be equivalent in Italian and Spanish in terms of discourse prominence. If we further assume that the discourse prominence of a constituent depends on its informational status and that, at the sentence level, this is encoded syntactically in both languages, the data could be taken as indirect evidence that, in the constructions tested, Spanish and Italian preverbal subjects have an equivalent informational status and occupy equivalent syntactic positions. With reference to the study on bilingual child language acquisition cited in the introduction (Sorace et al., 2009), the data presented here suggests that crosslinguistic influence between Italian and Spanish should not be immediately ruled out as an explanation for the differences found between Italian monolinguals and Spanish/Italian bilinguals with respect to the interpretation of Italian overt subjects.

344 | Francesca Filiaci

Finally, the last experiment tested the hypothesis that the different antecedent preferences of Italian and Spanish pronouns may depend at least in part on the characteristics of the verb agreement morphology and on language specific requirements imposing restrictions on the availability of null subjects. The results do not support the hypothesis, at least as it was formulated here, since they did not provide any evidence that, within each language, the ambiguity of the verb morphology has an impact on the resolution preferences of the overt pronoun. On the other hand, there is still a possibility that the hypothesis may be valid across languages even if the effects are not visible within each language. This question of course remains open for further research.

References Alexiadou, A. and Anagnostopoulou, E. 1998. Parametrizing Agr: Word order, V-movement and EPP-checking. Natural Language and Linguistic Theory, 16:491–539. Almor, A. 1999. Noun-phrase anaphora and focus: The informational load hypothesis. Psychological Review, 106(4):748–765. Almor, A. 2000. Constraints and mechanisms in theories of anaphor processing. In Crocker, M., Pickering, M., and Clifton, C., editors, Architectures and Mechanisms for Language Processing, pages 341–354. Cambridge University Press, England. Almor, A. and Nair, V. A. 2007. The form of referential expressions in discourse. Language and Linguistics Compass, 1(1-2):84–99. Ariel, M. 1990. Accessing Noun Phrase Antecedents. Routlege, London. Ariel, M. 2006. Accessibility theory. In Brown, K., editor, Encyclopedia of language and linguistics, pages 15–18. Elsevier, Oxford. Arnold, J. E. 2001. The effect of thematic roles on pronoun use and frequency of reference continuation. Discourse processes, 21(2):137–162. Borer, H. 1989. Anaphoric Agr. In Jaeggli, O. and Safir, K. J., editors, The Null Subject Parameter. Kluwer Academic Publishers, Dordrecht. ˆ pronominals in optimalBresnan, J. 1997. The emergence of the unmarked pronoun: Chichewa ity theory. Berkeley Linguistics Society: Proceedings of the Annual Meeting, 23spe:26–46. Bresnan, J. 2001. The emergence of the unmarked pronoun. In Legendre, G., Grimshaw, J., and Vikner, S., editors, Optimality-theoretic Syntax, pages 113–142. The MIT Press, Cambridge, MA. Cardinaletti, A. 1997. Subjects and clause structure. In Haegeman, L., editor, The New Comparative Syntax, pages 33–63. Longman, New York. Cardinaletti, A. and Starke, M. 1999. The typology of structural deficiency: A case study of the three classes of pronouns. In van Riemsdijk, H., editor, Clitics in the languages of Europe, pages 145–233. Mouton de Gruyter, The Hague. Carminati, M. N. 2002. The Processing of Italian Subject Pronouns. PhD thesis, University of Massachusetts Amherst, Amherst: MA.

Anaphoric subjects in Italian and Spanish

| 345

Casielles-Suárez, E. 2001. The syntax and semantics of preverbal topical phrases in Spanish. In Gutiérrez-Rexach, J. and Silva-Villar, L., editors, Current issues in Spanish syntax and semantics, pages 65–82. Mouton De Gruyter, Berlin/New York. Casielles-Suárez, E. 2004. The Syntax-Information Structure Interface: Evidence from Spanish and English. Routledge, New York/London. Casielles-Suárez, E., Andruski, J., Sahyang, K., Nathan, G., and Work, R. 2006. Syntactic and discourse features of subjects in child Spanish: evidence form Spanish/English bilingual acquisition. BUCLD 30, 1:72–83. Chomsky, N. 1982. Some Concepts and Consequences of the Theory of Government and Binding. MIT Press, Cambridge MA. Chomsky, N. 1995. The Minimalist Program. MIT Press, Cambridge, MA. Cohen, J. D., MacWhinney, B., Flatt, M., and Provost, J. 1993. Psyscope: A new graphic interactive environment for designing psychology experiments. Behavioral Research Methods, Instruments, and Computers, 25(2):257–271. Enríquez, E.-V. 1984. El pronombre personal sujeto en la lengua Española hablada en Madrid, volume 4. Consejo Superior de Investigaciones Científicas, Instituto Miguel de Cervantes, Madrid. Filiaci, F. 2011. Anaphoric Preferences of Null and Overt Subjects in Italian and Spanish: a Cross-linguistic Comparison. PhD thesis, The University of Edinburgh, Edinburgh. Filiaci, F., Sorace, A., and Carreiras, M. 2013. Anaphoric biases of null and overt subjects in Italian and Spanish: a cross-linguistic comparison. Language and Cognitive Processes. Gernsbacher, M. 1989. Mechanisms that improve referential access. Cognition, 32(2):99–156. Givón, T. 1979. From discourse to syntax: grammar as a processing strategy. In Givón, T., editor, Syntax and Semantics: vol.12: Discourse and syntax, pages 81–112. Academic Press, New York. Grosz, B. J., Joshi, A. K., and Weinstein, S. 1995. Centering: A framework for modelling the local coherence of discourse. Computational Linguistics, 21(2):203–225. Gundel, J. K., Hedberg, N., and Zacharski, R. 1993. Cognitive status and the form of referring expressions in discourse. Language, 69(2):274–307. Hemforth, B. and de la Fuente, I. 2014. Reference as a side effect of discourse expectations: The case of focus-sensitive particles. CUNY 2014. Hirotani, M., Frazier, L., and Rayner, K. 2006. Punctuation and intonation effects on clause and sentence wrap-up: Evidence from eye movements. Journal of Memory and Language, 54(3):425–443. Huang, Y. 1994. The syntax and pragmatics of anaphora. CUP, Cambridge. Huang, Y. 2000. Anaphora: a cross-linguistic study. Oxford University Press, Oxford. Jaeggli, O. and Safir, K. J. 1989. The null subject parameter and parametric theory. In Jaeggli, O. and Safir, K. J., editors, The Null Subject Parameter. Kluwer, Dordrecht. Just, M. A. and Carpenter, P. A. 1980. A theory of reading: From eye fixations to comprehension. Psychological Review, 87(4):329–354. Kehler, A., Kertz, L., Rohde, H., and Elman, J. L. 2008. Coherence and coreference revisited. Journal of Semantics, 25:1–44. Ordóñez, F. and Treviño, E. 1999. Left dislocated subjects and the pro-drop parameter: A case study of Spanish. Lingua, 107:38–68. Platzack, C. 2004. Agreement and null subjects. 19th Scandinavian Conference of Linguistics, 31.2.

346 | Francesca Filiaci

Prince, A. and Smolensky, P. 2004. Optimality: Constraint interaction in generative grammar. Blackwell, Malden, MA. Rizzi, L. 1982. Issues in Italian syntax. Foris, Dordrecht. Rizzi, L. 1986a. Null objects in Italian and the theory of pro. Linguistic Inquiry, 17(3):501–557. Rizzi, L. 1986b. On the status of subject clitics in romance. In Jaeggli, O. and Silva-Corvalán, C., editors, Studies in Romance linguistics, pages 391–419. Foris, Dordrecht. Rizzi, L. 1996. Residual verb second and the wh criterion. In Belletti, A. and Rizzi, L., editors, Parameters and functional heads: essays in comparative syntax, pages 63–90. OUP, New York. Rizzi, L. 1997. A parametric approach to comparative syntax: properties of the pronominal system. In Haegeman, L., editor, The New Comparative Syntax, pages 268–285. Longman, New York. Rohde, H., Levy, R., and Kehler, A. 2011. Anticipating explanations in relative clause processing. Cognition, 118:339–358. Rohrbacher, B. W. 1999. Morphology-driven syntax: a theory of V to I raising and pro-drop, volume 15. John Benjamins, Amsterdam. Serianni, L. and Castelvecchi, A. 1991. Grammatica Italiana, Italiano comune e lingua letteraria. UTET, Torino. Sorace, A., Serratrice, L., Filiaci, F., and Baldo, M. 2009. Discourse conditions on subject pronoun realization: testing the linguistic intuitions of older bilingual children. Lingua, 119(3):460–477. Sperber, D. and Wilson, D. 1995. Relevance: communication and cognition. Blackwell, Oxford, UK. Stevenson, R., Knott, A., Oberlander, J., and McDonald, S. 2000. Interpreting pronouns and connectives: Interaction among focusing, thematic roles and coherence relations. Language and Cognitive Processes, 15(3):225–262. Stevenson, R. J., Crawlay, R. A., and Kleinman, D. 1994. Thematic roles, focus and the representation of events. Language and Cognitive Processes, 9(4):519–548. Taraldsen, K. T. 1978. On the nominative island condition, vacuous application and the thattrace filter. Unpublished ms. Warren, T., White, S., and Reichle, E. 2009. Investigating the causes of wrap-up effects: evidence from eye movements and e-z reader. Cognition, 111(1):132–137.

Kyeong-min Kim and Chung-hye Han (Simon Fraser University)

Inter-speaker variation in Korean pronouns Abstract: This paper investigates the bound variable construal of Korean overt third-person pronouns. There is no general agreement in the literature concerning the interpretative status of ku ‘he’ (and kunye ‘she’). Two experimental studies were conducted to explore this issue, and the findings demonstrated that some speakers of Korean consistently allowed ku to induce a bound variable interpretation while others consistently did not. This result is highly suggestive of interspeaker variation regarding the interpretation of ku. Following Han, Lidz, and Musolino (2007), we argue that due to the paucity of input concerning overt pronouns, child learners of Korean must randomly choose between the “demonstrative grammar” and the “pronominal grammar” for ku, only the latter of which licenses a bound variable construal. This thus results in two groups of speakers in the Korean speech community regarding the grammar of ku. Adopting the syntactic typology of pronouns in Déchaine and Wiltschko (2002), we then propose that two types of ku, pro-DP ku and pro-ɸP ku, co-exist, with distinct binding-theoretic status. Keywords: Korean third-person pronouns, bound variable construal, inter-speaker variation, two-grammar hypothesis, demonstrative grammar, pronominal grammar

1 Introduction There is much disagreement in the literature as to whether Korean third-person pronoun ku ‘he’ (and its feminine counterpart kunye ‘she’) can be interpreted as a bound variable. That is, for some researchers the pronoun ku can be construed as a variable bound by the quantificational expression motwu ‘everyone’ in sentences such as (1a) and (1b) (Kang 1988; Suh 1990; Noguchi 1997; Koak 2008) while for others, it cannot be so construed (Hong 1985; Choe 1988; Kanno 1997; Kang 2000).¹ (1)

a. Motwu1 -ka ku1 -uy emeni-lul salangha-n-ta. everyone-nom he-gen mother-acc love-prs-decl ‘Everyone1 loves his1 mother.’

1 The Yale Romanization is used for transcribing the Korean data.

348 | Kyeong-min Kim and Chung-hye Han

b. Motwu1 -ka ku1 -ka chwukkwu-lul cal ha-n-tako everyone-nom he-nom soccer-acc well do-prs-comp sayngkakha-n-ta. think-prs-decl ‘Everyone1 thinks that he1 plays soccer well.’ The co-existence of these two conflicting intuitions suggests the possibility that variation across native speakers of Korean may exist regarding the availability of a bound variable construal for ku. If this disagreement is indeed a reflection of inter-speaker variation, the analyses proposed in the previous studies listed above would be problematic in that none of their conclusions, whatever they are, correctly represents Korean in general. Furthermore, one pronominal item with two contradictory binding possibilities would pose a challenge to the existing binding theories in the generative tradition (e.g., Chomsky 1981; Reinhart 1983, 1986).² The objective of this paper is two-fold. First, we aim to provide a solid empirical basis for elucidating the precise picture of Korean third-person pronouns, with respect to the availability of bound variable construal. We conducted two experimental studies to meet this objective, and found that some speakers of Korean consistently accepted the use of ku as a bound variable while others consistently did not, suggesting an existence of inter-speaker variation regarding the interpretative status of ku. Second, we intend to provide a principled account for why the variation phenomenon arises and how it can be captured in binding-theoretic terms. Taking into consideration the historical background and present status of ku, we have been led to the conclusion that child learners of Korean may not have sufficient and clear evidence from the input data regarding the grammar of ku. We propose, following Han, Lidz, and Musolino (2007), that due to such a lack in input, they must randomly choose between two competing grammars of ku, only one of which allows a bound variable construal. This thus results in two groups of speakers in the Korean speech community. We then attempt to provide theoretical support for this line of analysis by adopting the pronominal typology in Déchaine and Wiltschko (2002), wherein each type of pronoun has a distinct binding-theoretic status. The remainder of this paper proceeds as follows. In Section 2, we briefly review some of the previous approaches to the bindability of pronouns in various languages, and evaluate whether and how they can be applied in elucidating Ko-

2 Following Tanya Reinhart’s theory of anaphora (Reinhart 1983, 1986; Grodzinsky and Reinhart 1993), we assume that there are two types of pronominal anaphora, binding and coreference, and the binding theory (or the grammar) only governs binding.

Inter-speaker variation in Korean pronouns |

349

rean pronouns. Section 3 presents the methodology of the two experimental studies, along with the finding that there is inter-speaker variation regarding the status of ku. An account of the empirical findings and a discussion of theoretical implications are presented in Section 4. Section 5 concludes with an overall summary and a discussion of future research.

2 Previous views on bindability of pronouns In this section, we discuss the competition-based, constraint-based, and structural approaches.

2.1 Competition-based approach The core concept of the competition-based approach is that an overt pronoun in pro-drop languages such as Spanish and Japanese cannot function as a bound variable if a null pronoun (henceforth pro) is available in the same configuration (Montalbetti 1984; Kanno 1997, among others). This is illustrated in (2) and (3).³ (2) Spanish Nadie1 cree que pro1 /él∗1 es inteligente. nobody believe.prs comp pro/he cop intelligent ‘Nobody1 believes that he1 is intelligent.’ (Montalbetti 1984: 93) (3) Japanese it-ta no? Dare1 -ga pro1 /kare∗1 -ga sore-o mi-ta-to who-nom pro/he-nom that-acc see-pst-comp say-pst q ‘Who1 said that he1 saw that?’ (Kanno 1997: 266) 3 Montalbetti’s (1984: 94) actual proposal is put forth with the name Overt Pronoun Constraint (OPC), as follows. (i)

Overt Pronoun Constraint: An overt pronoun cannot link to a formal variable if and only if the alternation empty/overt obtains.

In (3), the Japanese long-distance anaphor zibun ‘self’ can appear at the embedded subject position, and thus could also be considered as competing with kare ‘he’ for a bound variable interpretation (see Huang 1991).

350 | Kyeong-min Kim and Chung-hye Han

In non-pro-drop languages such as English, on the other hand, an overt pronoun can readily be construed as a bound variable, as implied from the English translations above, because these languages lack pro, and thus there would be no competition for a bound variable interpretation. The fundamental problem with the competition-based approach is that it implicitly assumes the unbindability of overt pronouns to be a universal characteristic of all pro-drop languages (Kanno 1997; White 2003, among others), and thus can be falsified by even a single counter-example. In fact, the observed variable status of ku, an overt pronoun in Korean (a pro-drop language), undermines its tenability. Additionally, in certain structural positions where no competition exists with null pronouns, overt pronouns would be predicted to be capable of functioning as bound variables. This prediction is, however, not borne out. In Japanese, for instance, pro cannot be bound or be coreferential at the embedded object position (Noguchi 1997), but kare still cannot receive a bound variable interpretation in this configuration, as illustrated in (4).⁴ (4) Daremo1 -ga Mary-ga pro∗1 /kare∗1 -o sitteiru-to it-ta. everyone-nom Mary-nom pro/he-acc know-comp say-pst ‘Everyone1 said that Mary knew him1 .’ (Noguchi 1997: 774) It can be thus concluded that the notion of competition cannot capture the precise distribution and interpretation of overt pronouns in pro-drop languages.

2.2 Constraint-based approach Several constraint-based explanations have been given in the literature to handle the data of pro-drop languages discussed above. For instance, the following constraints have been proposed for the general observation that Japanese kare cannot be construed as a bound variable. (5) a. Kare must be operator-free (Katada 1991: 307). b. Kare must be A󸀠 -free (Aoun and Hornstein 1992: 5).

4 One might argue that in (4), pro and kare are not acceptable since zibun ‘self’ is fine in this context. This “strict” competition-based argument, however, fails to explain why both pro and zibun then should be available in (3). See Kang (1988: 195) for a similar argument.

Inter-speaker variation in Korean pronouns |

351

Hong (1985) also proposes a constraint on pronominal binding, which is subject to parametrization, as stated in (6a). Adopting Chomsky’s (1986 [1984]) notion of local binding (6b), she claims that while in non-pro-drop languages an overt pronoun can be A󸀠 -bound as long as it is not locally A󸀠 -bound, an overt pronoun in pro-drop languages cannot be A󸀠 -bound at all. This parametric difference is illustrated in Japanese (7a) and English (7b), the intended interpretations of which are identical. (6) a. Constraint on pronominal binding: Overt pronouns must be A󸀠 -free at LF in pro-drop languages. Overt pronouns must be locally A󸀠 -free at LF in non-pro-drop languages. (Hong 1985: 95) b. Local binding: X locally binds Y if X binds Y and there is no Z such that X binds Z and Z binds Y. (Chomsky 1986 [1984]: 164–165) (7) a. * Daremo-ga1 [t1 [kare1 -no hahaoya-o] aisite-iru] (not A󸀠 -free) b. Everyone1 [t1 loves [his1 mother]] (non-locally A󸀠 -bound, so locally A󸀠 -free) The problem with the constraint-based treatments is that a proposed constraint is only a circular description of the given phenomenon, which in turn invites another question (Noguchi 1997; Elbourne 2005). Under Aoun and Hornstein’s (1992) proposal, for instance, the unbindability of kare is attributed to the constraint that kare must not be A󸀠 -bound, but it then raises another question: why must kare not be A󸀠 -bound? Regarding the issue of the inconsistent binding theoretic status of the Korean pronoun ku, the main concern of this paper, it would not be satisfactory to simply suggest a constraint such as “Korean ku may or may not be A󸀠 -free,” which would again raise a question as to why it should be so.

2.3 Structural approach The structural approaches of Noguchi (1997) and Déchaine and Wiltschko (2002) have in common the assumption that different binding properties of pronouns in various languages derive from their different structural properties. However, it is crucial to note that they fundamentally differ on whether the binding theory

352 | Kyeong-min Kim and Chung-hye Han

is sensitive to differences in the internal structures of pronouns or differences in their syntactic projections or categories. First, consider (8) from Noguchi (1997). b. N-pronoun

(8) a. D-pronoun

D

DP llRRRRR l l RR lll D NP

he

N

DP

kare According to Noguchi’s (1997) proposal, there are two types of pronouns in languages: D-pronouns that behave like a determiner and N-pronouns that behave like a lexical noun, which English he and Japanese kare respectively fall into. We see above that within a DP, the D-pronoun he is placed at D and the N-pronoun kare at N, each of which is thus claimed to have a distinct internal structure. Assuming the hypothesis that binding only applies to functional (not lexical) items such as determiners, he proposes that D-pronouns can be bound as well as be coreferential while N-pronouns can only be coreferential. From a purely syntactic point of view, his structural approach can be said to imply that the binding property of a pronoun is determined by its internal structure – whether the pronoun is placed at D or N within a DP. Déchaine and Wiltschko (2002), on the other hand, propose that there are three types of pronouns in languages: pro-DP, pro-ɸP, and pro-NP.⁵ (9) a. Pro-DP DP R lll RRRRR l l R ll D ɸP lRR lll RRRRR l l l tú ɸ NP tl’o

c. Pro-NP

b. Pro-ɸP ɸP R lll RRRRR l l R ll ɸ NP newí7-s/he

0

NP N kare

0

5 In Déchaine and Wiltschko (2002: 410), ɸP is considered to correspond to “any intermediate functional projection that intervenes between D and N” and encodes ɸ-features (e.g., number, gender, and person).

Inter-speaker variation in Korean pronouns |

353

Assuming that the grammar can only see the maximal projection of a pronoun for binding, Déchaine and Wiltschko (2002) argue that the three pronoun types project different syntactic categories, DP, ɸP, and NP, each exhibiting distinct binding-theoretic properties: pro-DP, as a definite description, functions like an R-expression, subject to Condition C; pro-ɸP behaves as a (free or bound) variable, corresponding to the “standard Condition B pronoun”; pro-NP is inherently a constant, undefined regarding its binding-theoretic status. In their approach, accordingly, only pro-ɸPs (e.g., Shuswap newí7-s ‘he or she’ and English he) can enter into binding while pro-DPs (e.g., Halkomelem tú-tl’o ‘he or she’) and pro-NPs (e.g., Japanese kare ‘he’) cannot. In the structural approaches, whether a pronoun can function as a bound variable is accounted for by the essential nature of the pronoun itself rather than by the “external” factors such as competition with its alternatives (competitionbased approach) or the nature of its antecedents (constraint-based approach). Given the earlier discussion on the problems with the other two approaches, the structural approaches seem to provide a more principled explanation on the issue in question. However, the observed contrasting binding possibilities of the Korean pronoun ku is still difficult for the structural approaches to deal with, since they assume that the binding property of a given pronoun correlates (in one-to-one fashion) with the type it belongs to. If there indeed exists inter-speaker variation regarding the bound variable use of ku, a possible approach compatible with the structural approaches would be to postulate that a single pronoun may fall under two distinctive structural types. In Section 4, we will adopt this line of approach and attempt to capture the aforementioned variation of ku by postulating two different types of ku, each with distinct binding properties.

3 Experimental investigation As noted in the introduction, much disagreement exists as to the availability of a bound variable construal for the Korean pronoun ku. However, there appears to be a lack of experimental studies in the literature to clarify the true nature of ku. We thus conducted two truth-value judgment task experiments (Experiment 1 and Experiment 2). What we found from these experiments is that while some speakers of Korean consistently allowed ku to be construed as a bound variable, others consistently did not. This empirical finding strongly suggests the presence of interspeaker variation regarding the status of ku. In the remainder of this section, the methods and findings of the two experiments are presented.

354 | Kyeong-min Kim and Chung-hye Han

Before moving on, it should be emphasized that in contrast to ku, the longdistance anaphor caki ‘self’ and pro can readily yield a bound variable interpretation in Korean, as illustrated in (10), indicating that native speakers of Korean are capable of establishing bound variable anaphora. Note also that it is commonly accepted that ku and pro can freely enter into (intra-sentential) coreference, as shown in (11).⁶ (10) a. Motwu1 -ka caki1 -uy/pro1 emeni-lul salangha-n-ta. everyone-nom self-gen/pro mother-acc love-prs-decl ‘Everyone1 loves his1 mother.’ b. Motwu1 -ka caki1 -ka/pro1 chwukkwu-lul cal ha-n-tako everyone-nom self-nom/pro soccer-acc well do-prs-comp sayngkakha-n-ta. think-prs-decl ‘Everyone1 thinks that he1 plays soccer well.’ (11) a. Minswu1 -ka ku1 -uy/pro1 emeni-lul salangha-n-ta. Minswu-nom he-gen/pro mother-acc love-prs-decl ‘Minswu1 loves his1 mother.’ b. Minswu1 -ka ku1 -ka/pro1 chwukkwu-lul cal ha-n-tako Minswu-nom he-nom/pro soccer-acc well play-prs-comp sayngkakha-n-ta. think-prs-decl ‘Minswu1 thinks that he1 plays soccer well.’

3.1 Experiment 1 Experiment 1 was conducted (i) to investigate whether the Korean pronoun ku can be construed as a bound variable, and subsequently, (ii) to see if there is indeed inter-speaker variation regarding the interpretative status of ku.⁷

6 Caki ‘self’ is also available at the position of ku and pro in (11), and is construed as a bound variable (Cho 1996; Kim and Yoon 2009; Han and Storoshenko 2012). 7 The study based on Experiment 1 was presented at the 22nd Japanese/Korean Linguistics Conference at the National Institute for Japanese Language and Linguistics, Japan, in October 2012.

Inter-speaker variation in Korean pronouns |

355

3.1.1 Task A truth-value judgment task was employed in Experiment 1.⁸ First, sentences describing a context followed by a target sentence were presented to participants on a computer screen. The participants were then asked to indicate if the target sentence truthfully depicts the given context by assigning 1 for True and 0 for False.

3.1.2 Participants 18 adult native Korean speakers living in Vancouver, Canada, participated in Experiment 1.⁹ Socio-linguistic factors such as gender, age, and social status were assumed to not be associated with bound variable anaphora, and thus not controlled in this experiment. In order to minimize the probable L2 influence on L1 judgment, however, participants were required to have spent no more than 3 years living in countries other than Korea and to have never attended any nonKorean schools before the age of 16.

3.1.3 Design and materials Each target sentence contained a proper name or motwu ‘everyone’ as a matrix subject and pro or ku as an embedded subject, and each context was biased towards an interpretation of the embedded subject as anaphorically dependent on

8 The truth-value judgment task is an experimental methodology for investigating whether a certain interpretation assigned to a sentence can be licensed by a participant’s grammatical competence (Crain and McKee 1985; Crain and Thornton 1998; Gordon 1998). In a truth-value judgment task, a discourse context is provided in a story acted out in front of a participant usually using toys, and the participant is asked to judge whether the target sentence is true in the provided context. The participant’s judgement, acceptance or rejection, is then interpreted as revealing the state of his or her grammar regarding the linguistic phenomenon being investigated. The truth-value judgment task was originally designed to assess children’s linguistic competence. However, since this method has several advantages, such as reducing performance variables (Gordon 1998), it has also been popularly adopted in the literature to investigate adults’ grammars. 9 The participants were aged between 19 and 29 (average age: 22 years). Most of them were undergraduate students of Simon Fraser University, and the rest were ESL students or visitors from Korea with working holiday visas. All experimental sessions were conducted in a classroom at the Burnaby campus of Simon Fraser University, and each participant received $ 5 for his or her participation.

356 | Kyeong-min Kim and Chung-hye Han

the matrix subject. Hence, two within-subjects factors with two levels each were tested: Subject Type (the matrix subject is referential or quantificational) and Pronoun Type (the embedded subject is a null or overt pronoun). The test conditions are summarized in Table 1. Table 1: Test conditions in Experiment 1. Subject Type

Pronoun Type

Referential

Null Overt

Quantificational

Null Overt

The condition of interest was the Quantificational-Overt condition as our concern was whether ku can be interpreted as a bound variable. The other three conditions thus functioned as controls given that in Korean, pro can be coreferential or bound, and ku can be (at least) coreferential. It should also be noted that the experimental design made it possible to additionally test the tenability of the competition-based approach discussed in Section 2, through a comparison of the participants’ responses in Quantificational-Null and Quantificational-Overt conditions. (12) illustrates a test trial for the Quantificational-Overt condition.¹⁰ In the corresponding Referential-Null trial, motwu ‘everyone’ and ku in the target sentence were replaced with one of the characters’ names in the given context and pro, respectively (e.g., Hanswu-ka nongkwucang-eyse pro nongkwu-lul cal ha-n-tako malha-yess-ta. ‘Hanswu said at a basketball court that he plays basketball well.’). (12) a. Context: Hanswu, Cinswu, and Minswu were playing basketball at a basketball court. Hanswu said that self (caki) plays basketball well. Cinswu also said that self (caki) plays basketball well. Minswu also said that self (caki) plays basketball well.¹¹

10 Each condition was repeated four times using different background scenes with distinctive characters, places, and topics. For the names of characters, typical Korean male names were used. 11 Only English translations of the context sentences are presented here to save space.

Inter-speaker variation in Korean pronouns |

357

b. Target sentence: Motwu-ka nongkwucang-eyse ku-ka nongkwu-lul cal everyone-nom basketball.court-at he-nom basketball-acc well ha-n-tako malha-yess-ta. do-prs-comp say-pst-decl ‘Everyone said at a basketball court that he plays basketball well.’

3.1.4 Procedure The experiment was implemented using the WebExp software package, a webbased program for conducting linguistic experiments (Keller et al. 2009), and was run on a personal computer. The experimental session began with 3 practice trials to train the participants to fully understand their task. After the practice trials, 16 test trials (4 trials per condition) and 24 filler trials followed in a uniquely generated random order. The participants were asked to enter 1 if the target sentence on the last line truthfully described the context given in the preceding sentences, and 0 if it did not (see Figure 1). The crucial assumption was that the participants’ assignment of 1 or 0 in the Quantificational-Overt corresponds to their acceptance or rejection of the bound variable reading of ku.

Figure 1: Screen shot of a test trial in Experiment 1.

358 | Kyeong-min Kim and Chung-hye Han

3.1.5 Findings Figure 2 summarizes the mean acceptance rates by condition: 92 % and 81 % in the Referential-Null and -Overt conditions, and 96 % and 54 % in the QuantificationalNull and -Overt conditions. Data were fitted to generalized linear mixed-effects models using the lmer function in the R statistical software (Bates et al. 2012) to analyze the participants’ responses as a function of pronoun type and subject type, with participant and sentence as random effects. First, the analysis revealed a main effect of pronoun type (coefficient estimate = −4.18, s.e. = .77, z = −5.43, p < .001): regardless of subject type, speakers are significantly more likely to accept sentences with null pronouns than overt pronouns. Second, a significant interaction between pronoun type and subject type was found (coefficient estimate = 2.87, s.e. = .95, z = 3.02, p < .01): for overt pronouns, speakers are more likely to accept referential subject as antecedents than quantificational subjects, while accepting both quantificational and referential subjects as antecedents for null pronouns.

0

20

40

% 60

80

100

Null Overt

Referential

Quantificational

Figure 2: Mean acceptance rates by condition in Experiment 1.

Pairwise comparisons of the mean acceptance rates using Tukey’s tests revealed that in the quantificational subject condition, null pronouns are significantly more likely to be accepted than overt pronouns (p < .001) while in the referential subject condition, the acceptance rates of null and overt pronouns are not significantly different from each other. The analysis also revealed that for overt pronouns, referential subjects are significantly more likely to be accepted as their antecedents than quantificational subjects (p < .001), but for null pronouns, no

Inter-speaker variation in Korean pronouns |

359

6 4 0

2

Number of participants

8

10

significant difference is found between the acceptance rates of quantificational and referential subjects. Taken together, the results suggest that the acceptance rates in the Referential-Null, Referential-Overt, and Quantificational-Null conditions (but not the Quantificational-Overt condition) are uniformly high. In fact, this is predicted from and compatible with the earlier discussion on coreference and bound variable anaphora in Korean: pro can enter into both coreference and binding, and ku can enter into coreference. This result strongly supports that the experimental design was appropriate for assessing participants’ grammars of anaphora, thus ensuring that the 54 % acceptance in the Quantificational-Overt condition, the main concern of the current research, was indeed a result from ku being interpreted through binding. A closer inspection of the data from each participant was conducted to further identify the true nature of the acceptance rate in the QuantificationalOvert condition. That is, all participants (n = 18) were distinguished into three different groups according to their individual acceptance rates on the bound variable interpretation for ku, as demonstrated in Figure 3: accept (more than or equal to 75 % acceptance), ambivalent (equal to 50 % acceptance), and reject (less than or equal to 25 % acceptance). Consequently, a bimodal distribution of responses was observed: the participants nearly always accepted or nearly always rejected the bound variable interpretation for ku. In other words, the 54 % in the Quantificational-Overt condition is not derived from each participant accepting the bound variable reading of ku 54 % of the time, but from 54 % of the participants accepting it. This finding supports the presence of between-speaker (not within-speaker) variation regarding the interpretative status of ku in Korean.

Reject

Ambivalent

Accept

Figure 3: Bimodal distribution of responses to bound variable construal of ku in Experiment 1.

360 | Kyeong-min Kim and Chung-hye Han

Experiment 1 was also designed to investigate whether there is a competition between ku and pro for a bound variable interpretation. Under the competitionbased approach (e.g., Montalbetti 1984), an overt pronoun cannot be bound if a null pronoun can be bound in the same context. Given that the acceptance rate in the Quantificational-Null condition is near 100 %, the 54 % acceptance in the Quantificational-Overt condition vitiates the validity of the competition-based approach in a strict sense, as it would predict (near) 0 % acceptance of the bound variable interpretation of ku. Additionally, in the competition-based approach, a clear split found between speakers regarding the acceptance of bound variable ku would mean that there are two groups of speakers: those who accepted only the bound variable reading of pro belong to the group who is sensitive to the putative “pro-ku competition mechanism,” and those who accepted the bound variable reading of ku as well as pro belong to the group who is not sensitive to the mechanism at all. This is, however, an undesirable explanation which raises a complicated question as to what would cause such differences in speakers’ sensitivity to the competition mechanism. Given the other problems with the competition-based approach raised earlier, it is difficult to conclude that ku indeed competes with pro for a bound variable interpretation.

3.2 Experiment 2 Experiment 2 was motivated by Kang’s (1988) observation that the availability of a bound variable construal of ku may differ depending on clause type. He reports that a bound variable reading of ku is fine in a single clause, as in (13a), while it is very marginal across a clause boundary, as in (13b). Note that only sentences with embedded clauses similar to (13b) were used as test stimuli in Experiment 1.¹² (13) a.

Motwu1 -ka ku1 -uy emeni-lul coaha-n-ta. everyone-nom he-gen mother-acc like-prs-decl ‘Everyone1 likes his1 mother.’

(Kang 1988: 195) b. ?? Motwu1 -ka ku1 -ka hyenmyengha-tako sayngkakha-n-ta. everyone-nom he-nom wise-comp think-prs-decl ‘Everyone1 thinks that he1 is wise.’ (Kang 1988: 194)

12 Kang (1988: 195) maintains that ku generally can be interpreted as a bound variable while cases such as (13b) are exceptions, being controlled by some unknown pragmatic factors.

Inter-speaker variation in Korean pronouns |

361

Experiment 2 was thus conducted (i) to test if there is indeed a difference between the bound variable construal of ku in a single clause and across clauses, and subsequently, (ii) to see if the findings from Experiment 1 (the inter-speaker variation regarding the status of ku) could be further substantiated.¹³

3.2.1 Task The same version of the truth-value judgment task was employed as in Experiment 1. The participant’s task was to judge whether the target sentence describes the given context truthfully.

3.2.2 Participants 37 adult native speakers of Korean were tested.¹⁴ All participants met the same requirements for eligibility for participation as in Experiment 1.

3.2.3 Design and materials Each target sentence contained motwu ‘everyone’ as the matrix subject, and ku as a possessor in a simple clause or as a subject in an embedded clause. Each given context forced ku to be interpreted as anaphorically dependent on the matrix subject. The experiment thus had one within-subjects factor with two levels, resulting in two different conditions: Simple condition (ku is a possessor) and Embedded condition (ku is an embedded subject). (14) illustrates a test trial for the Simple condition. The test trials for the Embedded condition were drawn from those for the Quantificational-Overt condition in Experiment 1 (e.g., Motwu-ka nongkwucang-eyse ku-ka nongkwu-lul cal ha-ntako malha-yess-ta. ‘Everyone said at a basketball court that he plays basketball well.’).

13 The study based on Experiment 2 was presented at the Impact of Pronominal Form on Interpretation Workshop at the University of Tübingen in November 2013. 14 The participants were aged between 20 and 26 (average age: 24 years). Most of them were ESL students, and the rest were visitors from Korea with working holiday visas. All experimental sessions were conducted in the library of Simon Fraser University at Harbour Centre, Vancouver, and each participant was paid $ 10 for his or her participation.

362 | Kyeong-min Kim and Chung-hye Han

(14) a. Context: Hanswu, Cinswu, and Minswu were playing basketball at a basketball court. Hanswu drank self’s (caki) beverage. Cinswu also drank self’s (caki) beverage. Minswu also drank self’s (caki) beverage. b. Target sentence: Motwu-ka nongkwucang-eyse ku-uy umlyoswu-lul masi-ess-ta. everyone-nom basketball.court-at he-gen beverage-acc drink-pst-decl ‘Everyone drank his beverage at a basketball court.’

3.2.4 Procedure Experiment 2 used the same procedure as in Experiment 1, except that 8 test trials (4 trials per condition) and 24 filler trials were presented instead.

3.2.5 Findings

Simple Embedded

0

20

40

%

60

80

100

Figure 4 summarizes the mean acceptance rates by condition: 43 % in the Simple condition and 36 % in the Embedded condition. As in Experiment 1, generalized linear mixed-effects models were used to analyze the participants’ responses as a function of clause type, with participant and sentence as random effects. The analysis revealed no significant effect of clause type, indicating that the accep-

Figure 4: Mean acceptance rates by condition in Experiment 2.

Inter-speaker variation in Korean pronouns |

363

0.6 0.4 0.0

0.2

Simple

0.8

1.0

tance rates in the Simple and Embedded conditions are not significantly different from each other. Additionally, linear regressions testing the correlation between the participants’ acceptance rates in both conditions revealed that the correlation coefficient is significantly different from zero (R2 = .62, t = 7.49, p < .001), indicating that those participants who accepted the bound variable reading of ku did so for both simple and embedded clause types, and those who rejected the bound variable reading did so for both clause types (see Figure 5). Taken together, it can be concluded that there is no significant difference between the availability of a bound variable construal of ku in a single clause or across a clause boundary, contradicting Kang’s (1988) observation that clause type may affect the availability of a bound variable construal of ku.

0.0

0.2

0.4

0.6

0.8

1.0

Embedded

Figure 5: Correlation between acceptance rates in Simple and Embedded conditions.

As in Experiment 1, the acceptance rates of each participant in both conditions were looked into, and we found a tendency that some speakers of Korean consistently accepted the bound variable interpretation of ku while others consistently rejected it, as illustrated in Figure 6. In other words, the 43 % and 36 % acceptance rates in the Simple and Embedded conditions are derived from 43 % and 36 % of the participants accepting the bound variable reading of ku, not from each participant accepting it 43 % and 36 % of the time. This strongly supports the existence of inter-speaker variation regarding the status of ku, thus substantiating the findings from Experiment 1.

25

364 | Kyeong-min Kim and Chung-hye Han

15 10 0

5

Number of participants

20

Simple Embedded

Reject

Ambivalent

Accept

Figure 6: Bimodal distribution of responses to bound variable construal of ku in Experiment 2.

4 Discussion From the two experimental studies, we found evidence supporting the existence of inter-speaker variation in the availability of a bound variable construal for Korean third-person pronouns. Thus, it seems that the lack of consensus regarding the interpretative status of ku in the literature is indeed a reflection of this interspeaker variation. An important discussion question arises at this point: why does the variation phenomenon emerge? As a preliminary step in providing a principled answer to this question, we first consider the historical background and the present status of ku from a language acquisition perspective. The Korean pronoun ku has a somewhat idiosyncratic status among pronouns across languages: in contrast to pronouns in other languages such as English he, it is a rare pronominal item which was “artificially” introduced into the language by a certain group of speakers. It is generally accepted by Korean historical linguists that ku was first used by Korean writers in the early 1920s. It is worth noting the following excerpt from the autobiographical essay by Dong-In Kim (1900–1951), who is considered a pioneering novelist in using ku as a third-person pronoun (An 2008, among others).¹⁵

15 Ku was first used as a gender-neutral pronoun. Kunye ‘she’ was later coined by those writers who were using ku as a masculine pronoun and thus needed a corresponding feminine form (An 2008).

Inter-speaker variation in Korean pronouns |

365

I sometimes felt annoyed and tired with repeating the same names to refer to a character in a novel. I wanted to use a lexical item corresponding to English he or she, but Korean has nothing like that […] among the candidates for a third-person pronoun, ku ‘that’ seemed to be nicer conventionally than ce ‘that’ and others, so I just chose and used it without any hesitation. (Kim 1948; quoted in An 2008)¹⁶

As implied from above, the pronominal ku historically originated from and is thus homophonous with the demonstrative ku ‘that,’ which forms one part of the Korean tripartite demonstrative system, as illustrated in (15). The demonstrative ku can refer to an object close to hearer or invisible (but known) to both hearer and speaker while i ‘this’ and ce ‘that,’ respectively, refer to an object close to speaker and an object distant from both hearer and speaker (Hoji 1990; Sohn 1999). (15) a. i chayksang ‘this desk’ b. ku chayksang ‘that desk’ c. ce chayksang ‘that desk’ The use of ku as a personal pronoun seems to have been introduced to Korean speakers about 90 years ago, and it is thus a relatively recent linguistic evolution in Korean language while the use of ku as a demonstrative determiner has a much longer history (Kang 1988). This historical background of pronominal ku may have resulted in its unique status in present-day Korean. That is, ku is seldom used as a personal pronoun in colloquial speech, either as a coreferential pronoun or as a bound variable. Rather, it is predominantly used in written texts (Suh 1990; Han 2006; An 2008). Accordingly, native speakers of Korean are scarcely exposed to the pronominal usage of ku at an early age: rather, they generally first encounter it in written forms in essays or novels at school age. Even in the written texts, however, the pronominal ku tends to be avoided: in fact, repetition of full noun phrases is preferred in anaphoric contexts (Lee et al. 1997). Moreover, even when the pronominal ku is used, it mostly takes a discourse referent rather than an intrasentential one (Suh 1990). Given this situation, we have reached a conclusion that child learners of Korean may not receive sufficient or clear evidence from the input data regarding the grammar of ku.¹⁷

16 This quote was originally written in Korean, and translated here into English by the authors. 17 As noted in Section 3, this does not necessarily mean that the child learners are not able to establish bound variable anaphora as they might already have acquired it easily with the longdistance anaphor caki ‘self’ or pro.

366 | Kyeong-min Kim and Chung-hye Han

Following the two-grammar hypothesis by Han, Lidz, and Musolino (2007), we propose that such a paucity in input concerning the pronominal ku may be the source of the inter-speaker variation regarding the availability of its bound variable construal. According to Han et al., when the primary linguistic data that child learners of a given language are exposed to is compatible with (at least) two competing grammars, along with the lack of relevant input data that would assist them to choose between the two, they have to choose one grammar at random. As a consequence, some learners may acquire one grammar and others may acquire another, thus resulting in the presence of two groups of speakers in the given language community.¹⁸ We argue that the situation of the pronominal ku is similar to the situation described in Han et al. (2007). The input data regarding ku to child learners of Korean is compatible with two competing grammars, the pronominal grammar and the demonstrative grammar. The relevant input to aid them to choose one grammar over the other is scarce, and so they must choose one grammar at random. Accordingly, some may acquire the pronominal grammar for ku, in which case it can be construed as a bound variable and be referential as well, and others may acquire the demonstrative grammar for ku, in which case it can only be referential. We suggest that this is how the inter-speaker variation regarding the bound variable construal of ku emerges and exists in Korean. If our arguments based on the two grammar hypothesis are indeed valid, it should be admitted that there exist two kinds of pronoun ku: one kind of ku which can be bound and the other which cannot be bound. If this is so, another important question arises: how can the two kinds of ku with contrasting binding properties be captured in binding-theoretic terms? We attempt a proposal based on the work of Déchaine and Wiltschko (2002). In their structural approach, they propose three types of pronouns which respectively have different maximal projections: pro-DP, pro-ɸP, and pro-NP. This distinction of pronouns is fundamentally distinguished from Noguchi’s (1997) approach, wherein two types of pronouns (D-pronoun and N-pronoun) are placed at different syntactic heads, D and N, but are uniformly of the same syntactic category DP. It has been discussed that such a fundamental contrast between the two structural approaches is a reflection of their different underlying views on whether distinct binding properties must be represented by distinct syntactic projections

18 Han et al.’s proposal is based on the experimental evidence of the inter-speaker variation in judgment of negation and quantifier scope in Korean, which is argued to be compatible with two competing grammars: verb-raising grammar and non-verb-raising grammar. As Korean is a headfinal language, they argue, the input data provide little indication of (non-)verb raising to child learners, and the leaners thus have to choose one grammar at random.

Inter-speaker variation in Korean pronouns |

367

or distinct internal structures. We follow Déchaine and Wiltschko’s (2002:409) argument that the grammar (or syntax) cannot see the inside of a DP and therefore, internal structural differences of pronouns cannot lead to their “external” differences (i.e., binding properties). We therefore assume that the binding theory cannot be considered to be sensitive to the internal structure of a given pronoun when its outer layer is a DP (cf. Koak 2008). Adopting Déchaine and Wiltschko’s (2002) typology of pronouns, we propose that the two kinds of ku present distinct binding properties because they are of different syntactic categories. First, one of the two kinds is of category ɸP, as illustrated in (16), with the binding-theoretic status of an ordinary Condition B pronoun such as English he, and it thus can be used as a bound or free variable. (16) Pro-ɸP ku ɸP T jjjj TTTTTTT j j j T jj ɸ NP ku

0

The other ku is of DP, as illustrated in (17), and has the binding-theoretic status of an R-expression (see Kwon and Polinsky 2011 for a similar argument that overt pronouns in Korean function as R-expressions). The pro-DP ku, subject to Condition C, cannot be construed as a bound variable, and can only refer.¹⁹

19 Under the assumption that the Korean demonstratives, ku ‘that’, ce ‘that over there’ and i ‘this’, are heads of DP (Kang 2001; Chang 2009, among others), the pro-DP ku shares the same syntactic structure with the demonstrative ku, and thus can be claimed to be a demonstrative pronoun (a demonstrative determiner used pronominally with an empty NP). This pro-DP analysis of ku can be cross-linguistically supported by German demonstrative pronouns, which are argued to be pro-DPs in Déchaine and Wiltschko (2002) and Baltin et al. (2013). In (spoken) German, the definite articles der, die, and das can be used as demonstrative pronouns that are inflected for case, gender, and number (Wiltschko 1998; Bosch et al. 2003). Much like the proposed pro-DP ku, the German demonstrative pronouns can refer to individuals as in (i) and (ii), but cannot be construed as bound variables as in (iii). Note that in contrast to the demonstrative pronouns, the corresponding personal pronouns in German (ihn, sie, er) can be bound as well as referential, which are analyzed as pro-ɸPs in Déchaine and Wiltschko (2002) and Baltin et al. (2013). (i)

Maria hat ihn/den gesehen. Mary have.prs him/dem see.prf ‘Mary has seen him.’ (Wiltschko 1998: 144)

368 | Kyeong-min Kim and Chung-hye Han

(17) Pro-DP ku DP T jjjj TTTTTTT j j j T jj D ɸP jTT jjjj TTTTTTT jjjj ku ɸ NP 0 To summarize the proposal, given the lack of relevant input data, child learners of Korean must choose at random either the pro-DP ku or the pro-ɸP ku: some may acquire the pro-ɸP ku, which complies with the pronominal grammar and thus can be bound as well as be referential; others may acquire the pro-DP ku, which complies with the demonstrative grammar and thus can only be referential.

5 Conclusion To conclude, using an experimental methodology, we have demonstrated a novel empirical finding that there exists inter-speaker variation in the availability of a bound variable construal of the Korean third-person pronoun ku, and have attributed the inconsistency in the literature regarding its status to this interspeaker variation. Adopting the syntactic typology of pronouns in Déchaine and Wiltschko (2002), we have proposed that the variation phenomenon is a reflection of the co-existence of two types of ku, pro-DP ku that complies with the demonstrative grammar, and pro-ɸP ku that complies with the pronominal grammar. Following Han et al.’s (2007) two grammar hypothesis, we argued that the source of the variation is the input data to child learners of Korean that provide little evidence to aid them to choose one grammar over the other.

(ii)

Paul sah eine Frau hereinkommen. Sie/Die trug einen schwarzen Mantel. Paul see.pst a woman enter she/dem wear.pst a black coat ‘Paul saw a woman enter. She was wearing a black coat.’ (Bosch et al. 2003: 1)

(iii) Jeder Mann1 glaubt, daβ er1 /*der1 dumm ist every man believe.prs comp he/dem stupid cop.prs ‘Every man1 believes that he1 is stupid.’ (Wiltschko 1998: 144)

Inter-speaker variation in Korean pronouns |

369

It follows from our proposal that, in contrast to personal pronoun ku, the grammar of demonstrative ku should be uniform, as child learners of Korean are provided with sufficient input regarding demonstrative ku. An analysis consistent with our proposal on personal pronoun ku given here would be to uniformly place demonstrative ku in the pro-DP category of Déchaine and Wiltschko’s typology. This then predicts that all Korean speakers should disallow a bound variable reading for definite descriptions with demonstrative ku. In light of observations found in the literature that English and other languages (Evans 1977; Nishigauchi 1990; Noguchi 1997; Elbourne 2008) allow a bound variable reading for definite descriptions headed by a demonstrative in certain contexts, it remains an important task to test for the availability of such a reading in Korean. We leave this for future research. Acknowledgment: We thank Nancy Hedberg and Keir Moulton for their invaluable comments and constructive feedback on the research reported here. We are also grateful to Meghan Jeffrey, Queenie Chan, and the audience at the Impact of Pronominal Form on Interpretation Workshop, which was held at the University of Tübingen on November 15–17, 2013, for their insightful comments and questions. All errors are ours. This work was partially supported by SSHRC 410-2007-2169 and SSHRC 435-2014-0161 to Han.

References An, So jin. 2008. Sowui saminching taymyengsa ’ku,kunye’uy kinungey tayhaye [The function of third-person pronouns ‘ku, kunye’]. Hankwuke Hakhoy [Korean Linguistics] 38. 145–164. Aoun, Joseph & Norbert Hornstein. 1992. Bound and referential pronouns. In James C.-T. Huang & Robert May (eds.), Logical structure and linguistic structure, 1–23. Dordrecht: Kluwer. Baltin, Mark, Rose-Marie Déchaine & Martina Wiltschko. 2013. The structural heterogeneity of pronouns. Paper presented at the 2013 LSA Annual Meeting, Boston, MA. Bates, Douglas, Martin Maechler & Ben Bolker. 2012. lme4: Linear mixed-effects models using S4 classes, R package, version 0.999999-0. http://CRAN.R-project.org/package=lme4. Bosch, Peter, Tom Rozario & Yufan Zhao. 2003. Demonstrative pronouns and personal pronouns. German der vs. er. In Proceedings of the EACL 2003. Workshop on the computational treatment of anaphora, Budapest. Chang, Soo-Jung. 2009. Nominal structure and interpretation. Athens, GA: University of Georgia dissertation. Cho, Dong-In. 1996. Anaphor or pronominal?. Language Research 32(4). 621–636. Choe, Hyon-Sook. 1988. Restructuring parameters and complex predicates: A transformational approach. Cambridge, MA: MIT dissertation. Chomsky, Noam. 1981. Lectures on government and binding. Dordrecht: Foris.

370 | Kyeong-min Kim and Chung-hye Han

Chomsky, Noam. 1986 [1984]. Knowledge of language: Its nature, origin, and use. New York: Praeger. Crain, Stephen & Cecile McKee. 1985. The acquisition of structural restrictions on anaphora. In Stephen Berman, Jae-Woong Choe & Joyce McDonough (eds.), Proceedings of NELS 15, 94–110. Amherst, MA: GLSA. Crain, Stephen & Rosalind Thornton. 1998. Investigations into Universal Grammar: A guide to experiments on the acquisition of syntax and semantics. Cambridge, MA: MIT Press. Déchaine, Rose-Marie & Martina Wiltschko. 2002. Decomposing pronouns. Linguistic Inquiry 33(3). 409–442. Elbourne, Paul. 2005. Situations and individuals. Cambridge, MA: MIT Press. Elbourne, Paul. 2008. Demonstratives as individual concepts. Linguistics and Philosophy 31(4). 409–466. Evans, Gareth. 1977. Pronouns, Quantifiers, and Relative Clauses (I). Canadian Journal of Philosophy 7(3). 467–536. Gordon, Peter. 1998. The truth-value judgment task. Cambridge, MA: MIT Press. Grodzinsky, Yosef & Tanya Reinhart. 1993. The innateness of binding and coreference. Linguistic Inquiry 24(1). 69–101. Han, Chung-hye, Jeffrey Lidz & Julien Musolino. 2007. V-raising and grammar competition in Korean: Evidence from negation and quantifier scope. Linguistic Inquiry 38(1). 1– 47. Han, Chung-hye & Dennis R. Storoshenko. 2012. Semantic binding of long-distance anaphor caki in Korean. Language 88(4). 764–790. Han, Na-Rae. 2006. Korean zero pronouns: Analysis and resolution. Philadelphia, PA: University of Pennsylvania dissertation. Hoji, Hajime. 1990. On the so-called overt pronouns in Japanese and Korean. In Eung-Jin Baek (ed.), Seventh International Conference on Korean Linguistics, 61–78. Osaka: International Circle of Korean Linguistics and Osaka University of Economics and Law. Hong, Sungshim. 1985. A and A󸀠 binding in Korean and English: Government-Binding parameters. Storrs, CT: University of Connecticut dissertation. Huang, C. -T. James. 1991. Remarks on the status of the null object. In Robert Freidin (ed.), Principles and parameters in comparative grammar, 56–76. Cambridge, MA: MIT Press. Kang, Myung-Yoon. 1988. Topics in Korean syntax: Phrase structure, variable binding and movement. Cambridge, MA: MIT dissertation. Kang, Myung-Yoon. 2001. DPwa kwukeuy myengsakwu [DP hypothesis and Korean nominal phrases]. Hankwukehak [Korean Linguistics] 13. 15–51. Kang, Nam-Kil. 2000. Reflexives and the Linking Theory in Universal Grammar. Oxford, UK: University of Oxford dissertation. Kanno, Kazue. 1997. The acquisition of null and overt pronominals in Japanese by English speakers. Second Language Research 13(3). 265–287. Katada, Fusa. 1991. The LF representation of anaphors. Linguistic Inquiry 22(2). 287–313. Keller, Frank, Subahshini Gunasekharan, Neil Mayo & Martin Corley. 2009. Timing accuracy of web experiments: A case study using the WebExp software package. Behavior Research Methods 41(1). 1–12. Kim, Dong-In. 1948. Mwuntan samsipnyenuy cachwui [30 years of my writing life]. Seoul: Sinchenci. Kim, Ji-Hye & James H. Yoon. 2009. Long-distance bound local anaphors in Korean: An empirical study of the Korean anaphor caki-casin. Lingua 119(5). 733–755.

Inter-speaker variation in Korean pronouns |

371

Koak, Heeshin. 2008. A morpho-syntactic approach to pronominal binding. University of Pennsylvania Working Papers in Linguistics 14(1). 227–240. Kwon, Nayoung & Maria Polinsky. 2011. Anaphoric inventories and bound variable interpretation: Evidence from Korean. Poster presented at the 24th annual CUNY conference on human sentence processing, Stanford University, CA. Lee, Ik-Sup, Sang-Uk Lee & Wan Chae. 1997. Hankwukuy ene [Korean language]. Seoul: Singu Mwunhwasa. Montalbetti, Mario. 1984. After binding: On the interpretation of pronouns. Cambridge, MA: MIT dissertation. Nishigauchi, Taisuke. 1990. Quantification in the theory of grammar. Dordrecht: Kluwer. Noguchi, Tohru. 1997. Two types of pronouns and variable binding. Language 73(4). 770–797. Reinhart, Tanya. 1983. Anaphora and semantic interpretation. London: Croom Helm. Reinhart, Tanya. 1986. Center and periphery in the grammar of anaphora. In Babara Lust (ed.), Studies in the acquisition of anaphora, vol. 1, 123–150. Dordrecht: Reidel. Sohn, Ho-Min. 1999. The Korean language. Cambridge: Cambridge University Press. Suh, Jinhee. 1990. Scope phenomena and aspects of Korean syntax. Los Angeles, CA: University of Southern California dissertation. White, Lydia. 2003. Second language acquisition and Universal Grammar. Cambridge: Cambridge University Press. Wiltschko, Martina. 1998. On the syntax and semantics of (relative) pronouns and determiners. The Journal of Comparative Germanic Linguistics 2(2). 143–181.

Hendrikje Ziemann and Esther Ruigendijk (University of Oldenburg)

L2 speakers’ processing of reflexives and personal pronouns A self-paced reading study of German learners of Dutch Abstract: We investigated the processing of pronominal elements in Dutch by speakers of German who were second language learners of Dutch. We tried to find out which strategies second language learners employ in processing pronominal elements – syntactic operations or discourse operations. In order to find an answer to this question we carried out a self-paced reading study in which we compared reading times to reflexives and personal pronouns in different sentence structures. The results indicate that L2 speakers’ processing indeed differs from L1 speakers. We argue that our results support the idea that different processing strategies were applied in L1 and L2 processing as has been argued for by Ullman (2001) and Clahsen and Felser (2006). Keywords: personal pronouns, reflexives, second language acquisition, sentence processing, German, Dutch.

1 Introduction Second language (L2) speakers’ sentence processing has been suggested to differ from that of native (L1) speakers. Ullman (2001, 2006) and Clahsen and Felser (2006), for instance, posit that L2 speakers are unable (or less able) to use syntax for sentence processing and instead rely on non-syntactic operations based on lexical or discourse information, contrary to L1 speakers. One linguistic domain that would be relevant for testing the ideas of Ullman (2001) and Clahsen and Felser (2006) is that of pronominal elements, since their interpretation has been argued to depend on different linguistic operations (e.g. Reinhart and Reuland 1993, Reuland 2011). Reflexives, like Dutch zich (himself) can be computed through syntactic operations (feature checking), whereas personal pronouns, like

374 | Hendrikje Ziemann and Esther Ruigendijk

him are interpreted through discourse operations (coreference, see for more details below).¹ To contribute to the issue of L2 speakers’ sentence processing, we therefore investigated German speakers’ processing of pronominal elements in their L2 Dutch with a self-paced reading study. We concentrated on reflexives zich and personal pronouns hem in structures in which the antecedent and pronominal element were in a co-argument relation and in structures in which the pronominal element was part of a locative PP (see (1a) and (1b)). (1)

a. Pronominal elements in a co-argument relation Dutch: Jani waste zichi /hem∗i/j . German: Jani waschte sichi /ihn∗i/j . Jan washed SE/HIM ‘Jan washed himself/him’ b. Pronominal elements in a PP Dutch: Jani legde het boek naast zichi /hemi/j . German: Jani legte das Buch neben sichi /ihn∗i/j . Jan put the book next to SE/HIM ‘Jan put the book next to himself/him’

According to Reuland (2001) the reflexive in (1a) should be processed by a syntactic operation, whereas the reflexive in (1b) should be computed by a discourse operation (see below for an explanation). Reuland (2001, 2011) proposes an economy hierarchy for these different types of dependencies. Syntactic dependencies are the most economical in this hierarchy, since they depend on only one operation, whereas discourse dependencies are relatively more costly, since they need more operational steps. Syntactic operations, more specifically those establishing a syntactic dependency with the reflexive zich as in (1a), should thus be less costly and – as we argue – less time consuming, hence processed quicker, than the discourse operations needed for establishing a dependency with zich in (1b). L1 speakers are therefore expected to show a difference in processing of reflexives in (1a) and (1b), in that the reflexive in (1a) should be easier and quicker to process than the one in (1b). But if Ullman (2001) and Clahsen and Felser (2006) are right such a difference should not be found for L2 speakers and reaction times (RT) should be the same for the pronominal elements in all sentence types.

1 We use the term pronominal element to refer to both reflexives and personal pronouns. We will call a pronominal element a reflexive or a personal pronoun when we need to distinguish between them.

L2 speakers’ processing of reflexives and personal pronouns |

375

Although our main interest lies on pronoun processing strategies in the L2 in general, apart from that, Dutch provides an interesting ground for investigating pronominal elements. This is because German and Dutch have small differences in the pronominal dependencies that are possible. An example can be found in (1b), where the personal pronoun as well as the reflexive in Dutch may refer to the subject Jan. In German, however, only a reflexive may be used to refer to the subject Jan whereas the pronoun ihn may only refer outside the clause. The question is how this difference affects pronoun processing by L2 speakers. It might be the case that there is interference from L1 German into L2 Dutch. This interference could lead to processing costs and hence higher reaction times for personal pronouns compared to reflexives in PPs for the L2 speakers. With this study we try to answer the question which strategies/operations German L2 speakers of Dutch employ to process reflexive and personal pronouns in Dutch. First, we will present and discuss theories on the binding of pronominal elements that form the basis of our hypotheses. After that, we will report studies of L1 and L2 pronoun processing which we will connect to the linguistic theory. Then we will present our reaction time study, the method and results and finally discuss it in light of the model of Reuland (2001) on referential dependencies, as well as theories on second language sentence processing such as the declarative and procedural model by Ullman (2001) and the shallow structure hypothesis by Clahsen and Felser (2006).

2 Binding and the processing of pronominal elements In Primitives of Binding, Reuland (2001) presents a linguistic model for the establishment of pronominal dependencies, which can be formed at different linguistic levels (depending on form and structure): through syntax (feature checking), semantics (bound-variables) and/or discourse (coreference) operations. Interpreting a pronominal element through syntax involves fewer computational steps than e.g. a discourse operation involves, and should hence be more economical. Before we discuss this economy hierarchy in more detail, we present Principle A and B of Chomsky’s (1981) Government and Binding Theory and Reflexivity by Reinhard and Reuland (1993), since Reuland’s (2001) economy hierarchy for pronominal dependencies builds upon the ideas of these earlier theories. To account for the distribution of reflexives and personal pronouns as in (2) and (3), Chomsky (1981) formulated Principle A and B.

376 | Hendrikje Ziemann and Esther Ruigendijk

Binding Principles (Chomsky, 1981) Principle A: A reflexive pronoun² is bound within its governing category³ Principle B: A pronoun is free in its governing category Principle A describes the dependency of a reflexive as in (2). According to this principle zichzelf must be bound by Ellie because they are in the same governing category. In (3) the personal pronoun haar cannot be bound by Ellie as both appear in the same governing category in which a pronoun cannot be bound. (2) Reflexive in a co-argument relation Elliei tekent zichzelfi . Ellie draws SELF ‘Ellie is drawing herself’ (3) Personal pronoun in a co-argument relation Elliei tekent haar∗i . Ellie draws HER ‘Ellie is drawing her’ Chomsky’s Principle A and B clearly differentiate between the two types of pronominal elements and predict a complementary distribution: when a reflexive can be used for establishing a referential dependency, a personal pronoun is not allowed and vice versa. However, if we look at (4) a problem arises for these principles. The pronominal elements zich and hem are not in complementary distribution and can both be used to refer to the subject Jan. (4) Jani legde het boek naast zichi /hemi/j . Jan put the book next to SE/HIM ‘Jan put the book next to himself/him’ Another example where Principle A and B have problems describing the pronominal distribution is demonstrated in (5) and (6). In languages like Dutch, there are two kinds of reflexives. These are the so-called SE (zich) and SELF (zichzelf)-

2 In fact Chomsky (1981) used the term ‘anaphor’. For reasons of consistency, we stick to the term reflexive here, referring to the same thing. 3 A governing category is a clause consisting of a governor which c-commands another element. In (2) the governor is Ellie and it c-commands zich.

L2 speakers’ processing of reflexives and personal pronouns | 377

anaphors (here: reflexives). In (5) only zich is allowed, but never zichzelf and in (6) zichzelf is correctly used, and zich is not possible. The precise nature of the difference between zich and zichzelf will be illustrated below. For now, it is sufficient to know that Principle A cannot explain the difference between these two types of reflexives. (5) Reflexive in a co-argument relation Jani schaamde zichi /*zichzelfi . Jan ashamed SE/SELF ‘Jan ashamed himself’ (6) Reflexive in a co-argument relation Jani haatte zichzelfi /*zichi . Jan hated SELF/SE ‘Jan hated himself’ As should become clear, Chomsky’s Principle A and B are not able to capture all instances of pronominal reference. The Reflexivity theory (Reinhard and Reuland 1993) started out with these problems in reformulating binding phenomena into Condition A, and Condition B. Furthermore, they discussed the Chain Condition. With these conditions, they argued, it is possible to account for the phenomena in (2) to (6). Binding Conditions from Reflexivity (Reinhart and Reuland 1993). Condition A: A reflexive-marked syntactic predicate is reflexive. Condition B: A reflexive semantic predicate is reflexive marked. General Condition on A-Chains: A maximal A-chain (α1 , . . . , α n ) contains exactly one link – α1 – that is both +R and case-marked. We will concentrate on Condition B and the Chain condition here. Condition B says that a predicate has to be marked reflexively in order to be interpreted reflexively. If not marked reflexively the predicate cannot be interpreted reflexively. Reflexive marking of predicates can be done in two ways. The first possibility is that a predicate is lexically reflexive (as in (5)), and the second that one of the arguments of the predicate is a reflexive marker⁴, as in (6). Unlike Chomsky’s Principle A and B,

4 Note that Reinhart and Reuland (1993) use the term ‘SELF-anaphor’ here, for reasons of consistency we stick to the term reflexive. Note also that Reinhart and Reuland argue that simple reflexives (SE-anaphors) are not reflexive markers.

378 | Hendrikje Ziemann and Esther Ruigendijk

Condition B only applies to semantic predicates and hence to co-arguments (we refer to Reinhart and Reuland (1993) for a full discussion of why this should be the case). Pronominal elements appearing in other structures are not governed by this condition, but by the Chain Condition instead. Contrary to Chomsky (1981), Reflexivity differentiates between different kinds of reflexives. A SE-anaphor (Simple Expression, such as zich in Dutch) is a simple reflexive, which does not mark a predicate reflexively. A SELF-anaphor (such as Dutch zichzelf, English himself ), however, is a complex reflexive, which can mark a predicate reflexively. Looking at (5) zich is allowed to refer to Jan, but not zichzelf and in (6) it is the other way around. This is because schamen in (5) is lexically reflexive and haten in (6) is not. Thus in (5) zichzelf is not needed to mark the predicate reflexive as it is already lexically reflexive marked, but in (6) zichzelf has to be used to mark the predicate reflexively. Otherwise it would be unmarked and could not be interpreted reflexively. In (2) the reflexive zichzelf and the antecedent Ellie are co-arguments of the predicate tekenen. The two elements are co-indexed (i.e. form a reflexive predicate) and therefore the predicate must be marked reflexively. Since the verb tekenen is reflexive marked by the reflexive zichzelf, Condition B is met. In (3), the personal pronoun haar and the antecedent Ellie are co-arguments, but the personal pronoun does not reflexive mark the predicate and hence Condition B is violated. A reflexive interpretation is ruled out. In (4) Condition B does not apply, because here the pronominal element and antecedent are not co-arguments of the same predicate. In (4) the pronominal element is an argument of the preposition naast. The PP is an argument of the predicate leggen. Thus in (4) the pronominal element is not directly an argument of the semantic predicate. This means that Condition B does not apply in these structures and thus cannot be violated. As a consequence the reflexive and the personal pronoun may both refer to the subject Jan. Condition B also does not explain cases like (7), where the pronominal element and the antecedent are also not co-arguments of the same predicate. Therefore, Reinhard and Reuland (1993) adapted the Chain Condition to be able to explain structures like (7). Here the Chain Condition rules out that the personal pronoun hem refers to the subject Jan, since both elements are specified for Case and are referentially independent (+R). The reflexive zich, being referentially dependent (−R), instead can be used to refer to the subject Jan. The Chain Condition also applies in (3). The reflexive pronoun is not ruled out by the Chain Condition because it is (−R), but the personal pronoun is ruled out by the chain condition (as well as by Condition B) because both Ellie and haar are (+R).

L2 speakers’ processing of reflexives and personal pronouns | 379

(7) Jani zag zichi /*hemi dansen. Jan saw SE/HIM dance ‘Jan saw himself/him dance’ As we already said Condition B does not apply in structures like (4), hence the pronoun can refer to the subject antecedent. The question is why the Chain Condition does not rule out this local dependency for the personal pronoun. To cite Reinhart and Reuland (1993, p. 693): ‘the precise definition of the A-Chain domain is a topic of much current research and debate’. They use the approach on chains by Chomsky (1986) which entails that an independent P (such as naast in (4)) is a Minimality Barrier which means that the pronoun does not form a chain with the subject and hence the Chain Condition does not apply (we refer to Reinhart and Reuland, footnotes 32 and 45 for more details). Interestingly, pronominal elements in locative PPs in German unlike in Dutch and English are in complementary distribution (see (8)), as already mentioned in the introduction. (8) Jani legte das Buch neben sichi /*ihni . Jan put the book next to SE/HIM ‘Jan put the book next to himself/him’ This difference between German and Dutch could be caused by the presence of case in German (and not Dutch). Reuland (2001, 2011) argues that – for reasons of case assignment – in German the preposition neben and the verb legen have a much closer relationship than in Dutch and English, which results in a coargument-like relation between the prepositional phrase neben sich/ihn and the subject Jan. If there is such a relation, Condition B in German would apply and rule out the personal pronoun but not the reflexive (see Ruigendijk and Schumacher (2011) for an explanation along those lines for processing differences between German and Dutch speakers). Reuland (2001) uses the differences between reflexives and personal pronouns and syntactic, semantic, and discourse operations to formulate his economy hierarchy. He assumes that syntactic operations, bound variable operations, and discourse operations all differ in costs. A syntactic operation should be the least costly operation because it involves two steps of assigning a variable and forming a chain. A chain has to be formed between the reflexive and the antecedent and this chain creates one object. The bound variable interpretation should involve three steps of representation, namely the formation of two chains, which both assign an identical variable to the object. A discourse operation

380 | Hendrikje Ziemann and Esther Ruigendijk

consists of four steps. Here, two different chains are created and two variables are assigned to the objects. Thus, a discourse operation is the most costly and a syntactic operation the most economic. In other words, it should be relatively easier, or to put it in processing terms: quicker, to interpret a pronominal element through syntax compared to the other two types of operations. As Reuland (2001) argues a syntactic operation is possible for reflexives in coargument relations (for example (5)) because the reflexive is an argument of the predicate and forms a chain with the antecedent. Reuland (2001) assumes that in these cases a syntactic operation is used for interpretation, and all other operations will be blocked by the economy hierarchy because their computations would be more costly. The reflexives in a PP ((4) and (8)) cannot be interpreted by syntax only, since they are not co-arguments with the antecedent, and should instead be interpreted by a bound variable or discourse interpretation. Personal pronouns always have to be processed through discourse (or semantic variable binding) regardless of whether they appear in a co-argument relation with the antecedent or not. To sum up, according to the theoretical assumptions formulated by Reuland (2001) a reflexive pronoun in a co-argument structure should evoke fewer costs compared to personal pronouns and reflexives in a non-co-argument structure. We define processing cost here as the time taken to process a pronominal element, which can – albeit indirectly – be measured through reaction time studies, in our case self-paced reading.

3 Processing of pronouns and reflexives in L1 and L2 Studies focussing on the processing of pronominal elements by L1 speakers showed that reflexives in a co-argument relation were processed more rapidly than personal pronouns in a co-argument relation (Nicol and Swinney 1989, Hendriks, Banga, van Rij and Cannizzaro 2011), thus supporting Reuland’s (2001) economy hierarchy for establishing referential dependencies. Also, Burkhardt (2005), and Schumacher and others (2010) showed that English and Dutch L1 speakers take indeed longer in processing a reflexive pronoun in a PP which needs a discourse operation, compared to a reflexive in a co-argument relation, which needs a syntactic operation only. This supports the idea that syntax is less costly and hence computed faster than discourse. Contrary to Burkhardt (2005), Ruigendijk and Schumacher (2011) could not replicate these results for German, which they argue is probably due to cross-linguistic differences between these

L2 speakers’ processing of reflexives and personal pronouns | 381

languages, as we already discussed before (see (1b) and (4) vs. (8)). This supports our idea that these cross-linguistic differences influence processing, and hence might affect L2 speakers as well. The comprehension of reflexive pronouns in L2s like English, or languages, which are freer in the assignment possibilities of pronominal elements like Japanese, was examined in several studies (Hirakawa 1990, Thomas 1991, Matsumura 1994, Sorace and Filiaci 2006, Demirci 2000 and White, Bruhn-Garavito, Kawasaki and Prevost 1997). These studies indicated that L2 speakers tend to assign pronominal elements to antecedents as they do in their L1. Thus there seems to be an influence from the mother tongue. Some of these studies also found that L2 speakers with high proficiency rely more on the properties of the second language than on the first one (e.g. Thomas, 1991). An eye-tracking study of Roberts, Gullberg and Indefrey (2008) showed that German and Turkish L2 speakers of Dutch slow down compared to L1 speakers in reading of personal pronouns in co-argument relations. Felser et al. (2009) and Felser and Cunnings (2012) found longer reading times for reflexives in a coargument relation (9) for which two possible antecedents where available (Adam and Ian, both masculine like the reflexive), than in sentences were only one of two possible antecedents matched the reflexive pronoun’s features as in (10), where only Richard matches the reflexive in gender. The authors argued that L2 speakers evaluate the two possible antecedents, which means that they do not rely on syntax only, since applying Principle A (a syntactic principle) would have resulted in the correct interpretation. The fact that processing the pronoun in (9) took longer, shows that L2 speakers used discourse operations (considering both Adam and Ian as possible antecedent). Native speakers do not show this difference, since they can use syntax, and hence apply Principle A immediately, which means no other antecedents in discourse are even considered, according to Felser et al. (2009) which provides evidence for differences between L1 and L2 processing. (9) Adam believes that Ian blames himself. (10) Jane noticed that Richard had cut himself with a very sharp knife. Felser et al. (2009) argue that these results support the Shallow Structure Hypothesis (SSH), first introduced by Clahsen and Felser in (2006). As mentioned in the introduction, L2 sentence processing has been argued to be different from L1 sentence processing. To explain this difference, Ullman (2001) proposed the Declarative and Procedural Model (DP-model), and Clahsen and Felser (2006) presented the SSH, which, discuss somewhat similar ideas that we can use for our study (note that the theories do differ with respect to L2 processing and learn-

382 | Hendrikje Ziemann and Esther Ruigendijk

ing, but not in ways that are relevant here). According to the DP model, in native speakers lexical knowledge is part of declarative memory and grammar is part of procedural memory. Ullman (2001) assumes that procedural memory is less available or even inaccessible for L2 speakers acquiring their L2 after childhood, which might be caused by “greater age-of-exposure sensitivity of grammar than of lexicon” (Ullman 2001: 108). Syntax is therefore probably not processed by this memory system in L2 speakers. Ullman suggests that instead declarative memory takes over the functions of procedural memory for language learning, i.e. for syntax. The problem of this switch is that declarative memory is not as suited for processing syntax as procedural memory. Following this, the DP-model has implications for L1 and L2 speakers’ sentence processing. It assumes that there should be qualitative differences in processing between L1 and L2 speakers with regard to syntactic processing. Clahsen and Felser (2006) too assume that processing differences between L1 and L2 speakers are due to different processing strategies. According to their SSH, L2 speakers are not able to represent syntax as L1 speakers do, because their grammatical representations remain shallow. Shallow means that “L2 learners essentially compute predicate–argument structure representations of the input that capture thematic roles and other aspects of lexical–semantic structure, but which lack hierarchical detail and more abstract elements of syntactic structure” (Clahsen and Felser 2006: 32).

4 Hypotheses Psycholinguistic research conducted on processing of pronominal elements underpinned the dissociation between L1 and L2 speakers. The results of Nicol and Swinney (1989), Burkhardt (2005), Sturt (2003) Hendriks et al. (2011), Roberts et al. (2008), Felser et al. (2009) and Felser and Cunnings (2012) suggest that L1 and L2 speakers employ different processing strategies in the processing of a pronominal element and thus support the assumption of the DP-model and SSH with regard to pronominal processing. Whereas L1 speakers indeed seem to rely on both syntactic operations and discourse operations in interpreting pronominal elements, L2 speakers only seem to make use of discourse operations. Based on Reuland (2001, 2011) and Hendriks et al. (2011) and Burkhardt (2005) we therefore expect that reflexives in a co-argument structure are processed syntactically in L1, whereas reflexives in PPs and personal pronouns are interpreted by discourse operations. This should be reflected by shorter reaction times for reflexives in co-argument structures compared to other pronominal elements.

L2 speakers’ processing of reflexives and personal pronouns |

383

Based on the models by Ullman (2001) and Clahsen and Felser (2006) L2 speakers are expected not to use syntactic operations and hence should show equal reaction times for reflexive and personal pronouns in co-argument structures and in PPs. Finally, as also found by e.g. Sorace and Filiaci (2006), or White, et al. (1997), we expect interference effects of L1 (German) on L2 Dutch. Interference is assumed to impact the processing of personal pronoun in PPs in that higher reaction times are expected compared to reflexives in a PP.

5 Method 5.1 Participants 23 native speakers of Dutch participated in a self-paced reading study (mean age 21.4 years, range 19–29 years, 2 men and 21 women). They served as the L1 group. All of them grew up monolingually in the Netherlands⁵. They were all students at the University of Nijmegen, the University of Amsterdam or the University of Groningen. The L2 group consisted of 16 native speakers of German, 12 women and 4 men, who lived in the Netherlands for at least a half year (mean age 22.9 years, range 20–26 years). They grew up monolingually and started learning Dutch as their L2 at school or university, which means not before the age of 15⁶. None of the participants suffered from a neurological disorder, dyslexia or hearing problems.

5.2 Material The self-paced reading study consisted of 120 filler items and 105 test items. These 105 test items were divided into five conditions. In two conditions either a reflexive or a personal pronoun was tested in a non-co-argumenthood relation. In these non-co-argumenthood structures all words were the same except for the pronominal element (see Ia). In two other conditions, the pronominal elements were in a co-argumenthood relation and again the only lexical difference was the choice of the pronominal element (Ib).The fifth condition will not be discussed in this article.⁷ This way, we report here 4 conditions, with 21 items each. 5 The L1 speakers learned English at school as their first L2. 6 All L2 speakers learned English at school not before the age of 10. 7 In the last remaining condition a reflexive pronoun was in a co-argumenthood relation with only one possible antecedent, to compare it to Ib with two possible antecedents. For reasons of

384 | Hendrikje Ziemann and Esther Ruigendijk

Within the conditions, there were no differences in the number of syllables and words until the pronominal element. The verbs used in this study were matched for frequency based on Celex (http://celex.mpi.nl/, 14.8.2015). Also the lexical content of the lead-in of each condition was hold constant (see Table 1). All items were divided into segments in the same way. The pronominal element was in the ninth segment for all experimental items. This was the critical region (see Table 1: in bold). The segment immediately before the pronominal element was either a preposition, as was the case in the non-co-argumenthood structures, or a verb, as in the co-argumenthood structures. This is our precritical region (in Table 1: italics). After the pronominal element, we had two postcritical regions (in Table 1: underlined). The first postcritical region always was a conjunction and the second postcritical region consisted of a determiner phrase or a prepositional phrase. We added these two post critical region to our analysis, since several studies have shown that effects may not occur immediately, but only one or even several words after the pronominal element (see e.g. Korneef & van Berkum, 2006, Felser & Cunnings, 2012). The segmentation of the items is marked by | in Table 1. Fillers were constructed in the same way as the items. Fillers served to distract the attention of the participants from the goal of this research. Table 1: Example of items. Lead-in sentence De kapper en de visagist| werkten| in de kapsalon.| The hairdresser and the stylist were working in the hairdresser’s. Non-co-argument Ia

De kapper| verschoof| een grote| kapperstoel| naast| zich/hem| zodat| er beter| kon worden gewerkt.| The hairdresser moved a big barber’s chair next to himself/him so that one could work more easily.

Co-argument with two grammatical antecedents Ib

De kapper| die| graag| dingen uitprobeerde| schoor| zich/hem| zodat| de nieuwe aftershave| kon worden uitgeprobeerd.| The hairdresser who liked to try things shaved himself/him so that the new aftershave could be tested.

space, this latter condition is not discussed in this paper, instead we concentrate on the difference between pronominal element (zich/hem) and structure (co-argument- non-coargument).

L2 speakers’ processing of reflexives and personal pronouns |

385

5.3 Procedure The items were divided in two sessions that took place on two days. Each session lasted about 45 minutes. The participants were told that the sentences were not presented as a whole but in segments in the centre of the screen. The first segment was always given automatically, but each following segment only appeared after the participant pressed the space button. They were asked to read and press as quickly as possible. After pressing this button the segment on the screen disappeared and the next segment appeared. When the sentence was finished, a yes/no question was asked like Verschoof de kapper een stoel? (‘did the hair dresser move a chair?) after reading (Ia in Table 1), which tested the comprehension of the sentence to ensure that participants not just clicked through the sentence without reading for comprehension. At least 80 percent correct answers were required for inclusion in our statistical analyses.

5.4 Analysis Before we ran Anovas, we rejected all items from our analyses where the participants answered the comprehension questions incorrectly. For the German group 89,9 % of the data remained and for the Dutch group 89,2 %. RTs for the important regions per condition (precritical, critical and postcritical) were also removed from analyses if they were below or above 2 SDs of the mean for that region from the group. We ran an Anova per sentence segment with group, type of pronominal element and type of structure as variables. This way, we tested whether the groups behaved differently from each other on the different pronominal types and sentence structures.

6 Results Figure 1 presents the results or our self-paced reading study. As can be seen, overall RTs are not the same for the different parts of the sentence (more detailed results can be found in the Appendix.) At the precritical region (Figure 1, upper left graph), that is immediately before the pronominal element, as expected we did not find effects of structure, or pronominal element (F(1, 37) = 1.08, p = .31, F(1, 37) = 2.32, p = .14 respec-

386 | Hendrikje Ziemann and Esther Ruigendijk

Figure 1: Mean reaction times (RT) in ms, for L1 and L2 speakers on the precritical region, critical region, and the two post critical regions. PRO = personal pronoun, REF = reflexive, CO-ARG = co-argument structure, and PP = non-co-argument structure.

tively), nor any interaction (p > .05)⁸. We did find an overall effect of group, in that L2 speakers showed higher RTs than L1 speakers (F(1, 37) = 6.87, p < .05). At the critical region (Figure 1, upper right graph), the reflexive or personal pronoun, we again found the same group effect (F(1, 37) = 9.51, p < .01). Here, we found a main effect for pronominal element as well (F(1, 37) = 4.91, p < .05), caused by an overall higher mean RT for the personal pronoun than the reflexive. We also found an interaction of group and structure (F(1, 37) = 5.53, p < .05). L1 speakers processed pronominal elements faster in the co-argument structures compared to the non-coargument structures, whereas for L2 speakers it seemed to be the other way around. No main effect for sentence structure was found (F(1, 37) = 2.01, p = .16), and no other interactions were found. At the first post critical region, the word immediately after the pronominal element (Figure 1, bottom left graph), we again found an effect of group, the L2 speakers having higher RTs overall (F(1, 37) = 5.00, p < .05). Here we found an interaction of structure and pronominal element as well, indicating that personal pronouns in a co-argument structure were processed the slowest (F(1, 37) = 4.65, p < .05). Also a marginally significant effect of pronominal element (F(1, 37) = 3.98, p = .053) was found, caused by overall higher RTs for personal pronouns.

8 For reasons of space we do not report all details on non-significant interactions here.

L2 speakers’ processing of reflexives and personal pronouns | 387

Finally, for the second post critical region (Figure 1, bottom right graph), we did not find an effect of group (F(1, 37) = .69, p = .41). A main effect of pronominal element was found (F(1, 37) = 17.48, p < .001), as well as three-way interaction between group, pronominal type and sentence structure (F(1, 37) = 15.53, p < .001). This was most likely caused by the difference between the two groups on the processing of reflexives in co-argument and non-co-argument sentences, in that L1 speakers, but not L2 speakers showed the lowest RTs for the reflexive in the co-argument structure. To disentangle this interaction, we follow-up with by-group analyses. These analyses revealed an effect of structure as well as an interaction of structure and pronominal element in the L1 group (F(1, 22) = 7.76, p < .05 and F(1, 22) = 21.24, p < .001), caused by lower RTs on reflexives in the co-argument condition compared to the three other conditions. The German L2 group also showed an interaction between structure and pronominal element (F(1, 15) = 6.60, p < .05), but in their case this was caused by overall higher RTs for the pronominal element in the non-co-argument condition compared to the three other conditions.

7 Discussion We investigated the processing of reflexives and personal pronouns by German L2 speakers of Dutch. The question we tried to answer was which operations L2 speakers use in order to interpret pronominal elements in their L2. Our results firstly support the idea that L2 speakers differ from L1 speakers in their processing of pronominal elements. Furthermore, the results confirm earlier studies that showed that L1 speakers process reflexives in a co-argument structure differently from reflexives in a non-co-argument structure as well as differently from personal pronouns. Finally, we found evidence that there was a cross-linguistic influence in the processing of personal pronouns in PPs for the L2 group. Now, we will discuss the results of our study in more detail. Our first hypothesis concerned Reuland’s model of an economy hierarchy for pronominal dependencies in L1. He assumes that the establishing of dependencies for pronominal elements is in principle possible via a syntactic or discourse operation⁹. However, he assumes differences in “costs” in that a syntactic operation is more economical than a discourse operation, because it requires fewer computational steps. Translating this assumption to the reaction time data of our experiment, in the

9 We ignore semantic operations (bound variable interpretation) here, since they are not relevant for our study, but of course, these too are in principle possible.

388 | Hendrikje Ziemann and Esther Ruigendijk

sense that lower RTs are assumed to reflect lower costs, we found that this holds true for the L1 speakers. The processing of reflexives in a co-argument relation was faster than the processing of reflexives in PPs or than personal pronouns in a co-argument relation or PPs in the second post critical region, thus supporting earlier data from Schumacher and others (2010). This difference in processing between reflexives in a co-argument relation and in a PP and between reflexives and personal pronouns lies in the different options available for the interpretation of a pronoun. As Reinhard and Reuland (1993) discuss in their Reflexivity paper the syntactic environment in which the pronominal element appears influences how it can be processed. Specifically, reflexives in a co-argument relation can be interpreted through a syntactic operation because of this co-argument relation between the antecedent and the reflexive and because of the feature underspecification of the reflexive. The reflexive gets its features checked through chain formation, a purely syntactic operation (see also Reuland, 2001). In a PP, there is no co-argument relation between the reflexive and the antecedent, thus no syntactic operation is available for establishing a dependency with the reflexive. Instead, discourse operations play a role here, which makes processing the reflexive in a PP more costly than in a co-argument relation. The difference we found between the reflexives and personal pronouns in a co-argument structure in both the critical and the first post critical region for L1 speakers indicates that processing of the reflexives is indeed less costly. Even though the pronominal elements appeared in the same structure, meaning they are co-arguments to the antecedent, a syntactic operation was not possible for the personal pronoun, because of the full feature specification of the personal pronoun. In such cases a more costly discourse operation is used to interpret the pronominal element (see Reuland, 2001). Now, we will discuss the results of the L2 speakers. We will do this by adopting the idea by Ullman (2001) and Clahsen and Felser (2006) who argue that L2 sentence processing is different from L1 processing. As Ullman (2001) postulates L2 speakers should not be able to access procedural memory, which is used for processing of grammar in native speakers. Instead declarative memory is employed. But this memory is originally responsible for computing world knowledge, lexical information etc., and hence not suitable for processing grammar. Similarly, Clahsen and Felser (2006) argue in their SSH that L2 speakers cannot use a full syntactic representation. If Ullman (2001) and Clahsen and Felser (2006) are right in assuming such a failure in the use of the procedural memory, or full syntactic representations for L2 speakers then our results should not show a difference in processing between reflexives and personal pronouns for this group. First, our results show that overall our L2 speakers had higher RTs than the L1 speakers, which does not come as a surprise, and can be interpreted as overall higher processing costs for L2 processing. Furthermore the L2 speakers did

L2 speakers’ processing of reflexives and personal pronouns | 389

not show the same effect as the L1 speakers in the second post critical region with lower RTs for the coargument reflexive, which we would have expected if they had been able to process these elements syntactically, as L1 speakers. Furthermore, L2 speakers seem to be faster in processing pronominal elements in general in the non-coargument PP structures than in the coargument structures in the critical region. This may at first seem somewhat counterintuitive. It might be explained though by the fact that in L2 processing at this early stage in the sentence, only phrase structure building operations take place, i.e. for the coargument structure integrating the object (i.e. the pronominal object) into the structure and linking it to the verb’s argument structure. If this is the case, one could argue that in fact, for the non-coargument structure, less phrase structure has to be built at that stage than for the coargument structure, namely ‘only’ a PP, including a preposition and the pronominal element. Effects of pronominal dependencies may be found somewhat later in the sentence. This is why we included spill-over regions in the first place. Of course this explanation is still a bit speculative and needs to be supported by more studies. The different processing operations just discussed for L1 speakers do not seem to play a role for L2 speakers. Hence processing is more similar for all pronominal elements across all structures and we argue that the only operation applied to process pronominal elements is discourse, supporting Ullman’s (2001) and Clahsen and Felser’s (2006) theories for L2 sentence processing. The last issue we need to address was whether there would be a cross-linguistic influence in processing personal pronouns in PPs for the L2 group (12). (11) Jani legde het boek naast zichi /hemi/j . ‘Jan put the book next to himself/him’ This is what we found. The L2 speakers showed an effect for personal pronouns in PPs in the sense that this pronominal element showed the highest RTs in comparison to the three other conditions in the second postcritical region. This effect could be caused by the differences between German and Dutch, in that in PPs the personal pronoun in Dutch may be bound by the subject Jan, which is impossible in German. If the L2 speakers were guided by the knowledge of pronominal dependencies by their mother tongue, then this could result in longer RTs for this condition. The pronoun could then first be processed as a kind of grammatical violation. However, if we follow the argumentation of Ruigendijk and Schumacher (2011) who argue for a co-argumentlike relation of the antecedent and the pronominal element in PPs for German, then personal pronouns in PP should not show longer RTs than personal pronouns in a co-argument relation. This would then mean that something else is influencing the processing of pronominal elements in

390 | Hendrikje Ziemann and Esther Ruigendijk

PPs. Possibly there is influence from the L2 Dutch as well. This means that maybe the L2 speakers consider the knowledge from their mother tongue in the processing of PPs but that they were also influenced by their knowledge of the L2. This ‘confusion’ could then lead to higher RTs for the personal pronouns in PPs. In order to get a better understanding of what is happening in L2 speakers when they are processing pronominal elements in these structures we are now running an eyetracking study, which enables us to not only examine processing time, but also (through regressions) what possible antecedents are considered for the pronominal element during reading by the L2 speakers. By doing this, we hope to find out in more detail how pronominal elements are computed and we hope to be able to say more about the types of operations used for pronominal processing in L2.

8 Conclusion The goal of our study was to find out which operations German L2 speakers employ in order to process pronominal elements in their L2 Dutch and in how far they differ from L1 speakers. Three major findings arose. Namely, that L1 speakers used different operations in processing reflexive and personal pronouns, we argue: syntactic versus discourse operations, whereas L2 speakers do not show these difference, and hence we argue that they only made use of discourse operations. Also we found a cross-linguistic influence in that processing seems to be guided by constraints provided by the mother tongue. Our results support the economy hierarchy of referential dependencies (Reuland 2001) in that syntactic processing used for reflexive interpretation in coargument structures requires less time than discourse processing at least for L1 speakers. Moreover, this study also supports L2 psycholinguistic models in that L2 sentence processing is different from L1 sentence processing.

References Burkhardt, Petra 2005. The syntax-discourse interface: representing and interpreting dependency. Amsterdam: John Benjamins B.V. Chomsky, Noam 1981. Lectures on government and binding. Dordrecht: Mouton de Gruyter. Chomsky, Noam 1986. Barriers. Cambridge: MIT Press. Clahsen, Harald & Claudia Felser 2006. Keynote article. Grammatical processing in language learners. Applied Psycholinguistics 27, 3–42. Demirci, Mahide 2000. The role of pragmatics in reflexive interpretation by Turkish learners of English. Second Language Research 16, 325–353.

L2 speakers’ processing of reflexives and personal pronouns | 391

Felser, Claudia & Ian Cummings 2012. Processing reflexives in a second language: The timing of structural and discourse-level constraints. Applied Psycholinguistics 33, 571–603. Felser, Claudia, Mikako Sato & Nicholas Bertenshaw 2009. The on-line application of binding Principle A in English as a second language. Bilingualism: Language and Cognition 12, 485–502. Hendriks, Petra, Arina Banga, Jacolien van Rij, Gisi Cannizzaro & John Hoeks 2011. Adults’ on-line comprehension of object pronouns in discourse. In Angela Grimm, Anja Müller, Cornelia Hamann & Esther Ruigendijk (eds), Production-comprehension asymmetries in child language. De Gruyter Mouton, Berlin, 193–216. Hirakawa, Makiko 1990. A study of L2 acquisition of English reflexives. Second Language Research 6, 60–85. Koorneef, Arnout, W. & van Berkum, Jos, J.A. 2006. On the use of verb-based implicit causality in sentence comprehension: Evidence from self-paced reading and eye tracking. Journal of Memory and Language, 54 (4), 445–465. Matsumura, Masanori 1994. Japanese learners‘ acquisition of the locality requirement of English reflexives. Studies in Second Language Acquisition 16, 19–42. Nicol, Janet & David Swinney 1989. The role of structure in coreference assignment during sentence comprehension. Journal of Psycholinguistic Research 18, 5–19. Reinhart, Tanya & Eric Reuland 1993. Reflexivity. Linguistic Inquiry 24, 657–720. Reuland, Eric 2001. Primitives of Binding. Linguistic Inquiry 32, 439–492. Reuland, Eric 2011. Anaphora and Language Design. Cambridge, MA: MIT Press. Roberts, Leah, Marianne Gullberg & Peter Indefrey 2008. Online pronoun resolution in L2 discourse. L1 influence and general learner effects. Sudies in Second Language Acquisition 30, 333–357. Ruigendijk, Esther & Petra Schumacher (2011). Reaction time data reveal variation in reference assignment processes in Germanic languages. In Iris Hendrickx, Antonio Branco, Sobha L. Devi & Ruslan Mitkov. (eds.) Proceedings of DAARC 2011 – The 8th Discourse Anaphora and Anaphora Resolution Colloquium. Lisbon: Edições Colibri, 127–138. Schumacher, Petra, Maria M. Piñango, Esther Ruigendijk, & Sergey Avrutin. (2010). Reference Assignment in Dutch: Evidence for the Syntax-Discourse Divide. Lingua 120, 1738–1763. Sorace, Antonella & Francesca Filiaci 2006. Anaphora resolution in near-native speakers of Italian. Second Language Research 22, 339–368. Sturt, Patrick 2003. The time-course of the application of binding constraints in reference resolution. Journal of Memory and Language 48, 542–562. Thomas, Margaret 1991. Universal grammar and the interpretation of reflexives in a second language. Language 67, 211–239. Ullman, Michael T. 2001. The neural basis of lexicon and grammar in first and second language: The declarative/procedural model. Bilingualism: Language and Cognition 4, 105–122. Ullman, Michael T. 2006. Commentaries. Applied Psycholinguistics 27, 43–105. White, Lydia, Joyce Bruhn-Garavito, Takako Kawasaki, Joe Pater & Philippe Prevost (1997). The researcher gave the subject a test about himself: Problems of ambiguity and preference in the investigation of reflexive binding. Language Learning 47, 145–172. Webcelex: http://celex.mpi.nl/, 14.8.2015

392 | Hendrikje Ziemann and Esther Ruigendijk

Appendix Mean RTs in ms (SD in brackets) for structure and pronominal element type for the L1 group. L1 group

precritical region

critical region zich/hem

first postcritical region

second postcritical region

personal pronoun CO-ARG reflexive CO-ARG personal pronoun PP reflexive PP

471 ms 445 ms 457 ms 448 ms

436 ms (118) 402 ms (68) 431 ms (80) 421 ms (80)

459 ms (116) 424 ms (103) 428 ms (78) 426 ms (86)

534 ms 477 ms 509 ms 523 ms

(94) (75) (75) (88)

(140) (133) (124) (135)

RTs in ms (SD in brackets) for structure and pronominal element type for the L2 group. L2 group

precritical region

critical region zich/hem

first postcritical region

second postcritical region

personal pronoun CO-ARG reflexive CO-ARG personal pronoun PP reflexive PP

524 ms (124) 512 ms (106) 535 ms (94) 558 ms (78)

521 ms (94) 520 ms (81) 495 ms (108) 488 ms (80)

516 ms (121) 496 ms (96) 494 ms (111) 500 ms (81)

556 ms (118) 539 ms (101) 574 ms (115) 499 ms (74)

Index AAE see Anaphor Agreement Effect Agree 58–63, 90 – Agree chain 62, 66 agreement 39, 136, 171, 319 – agreement chain 145–147 – mismatches 136, 137 – past agreement 32 – semantic agreement 154 – Φ-agreement 77 – see also morphology American Sign Language 222 anaphor 14, 64, 77, 354, 378 Anaphor Agreement Effect 77 anaphora 77, 227, 257, 296, 320 – resolution of anaphoric subjects 320 article 231–235, 252 at-issue 251, 271–275 attitude 87, 179, 216, 257

– demonstrative pronouns 189, 242, 252 – determinative use 258 – emotive use 257 – recognitional use 256 – see also distal, proximal dependency 42, 66, 289, 374 Derived Kind Predication 120 description 290 – definite description 17, 198, 230, 353 distal 238, 251 donkey pronoun 192 – see also e-type binding Dravidian 78 Dutch 163, 373 Dynamic Syntax 288

bare plural 112 binding 54, 92, 144, 171, 189, 222, 348, 375 – binding theory 65, 180, 199, 351 – Condition A 180, 377 – Condition B 241, 353, 377 – Condition C 242, 353 – Principle A 336, 376 – Principle B 337, 376 – Principle C 199 bound variable 1, 14, 56, 169, 199, 242, 347, 375

fake indexicals 14 fake past 17 FID see Free Indirect Discourse finite control 54 Free Indirect Discourse 211

case 78, 135, 180, 379 chain condition 66, 377 clitic 108, 160, 326 co-argumenthood 383 coreference 46, 56, 94, 108, 145, 180, 194, 354, 374 definiteness 233, 251 deixis 236, 254 demonstrative 222, 223, 257 – anamnestic use 256 – anaphoric use 256 – demonstrative grammar 366

e-type binding 241 – see also donkey pronoun

German 39, 136, 189, 238, 251, 336, 374 impersonal 35, 59, 109 index 59, 108, 169, 222 indirect modification 116 inter-speaker variation 348 Italian 78, 107, 319 kind 118–122 Korean 228, 347 L2 373 left periphery 91 locus 225 mental distance 251 mismatch 89, 136 morphological ambiguity hypothesis 338

394 | Index

morphology – agreement morphology 319, 320 – past tense morphology 21 – verbal morphology 319 – Φ-morphology 85 multi-functionality 15

PRO 36, 56, 183 pro-NP 108, 352 pro-PP 109 pronominal grammar 366 proximal 238, 252 quantification 172, 196, 229, 302

ne 107 not-at-issue 251 null pronouns 53, 90, 162, 349 – null and overt subjects 319 ones 110 personal pronoun 1, 39, 108, 157, 189, 221, 253, 321, 365, 373 – strong pronoun 158, 321 – weak pronoun 46, 159, 194, 321 – 1st/2nd person 1, 17, 34, 135, 223 perspective 91, 189 – perspectival centre 189 phase 54, 86 Phase Impenetrability Condition 66, 67 pointing 223, 253, 287 Position of Antecedent Strategy 323 possessive pronoun 144 pragmatics 288 presupposition 169, 198, 230 – projection 169 pro 36, 90, 131, 162, 321, 349 – pro-drop 77, 319, 349 – see also null pronouns

reflexive 1, 42, 84, 144, 180, 200, 374 relative clause 116, 135, 200, 258 – non-restrictive relative clause 135 – restrictive relative clause 153, 258 relative pronoun 135 relevance 293 resumptive pronoun 136 Russian 42, 53 self-paced reading 320, 374 sentence processing 373 sign language see American Sign Language Spanish 319, 349 subject 37, 53, 80, 113, 135, 180, 190, 231, 253, 298, 319, 355, 375 subject pronouns 58, 145, 322 Tamil 78 topicality 194 two-grammar hypothesis 366 Φ-feature 54, 90, 169 – gender 47, 60, 95, 113, 150–157, 169, 322 – number 60, 89, 113, 142, 169, 226, 263, 319