Sinn und Bedeutung 15 - [PDF Document] (2024)

Sinn und Bedeutung 15 - [PDF Document] (1)

Sinn und Bedeutung 15

Proceedings of the 2010 Annual Conference ofthe Gesellschaft für Semantik

Edited by

Ingo ReichEva HorchDennis Pauly





Sinn & Bedeutung – the annual conference of the Gesellschaftfür Semantik – aims to bring together both establishedresearchers and new blood working on current issues in naturallanguage semantics, pragmatics, the syntax-semantics interface,the philosophy of language or carrying out psycholinguisticstudies related to meaning.

Every year, the conferencemoves to a different location in Europe.

The 2010 conference – Sinn & Bedeutung 15 – took place onSeptember 9 - 11 at Saarland University, Saarbrücken, organizedby the Department for German Studies.

universaarUniversitätsverlag des SaarlandesSaarland University PressPresses Universitaires de la Sarre

semnatik_ir_cover_31-395:Layout 1 21.07.2011 09:40 Seite 1

Sinn und Bedeutung 15 - [PDF Document] (2)

Ingo Reich, Eva Horch, Dennis Pauly (Eds.)

Sinn und Bedeutung 15

Proceedings of the 2010 Annual Conferenceof the Gesellschaft für Semantik

universaarUniversitätsverlag des SaarlandesSaarland University PressPresses Universitaires de la Sarre

Sinn und Bedeutung 15 - [PDF Document] (3)

© 2011 universaarUniversitätsverlag des SaarlandesSaarland University PressPresses Universitaires de la Sarre

Postfach 151150, 66041 Saarbrücken

ISBN 978-3-86223-038-9 gedruckte AusgabeISBN 978-3-86223-039-6 Online-AusgabeURN urn:nbn:de:bsz:291-universaar-300

Projektbetreuung universaar: Isolde Teufel

Satz: Eva Horch, Dennis Pauly, Ingo ReichUmschlaggestaltung: Julian Wichert

Gedruckt auf säurefreiem Papier von Monsenstein & Vannerdat

Bibliografische Information der Deutschen Nationalbibliothek:Die Deutsche Nationalbibliothek verzeichnet diese Publikation in der DeutschenNationalbibliografie; detaillierte bibliografische Daten sind im Internet über<> abrufbar.

Sinn und Bedeutung 15 - [PDF Document] (4)


Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

Specificity, Referentiality and Discourse Prominence: German IndefiniteDemonstrativesKlaus von Heusinger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

On the Semantics of Existence PredicatesFriederike Moltmann . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

How the Emergence of Propositions Separates Strict Interfaces fromGeneral InferenceTom Roeper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

Appositives in Modal ContextsKjell Johan Sæbø . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

An Unaccusativity Diagnostic at the Syntax-Semantics Interface:there-Insertion, Indefinites and Restitutive againArtemis Alexiadou, Florian Schäfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

Optatives: Deriving Desirability from Scalar AlternativesMaría Biezma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

On French un même and AntispecificityIsabelle Charnavel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

Performative Verbs and Performative ActsCleo Condoravdi, Sven Lauer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

Identity and Definiteness in Chinese wh-ConditionalsStephen Crain, Qiong-Peng Luo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165

Total vs. Partial Adjectives: Evidence from ReduplicationMojmír Docekal, Ivona Kucerová . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

Focus Marking via GesturesCornelia Ebert, Stefan Evert, Katharina Wilmes . . . . . . . . . . . . . . . . . 193

Sinn und Bedeutung 15 - [PDF Document] (5)


Imperatives as Future PlansRegine Eckardt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209

Focus Influences the Presence of Conditional Perfection: ExperimentalEvidenceMarie-Christine Farr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225

Stative Passives and Event KindsBerit Gehrke . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241

Events in Adjectival PassivesHelga Gese . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259

Facts and Ideals: On the Role of doch in Conditionals and OptativesPatrick Grosz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275

Conventional and Free Association with Focus in Ngamo (West Chadic)Mira Grubic, Malte Zimmermann . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291

The Concept of Semantic Phase and the Different Readings of ‘again’Sabine Gründer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307

Degree Modification in Russian Morphology: The Case of the Suffix-ovatOlga Kagan, Sascha Alexeyenko . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321

On the Relation between Coherence Relations and AnaphoricDemonstratives in GermanElsi Kaiser . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337

Lexikal vs. Pragmatically Derived Interpretations of NumeralsMasaaki Kamiya, Akemi Matsuya . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353

Optional and Obligatory Modal SubordinationPeter Klecha . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365

Direction and Obviation in Plains Cree: A Referent Systems ApproachUdo Klein, Marcus Kracht . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381

Thematic Roles and the Interpretation of one-another ReciprocalsChris LaTerza . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397

Evidentials in Interrogatives: A Case Study of KoreanDongsik Lim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419

Sinn und Bedeutung 15 - [PDF Document] (6)


On the Encoding of the Definite/Indefinite Distinction in KaritianaAna Müller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435

On Temporal QuantificationEdgar Onea . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451

Focus and wh-Questions in MongolianEdgar Onea, Dolgor Guntsetseg . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 467

Relevance TopicsSophie Repp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483

Early Implicatures by Children and the Acquisition of ScalarImplicaturesStefanie Röhrig . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 499

Semantic Reconstruction and the Interpretation of ChainsEddy G. Ruys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515

Be Positive! Norm-Related Implications and BeyondGalit W. Sassoon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 531

English Comparative Correlatives, Conditionals, and Adverbs ofQuantificationE. Allyn Smith . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 547

How many Mosts?Stephanie Solt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 565

Psycholinguistic Evidence for Presuppositions: On-line andOff-line DataS. Tiemann, M. Schmid, N. Bade, B. Rolke, I. Hertrich,H. Ackermann, J. Knapp, S. Beck . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 581

Verification Strategies for Two Majority Quantifiers in PolishBarbara Tomaszewicz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 597

Accomplishments: Their Telos and Their StructureLucia M. Tovena . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 613

Nominal Reference in Two Classifier LanguagesTue Trinh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 629

Sinn und Bedeutung 15 - [PDF Document] (7)


On the Rise and Fall of DeclarativesTue Trinh, Luka Crnic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 645

Degree Possession Is a Subset Relation (As Well)Zhiguo Xie . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 661

Divergent ApproximatorsErin Zaroukian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 677

Sinn und Bedeutung 15 - [PDF Document] (8)

PrefaceThese proceedings contain all papers presented at the 15th annual conferenceof the Gesellschaft für Semantik, »Sinn & Bedeutung 15«, which took place onSeptember 9 - 11, 2010, at Saarland University, Saarbrücken.

The papers in this volume cover current issues in natural language semantics,pragmatics, the syntax-semantics interface, psycholinguistic studies related tomeaning, and the philosophy of language.

The »Sinn & Bedeutung 15« conference was organized by the Gesellschaftfür Semantik and the Department for German Studies at Saarland University,Saarbrücken.

We would like to thank the countless reviewers for providing detailed andfair reviews in due time, the members of the program committee for puttingtogether such a great program, and finally all supporting institutions, whosefunding made »Sinn & Bedeutung 15« possible in the first place.

These proceedings are also available electronically from Saarland UniversityPress (universaar) at

Saarbrücken, July 2011The editors

Local Organizers

Ulrike DemskeEllen Geibel-StutzNele HartungEva HorchSergey KulakovSabine Mayer-SaarAnna MönnichDennis PaulyChristian RamelliIngo ReichUlyana SenyukJulia Stark

Programm Committee

Claudia MaienbornCécile MeierIngo ReichMagdalena SchwagerArnim von StechowEde Zimmermann

Supporting Institutions

DFGMinisterium für Wirtschaft

und Wissenschaft desSaarlandes

Freunde der Universitätdes Saarlandes

KWTOxford University PressPalgrave MacmillanMouton de GruyterEmeraldDUDENPeter Lang Verlag

Sinn und Bedeutung 15 - [PDF Document] (9)

Sinn und Bedeutung 15 - [PDF Document] (10)

Specificity, Referentiality and Discourse Prominence: German Indefinite Demonstratives*

Klaus von Heusinger

University of Stuttgart [emailprotected]

Abstract. There are various notions of specificity, ranging from Fodor & Sag’s (1982) referentiality view to Givón’s (1983) discourse prominence view. Ionin (2006) discusses the relation between these two perspectives by analyzing the English indefinite this. She represents indefinite this as a referential operator in the sense of Fodor & Sag (1982), but also adds the felicity condition of “noteworthiness”. She notes that it is an open question how these two properties of indefinite this are linked to each other. Wright & Givón (1987) claim that the discourse prominence is primary and that referential properties are derived from it. I argue that the contrary holds: On the analysis of German indefinite demonstrative dies (‘this‘) and so’n (‘such-a’) I demonstrate how we can derive discourse properties of indefinite demonstratives from their referential properties.

1 Introduction Specificity is a semantic-pragmatic notion that distinguishes between different uses or interpretations of indefinite noun phrases. It is related to the communicative notion of “referential intention”. A speaker uses an indefinite noun phrase and intends to refer to a particular referent, the referent “the speaker has in mind”. This function of the indefinite has various consequences for sentence and discourse semantics. In this article I focus on two aspects of specificity: Fodor & Sag’s (1982) notion of referentiality and

* Part of this presentation is joint work with Sofiana Chiriacescu (so’n) and Annika Deichsel (indefinite dies). I am grateful to the audience of Sinn und Bedeutung for helpful comments, and to Sofiana Chiriacescu, Annika Deichsel, Cornelia Ebert, Donka Farkas, Ljudmila Geist, Jeanette Gundel, Dag Haug, Irene Heim, Elsi Kaiser, Hans Kamp, Edgar Onea, Albert Ortmann, Alice ter Meulen for constructive comments and considerable assistance on earlier versions of this paper. The research is supported by the German Science Foundation by a grant to the project C2: Case and referential context, as part of the Collaborative Research Center 732 Incremental Specification in Context at the University of Stuttgart. Furthermore, I gratefully acknowledge the support of the Fritz Thyssen Foundation and the VolkswagenStiftung (opus magnum).

In Reich, Ingo et al. (eds.), Proceedings of Sinn & Bedeutung 15, pp. 9–30. Universaar – Saarland University Press: Saarbrücken, Germany, 2011.

Sinn und Bedeutung 15 - [PDF Document] (11)

10 von Heusinger

Givón’s (1983) notion of discourse prominence as the central effect of specific indefinites. This two-sided behavior of specific indefinites was illustrated by the referential and discourse properties of indefinite this in English (Perlman 1969, Maclaran 1980, Prince 1981, Ionin 2006). The demonstrative this in English has an “indefinite” or “presentative” use, as in (1) and (2a). In (1) the noun phrase this man is clearly indefinite as it appears in an existential context. It is discourse- and speaker-new and Ionin (2006) argues that it is felicitously used if it introduces an interesting or “noteworthy” property into the discourse. The use of indefinite this in (2b) is not felicitous as the given information is not noteworthy, but rather expected (examples from Maclaran 1980 and Ionin 2006):

(1) There is this man who lives upstairs from me who is driving me mad because he jumps rope at 2 a.m. every night.

(2) a. I put a/this 1$ stamp on the letter and realized too late that it was worth a fortune.

b. I put a/*this 1$ stamp on the letter. I wanted to mail the letter to Europe.

Besides these discourse properties, Prince (1981) also discusses particular referential properties that are characteristic for specific or referential in-definites (Fodor & Sag 1982). Indefinite this always takes wide scope with respect to extensional operators, as illustrated in (3a). On the other hand, the indefinite noun phrase a poem in (3b) is ambiguous between a wide-scope reading and a narrow-scope reading, thus allowing for the inference that different students might have read different poems.

(3) a. He gave an A to every student who recited this poem by Pindar. (à Only one poem overall) b. He gave an A to every student who recited a poem by Pindar.

(à Possibly many poems)

Indefinite this in (4a) always allows an existential entailment or presupposition, while the ordinary indefinite article a does not. (5a) shows that it is a presupposition, since it allows an existential inference even under negation.

(4) a. Alice wanted to kiss this sailor boy. (à There was a sailor boy) b. Alice wanted to kiss a sailor boy. (-/-> There was a sailor boy)

(5) a. Mary didn’t buy this pink truck. (à There was a pink truck) b. Mary didn’t buy a pink truck. (-/-> There was a pink truck)

Sinn und Bedeutung 15 - [PDF Document] (12)

Specificity, Referentiality and Discourse Prominence 11

Fodor & Sag (1982) observe that the use of indefinite this is different from the use of the definite article in such contexts. The definite article pre-supposes familiarity of speaker and hearer with the associated referent, while the indefinite demonstrative only indicates familiarity of the speaker, but unfamiliarity of the hearer. It is the prototypical instance of a specific (or referential) indefinite noun phrase. Its definition (6) expresses that a specific or referential indefinite introduces a new discourse referent such that the speaker has a “unique individual in mind”. Heim (2011, ex. (56)) formulates Fodor & Sag’s (1982) original idea in a two-dimensional semantics with a context set c and an evaluation point i. The indexical or referential meaning of an indefinite only depends on the utterance context, as it is the case for regular indexical expressions. Ionin (2006) adds a felicity condition to this definition in order to motivate the use of such a referential indefinite, as in (7). The use of indefinite this is only felicitous if the speaker contributes a noteworthy property to the introduced referent.

(6) Referential indefinites (Fodor & Sag 1982, Heim 2011, ex. (56)) [[aref α]]c,i is defined only if there is a unique individual that the speaker

of c has in mind in c, and this individual is in [[α]]c,c; where defined, [[aref α]]c,i = this individual.

(7) Indefinite this (Ionin 2006: 187) A sentence of the form [sp α] φ expresses a proposition only in those

utterance contexts c where the following felicity condition is fulfilled: the speaker of c intends to refer to exactly one individual xc in c, and there exists a property u which the speaker considers noteworthy in c, and xc is both α and u in c. When this condition is fulfilled, [sp α] φ expresses that proposition which is true at an index i if xc is φ at i and false otherwise.

Ionin (2006) combines the two characteristics of indefinite this in her definition (7): (i) the “referential intention” of the speaker yielding the semantic property of high referential strength described above; and (ii) the noteworthiness property closely related to the pragmatic property of high prominence in the discourse. She discusses the relation between these two properties, but without conclusion. Wright & Givón (1987) focus on dis-course prominence and compare the grammaticalization of indefinite this with the indefinite article a. They argue that such indefinite articles first acquire a pragmatic discourse function and only then the referential function. Generalizing the empirical data they found, Wright & Givón (1987, 29) maintain the claim (8) and argue that grammaticalization starts with

Sinn und Bedeutung 15 - [PDF Document] (13)

12 von Heusinger

pragmatic discourse functions and then proceeds to semantic functions such as high referential strength.

(8) Implicational relation between pragmatic and semantic reference “If a nominal is prag-referential, then it is most likely to also be SEM – referential (but not vice versa)”

With this short overview on indefinite this we are now in a position to formulate the research questions of this paper with respect to specificity: (i) Are discourse prominence and referentiality two instantiations of specificity, (ii) if so, are they related, and (iii) if they are related do they exhibit the implicational relation in (8) or the opposite. I argue that (8) does not correctly describe the situation with indefinite this in English and with its two German equivalents. While the analysis of indefinite this in English might not be conclusive (cf. Ionin 2006) the comparison with the two German specific indefinite articles dies and so’n (< so+’n, ‘such+encl*tic indefinite article’) indicates that the semantic function is primary and the discourse prominence derived.

Section 2 summarizes different types of specificity and discusses the different ways to group these subtypes together. Section 3 presents the semantic analysis of specificity in terms of referential anchoring. Section 4 and 5 provide information about the different uses and functions of German indefinite dies and so’n. Section 6 discusses some of their discourse functions and section 7 focuses on their referential properties. Section 8 presents an analysis of the function of indefinite demonstratives and section 9 formulates a first hypothesis concerning the semantics of such demonstratives in terms of referential anchoring. Section 10 concludes with a brief summary and some new research questions.

2 Types of Specificity The notion of specificity is associated with various types of data and accounted for in different theories (see Farkas 1994, Ionin 2006, Kamp & Bende-Farkas 2006, and von Heusinger 2011 for an overview). I suggest to classify the various notions of specificity in seven types: (i) Specificity in opaque contexts (referential specificity) expresses a contrast between a reading that allows existential entailment (9a) and a reading that does not (9b); (ii) scopal specificity (often also including type (i)) refers to the ability of certain indefinites to escape scope islands like the conditional in (10a), that a universal quantifier cannot escape (10b); (iii) epistemic specificity ex-

Sinn und Bedeutung 15 - [PDF Document] (14)

Specificity, Referentiality and Discourse Prominence 13

presses the contrast between speaker’s knowledge (11a) and speaker’s ignorance (or indifference) (11b) about the referent of the indefinite.

(9) a. Paula believes that Bill talked to an important politician. (-> there is an important politician) b. Paula believes that Bill talked to an important politician. (but there is no important politician)

(10) a. If a friend of mine from Texas had died in the fire, I would have inherited a fortune. (possible reading: there is a friend of mine

and…) b. If each friend of mine from Texas had died in the fire, I would have

inherited a fortune. (not possible: for each of my friends, if one of them…)

(11) a. A student in Syntax 1 cheated in the exam. I know him: It is Jim Miller. b. A student in Syntax 1 cheated in the exam. But I do not know who it is.

(iv) specificity is sometimes associated with different types of familiarity such as d-linking, partitivity, and presuppositionality: the indefinite is part of an already introduced set, as in (12a), or not, as in (12b); (v) specificity is also related to topicality as in (13a), where the topical element can be understood as a specific expression, while (13b) only expresses an existential claim.

(12) a. 50 students entered the room. I knew two girls. b. 50 students entered the room. They greeted two girls (already in the


(13) a. Some ghosts live in the pantry; others live in the kitchen. b. There are some ghosts in this house.

There are two further notions of specificity that concern the forward referential potential of indefinites: (vi) specificity as noteworthiness assumes that the presentative this in (14) signals that the speaker intends to assert a noteworthy property of the referent, as in (14a), while (14b) is reported to be infelicitous since no such property is mentioned. (vii) specificity as discourse prominence refers to an aspect of discourse prominence, namely “referential persistence” or “topic shift”, i.e. the potential of an indefinite to introduce a referent that will be mentioned again and may even become a topic in the subsequent discourse.

Sinn und Bedeutung 15 - [PDF Document] (15)

14 von Heusinger

(14) a. He put a/this 31 cent stamp on the envelope, and only realized later that it was worth a fortune because it was unperforated.

b. He put a/#this 31 cent stamp on the envelope, so he must want it to go airmail.

(15) a. There was a king and the king had a daughter and he loved his daughter …

b. There was a king and # the season was very short and hot …

These different subtypes of specificity can be roughly categorized into larger groups as in Figure 1 with a referential notion, a familiarity notion and a dis-course prominence notion of specificity.


referential notion

(sentence semantics and pragmatics)

familiarity-based notion

(backward reference)

discourse prominence

(forward reference)


referential specificity


scopal specificity


epistemic specificity


partitivity, presup-

positionality, backgrounding






discourse prominence

Figure 1: Family tree of specificity

Researchers on specificity differ in their assumptions on (a) which subtype qualifies for specificity proper and (b) how many representations are necessary to cover these types. Fodor & Sag (1982) take the subtypes (i)-(iii) as the central notion of specificity and assume the single representation (6) for them. Farkas (1994) argues that (i)-(iv) are independent subtypes but with similar effects. She suggests different representations, which are, however, similar in the effect that they reduce the restrictor set of the indefinite. Kamp & Bende-Farkas (2006, submitted) assume that epistemic specificity is the central notion which is basically the same as (i) and from which we can

Sinn und Bedeutung 15 - [PDF Document] (16)

Specificity, Referentiality and Discourse Prominence 15

derive effects described under (ii). Prince (1981) and Ionin (2006) analyze indefinite this and show that (vi) is related to (i)-(iii). Givón (1983) focuses on the discourse prominence aspect (vii) and assumes the implicational hierarchy of object domains in Figure 2, according to which a discourse prominent expression implicates that the associated referent is intended by the speaker, and what is intended also has a reference in the world. It is the general scheme from which Wright & Givón’s (1987) more specialized implicational relation (8) above is derived.

discourse prominence > speaker’s intentions > reference in the “world”

Figure 2: Ranking of specificity types according to Givón (1984: 135)

3 Referential Anchoring Different contrasts associated with different kinds of specificity can be best unified by the following generalization: In its prototypical use, the concept of specificity is associated with the communicative notion of referential intention. Grammatical contrasts, such as specific articles, indefinite pro-nouns or differential object marking associated with this function are also used to express relations between discourse entities which do not express “referential intentions” in the literal sense. Rather, it seems that specificity is a grammaticalized means to structure the relations among discourse items: A specific indefinite is referentially anchored to a salient discourse participant or another discourse referent, i.e. “the referent of the specific expression is linked by a contextually salient function to the referent of another expression“ (von Heusinger 2002: 45). Under this account the context has to provide two parameters: the anchoring function and the anchor itself. The speaker has to be able to specify the anchoring function, while it must be unfamiliar for the hearer, the same way as the intended referent must be unfamiliar. Still the hearer has to represent the fact that there is an anchoring function. The anchor, however, must be familar to both speaker and hearer, which allows speaker and hearer to share the scopal properties of the indefinite. This concept of specificity is a refinement of Fodor & Sag’s (1982) original account in terms of referential (Kaplan-style) expressions. Below I present a sketch of the theory by stepwise modifying Fodor & Sag’s (1982) original proposal. They assume two semantic representations for exis-tential indefinites and referential indefinites, as in (16).

(16) a. [[aquant N]] = λQ. ∃x. [N(x) & Q(x)]

Sinn und Bedeutung 15 - [PDF Document] (17)

16 von Heusinger

b. [[aref N]] = is defined only if there is a unique individual that the speaker of the sentence has in mind, and this individual is N

The definition (16b) might be adequate for English indefinite this, but it has been shown that it is not sufficient to account for various other types of specific indefinites. It needs additional modifications affecting the parameters listed in (17):

(17) Modification of the original Fodor & Sag (1982) account (= (4b)) (i) replacing the uniqueness condition by an explicit anchoring

function (ii) allowing for other anchors than the speaker (iii) allowing for different content of the anchoring function

The uniqueness condition in definition (16b) is ‘built in’ by a function from the anchor to the referent: f(anchor) = referent. The second modification concerns potential anchors, which can be the speaker in (18), but also some other attitude holder in (19). But we can abstract even further, as the anchor can also be realized by a quantifier phrase, as in (20).

(18) Paula believes that Bill talked to an important politician.

(19) a. George: “I met a certain student of mine today.” b. Jack: “George said that he met a certain student of his today.”

(20) Every husband had forgotten a certain date – his wife’s birthday.

The third modification affects the status of the content of the anchoring function. The anchor must in principle be familiar to both speaker and hearer, i.e. it must be contextually given or accessible. The content of the anchoring relation must be hearer-new in order to distinguish between specific indefinites and definites. (21) and (22) demonstrate that the exact definition of the function may even be unknown to the speaker (see Enç 1991: 20 for discussion):

(21) The teacher gave every child a certain task to work on during the afternoon.

(22) Each reporter was assigned to a certain politician by the editor of the paper.

We can summarize the modifications and give the informal definition of referential anchoring in (23):

(23) Informal definition of specificity in terms of referential anchoring

Sinn und Bedeutung 15 - [PDF Document] (18)

Specificity, Referentiality and Discourse Prominence 17

A specific indefinite a N is represented by an anchoring function f from an anchor to an individual and this individual is N. Both the anchor as well as the anchoring function must be given in the context

a) anchor is speaker- and hearer-given b) content of anchoring function is hearer-new

Von Heusinger (2002 based on earlier work) cashes out the idea of referential anchoring in terms of parameterized or Skolemized choice functions, also known from Kratzer (1998) and Chierchia (2001, 2005). The idea is that the indefinite article can translate into the complex pronominal element fx with x being a parameter that might be bound by some context agent or some quantifier phrase that has wider scope than the indefinite. The function f applied to the anchor yields a choice function that is applied to the set denoted by the descriptive content of the indefinite yielding the referent, as in (24) adapted from Roberts (2007) (for alternative treatments of this idea see Kamp & Bende-Farkas (to appear), Onea & Geist 2010).

(24) Referential anchoring with parameterized choice functions i) complex pronominal element fx ii) x parameter (= anchor), the argument of f, binding is pragmatically

given a) might be bound by some context agent (speaker etc.) b) might be bound by a wider scope QP to yield intermediate scope

iii) f(x): a choice function that takes a set denoted by DC as its argument and yields an element of that set

In summary, the concept of referential anchoring provides a consistent account of specificity. It links the notion of referential intention to a semantic representation with an anchoring function and an anchor. The anchor must be familiar to speaker and hearer, while the content of the function must not be familiar to the hearer (and is generally familiar to the speaker). Still the hearer has to establish a permanent representation for the specific indefinite, based on the assumption of the existence of such an anchoring function. Thus, this account ties in with the other concepts of specificity, including familiarity-based and discourse-based concepts, discussed in the next sections.

4 Indefinite dies German has a proximal demonstrative dieser, diese, dies(es), and a not very productive distal jener, jene, jenes. The proximal demonstrative has various functions, the most important of which are listed in Table 1:

Sinn und Bedeutung 15 - [PDF Document] (19)

18 von Heusinger

a) deictic discourse status further

characterization ex.

b) anaphoric speaker- and hearer-known perceivable in situation


c) discourse deictic

speaker- and hearer-known discourse-given (26)

d) recognitional speaker- and hearer-known reference to discourse (items)


e) indefinite speaker- and hearer-known, but discourse-new

shared (personal) knowledge


f) emotional speaker-known hearer-new, discourse-new

unaccented (29)

g) deictic speaker-known emotional / social distance


Table 1: usages of dies in German

(25) Nimm diesen Apfel. Take this apple.

(26) Es war einmal ein König. Dieser König hatte eine Krone. Once upon a time there was a king. This king had a crown.

(27) Er sagte: „Ich liebe Dich“, und mit diesen Worten ging er. He said „I love you“, and with these words he left.

(28) Weißt du was mit diesem Telefon passiert ist, das immer in deinem Zimmer war?

Do you know what happened to that (dieses) phone that used to be in your room?

(29) Gestern kam ich in eine Bar und da war dieser Fremde, der mich die ganze Zeit anstarrte.

Yesterday I walked into a bar and there was this stranger who stared at me all the time.’

(30) Und dann traf ich diesen Nachbarn von dir. And then I met this neighbour of yours.

Sinn und Bedeutung 15 - [PDF Document] (20)

Specificity, Referentiality and Discourse Prominence 19

The deictic use (25), the anaphoric one (26) and the discourse deictic one (27) are expected from the general function of demonstratives. They are clearly definite and discourse-given. The recognitional (or “anamnestic”) function (27) is discourse-new, but speaker- and hearer-given, i.e. definite. The indefinite (or presentative) use in (29) is speaker-given, but discourse-new and hearer-new. Lakoff (1974) describes an emotional use of the demonstrative in (30) and subsumes the indefinite use under it. However, I maintain that the indefinite use is independent as it is the case for the English indefinite this. The recognitional and indefinite uses are somewhat informal, but still to be found in written texts.

5 Indefinite so’n German provides another indefinite demonstrative, namely the form so’n, which derives from the demonstratives for properties so ‘such’ and the reduced and encl*tic indefinite article ‘n. It can substitute most, if not all, instances of indefinite dies in German. This form is rarely found in formal language, but quite frequent in informal registers. The spelling varies between so’n and son. Both forms can be found in the literature as well as in written versions of informal registers. It is controversial whether it constitutes an independent determiner with its own semantics, or is a merged form with a compositional semantics of demonstrative and indefinite article plus some pragmatic rules (as in the case of German prepositions with weak definite articles). Henn-Mennesheimer (1986) and Lenerz & Lohnstein (2004) assume that it consists of two underlying forms, while Hole & Klumpp (2004) maintain that it constitutes one form. They argue that so’n shows a plural paradigm in (31), which cannot be explained by a merged form since the indefinite article in German has no plural form (See also Chiriacescu 2011, von Heusinger (to appear) for more examples and a detailed discussion. Note that we could not find examples for the genitive):

(31) Paradigm of so’n Sg. Pl. Nom. so’n Pullover Nom. so’ne Pullover Gen. so’nes Pullovers Gen. so’ner Pullovers Dat. so’nem Pullover Dat. so’nen Pullover Akk. so’nen Pullover Akk. so’ne Pullover

German has a second demonstrative for properties, namely solcher, solche, solch(es), which behaves in function and distribution like English such. So is

Sinn und Bedeutung 15 - [PDF Document] (21)

20 von Heusinger

more flexible as it does not take nominal inflection. It has various functions, but we focus on usages with adjectives and unmodified nouns as in Table 2:

notion discourse status further

characterization ex.

a) gradable + deictic speaker- and hearer- known

refers to a grade (32)

b) gradable + anaphoric speaker- and hearer- known

refers to a grade (33)

c) deictic speaker- and hearer- known

refers a to a property or to a type (kind)


d) anaphoric speaker- and hearer- known

refers a to a property or to a type (kind)


e) intensifier speaker-known shifts the standard upwards (only with gradable nouns)


f) „hedging“ noun for exact description is unknown

denotation is extended (37)

g) indefinite speaker-known, hearer-new, discourse-new

unaccented (38)

h) emotional speaker-known emotional / social distance


Table 2: usages of so in German (Ehlich 1986, Eisenberg 1994, Umbach & Ebert (to appear))

(32) Ana ist so groß. Ana is so tall.

(33) Ana ist 1,80m groß. Maria ist auch so groß. Ana is 1.80m tall. Mary is also so tall.

(34) Er hat so ein Auto. He has such-a car.

(35) Maria hat ein Auto mit Heckklappe. So ein Auto hat er auch. Mary has a car with a hatchback. Such-a car does he also have.

(36) Er ist so ein Kind / so ein Pedant. He is such a child/ pedant.

Sinn und Bedeutung 15 - [PDF Document] (22)

Specificity, Referentiality and Discourse Prominence 21

The deictic and anaphoric uses of demonstrative so are illustrated with adjectives and nouns in (32) – (35). So refers to a grade if applied to adjectives and to a situationally or anaphorically given property if applied to nouns. It can trigger an intensified reading if applied to (unmodified) nouns that are inherently gradable, as in (36). So can trigger different kinds of hedging processes, as in (37), where it signals that the client can identify the referent, but does not have the correct lexical item at hand. With semantically bleached nouns like type, man, guy or with nouns in their typical environment, as in (38), neither an intensified nor a hedging function seems appropriate. Rather the form signals that a speaker-known, but hearer- and discourse-new referent is introduced. So’n also shows an “emotional” use, as in (39). In the following we focus on the indefinite function of so’n.

(37) Kunde im Geschäft: „Haben Sie so eine Klammer?“ Client in shop: Do you have such a clip?

(38) Da gibt’s so’nen Lehrer in meiner Schule in den ich verliebt bin. There is such -a teacher in my school whom I’m in love with.

(39) Peter hat so’n Hund gekauft. Peter bought such-a dog.

6 Discourse Properties Prince (1981) and Ionin (2006) report that indefinite this in English is only felicitous if the referent is taken up in the discourse and a noteworthy property is asserted with respect to it. Givón (1983) presents a quantitative study on the referential persistence of referents introduced by indefinite this. Here I can only report a first impression from corpus searches and the results from a pilot study on discourse prominence. A referent introduced by in-definite dies or indefinite so’n is typically picked up in the subsequent discourse, as in (40) and (41). Please note that the two indefinite determiners can replace each other and can also be replaced by the indefinite article (for more examples see Chiriacescu 2011, von Heusinger (to appear)).

(40) Da war dieser Typ aus Deutschland, den ich in einem Hostel auf der neuseeländischen Insel Waiheke kennen lernte. Vielleicht hieß er Wolfgang, vielleicht Volker - nicht so wichtig. Ich erinnere mich nur, dass er nett, ... (Cosmas)

‘There was this guy from Germany who I got to know in a hostel on the New Zealand island Waiheke. May be he was called Wolfgang, maybe Volker – not that important. I just remember that he was nice…

Sinn und Bedeutung 15 - [PDF Document] (23)

22 von Heusinger

(41) In unsrer Stadt gibts so’n Mann er ist nicht irgendeiner, nein, er ist unser neuer Bürgermeister, er sieht aus wie ein Vogelstrauß. (Google) ‘In our city there is such-a man, he is not anyone, no, he is our new mayor, he looks like an ostrich.’

I distinguish three different types of discourse prominence that can be measured. (cf. Givón 1983, Chiriacescu & von Heusinger 2010): (i) referen-tial persistence or the number of anaphoric expressions referring back to the discourse referent; (ii) topic shift potential or the distance between the discourse referent and its use as topic in the subsequent discourse, and (iii) discourse activation or the level of activation that determines the DP-type of the next anaphoric expression (according to the Givenness Hierarchy of Gundel et. al. 1993). Sofiana Chiriacescu and Annika Deichsel performed a pilot sentence continuation test. 10 subjects read the small fragment in (42) and were asked to continue it with five sentences. We then analyzed the five continuation sentences according to the three discussed parameters. I present the results from referential persistence (see Chiriacescu 2011 and Deichsel 2011 for more results). We counted the absolute number of all referential items that were anaphorically linked to the expression ein / so’n / dieser Mann in Figure 3.

(42) Das Essen in dem Restaurant war wirklich total lecker, aber ziemlich teuer. Als ich nach fünf Gängen beim Dessert war, hab’ ich gesehen, wie ein / so’n / dieser Mann Sekt bestellte.

‘The meal in the restaurant was really excellent, but quite expensive. When I got to dessert after five courses, I saw that a / such-a / this man ordered champagne.’

Sinn und Bedeutung 15 - [PDF Document] (24)

Specificity, Referentiality and Discourse Prominence 23

Figure 3: Referential persistence of ein / so’n / dieser Mann in (42)

The sum of all items referring back to the indefinite in the five sentences provided by the 10 subjects is 29 for indefinite dies, 24 for indefinite so’n, but only 8 for the indefinite article ein. This clearly indicates a difference be-tween indefinite dies and the indefinite article ein, with indefinite so’n taking an intermediate position. We can safely conclude that both indefinite demon-stratives signal discourse prominence.

7 Referential Properties The two indefinite demonstratives also show particular referential properties. I can only provide a few test sentences concerning referential specificity in (43) and (44), scopal specificity in (45) and epistemic specificity in (46). Table 4 summarizes the results of my own judgments and of the judgments of some informants (we also did a pilot questionnaire which confirmed the first intuitions – see Chiriacescu 2011 for so’n and Deichsel 2011 for dies).

The indefinite article in (43) and (44) allows a referential and a non-referential reading, with a preference for the latter. The use of indefinite dies is only compatible with the referential reading, while so’n has a preference for a referential reading, but is compatible with a non-referential reading, which is particularly obvious in (44).

(43) Eva will einen / so’n / diesen Film über Eliade sehen.

Sinn und Bedeutung 15 - [PDF Document] (25)

24 von Heusinger

Eva wants to watch a / such-a / this movie about Eliade.

(44) Maria will einen / so’nen / ?diesen Prinz auf einen weißen Ross heira-ten.

Mary wants to marry a / such-a / this prince on a white horse.

The indefinite article ein in (45) signals a preference for narrow scope, the in-definite dies always indicates wide scope, and the indefinite so’n either allows a wide-scope reading or a narrow-scope reading of the referent (co-variation with the universal quantifier), but then with a wide-scope reading of a property that is characteristic for all referents. We find similar intuitions for the epistemic reading in (46): The indefinite article allows both readings (with a preference for a non-specific reading), the indefinite dies only the (epistemic) specific reading, and the indefinite so’n clearly signals the specific reading, but may also by compatible with the non-specific reading, however, intuitions are unclear and blurred by other functions of so+n.

(45) Jeder meiner Kollegen hat ein / dies / so’n Buch von Eliade gelesen. Each of my colleagues read a / this / so-a book by Eliade.

(46) Ein / so’n / dieser Student in der Einführung hat beim Examen ge-schummelt.

A / this / such a student in the introduction has cheated in the exam.

Table 4: Referential properties of ein, so’n, dieser

This brief overview of the referential properties of the two indefinite demonstratives clearly indicates a high referential strength in contrast with the indefinite article ein, but also some differences between dies and so’n. Dies is more like the English indefinite this, always referential, scopal and epistemic specific, whereas so’n shows more variation which has to be inves-tigated in more detail.

Referential specificity (43) + (44)

Scopal specificity (45)

Epistemic specificity

(46) ein non-ref > ref narrow > wide non-spec > spec

so’n ref > (non-ref) wide > narrow (with wide scope for a property)

spec (*non-spec)

dieser ref (*non-ref) wide (*narrow) spec (*non-spec)

Sinn und Bedeutung 15 - [PDF Document] (26)

Specificity, Referentiality and Discourse Prominence 25

8 Demonstration and Topic Shift I assume a Kaplan- (1977/1989) style semantics for demonstratives according to which a demonstrative expression refers directly to its referent. The expression needs an accompanying demonstration (ostension), which raises the attention of the hearer to the intended referent. This semantics can be applied to demonstrative dies referring to entities, and to demonstrative so, referring to properties, as in (47) (see also Roberts 2002).

(47) Deictic readings of demonstratives a. [[dies N]] = is defined only if there is a demonstration d focussing on

(raising the attention to) a unique referent such that the referent is N (some additional conditions that the referent must be close to speaker etc.)

b. [[so’n N]] = is defined only if there is a demonstration d focussing on (raising the attention to) a unique property P and there is a referent x such that x is N and P. (some additional conditions that the referent must be close to speaker etc.)

Demonstratives are used without demonstration in their anaphoric use. They even introduce new discourse items as discussed in this article. It seems that the act of demonstration to a visible or perceivable object is shifted to the intention of the speaker towards a referent, which is unknown to the addressee. We can formulate a preliminary hypothesis that demonstrative nouns raise the attention of the hearer towards a new discourse item in (i) the (visible) situation, (ii) in the previous text, or (iii) in the subsequent text. We modify (47) to an informal definition of indefinite readings of demonstratives in (48):

(48) Indefinite readings of demonstratives a. [[diesindef N]] = is defined only if there is an intention of the speaker

to focus on (to raise the attention to) a unique referent such that the referent is N.

b. [[so’nindef N]] = is defined only if there is an intention of the speaker to focus on (to raise the attention to) a unique property P and there is a referent x such that x is N and P.

If the hearer recognizes the referential intention of the speaker, the hearer will establish a permanent discourse representation for the introduced referent: (i) indefinite dies: for an individual discourse referent; (ii) so’n: for the intended property and therefore also for the individual that falls under that property.

Sinn und Bedeutung 15 - [PDF Document] (27)

26 von Heusinger

This provides the link between a noteworthy property (see Ionin 2006) and the prominence of the discourse referent (Givón 1983).

9 Referentiality and Discourse Prominence In the last section we have sketched the shift from a deictic or anaphoric use of a demonstrative to an indefinite use. The idea is that one of the fundamental functions of demonstratives is to raise the attention towards a new referent. In this section I want to propose the link to the semantics of specific indefinites spelling out the relation between the referential properties and the discourse properties of the indefinite demonstratives in German (and English). Definition (48) for the indefinite reading of demonstratives includes as one of its central conditions the speaker’s intention. We have seen earlier that definition (23), repeated as (49), for specificity semantically represents this intention as an anchoring function between an attitude holder (or some other discourse referent) and the intended referent. This anchor is speaker- and hearer-given, but the content of the anchoring function is hearer-new, and therefore the intended referent is new, too.

(49) Informal definition of specificity in terms of referential anchoring A specific indefinite a N is represented by an anchoring function f from

an anchor to an individual and this individual is N. Both the anchor as well as the anchoring function must be given in the context

a) anchor is speaker- and hearer-given b) content of anchoring function is hearer-new

If we use this definition for representing the informal concept of “referential intention” in (48) we can formulate (50) for a semantics of indefinite demon-stratives. For both the anchor must be the speaker (thus reflecting the “demonstrative” or indexical nature). For dies the anchoring function yields the intended referent, while for so’n the anchoring function yields a property with which we uniquely identify the referent (reflecting the original nature of so as a demonstrative of properties). Thus indefinite so’n only indirectly pro-motes a referent to high referential strength and high discourse prominence.

(50) Indefinite readings of demonstratives a. [[diesindef N]] = is defined only if there is an anchoring function

from the speaker to an object such that the object is N. b. [[so’nindef N]] = is defined only if there an anchoring function from

the speaker to a property P such that the referent is N and that there is a referent x such that x is N and P.

Sinn und Bedeutung 15 - [PDF Document] (28)

Specificity, Referentiality and Discourse Prominence 27

We can conclude that the referential property of “referential intention” is the core meaning of indefinite demonstratives. It is best represented by refer-ential anchoring. We can then derive from this core-meaning the discourse function of “raising attention” as illustrated in section 8. We can also account for the differences between the two indefinite demonstratives – they are of different semantic types, they are represented by different semantic repre-sentations (giving rise to different semantic properties) and thus they also trigger different kinds of discourse prominence as reported above.

10 Summary I have shown that German has two additional indefinite articles, indefinite dies and (informal) so’n corresponding to English indefinite this. Both articles have grammaticalized from deictic expressions. They are different: indefinite this derives from the demonstrative for objects, while so’n derives from the demonstratives for properties. They are used to signal discourse prominence and they show a strong tendency for a referential meaning (wide scope with respect to other operators, rigid reference etc.). Indefinite dies has a higher discourse prominence and a higher referential strength than indefinite so’n, which only indirectly promotes the referent to high referential strength and high discourse prominence. I have argued that the shift from the deictic or anaphoric use of a demonstrative to an indefinite use can be best explained by a semantics of “referential anchoring”. This semantics makes it possible to represent the “referential intention” of the speaker in an adequate way, qualifying for a specific interpretation of these demonstratives. The specific semantics then allows for discourse prominence, e.g. for referential persistence. In this view the referential properties are primary and the discourse properties are derived, contradicting Wright & Givón’s (1987) claim. This picture needs more empirical coverage and a carefully worked out semantic representation, which are two challenging research issues.

References Chiriacescu, Sofiana. 2011. Topic shifters. Dissertation. Universität Stuttgart. Chiriacescu, Sofiana & Klaus von Heusinger. 2010. Discourse prominence

and pe-marking in Romanian. The International Review of Pragmatics 2. 298–322.

Chierchia, Gennaro. 2001. A puzzle about indefinites. In: C. Checchetto, G. Chierchia & M.-T. Guasti (eds.). Semantic Interfaces: Reference, Anaphora and Aspect. Stanford: CSLI Publications. 51–89.

Sinn und Bedeutung 15 - [PDF Document] (29)

28 von Heusinger

Chierchia, Gennaro. 2005. Definites, locality, and intentional identity. In: G.N. Carlson & F. J. Pelletier (ed.). Reference and Quantification. The Partee Effect. Stanford/Calif.: CSLI Publications. 143–178.

Deichsel, Annika, 2011. The Semantics and Pragmatics of Indefinite Demonstratives. Ms. Universität Stuttgart

Ehlich, Konrad. 1986. So- Ueberlegungen zum Verhaeltnis sprachlicher Formen und sprachlichen Handelns, Allgemein und an einerm widerspenstigen Beispiel. In: I. Rosengren (eds.). Sprache und Pragmatik. Lunder germanistische Forschungen 55. 279–298.

Eisenberg, Peter. 1994. Grundriss der deutschen Gramatik. Stuttgart: Metzler.

Endriss, Cornelia. 2009. Quantificational Topics. A Scopal Treatment of Exceptional Wide Scope Phenomena. Dordrecht: Springer.

Enç, Mürvet. 1991. The semantics of specificity. Linguistic Inquiry 22. 1–25. Farkas, Donka. 1994. Specificity and scope. In: L. Nash & G. Tsoulas (eds.).

Actes du Premier Colloque Langues & Grammaire. Paris. 119–137. Fodor, Janet & Ivan Sag. 1982. Referential and Quantificational Indefinites.

Linguistics and Philosophy 5. 355–398. Givón, Talmy. 1983. Topic Continuity in Discourse. An Introduction. In: T.

Givón (ed.). Topic Continuity in Discourse. A Quantitative Cross-Language Study. Amsterdam; Philadelphia: Benjamins. 1–41.

Gundel, Jeanette, Nancy Hedberg & Ron Zacharski. 1993. Cognitive Status and the Form of Referring Expressions in Discourse. Language 69. 274–307.

Heim, Irene. 2011. Definiteness. In: Klaus von Heusinger & Claudia Maienborn & Paul Portner (eds.). Semantics. An International Hand-book of Natural Language Meaning. Vol. 2. Berlin; New York: de Gruyter.

Henn-Memmesheimer, Beate. 1986. Nonstandardmuster. Ihre Beschreibung und das Problem ihrer Arealität. Tübingen: Niemeyer.

von Heusinger, Klaus. 2002. Specificity and definiteness in sentence and discourse structure. Journal of Semantics 19. 245–274.

von Heusinger, Klaus. 2011. Specificity. In: Klaus von Heusinger & Claudia Maienborn & Paul Portner (eds.). Semantics. An International Handbook of Natural Language Meaning. Vol. 2. Berlin; New York: de Gruyter.

von Heusinger, Klaus. to appear. Referentialität, Spezifizität, und Diskursprominenz im Sprachvergleich. In. Lutz Gunkel (ed.). Deutsch im Sprachvergleich – Grammatische Kontraste und Konvergenzen. Berlin: de Gruyter.

Sinn und Bedeutung 15 - [PDF Document] (30)

Specificity, Referentiality and Discourse Prominence 29

Himmelmann, Nikolaus P. 1996. Demonstratives in Narrative Discourse. A Taxonomy of Universal Uses. In: Barbara Fox (ed.). Studies in Anaphora. Amsterdam: John Benjamins. 203–252.

Hole, Daniel & Gerson Klumpp. 2000. Definite type and indefinite token: the article son in colloquial German. Linguistiche Berichte 182. 231–244.

Ionin, Tania. 2006. This is definitely specific: specificity and definiteness in article systems. Natural Language Semantics 14. 175–234.

Kamp, Hans & Ágnes Bende-Farkas. 2006. Specific Indefinites: Anchors and Functional Readings. Ms. Universität Stuttgart.

Kamp, Hans & Ágnes Bende-Farkas. submitted. Epistemic specificity from a communication-theoretical perspective. Journal of Semantics.

Kaplan, David. 1977/1989. Demonstratives. An Essay on the Semantics, Logic, Metaphysics, and Epistemology of Demonstratives and Other Indexicals. In: J. Almog, J. Perry & H. Wettstein (eds.). Themes from Kaplan. Oxford: Oxford University Press. 481–563.

Kratzer, Angelika. 1998. Scope or pseudoscope? Are there wide-scope indefinites? In: S. Rothstein (ed.). Events and Grammar. Dordrecht: Kluwer. 163–196.

Lakoff, George. 1974. Remarks on this and that. Chicago Linguistic Society (CLS) 10. 345–356.

Lenerz, Jürgen & Horst Lohnstein. 2004. Nur so-Strukturaspekte der Ver-gleichskonstruktion. In: Franz Josef d’Avis (ed.). Deutsche Syntax: Empirie und Theorie. Symposium Göteborg, 13–15 Mai 2004. Acta Universitatis Gothoburgensis. 81–103.

Maclaran, Rose. 1980. On two asymmetrical uses of the demonstrative determiners in English. Linguistics 18. 803–820.

Onea, Edgar and Ljudmila Geist. 2010. Indefinite Determiners and Referential Anchoring. Ms. Universität Göttingen and Universität Stuttgart.

Prince, Ellen. 1981. On the Inferencing of Indefinite-this NPs. In: A. Joshi & B. Webber & I. Sag (eds.). Elements of Discourse Understanding. Cambridge: CUP. 231–250.

Roberts, Craige. 2002. Demonstratives as Definites. In: K. van Deemter & R. Kibble (eds.). Information Sharing: Reference and Presupposition in Language Generation and Interpretation. Stanford: CSLI Publication. 89–136.

Roberts, Craige. 2007. Specificity and Incomplete Descriptions. Handout UMass/Amherst, March 27, 2007.

Umbach & Ebert. (to appear). German demonstrative so – intensifying and hedging effects. In: H. C. Schmitz & H. Zeevat (eds.). Proceedings of

Sinn und Bedeutung 15 - [PDF Document] (31)

30 von Heusinger

the ESSLLI X Workshop on formal and experimental approaches to discourse particles and modal adverbs.

Wespel, Johannes. 2006. Indefinite demonstratives. Ms. Universität Stuttgart. Wright, S. & Givón, Talmy. 1987. The pragmatics of Indefinite Reference:

Quantified Text-Based Studies. Studies in Language 11. 1–33.

Sinn und Bedeutung 15 - [PDF Document] (32)

On the Semantics of Existence Predicates

Friederike Moltmann IHPST (Paris1/ENS/CNRS), Paris [emailprotected]

Abstract. The most common philosophical view about the notion of existence is that it is a second-order property or existential quantification. A less common view is that existence is a (first-order) property of 'existent' as opposed to 'nonexistent' (past or merely intentional) objects. An even less common view is that existence divides into different 'modes of being' for different sorts of entities. In this paper I will take a closer look at the semantic behavior of existence predicates in natural language, such as 'exist', 'occur', and 'obtain', arguing that existence predicates in natural language support the two less common philosophical views. I will develop explicit analyses of existence predicates in their time-relative and space-relative uses which will explain why they apply to some kinds of entities, but not others.

1 Introduction The most common philosophical view of existence is that existence amounts to existential quantification (the Quinean tradition) or is a second-order concept (the Kant-Frege tradition). A less common philosophical view is that existence is a first-order property distinguishing between nonexistent (past, possible, or merely intentional) objects and existing objects. An even less common philosophical view is that existence divides into different ‘modes of being’ for different kinds of entities (a view held, for example, by Aristotle, Heidegger, Sarte, Ryle, and more recently McDaniel).1 The aim of the present paper is to take a closer look at how the notion of existence is in fact expressed in natural language. In natural language, it appears, existence is not so much expressed by quantification, which can be shown to be neutral as regards any distinction between existent and nonexistent objects that one might draw. Rather existence is expressed by predicates, and that is, first-order predicates. Furthermore, there is, at least in English, no single existence predicate, but rather at least three: exist, occur (or related predicates such as

1 For a recent defense of the Quinean view, see van Inwagen (1998). For a recent defense of a view of existence dividing into different modes of being see McDaniel (to appear a, b).

In Reich, Ingo et al. (eds.), Proceedings of Sinn & Bedeutung 15, pp. 31–54. Universaar – Saarland University Press: Saarbrücken, Germany, 2011.

Sinn und Bedeutung 15 - [PDF Document] (33)

32 Moltmann

happen or take place), and obtain. The semantic behavior of such existence predicates (regarding the kinds of entities that they can apply to and the sorts of adverbial modifiers they allow) reveals a notion of existence that divides into at least three different kinds of modes being, reflecting the distinction between endurance and perdurance, as well as their space-related analogues, but also the particular mode of being of such entities as states, facts, conditions, and laws. Existence predicates reflect such distinctions primarily in their time- and space-related uses, but in fact the location-independent uses of existence predicates should best be understood as derivative upon their location-dependent uses.

I will first clarify the notion of an existence predicate itself as well as some fundamental semantic differences in natural language between constructions of existential quantification and existence predicates. I then discuss the semantics of the three existence predicates exist, occur, and obtain in greater detail.

2 The Notion of an Existence Predicate While many philosophers take existence to amount to just existential quantification, it has hardly escaped philosophers’ attention that natural language displays existence predicates, in particular, of course, the predicate exist. A number of philosophers, most notably Frege, have taken exist to be a special, second-order predicate, applying to concepts rather than individuals. However, from the point of view of natural language semantics, exist clearly is a first-order predicate.2 It does not require, like putative second-order predicates, predicative terms, but rather requires expressions in subject position that act as singular terms, as in (1a, b) and (2a, b), or that act as quantifiers binding individual variables, as in (1c):3

(1) a. The president of France exists. b. Mars exists.

2 Philosophers that have argued that exist is a first-order predicate include Miller (1975, 1986, 2002), Salmon (1987, 1989), and McGinn (2000). 3 Exist also allows for bare plurals or mass nouns in subject position, in which case it does seem to express existential quantification: (i) a. Giraffes exist. b. Gold exists. However, as will be shown later, bare plurals and mass nouns in the subject position of exist-sentences are in fact kind-referring, and thus singular terms. This means that exist is a first-order predicate in sentences like (i-a, b) as well.

Sinn und Bedeutung 15 - [PDF Document] (34)

Existence Predicates 33

c. Some planet exists.

(2) a. The king of France does not exist. b. Vulcan does not exist.

When exist is viewed as a first-order predicate in positive existence statements as in (1a, b, c), it is generally taken to express a trivial or almost trivial property, the property every entity has or, less trivially, the property that every present and actual entity has. The occurrence of exist in negative existentials, as in (2a, b), is more difficult to handle. The main focus of this paper is the application of time- or space-relative uses of existence predicates to particular kinds of entities, but for an appropriate discussion of existence predicates a few words are necessary concerning negative existentials.

There is a significant debate of how to analyse negative existentials with singular terms, while treating exist as a first-order predicate. On one view, the subject term in negative existentials is an empty term, exist expresses the trivial property everything has, and negation is taken to be external negation. On that view, (2b) will mean: ‘it is not true that Vulcan exists’. On another view, the Meinongian view, the subject term in a true negative existential always stands for an entity, but a ‘nonexistent’ entity, an entity of which exist is false. There is also a third, hybrid view, that of Salmon (1987, 1998), on which the subject term in true negative existentials sometimes stands for an object of which exist is not true, namely an object that has existed only in the past or a merely possible object. If the subject of the negative existential is a fictional term, though, Salmon takes it to be empty, with negation then being external negation.

Negated existence predicates in existentially quantified sentences pre-sent a particularly interesting phenomenon which appears to give support for the Meinongian view. Meinongians have long argued that existential quan-tification, unlike predication with exist, is not existentially committing.4 This is displayed by the following ‘Meinongian’ statement:

(3) There are things that do not exist.

The Meinongian statement in (3) can hardly serve as a piece of ‘linguistic evidence’ for the Meinongian view, though: a sentence like (3) serves to express a philosophical position, rather than being a ‘natural’ sentence of natural language, a sentence that can be used without thereby making explicit a philosophical conviction. But there are constructions in natural language that appear to involve intentional objects as semantic values of subject terms 4 Recent Meinongians include Parsons (1980) and Priest (2005).

Sinn und Bedeutung 15 - [PDF Document] (35)

34 Moltmann

in negative existence statements. The relevant sentences are entirely natural in the sense that they hardly sound like the expression of a philosophical view. These are examples:

(4) a. Some people John mentioned do not exist. b. Some things John thought of do not exist.

In (4a, b) the subject consists of a definite description formed, crucially, with an intentional verb, such as mention or think of. Such verbs appear to take intentional ‘nonexistent’ objects as arguments when the intentional act they describe is not successful, and these entities appear to be the ones the existential quantifiers in (4a,b) range over.

If intentional ‘nonexistent’ objects are involved in existentially quan-tified negative existentials as in (4), then they may just as well be involved in negative existentials with apparent empty proper names and descriptions associated with a failed or pretend act of reference (and in particularly des-criptions formed with intentional verbs as in the woman John mentioned does not exist). However, this paper is not the place to defend such a view further.5 What is important in the present context is simply the difference displayed by existential quantification and existence predicates in natural language.

The apparent ability of natural language quantifiers to range over non-existent objects is not the only respect in which existential quantification in natural language differ from existence predicates. There is a further fun-damental difference between sentences with an existence predicate and with existential quantification. Existential quantifiers in natural language may range over entities of any kind whatsoever, but existence predicates in natural language are generally restricted to a particular kind of entity. Thus, exist is restricted to enduring objects (basically material objects), at least in its tensed form. For perduring entities, that is, events, occur (or happen or take place) is the appropriate existence predicate instead. Moreover, for entities of the sort of states or facts, abstract entities in a certain sense, there is a specific existence predicate in English, namely obtain.

The apparent variety of existence predicates in natural language raises the question of what makes a predicate an existence predicate in the first place. This question obtains importance in view of the fact that in the history of philosophy there have been a variety of views as to what ‘modes of being’ there are. Thus Sartre took conscious entities to engage in a distinctive mode of being and in the phenomenological tradition ‘being experienced’ was considered a mode of being (even the only one). The question thus is: is there 5 See McGinn (2000) for a philosophical defense of that view.

Sinn und Bedeutung 15 - [PDF Document] (36)

Existence Predicates 35

a criterion that tells us whether a predicate is an existence predicate in a linguistically relevant sense? There is in fact a very clear criterion for exis-tence predicates, namely the semantic behavior of predicates under negation. Negative existentials display the peculiarity that they can be true even if the subject does not refer to or stand for an actual object. Other predicates simply do not display that particular feature. Ordinary predicates when negated will in such a case lead to sentences that intuitively lack a truth value (such as the present king of France is bald).

Intentional verbs like think of, mention, or describe, as already men-tioned, arguably take intentional objects as arguments (John thought of the woman Bill mentioned). Unlike existence predicates, though, they do not sys-tematically yield truth when negated. Thus, the sentence below may very well be false, even if the woman John described does not exist:

(5) Bill did not think of the woman John described.

Ordinary (non-intentional) predicates thus are subject to the following condition:

(6) a. A (intransitive) predicate P is an ordinary predicate iff for any world w and time t, for any singular term T, if T does not stand for an actual entity in w, then neither [T not P]w, t = true nor [T not P]w, t = false.

By contrast, existence predicates are subject to the following condition (formulated so as to remain as neutral as possible regarding the treatment of negative existentials):

(6) b. Criterion for Existence Predicates An (intransitive) predicate P is an existence predicate iff for any world w and time t, for any singular term T, if T does not stand for a (present, actual) entity in w, then [T not P]w, t = true.

Applying the criterion to some putative existence predicates, exist obviously is classified as an existence predicate, as are occur and obtain given the possible truth of the sentences below:

(7) a. The event John mentioned did not occur. b. The situation John described did not obtain.

Live and be alive are putative existence predicates, but by the criterion they do not come out as such:

(8) The person Mary mentioned does not live / is not alive.

Sinn und Bedeutung 15 - [PDF Document] (37)

36 Moltmann

Live and is alive presuppose that the object they apply has been alive before; they do not specify the actual existence of an object of thought. The criterion also rules out as existence predicates a range of other predicates that according to particular historical philosophical views might be regarded as such, for example being experienced.

With this clarification of the notion of an existence predicate as such, let us now focus on the semantic differences among different existence predicates as well as their time- and space-related application.

3 The Difference between exist and occur Existence predicates when they occur in a positive sentence in a tensed form have a particular lexical meaning relating an entity in a certain way to its location in time. Let us first look at the two existence predicates exist and occur. We have already seen that tensed exist applies to ‘enduring’ objects (which I will call simply ‘objects’), whereas occur applies to ‘perduring’ object’, that is, events.

The categorial restrictions of exist and occur give support for ‘three-dimensionalism’ as opposed to ‘four-dimensionalim’ regarding material objects. On a four-dimensionalist view, events and objects are both space-time worms bearing the same relation to space and time: they are both ‘spread out’ in space and time. This means that they both occupy a spatio-temporal region by having parts that occupy the subregions of the spatio-temporal region. Given the four-dimensionalist view, it is hard to see how time-relative exist could be restricted to one sort of four-dimensional object rather than another. By contrast, on a three-dimensionalist view, objects and events will be fundamentally different kinds of entities, bearing fundamentally different relations to time: objects endure, whereas events perdure. These restrictions are easy to formulate on a three-dimensionalist view

The standard view of endurantism draws the distinction between en-during and perduring entities as follow (Lewis 1986, Hawley 2001): An enduring entity occupies a time t by being wholly present at each moment of t, whereas a perduring entity occupies a time t by some part of it being present at any moment of t. Endurance thus requires complete presence of an object at each moment of its lifespan. By contrast, perdurance requires only the presence of a part of an entity at any moment of its duration. The notions of endurance and perdurance correspond to time-relative existence and extension, in the following sense of Fine (2006): enduring entities exist at a time, which means they are completely presence at each moment of the time;

Sinn und Bedeutung 15 - [PDF Document] (38)

Existence Predicates 37

by contrast, perduring entities are extended over a time, which means only a part of them needs to be present at any given moment of the time. This way of drawing the distinction means that perduring entities can have temporal parts, whereas enduring entities cannot. Entities such as organisms, artifacts, and entities with lasting spatial boundaries, such as countries or stones, are among the enduring entities, whereas events are examples of perduring entities.

The distinction between enduring and perduring entities in terms of time-related existence and extension seems well suited for capturing the lexical meanings of exist and occur. However, exist cannot just express the property that holds of an object x at a time t iff the whole of x is present at each moment of t, and occur cannot just express the property that holds of an object at a time t iff for any moment t’ of t, some part of x is present at t’. This would allow both occur and exist to apply to momentaneous events. Exist and occur will in addition impose particular sortal constraints on the entities they can apply to. Exist imposes the condition that the entity it applies to not be able to have temporal parts, in virtue of the kind of entity it belongs to. Occur imposes the condition that the entity it applies be able to have temporal parts, in virtue of the kind it belongs to.

Thus the lexical meanings of tensed exist and of occur will be as follows:

(9) a. The lexical meaning of tensed exist For an entity x that cannot have temporal parts, for a time interval t, x ∈ [exist]t iff for every t’ < t, the whole of x is present at t’.

b. The lexical meaning of occur (preliminary formulation) For an entity x that can have temporal parts, for a time interval t, x ∈ [occur]t iff for every t’ < t some part of x is present at t’.

There is a problem, though, with giving the meaning of exist and occur as in (9), and that is that it fails to account for the fact that exist is a stative verb, whereas occur is an eventive verb. It is remarkable that in natural languages there is generally no stative existence predicate of events, even though ‘extension in time’ appears to be a state. English occur clearly is an eventive verb: it takes adverbials modifiers that are typical of event predicates, such as suddenly or quickly, modifiers that are impossible with stative predicates; moreover it allows for the progressive (the event is occurring right now), which again is typical of eventive predicates. Occur thus does not describe a state of extendedness over a time, but rather it describes a transition, or rather

Sinn und Bedeutung 15 - [PDF Document] (39)

38 Moltmann

a series of transitions, from an event part being at a time to another event part being at a subsequent time.

In order to account for this difference between exist and occur, it is useful to adopt the Davidsonian view of the semantics of verbs on which events (and states) act as implicit arguments of verbs. Given the Davidsonian view, the implicit event arguments of verbs will also be the referents of the corresponding nominalizations such as the occurrence of the murder or the existence of the house. Occur will thus express a two-place-relation between occurrences and (occurring) events and exist a two-place relation between states of existence and (existent) entities.

Occurrences appear to be a kind of temporal abstraction from the occurring events. This is indicated by the differences in the kinds of pre-dicates that occurrences and occurring events accept: occurrences, unlike occurring events, do not accept qualitative predicates:

(10) a. The murder was grisly / brutal. b. ??The occurrence of the murder was grisly / brutal.

Occurrences do allow for other types of event predicates, though, for example temporal and attitudinal predicates:

(11) The occurrence of the murder was sudden / unexpected.

Thus it appears that occurrences are events that retain none but the temporal features of the corresponding occurring events. Occurrences may therefore be viewed as transitions from the presence of a proper part of the occurring event at a time to the presence of a proper part of the occurring event at a subsequent time. I will take the ‘presence’ of an event part at a time to be a temporal trope: the relational trope that is the instantiation of the temporal at-relation (AT) in the event and the time. Within the Davidsonian semantics of events, the lexical meaning of occur can thus be reformulated as follows, where f is the function mapping an n-place relation and n entities to the instantiation of the relation in those entities:

(12) The lexical meaning of occur For a time t and events e and e’, <e, e’> ∈ [occur]t iff e = the sequence of transitions from f(AT, e’’, t’) to f’(AT, e’’’, t’’) for any subsequent times t’, t’’ < t for which there are event parts e’’ and e’’’ of e.

What about momentaneous events, a flash or a crack? The account applies here as well: the occurrence of such a momentaneous event is simply a single trope that is the manifestation of the temporal ‘at’-relation with respect to that event and the moment it occurs.

Sinn und Bedeutung 15 - [PDF Document] (40)

Existence Predicates 39

The Davidsonian semantics of verbs can also be applied to exist. Exist is a stative verb, satisfying standard linguistic criteria for stativity. Also its nominalization existence clearly describes a state rather than an event. Looking at the range of predicates that can apply to ‘existences’, though, it appears that existences are states of a particular sort: existences have fewer properties than one might have thought states of existence should have. For example, a state of existence, one might think, should have a spatial location, being located just where the object is during its existence. But sentences with exist do not allow for spatial modifiers:6

(13) a. *John exists in Germany. b. *The king of France existed in France.

Obviously, the location modifiers cannot be predicated of the state of existence, and thus existences, if they are Davidsonian event arguments, simply cannot have spatial location.

The location modifiers in (13a, b) may be understood in another way than as predicates of the Davidsonian event argument. In natural language semantics, two different kinds of adverbial modifiers in fact need to be distinguished: adverbials that act as adjuncts and adverbials that act as complements. If an adverbial modifier is obligatory, it will be a complement; if it is not, it may be adjunct (though it need not be). As an adjunct, an adverbial acts as a predicate of the state or event described. As a complement, it provides an argument of the relation expressed by the verb, and thus generally provides a component that is constitutive of the described event or state.7 Exist does not allow for location modifiers as complements either, when it applies to material objects. The reason is that material objects do not exist in space: for an entity e to exist at a spatial location l requires e to be completely present at each sublocation of l. But this is impossible for material objects. Material objects are rather extended in space: they occupy a space by having some part being at any sub-region of the space. Since material objects have spatial parts, they cannot be completely present at each sublocation of their location.

6 There is one important exception to this generalization, namely exist-sentences with bare plurals and mass nouns, which in fact are kind-denoting terms: (i) Giraffes exist only in Africa. I will turn to those in Section 3. 7 A variety of syntactic tests distinguish adjunct and complement adverbial modifiers, which I will not go into.

Sinn und Bedeutung 15 - [PDF Document] (41)

40 Moltmann

Recent linguistic semantic work on stative verbs can help give an answer to the question why exist resists location modifiers as adjuncts. It has long been observed that in fact most stative verbs do not allow for location modifiers and thus display what is known as the Stative Adverb Gap (Katz 2003). Among those verbs are know, own, and weigh:

(14) a. ??John knows mathematics in France. b. ??John owns a watch in Munich. c. ??John weighs 100 kilos in Germany.

In fact, such verbs also resist a range of other modifiers, such as manner modifiers, and they cannot form the complement of perception verbs:

(15) a. ??John owns a watch with a lot of effort. b. ??John knows French in a strange way.

(16) a. ??Mary heard John know French. b. ??Mary saw John weigh 100 kilos.

Exist in fact patterns with that class of verbs also in these two respects:

(17) a. *The house has existed for a few years in a strange way. b. *John saw the house exist for many years.

Maienborn (2007) has proposed an ontological account of the Stative Adverb Gap, tracing it to the particular nature of the state most stative verbs describe.8 Maienborn argues that stative verbs take states as arguments that are abstract in the sense of a Kimean conception of events or rather states. A Kimean account of states will consist in a specification of the existence and identity conditions of states on the basis of a property, an individual, and a time, as below.

(18) The Kimean account of states a. For a property P and an object o, f(P, o) obtains at a time t iff o ∈Pt. b. For properties P, P’, o, o’, f(P, o) = f(P’, o’) iff P = P’, o = o’.

The Kimean account of states amounts to an implicit definition of states. This means that states will have only those properties as intrinsic properties that are specified by the account itself. Thus, they will have properties of temporal duration, but not of spatial location. On a Kimean account, states will come

8 There is an alternative proposal concerning the Stative Adverb Gap, namely that of Katz (2003). Katz argues that stative verbs lack an event argument position. But see Maienborn (2007) for discussion.

Sinn und Bedeutung 15 - [PDF Document] (42)

Existence Predicates 41

out as abstract in the sense of not involving a particular manifestation; rather all there is to a Kimean state is what is specified by the account itself.9

The condition of complete presence that exist imposes requires some qualifications. It cannot be that strictly all the parts of an enduring object need to be present at any moment of the lifespan of the object; this should only hold for atemporal parts, not the parts that a material object may have only at some point during its lifespan (Fine 2006). Also the atemporal parts need to be understood appropriately. An object may have functional parts which can be constituted by different material at different times and thus are not strictly speaking material parts. Furthermore, the question arises whether complete presence should not also include essential features of an object, such as its configuration or form if it is essential or qualitative features. In any case, what defines existence at a time should be the recurrence of features and parts constitutive of an object throughout the time, not the presence of material parts as such.

4 Space-­‐Relative Existence Ordinary material objects generally cannot bear the existence relation to a spatial location because they cannot be completely present at the spatial sublocations. But there are entities for which space-relative existence is possible. Fine (2006) in fact argues for a generalization of existence at a time, as complete presence at a time, to existence at a spatial location, that is, complete presence at a location. Fine gives the example of a composite aroma of coffee and vanilla whose presence at a location, he argues, requires the presence of both the aroma of vanilla and the aroma of coffee. This example is not unproblematic from the present point of view, though: aromas do not go along very well with the existence predicate exist:

(19) a. ??The aroma exists in that room.

The reason why aromas do not go along with exist appears to be an ontological one. Aromas as particulars simply cannot be wholly present at different locations. Only aromas as kinds can, as in the examples below:10

9 Verbs that describe Kimean states contrast with verbs that describe what Maienborn calls ‘Davidsonian states’. The latter include verbs like stand, sit, and sleep. Concrete states do allow for location and manner modifiers and can be the object of perception. 10 Sounds and physical fields for Fine are other cases of entities involving complete presence at a given location. I find the example of sounds even more problematic than aromas. Sounds neither as particulars nor as kinds seem to accept existence predicates, including location-relative existence predicates in particular:

Sinn und Bedeutung 15 - [PDF Document] (43)

42 Moltmann

(19) b. This kind of perfume does not exist in France anymore. c. This kind of aroma only exists in oriental countries.

What are aromas as particulars? Arguably aromas as particulars are tropes without a bearer: they are mere spatio-temporally located features. Tropes in fact in general do not go along very well with space-relative existence pre-dicates:

(20) ??The greenness of the plants exists everywhere in the garden.

Space-relative exist with tropes is not possible because tropes as particulars (with or without a bearer) cannot be present at different locations at once.

But there are entities, particular sorts of abstract entities, with which space-relative exist is perfectly natural. One such case is languages. Lan-guages can be completely present at different places, and they do allow for location modifiers with exist:11

(21) This dialect does not exist in this region anymore.

It is easily explained why languages have space-relative existence. The lo-cation of a language is the region where the language is spoken, and every part of that region should count as a location of the language, and of course the entire language.

Other entities that display space-relative existence include condition-like entities such as situations and laws, which also allow for space-relative obtain, as will be discussed later.

Another important case of abstract entities displaying space-relative existence are kinds. This requires a little linguistic elaboration. It is a com-mon view among linguists that kinds can be the referents of bare mass nouns and plurals, in particular with predicates like widespread or rare:12

(22) a. Ants are widespread. b. White gold is rare.

Whereas kind reference of bare plurals and mass nouns with predicates like widespread is considered unproblematic, linguists are not unanimous that bare plurals and mass nouns are always kind-referring. Some linguists, in

(i) a. ?? The sound exists throughout the house. b. This kind of sound does not exist in modern opera houses anymore. 11 Thanks to Roger Schwarzschild for bringing this example to my attention. 12 Obviously these need not be natural kinds, but may include kinds of artifacts – in fact kinds of any sorts of entities.

Sinn und Bedeutung 15 - [PDF Document] (44)

Existence Predicates 43

particular Carlson (1977) and Chierchia (1998), held the view that bare mass nouns and plurals are (almost) always kind-referring (with the predicate being responsible for different readings). Other linguists hold the view that with some predicates, such as eat or buy bare plurals or mass nouns express existential quantification ranging over the instances of the kind (such as quantities or individuals). The present discussion fortunately can stay entirely neutral as regards those two linguistic views. There are a range of tests for kind reference that tell if a given occurrence of a bare mass noun or plural can only be kind-referring, and these tests show clearly that with exist a bare plural or mass noun must be kind-referring.

First, anaphora when they take a bare plural as subject of an exist-sentence as antecedent stand for the entire kind, not one of the instances that an existential quantifier would range over:

(23) a. Dinosaurs do not exist. But they once did exist. b. Three dinosaurs do not exist. *But they (three dinosaurs or other)

once did exist.

Moreover, bare plurals and mass nouns can be modified by a relative clause whose predicate is an instance-distribution predicate:

(24) Dinosaurs, which used to be widespread in Europe, do not exist anymore.

Also temporal modifiers of exist with bare plurals or mass nouns show kind reference:

(25) a. Dolphins still exist. b. Dinosaurs no longer exist.

The same holds for aspectual predicates such as continue or cease:

(26) a. Dinosaurs continued to exist. b. Dinosaurs ceased to exist.

Finally, bare mass nouns and plurals do not allow an interpretation on which they act as existential quantifiers taking wide scope over other quantifiers or negation in the sentence, unlike ordinary existentially quantified NPs:

(27) a. Dinosaurs do not exist anymore. (impossible as: for some dinosaurs x, x does not exist anymore)

b. Two dinosaurs do not exist anymore. (ok as: for two dinosaurs x, x does not exist anymore)

Sinn und Bedeutung 15 - [PDF Document] (45)

44 Moltmann

Thus, bare plurals and mass nouns in exist-sentences stand for kinds, in the relevant (linguists’) sense.

An important observation about exist-sentences with kind terms is that they also allow for terms referring to kinds of events:

(28) Great wars still exist.

The crucial observation now is that exist-sentences with kind terms, including terms for kinds of events, allow for location modifiers:13

(29) a. Giraffes still exist in Africa. b. Political protests do not exist in Bhutan.

What does the existence of a kind amount to, and the existence of a kind at a location in particular? Given the semantics of existence statements with kind terms, obviously, the existence of a kind as such means that there is an actual instance of the kind, and the existence of a kind at a location means that there is an instance of the kind at the location. That is, the existence of a kind at a location means that the kind is instantiated in an individual at the location (or at a part of the location).

Location-relative existence applied to kinds should also amount to complete presence at the relevant locations. But in what sense could a kind be completely present at a location, being instantiated in an individual at the location? It should somehow mean that all the parts of the kind are present at the location of the individual instantiating the kind. But what are the parts of a kind? One might think that the parts of a kind are the instances of the kind, a kind being a sort of plurality of all its instances (or all its possible instances). But this would give the wrong result since not all the instances can be at any location at which a kind is instantiated. In fact, the more common view about the parts of a kind is that the parts are the characteristics of the kind, that is, the attributes that together make up the ‘essence’ of a kind. Complete presence of a kind at a location would thus mean instantiation of all the attributes of the kind in an individual at the location.

There is one remaining puzzle concerning the space-relative existence of kinds. One might have thought that an existence statement locating a kind

13 There are in fact two linguistically relevant notions of kind: referents of bare plurals or mass nouns and kinds as referents of definite singular kind terms as in (i) below. Only the former not the latter allow for space-relative exist, as seen in (ii): (i) The giraffe is mammal. (ii) a. Giraffes exist everywhere. b. ??The giraffe exists everywhere.

Sinn und Bedeutung 15 - [PDF Document] (46)

Existence Predicates 45

at a spatial region such as (29a) should state the complete presence of the kind at each sub-location of the location mentioned by the location modifier, just as a time-relative existence statement requires the complete presence of the individual at each moment of its lifespan. This is not the case in (29a), though, which does not require the complete presence of the kind ‘giraffes’ in each part of Africa. It is in fact sufficient that the kind be completely present just at the locations in Africa, and it suffices that there be just some instances of the kind in Africa.

It seems that this puzzle has less to do with space-relative existence for kinds as such than with the semantics of English locative sentences. A weak, existential condition seems to be part of the semantics of locative modifiers in general, for example in the sentence below:

(30) John resides in Munich.

In Munich specifies that John’s residence is located somewhere in Munich, not that it is located everywhere in Munich or all over Munich.14

Not just location modifiers with in, but also those with a variety of other spatial prepositions locate the described event or individual in fact just somewhere within the location mentioned. This is illustrated with location-relative existence statements below:

(31) a. Giraffes exist outside of Africa. b. Giraffes still exist near the coast.

Also the location modifiers in (31) do not give the precise location of the entity in question. Thus, for a term T, in T locates an entity somewhere in the location that T refers to, outside T locates it somewhere outside that location, and near T locates it somewhere ‘near’ that location.

Only special locational modifiers such as throughout and all over require that every part of the location mentioned is where the entity or event in question is located:

(32) a. Giraffes exist throughout Africa. b. Giraffes exist all over the world.

14 Existential quantification is in fact also involved in the semantics of temporal modifiers: (i) John resided in Munich last year. Last year requires John to have resided in Munich at some point in the last year, not throughout the last year.

Sinn und Bedeutung 15 - [PDF Document] (47)

46 Moltmann

This is because such location modifiers are in fact quantificational, containing an explicit (all in (32b)) or implicit (thoughout) quantifier ranging over the parts of the location.

A distinction thus needs to be made between the location mentioned by the location modifier and the strict location, the location that is in fact where exactly the entity or event in question in located. The complete-presence condition of exist needs to be fulfilled only with respect to the parts of the strict location, not the location mentioned.

With an ordinary location modifier, an existence statement concerning a kind requires just that the kind be instantiated in an individual at some sublocation of the location mentioned by the modifier, the strict location. But would this not require that the kind be present at each sublocation of the location of a relevant instance of the kind? This is certainly not the case. The location of an instance should in fact count as a minimal location for the kind. The reason for that is this. Kinds inherit their location from the location of their instances; they cannot have a location in any other way. Thus they could not possibly be located at a proper part of the location of an instance of the kind.

As a consequence of the possibility of the space-relative existence of kinds, kinds can be multiply located. That is, exist may locate a kind at different locations, which amounts to the kind being ‘entirely’ present in instances of the kind at those locations. This matches well the Aristotelian view of universals as being able to be multiply located, located just where the instances are located.

The existence of a kind at a location obtains in virtue of an entity that completely instantiates the kind being at the location. Note that this in-virtue condition does not require that the instances of the kind ‘exist’ at the location in question. A requirement that the instances exist at the location could not be fulfilled by instances that are enduring objects and thus cannot exist at locations. Moreover, such a requirement could not be fulfilled by instances that are events, since events do not ‘exist’ in the first place. Rather for the existence of a kind at a location l, it suffices for an instance to simply be at l. The relation of being at a location is applicable both to enduring objects and to events.

Kinds generally have properties in virtue of their instances exhibiting particular conditions. So far the examples involved kinds as referents of bare mass nouns and plurals. In the linguistic semantic literature, it has been argued that kinds in this sense obtain certain of their properties in a particularly strict sense from their instances. Ever since Carlson (1977), it has become a common view to take the application of so-called stage-level and

Sinn und Bedeutung 15 - [PDF Document] (48)

Existence Predicates 47

individual-level predicates to kinds to be obtained from predicates applying to individuals on the basis of existential and generic quantification over instances of the kind. Individual-level predicates generally are taken to be predicates that are true of individuals throughout their lifespan. Such predicates generally exhibit a generic reading when applied to kinds, as in (33a):

(33) a. Firemen are intelligent.

Stage-level predicates are generally taken to be predicates that are true only of a ‘temporal stage’ of an individual. Such predicates generally exhibit an existential reading when applied to kinds, as in (33b):

(33) b. Firemen are available.

On the Carlsonian view, individual-level predicates are lifted to kind pre-dicates on the basis of generic quantification over individuals and stage-level predicates on the basis of existential quantification, as below:15

(34) a. For an individual-level predicate P, for a kind x, x ∈ [Pkind] iff Gn y [y I x; y ∈ [P]].

b. For a stage-level predicate P, for a kind x, x ∈ [Pkind] iff ∃y [y I x; y ∈ [P]].

Given the Carlsonian view, the question arises, should exist be classified as a stage-level predicate or as an individual-level predicate? Exist when applied to kinds of concrete objects clearly triggers an existential not a generic interpretation. But exist is not easily classified as a stage-level predicate. Exist necessarily holds of an entity throughout its life span. A characterization of stage-level predicates as predicates expressing accidental properties seems to do better. Concrete entities generally exist only accidentally, not essentially. Since this paper does not provide the space for an in-depth discussion of the stage-level/individual-level distinction as such, let us just note that exist, if anything, goes along with the class of predicates generally classified as ‘stage-level’, rather than those classified as ‘individual-level’.

Exist as a stage-level predicate applying to kinds could obviously not be obtained by the condition in (34b), by existential quantification over instances with the application of exist to particular instances. Location-relative exist as a kind predicate is not obtained from location-relative exist as a predicate applying to individuals. Rather location-relative exist involves 15 Gn is the generic quantifier, see Krifka et al. (1995).

Sinn und Bedeutung 15 - [PDF Document] (49)

48 Moltmann

existential quantification over instances fulfilling a ‘weaker’ condition than existence relative to the location in question.

Exist is not alone in that respect. The same also holds for certain other ‘stage-level’ predicates, such as recognize:

(35) John recognized gold (in virtue of coming across some instances).

In order for John to recognize gold, he must have ‘come across’ an instance, though ‘recognize’ would not apply to that instance. Recognize applies to a kind rather directly, requiring a weaker condition on an instance, that of having ‘come across’ an instance.16 To conclude, some stage-level predicates, including exist and recognize, need not apply themselves to an instances; rather only a weaker condition than that expressed by the predicate itself may be applied to an instance.

There is another reason not to trace the existential reading of exist with kinds to a Carlsonian account of stage-level predicates with bare mass nouns or plurals in general. The reading of exist involving existential quantification over instances is equally available for certain other kind terms than bare mass nouns and plurals, in particular demonstrative kind terms of the sort this flower or this animal:17

(36) a. This flower exists in many countries in Europe. b. This animal does not exist in this region anymore.

To summarize, space-relative exist can apply to kinds because of the particular nature of kinds, their ability to be completely present at different locations at once.

5 The Existence Predicate obtain We have seen that location-relative exist expresses the condition of complete presence at all the relevant sublocations. The same condition is in fact ex-pressed by location-relative obtain, even though obtain imposes different requirements on the entities it can apply to. Obtain, recall, is an existence

16 Another example of a stage-level predicate applying to a kind ‘directly’ is disappear, as in Geach’s (1968) example below: (i) Dinosaurs have disappeared. Here in fact no existential quantification is involved at all in the interpretation of the predicate. 17 Obviously demonstrative kind terms like this flower behave differently in that respect from definite kind terms of the sort the giraffe, see Fn. 10.

Sinn und Bedeutung 15 - [PDF Document] (50)

Existence Predicates 49

predicate that applies to entities such as states, situations, conditions, and facts. It does not apply to material objects, persons, or events:

(37) a. The state / situation / condition / fact obtains. b. *The house / The person / The event obtains.

The restriction imposed by obtain cannot be one to abstract objects: obtain does not apply, for example, to mathematical objects, properties, or pro-positions.

Obtain like exist is an abstract state verb. Moreover, it has a time-relative as well as a space-relative application, with entities such as states, situations, or conditions:

(38) a. The state / situation / condition still obtains. b. The state of emergency / The same situation / The same condition obtains in Arizona.

Time-relative and space-relative obtain does not apply to facts, though. The reason clearly is that facts are themselves already location-wise complete.

What characterizes the entities to which obtain is restricted is that they are constituted by certain conditions holding, that is, by certain properties or relations holding of an object or a number of objects. Let me call those entities condition-like entities and the conditions in question constitutive conditions. Some condition-like entities go along with canonical descriptions, that is, descriptions that display exhaustively the nature of the entity in question. Facts have canonical descriptions of the form the fact that S, states have canonical descriptions of the form the state of NPs being VP, and similarly conditions have standard descriptions of the sort the condition of NP’s being VP. It is the canonical description that explicitly displays the property or relation whose holding is constitutive of the fact, state, or condition. The states will thus be abstract states as described earlier. In fact, the four kinds of entities to which obtain applies are precisely the kinds of entities for which a Kimean account would be suitable (that is, the account Kim originally proposed for events discussed earlier).18 This account, recall, gives identity and existence conditions in terms of a property or relation holding of an object or a number of objects as well as possibly a location. (Of course the property of relation may itself be complex, involving connectives or quantifiers.)

18 This implies a Strawsonian view of facts, on which facts are not in the world, but abstractions from things going on in the world (Strawson 1950), rather than an Austinian view (Austin 1979).

Sinn und Bedeutung 15 - [PDF Document] (51)

50 Moltmann

Time-relative and space-relative obtain involves as its application condition a condition that like the application condition of location-relative exist, is the condition of complete presence at the relevant sublocations of the ‘strict location’. But in the case of obtain, the condition manifests itself in somewhat different ways. Obviously, for an entity e to ‘obtain’ at a time interval t, the constitutive conditions must hold of the objects in question at t, and in fact at all the moments of t. If not, the entity may not be a state, but rather a sort of event, involving a transition from one state to another, distinct state. Similarly, it is plausible that if obtain applies to an entity e relative to a strict spatial location l, this will require the fulfillment of the constitutive condition of e at relevant sublocations of l. Thus, the sublocation condition holds for location-relative obtain just as it did for location-relative exist.

It remains then the complete-presence condition. What could the complete presence of a situation, state or condition at a time or location consist in? That is, what could count as the parts of a condition-like entity? The objects and times from which condition-like entities are obtained (in the ‘Kimean’ way) certainly do not count as parts of such entities. This manifests itself in the fact that they are not treated as parts by part-related expressions of natural language: part of the situation, part of the condition, or part of the state can never ‘mean’ a participant or location of the situation, condition, or state. Furthermore, if the parts of condition-like entities include properties, it is hard to make sense of them being ‘present’ at a time. The only suitable candidates for involvement in the complete presence condition are in fact any constitutive sub-conditions. This corresponds well to the way the part-of construction is used in natural language: part of the condition, part of state, part of the situation can only make reference to constitutive subconditions. Thus, condition-like entities, unlike material objects, do not have spatial parts, and unlike events, they do not have temporal parts. Their only parts are constitutive subconditions. It is relative to them that complete presence at sublocations needs to be fulfilled when obtain applies. Thus for a situation, condition, or state e to obtain relative to a location means that all the constitutive subconditions of e are fulfilled at all the relevant sublocations of l. Because condition-like entities have neither temporal nor spatial parts they can be completely present, in the sense of complete obtaining, at different times as well as different places. In that sense, they ‘endure’ both throughout time and across space.

The closeness between obtain and exist manifests itself also in that exist can apply to all the entities to which obtain can apply. Thus, (39a) and (39b) (with location-relative exist) are fairly acceptable, unlike sentences with exist applying to events:

Sinn und Bedeutung 15 - [PDF Document] (52)

Existence Predicates 51

(39) a. The state / The situation / The condition / The fact exists. b. The same state / situation / condition still exists in some countries.

Location-relative obtain thus shares the condition of complete presence at the relevant sublocations with location-relative exist. The way obtain differs from exist resides in its sortal restriction to condition-like entities as well as the particular notion of ‘presence’ it involves: exist requires presence in the sense of spatial or temporal location, whereas obtain requires presence in the sense of a property being true of an object relative to a location. (What makes exist in (39a) and (39b) acceptable is obviously that the presence condition imposed by exist is extended so as to cover the ‘holding-at-a-location’ condition as well.)

Condition-like entities raise the question on what grounds their constitutive conditions hold, be it at a time, at a place, or absolutely. There are in fact two fundamentally different kinds of condition-like entities: those based on empirical facts (about the time or spatial location, or the world), and those based on normative conditions or conditions resulting from ‘declarations’ (which may or may not be restricted to a time or a spatial location). The state of someone’s mind or health is a state of the first kind, as are habits; a state of war, requirements etc are condition-like entities of the second kind. The first kind of state holds in virtue of what is taking place at the relevant location; the second kind of state holds by declaration or whatever may ground normative conditions. Thus for a condition-like entity to obtain at a location, either all the various things need not happen at the location in virtue of which the condition can be said to obtain or else the relevant condition, with all its subconditions needs to have been put into place for that location. Either way, the condition-like entity will need to enjoy endurance throughout the location as long as the constitutive condition holds. Note that both obtain and exist are applicable to normative condition-like entities, including laws:

(40) a. The law still obtains / exists in some countries. b. The requirement for a president still exists / obtains in many


6 Conclusion Existence predicates in English, I have argued, form a clearly characterized semantic class of predicates. In their location-relative use, exist and obtain have a nontrivial meaning, specifying the complete presence of an entity at the relevant sublocations. Occur, by contrast, tracks the temporal locations of

Sinn und Bedeutung 15 - [PDF Document] (53)

52 Moltmann

subevents of an event, describing an event that reflects the mere temporal structure of the occurring event.

There is one remaining question, and that is: what is the meaning of location-independent uses of existence predicates? It is important to note that location-independent uses of existence predicates still impose the same sorts of restrictions on the kinds of entities they can apply to. Location-independent exist like location-relative exist is inapplicable to events. Location-independent exist is also applicable to mathematical and other abstract objects, unlike location-dependent exist. Location-independent obtain, unlike location-dependent obtain, is applicable to facts. But it is still inapplicable to entities other than condition-like objects. The preservation of the sortal restrictions is an indication that the location-independent meaning of existence predicates is derivative upon their location-relative meaning.

How can the location-independent meaning then be derived from the location-dependent one? Time-independent exist applying to abstract objects may be obtained from time-relative exist by universal quantification over times. That is, exist is true atemporally of an object o if it is true of o at all times (which means o is completely present at all times). Exist with that meaning could not apply to events: it would require the complete presence of an event at all times, which is impossible.

Can the time-independent use of obtain can be derived from the time-relative use in that way as well, namely by universal quantification over times? In the case of obtain, this should mean complete fulfillment at all times of the constitutive conditions of the condition-like entity. Facts that are constituted by the holding of a property of objects at a particular time could not fulfill this condition, though. Thus, the time-independent use of obtain must be derived differently for that case. It is plausible that condition-like entities that are complete regarding the parameters of the holding of the constitutive condition are completely present at any time (and any space). Time-independent obtain still presupposes that the entity it applies to be condition-like and that is because it is derived from location-relative obtain, which which specifically relates to the constitutive conditions of the entities it applies to.

Can space-independent exist be derived from space-relative exist? This would require complete presence everywhere of entities that exist space-independently. Complete presence everywhere is of course impossible for

Sinn und Bedeutung 15 - [PDF Document] (54)

Existence Predicates 53

material objects (which are extended in space). Perhaps space-relative exist is in fact itself derived from time-relative exist by form of analogy.19

Overall, we have seen that tied to the notion of existence is fun-damentally that of complete presence throughout a location. This condition is inapplicable to some entities, such as events. It is in this sense that events might be said to have ‘being’ in a weaker sense. What is constitutive of objecthood is recurrence of the essential parts or features of an entity across locations, which is what the more fundamental notion of existence amounts to.

It has sometimes been argued that our linguistic intuitions about the verb exist should not be taken too seriously, for making either a semantic or a philosophical point, since exist is a relatively recent verb and tied to a more ‘technical’ use in philosophical contexts.20 I think this caution is unjustified. We have seen that our intuitions about the verb exist are in fact very systematic and allow for a natural explanation within a fairly well-established ontological view. Furthermore exist is not alone in displaying the relevant semantic behavior. In English, the existence predicate obtain behaves strikingly parallel to exist. The linguistic intuitions associated with exist thus are better viewed as displaying an important underlying concept of location-relative existence rather than peculiar features of a somewhat special lexical item.

References Austin, J. L. 1979. ‘Unfair to Facts’. Philosophical Papers. Oxford

University Press, New York. Carlson, G. 1977. ‘A Unified Analysis of the English Bare Plural’.

Linguistics and Philosophy, 413–457. Carlson, G. / J. Pelletier 1995. The Generic Book. Chicago UP, Chicago. Chierchia, G. 1998. ‘Reference to Kinds across Languages’. Natural

Language Semantics 6, 339–405. Fine, K. 2006. ‘In Defense of Three-Dimensionalism’. Journal of Philosophy,

699–714. Frege, G. Grundlagen der Arithmetik. Olms, Hildesheim. Geach, P. 1968. ‘What Actually Exists’. Proceedings of the Aristotelian

Society, Supp. Vol. 42, 7–16. Hawley, K. 2001. How Things Persist. Oxford UP, Oxford.

19 Other uses of exist applying to material objects may involve implicit existential quantification, as part of the interpretation of tense. 20 See in particular Simons (2006).

Sinn und Bedeutung 15 - [PDF Document] (55)

54 Moltmann

van Inwagen, P. 1998. 'Meta-ontology'. Erkenntnis 48, 233–250. Katz, G. 2003. ‘Event Arguments, Adverb Selection, and the Stative Adverb

Gap’. In E. Lang (eds.): Modifying Adjuncts, de Gruyter. Krifka, M. et al. 1995. ‘Genericity: An Introduction‘. In Carlson / Pelletier

(eds.). Lewis, D.K. 1986. On the Plurality of Worlds. Blackwell, Oxford. McGinn, C. 2000. Logical Properties. Oxford UP, Oxford. Maienborn, C. 2007. ‘On Davidsonian and Kimian States’. In I. Comorovski /

K. von Heusinger (eds.): Existence: Semantics and Syntax. Springer, 107–130.

McDaniel, K. to appear a. ‘A Return to the Analogy of Being’. Philosophy and Phenomenological Research.

McDaniel, K. to appear b. ‘Being and Almost Nothingness’. Nous. Miller, B. 1975. ‘In Defense of the Predicate Exist’. Mind 84, 338–354. Miller, B. 1986. ‘“Exists” and Existence’. The Review of Metaphysics 40,

237–270. Miller, B. 2002. ‘Existence’. Stanford Encyclopedia of Philosophy. Parsons, T. 1980. Nonexistent Objects. Yale UP, New Haven. Priest, G. 2005. Towards Nonbeing. Oxford UP, Oxford. Russell, B. 1905. ‘On Denoting’. Reprinted in R.C. Marsh (ed.): Logic and

Knowledge. Allen & Unwin, London. Salmon, N. 1987. ‘Existence’. Philosophical Perspectives 1, 49–108. Salmon, N. 1998. ‘Nonexistence’. Nous 32.3., 277–319. Sider, T. 2001. Four-Dimensionalism. Clarendon Press, Oxford. Simons, P. 2006. ‘Modes of Extension: Comments on Kit Fine’s ‘In Defence

of Three-Dimensionalism’. The Royal Institute of Philosophy 83, p. 17–20.

Strawson, P. 1950. ‘Truth’, Proceedings of the Aristotelian Society, reprinted in Strawson 1971, Logico-Linguistic Papers, Methuen, London, 1971.

Sinn und Bedeutung 15 - [PDF Document] (56)

In Reich, Ingo et al. (eds.), Proceedings of Sinn & Bedeutung 15,pp. 55–77. Universaar – Saarland University Press: Saarbrücken, Germany, 2011.

How the Emergence of Propositions Separates Strict Interfacesfrom General Inference∗

Tom RoeperUniversity of Massachusetts


Abstract. How does a child utilize inferences in acquisition and ultimately sepa-rate inferences from implicatures from semantic composition? How exactly doesa child recognize a truth-functional proposition? It is argued that the child beginswith rich, free inferences that are systematically replaced by syntax/semanticscompositional rules. Strict syntax/semantics interfaces are discussed and linkedto the syntax of subject auxiliary inversion and acquisition evidence of copying(Is Bill is busy?) and long-distance movement in opaque contexts (What did shesay she bought?). A connection between Tense, propositions, and Phase bound-aries is argued to be a critical syntax/semantics interface.

1 Acquisition Goals: the Syntax/Semantics InterfaceWhile acquisition has been conceived of largely in syntactic terms, it is intu-itively obvious that semantics and pragmatics motivates a child to unlock mys-terious adult utterances, not just their unusual syntactic character. But what,actually, are the steps a child takes if syntax, semantics, and pragmatics con-verge upon her? Where does she begin and how can she simplify the process?The original notion of the autonomy of syntax offered an illusory image that thechild could see through every linguistic situation to see the syntactic skeletonwithin. It cannot be so simple.

Our account will be deliberatively intuitive, dwelling upon imprecise no-tions of propositionality and presupposition, but it parallels in spirit the ideaof compositionality: each syntactic step has a semantic and pragmatic conse-quence. The hope is that an intuitive discussion helps us see just where morerigorous formalization is appropriate and where it may not be.

∗ Acknowledgements: Thanks to the Sinn and Bedeutung audience and audiences at IASC: in Ed-inburgh, Assymmetry conference in Montreal, Information Structure conference at Potsdam, andthe Ex-FA conference in Campinas, Harvard, and UMass where parts of this talk have been pre-sented and to Cedric Boeckx, Seth Cable, Noam Chomsky, Jill deVilliers, Kyle Johnson, WolframHinzen, Bart Hollebrandse, and Ian Roberts for helpful remarks. Errors are mine, not theirs.

Sinn und Bedeutung 15 - [PDF Document] (57)

56 Roeper

1.1 LearnabilityIf an interaction with semantics and pragmatics is assumed to be relevant, thenin classic terms, the set of possible grammars should go up exponentially andthe learnability problem is then dramatically increased. The child must identifythe optimal grammar not only in terms of syntactic derivation, but in terms ofthe set of meanings she map onto it as well. If structure X defines a set ofpossible meanings within the grammar, then each member of the set must beevaluated for presence in a particular grammar, not just the syntactic structure.It should be obvious that we must reverse that logic: the presence of semanticand pragmatic factors should serve to constrain the set of possible grammars,not expand them. How could that work? What prevents a child from makingarticles recursive: the the hat. One answer could be that syntax does not preventthis possibility, but that no semantics can be attached to a recursive articlethat would make sense: an object cannot be specific in two ways. In fact, apragmatic meaning is not that far off: the the hat I like could mean the hat ofthe hats that I like. So we would want the exact semantics of the syntax – thesemantics of recursion in fact – to be incompatible with this option.

If one takes the literature in formal semantics as a starting point, it is farfrom obvious how to make it work for the child. First we must see that there isa significant division between the inherent logic of cognition, and the speciallogic of a syntactic semantics. The difference is important if we wish to see ex-actly which acquisition hooks are actually used by the child. Formal semanticshas not, to my knowledge, explicitly sought to articulate such a distinction.1

1.2 Interface HypothesisLet us begin then, with an interface hypothesis which serves as a motivatingdesideratum:

(1) Strong Interface Hypothesis: Every step in acquisition must satisfysyntactic, semantic, and pragmatic criteria.2

An acquisition pragmatic principle is a natural consequence:

(2) Pragmatic Principle: Connect to Context as quickly as possible.

From an acquisition perspective, we need an hypothesis of this kind to explain,for instance, the fact that small children can project and interpret single words.

1 This article is written from an acquisition perspective and there well may be appropriate literatureof which I am unaware.2 The hypothesis is surely too strong because there may be morphological or movement (scram-bling) operations which fail to constitute a shift for each criterion.

Sinn und Bedeutung 15 - [PDF Document] (58)

Acquisition of Propositions 57

Let us elaborate a simple example to see how pragmatics and semantics canparticipate in syntactic triggering. In an early approach to the interface ques-tion, Lebeaux (2000) argued that semantics served to confirm syntactic hy-potheses which are derived from context. How does a child acquire the passive?If he can independently determine meaning from context, and then generate asyntax to match the meaning, then semantics/pragmatics confirms the syntax(as in Roeper (1982)). Suppose the child hears:

(3) The cheese was eaten by the mouse.

If the child knows semantically and pragmatically that it must be the mouse whoate the cheese, therefore cheese must be restored to a position after eat wherethat meaning is available. If UG syntax can find an operation – reverse objectto subject movement – which restores cheese to the object position, then themeaning will be consistent with common sense. Once acquired, but only then,the syntax can become autonomous and can generate a meaning inconsistentwith pragmatics (plausible real-world knowledge) such as:

(4) The mouse was eaten by the cheese.

The emergence of propositional representations can, as we shall see, be repre-sented from this interface perspective. In broad terms, children first:

(5) Represent propositions as events.

In general, the emergence of propositional interpretations can be profoundlyclouded by general inference capacities. For adults, strictly speaking, the in-ference is not warranted but often occurs. Ultimately, the child must separatelogical implicatures from insecure situational inferences. That is, she must cre-ate an autonomous semantic space – constrained by syntax – which blocks theinappropriate use of general inference.

(6) Move from rich inference to constrained compositional readings.

The differences between inference, implicature, and syntactic compositionalsemantics are part of the acquisition challenge. The immediate relevance ofinference is obvious to any parent whose child’s first word is “no”. It is oftennot clear exactly to what the “no” applies. When a child resists socks, does“no” mean:

(7) Don’t put any socks on me or don’t put those socks on me.

Sinn und Bedeutung 15 - [PDF Document] (59)

58 Roeper

And the child is confronted with the same conundrum when the parent says“no!”. We assume a situational salience of something in a “Common Ground”,but it is not only unwarranted, but the failure of inference is a reasonable moti-vation for a child to acquire an articulated grammar.

2 How Do Propositions Emerge?Propositions, as well as logical inferences and systematic implicatures, havea long history in philosophy and they are the subject of many definitions inPropositional Logic. At the same time, there are those (Hinzen 2007; Chomsky2007) who have suggested that the notion proposition needs to be delicatelyintroduced into linguistics and may not have a single logical definition. Ourdiscussion will follow this tradition. Nonetheless one goal must ultimately beto say how a child arrives at a notion of proposition that is useful in traditionalsyllogistic reasoning:

(8) John is a man. All men are mortal.⇒ Therefore John is mortal.

All indications are that this kind of explicit reasoning is far from nursery schoolchildren. At the same time, many implicit logical relations are undoubtedlypresent and a prerequisite for every stage of acquisition.We should note that our approach is orthogonal to formal semantics. Many ofthe rich and detailed distinctions found in formal semantics are natural fod-der for experimentation in the future. We aim to look at the “moving parts”in acquisition. This may leave the reader unsatisfied because we will avoid arigorous treatment which may be found in philosophical discussions. Many in-teresting questions arise, for instance, whether and when children grasp upwardand downward entailments for which there are ongoing experiments.Nonetheless, the most primitive first stage kinds of initial entailments in lan-guage seem to refer to how thinking of any kind operates. What belongs topresuppositions or entailments about thought in general, and what calls for aspecial mapping onto linguistic structure? To trace the acquisition path wisely,we need to draw this line correctly. Our account builds from the point wherea proposition is a syntactic projection that is distinct from a semantic prim-itive. Ultimately, it must all be embedded in a realistic version of children’spragmatic experience.

2.1 Syntactic Propositions and DeniabilityWhere should we begin to look for “moving parts”? One prominent feature ofpropositions is the notion of deniability which happens to be elicited preciselyby syntactic movement, inversion:

Sinn und Bedeutung 15 - [PDF Document] (60)

Acquisition of Propositions 59

(9) John is playing baseball⇒ is John playing baseball

The movement of the Tensed element cancels the property of assertion and au-tomatically introduces deniability via a yes/no question construction. Thereforeour definition of propositions is:

(10) Propositions are deniable: True or False.

We argue, consistent with classic syntactic claims, that this property resultsfrom a strict interface between syntax and semantics which pivots upon thisUG claim (Klein 2006; Schulz 2003):

(11) Universal Grammar: Tense projects a proposition.

This has led to concepts like “Propositional-Island Constraint” for tensed clau-ses. A child should, if UG is innate, grasp this notion of proposition, not im-mediately, but very early, as a UG reflex of the acquisition of Tense.Yes/no questions via inversion give overt manifestation of this:

(12) John is singing⇒ is John singing

Most of the meaning of yes/no questions, we claim, emerges automatically viaa Strict Interface, and therefore predictably, emerges before other propertiesof propositions, clouded by inference, but after entailments that are direct re-flections of cognitive organization. First we look at the child’s semantic andpragmatic environment and the kinds of propositional distinctions, hanging ina world of inferences, that a child must ultimately master.

2.2 Non-Assertion PropositionsOther forms of meaning that are stateable as propositions include:

(13) Entailments Implicatures Presuppositions

Each of these properties is close to situational pragmatic reasoning that doesnot depend upon language (at least in my estimation). Entailments are not sub-ject to deniablility (unless elevated to an overt assertion) – and they are prereq-uisite to the composition of any kind of meaning, linguistic or non-linguistic.Therefore I claim, that propositions are obscure, but in some measure they arepart of any form of knowledge, as when we presuppose the existence of anobject when we speak of it.

(14) Hypothesis: The capacity for “proposition” is innate and immediate,but not necessarily available to conscious deduction.

Sinn und Bedeutung 15 - [PDF Document] (61)

60 Roeper

Presuppositions and their triggers are notoriously difficult to pin down, but theyemerge as intuitively critical in what follows.3

2.3 EntailmentsSuch a notion of proposition is present in much of our thought, but not au-tomatically subject to any other linguistic operation like: denial, contrast, ordeductive consequence.For example if:

(15) John has a blue car.

It entails the proposition: John has a car. A sentence like:

(16) John likes the car.

has the existential presupposition that:

(17) A car exists.

But we do not expect that children – nor probably animals – lack this mentalcapacity. It underlies not only language but many actions. It is plausible to ar-gue that more sophisticated structures partake of this connection. For instance,understanding:

(18) John was sad that the Bruins lost.

involves sad [clause] where the truth of the complement is assumed (“Speaker-factivity”). It could, therefore, share a presuppositional basis with blue car.We predict therefore that adjective complements should be easier for cognitivereasons than attitude verbs that create opaque contexts. Thus:

(19) a. John was unaware that the Bruins lost.b. John did not think that the Bruins lost.

(19a, b) might seem pragmatically close in content, but the grammar deliversthe meaning of the complement in different ways, where (a) has speaker orCommon Ground factivity, but (b) is marked without a fixed truth value. Legere(2008) has shown that this is true in acquisition. With sentences roughly like ina carefully controlled experiment, 92% of 5yr olds understood (20a), but only62% understood (20b)

(20) a. She is happy she has a hat.

3 See Beaver & Geurts (2010) for an overview.

Sinn und Bedeutung 15 - [PDF Document] (62)

Acquisition of Propositions 61

b. She knows she has a hat.

This kind of data is consistent with diverse contexts under which presupposi-tions are triggered as the literature attests.

2.4 ImplicaturesLinguistically-based implicatures involve derived propositions usually by con-trast.

(21) a. John has some of the marbles.b. ⇒ John does not have all of the marbles.

Linguistically-based implicatures involve derived propositions usually by con-trast.

(22) a. John has some of the marbles.b. ⇒ John does not have all of the marbles.

Nevertheless implicatures, at least some of those relating to quantification, ap-pear to involve an extra semantic operation that takes more time and showsup later in acquisition (see work by Snedeker and colleagues (e.g. Panizza,Chierchia, Huang & Snedeker (in press) for recent discussion)). They occupya different propositional corner.

2.5 Entwined InferencesWhat stands in contrast to all of these forms of logical reasoning are rather sim-ilar situational inferences, which form a backdrop to all language and mingle insubtle ways with grammatical properties. Such inferences are ever-present butless constrained and not subject to logical verification. For instance, for (22),while some⇒ not all is fixed, I might infer that someone other than John has afew marbles. But such an inference is insecure: some of the marbles could beunder the couch, not in someone else’s possession. In general we make whatshould be a virtually self-evident claim:

(23) Acquisition shifts, systematically, from an over-reliance upon infer-ence to a reliance upon systematic semantics.

This view leads to the natural question: What is the system whereby the shiftoccurs: Does a child know when a situational interpretation corresponds tothe grammar or when it is an inference that exceeds the meaning grammarcarries? This is a significant acquisition problem whose solution will engagemany dimensions of linguistic theory.

Sinn und Bedeutung 15 - [PDF Document] (63)

62 Roeper

2.6 Inferences and GrammarBefore we proceed, we would like to demonstrate how intimately woven un-substantiated inferences can compromise systematic semantics in the eyes of achild. They are immediately available to children and confound the acquisitionpath. One can easily conclude the truth of both (24b, c) from the observationof (24a).

(24) a. Situation: John observes Bill drinking beer.b. John saw Bill drink beer. = Eventc. John saw that Bill drinks beer. = Proposition

The event seems to entail the proposition and the proposition seems to entail theevent, although neither conclusion withstands careful scrutiny (Higginbotham1983). But it is a part of how we commonly understand things, and court caseswill show it. Imagine this scene:

(25) Mary said “Bill saw John drink beer.”Now if someone were asked in court:“Did Mary say that John drank beer?”Most people would in fact say “yes”.

But linguists know that if you saw John drinking, but did not know what it was,someone might say: “You saw John drink beer – you just did not realize it wasbeer.” And if you only saw empty bottles afterwards, you might say: “you sawthat John drank beer” without entailing that you saw him drink beer.Some constructions are ambiguous:

(26) She saw the boys drink beer. = Event perceptionor Proposition = She saw (that) the boys drink beer.

If a child at first depends upon these inferences in the initial state (as do adultsunderstanding children), they can be legitimately confused as to whether theconstruction they heard was the representation of an event in a small clause ora proposition.This reasoning extends to False Belief environments that surround opaque ut-terances as well. We commonly allow inferences to overrule opacity in ourunderstanding as well. Consider this scene:

(27) Grandma asks Dad: “What did Mom buy at the store?”and he answers: “Plastic toys.”“How do you know?”Dad: “I heard a friend talking to Mom on a cell phone at the store and

Sinn und Bedeutung 15 - [PDF Document] (64)

Acquisition of Propositions 63

I asked the friend: “What did Mom say she bought at the store?” andthe friend said she had said “Plastic toys.”

Although under two recursive verbs of “saying” the complement is opaque andtechnically Dad does not necessarily know what she bought, this inference iscommonly accepted in normal communication as a basis of “knowing”. There-fore it is something a child easily hears and must, ultimately, learn to qualifyor discount in his construction of the sentence:

(28) What did Mom say that she bought __

Acquisition experiments aim very precisely at creating situations where adultswill apply a notion of opacity to the complement of say, but in real life we areoften entitled to make this inference.Suppose again this were a court case, one could imagine a defense lawyerseizing upon the distinction and asking a witness, who is under oath:

(29) A: “Do you know what she bought at the tobacco store?”B: “Yes, I know she bought drug paraphernalia.”A: “How do you know?”B: “She told me she bought drug paraphernalia.”A: “Has she ever lied to you?”B: “Yes.”A: “Then she could have been lying?”B: “She could have.”A: “Then in fact you do not know that she bought drug

paraphernalia, because you never saw it and shemight have been lying.”

In a court, reasoning of this kind, pivoting on the illegitimacy of common in-ferences is a standard technique to downgrade the value of testimony. Here isa putative adult caught in the fact that his statements, under oath, dependedupon an inference and not actual knowledge. What he actually knew was thatshe said she bought drug paraphernalia but this was allowed to shift to: Knowthat she bought drug paraphernalia. No one would be indicted for perjury forsuch an assertion: He said he knew what she bought, but he didn’t, thereforehe lied. Moreover, the information would not be disallowed as evidence. Thefact that she said she bought drug paraphernalia could be legitimately relevantto the judgement, in a juror’s mind, that she did buy drug paraphenalia.If adults are easily uncertain about the boundaries of semantic meaning, then itmust be a challenge for a child as well because his experience has inconsistent

Sinn und Bedeutung 15 - [PDF Document] (65)

64 Roeper

information. Sometimes “saying that” can be relied upon to introduce a truth,and sometimes not. It is no wonder that children must learn to disentangle themeaning from the inferences when such inferences are a part of normal life.There is subtle naturalistic evidence that children may take the verb say to carrya factive complement. In the adult language factive complements for verbs likeknow are deletable (b) as opposed to non-factive verbs like think (c) with whichwe generate the opaque complement via verbphrase-prominalization with theword so (d):

(30) a. A: John went outside.b. B: I know.c. B: *I think.d. B: I think so.

Children are known to do the same in conversation but with the non-factive saybeing treated like a factive, although adults do not do this:

(31) “Can we have cookies?”“Yeah, Mom said.”

In fact, the Childes database has examples with exclamations that feel non-adult, as if a factive complement has been deleted, which suggests that a childmight first understand the word that way.

(32) “You said!”

We turn now to an examination of acquisition data. Much of this data can beregarded as experimental hints – in a way like “??” grammatical judgments –that need to be subject to more careful scrutiny, as do many grammaticalityjudgments, but which nonetheless carve out the terrain that needs exploration.

3 The Emergence of Tense and PropositionalityWhere does a child begin? One-word utterances like

(33) “uh-oh”, “no”, “dat”, “fish”, “milk”, “juice”

have the same variety of interpretations, expanded by intricate inferences thatthey do for adults who say:

(34) “John!”, “No way!”, “Beer!”

and leave it to speaker/hearer inferences to fill in the communication. Parentand child must both exercise inferential capacities, which often run awry, pre-

Sinn und Bedeutung 15 - [PDF Document] (66)

Acquisition of Propositions 65

sumably providing the child with motivation to acquire more grammar in orderto be clear. Nonetheless, the efficiency of such inferences must be present earlyon to make the utility of one-word utterances a good starting place for children.The child who says “milk” is, often but not always, trying to express an imper-ative like “please give me some milk”. Early work in acquisition claimed thatchildren had full sentences here, but there is no more reason to believe that thanthe notion that adult exclamatives should be expanded as full sentences.Potts & Roeper (2005) argue that the same analysis holds for two-word utter-ances, which are like expressives for adults:

(35) Exclamations: You fool! You idiot! You jerk!

These utterances are not equivalent to a tensed sentence:

(36) You are an idiot.

The latter has a profoundly different force, just as a parent who screams at achild “You idiot!” is not dealing with the same meaning as a psychologist whosolemnly pronounces:

(37) Your son is an idiot.

Where the tensed element implies reference to evidence like intelligence testscores.Expressive exclamations cannot be cancelled or tagged or take articles:

(38) *You idiot, but it does not look that way.*You idiot, aren’t you.?*You an idiot.

3.1 The Pre-Propositional StageA child utterances often seem to carry exclamative force:

(39) “it big” or “dat here” or “Mommy sock”

They are therefore plausibly using the syntax of expressives, though his com-municative intent might be closer to an assertion, the notion that it is an asser-tion is then an inference.Therefore we argue that they are pre-propositional. They show no indicationsof a proposition, like tense-marking, tag-questions or even proto-tag questions(as far as we know):

(40) *It big, isn’t it?

Sinn und Bedeutung 15 - [PDF Document] (67)

66 Roeper

*It big, huh?

What drives us – in particular children – to abandon these rich inferences infavor of grammar? We are left with many uncertainties in a world of pure in-determinate inference. We seek the precision and reliability of assertions thathave an evidential relation to context.

3.2 The Semantic Side of Root InfinitivesWhat exactly happens when the child moves from (a) to (b):

(41) a. he big ⇒ b. he is big

He is taking both a big semantic and a big syntactic step. There is evidencevery early that precisely its deniable propositional character is quickly recog-nized, when a child uses Verum Focus in a context where someone claims thatsomething is not working:

(42) “ it IS working!” (Danilo Azcarate 3,4yrs)

Thus we have early evidence that a putatively UG-mediated and innate con-nection falls into place easily and quickly, much like children once standing upquickly learn to walk.A large literature on Root Infinitives argues specifically that children initiallylack Tense (see Wexler (to app.) for a summary). What the literature fails toaddress is the fact that propositions are carried by tense. Therefore until thechild can project a proposition – which we define as carrying deniability – thesemantics as well as the syntax can be inhibiting the move from expressionslike:

(43) a. Hände waschen [hands to wash]b. wäscht die Hände [washes the hands]

The child who moves from (43a) to (43b), not only raises the verb and addsa tense-marking, but adds a deniable proposition as well. It is notable that animportant form of language disorder is the failure to express tense. We takethat to be not only a syntactic, but a semantic failure, and the step toward tenserecognition to engage the strict interface between syntax and semantics.A close look at a child’s departures from the adult language indicate that thenotion of proposition is psychologically present.4 Why do we ask a yes/no

4 We have undertaken no analysis of languages that lack Tense marking. It is perfectly possible thatUG has a few innate avenues where propositionality is linked to syntactic markers or constructions.

Sinn und Bedeutung 15 - [PDF Document] (68)

Acquisition of Propositions 67

question with inversion?5 Why don’t we just add a word that carries that mean-ing, like whether:

(44) a. Is John here?b. *Whether John is here?6

It is really a strange fact that we can signal that a whole proposition is question-able by just moving an auxiliary. We argue that something deeper is at work:an operation on the notion proposition.We argue that the interpretation arises from what we call a syntax/semanticsStrict Interface. An application of this principle leads to Chomsky’s StrongMinimalist Thesis:

(45) Strong Minimalist Thesis (SMT):Each syntactic Phase undergoes semantic interpretation.

(Chomsky 2005, 2008)

It now becomes plausible that if something is removed from the Phase – liketense-marking – then it cannot be interpreted, which I call Vacate Phase. Thus ifa clause is a Phase, then moving something out of it, blocks an interpretation.Movement is motivated to create new discourse meaning, like questions, butalso to avoid meaning.Consider the contrast between (46a) and (46b), two questions around the sameproposition:

(46) a. Can you t play baseball?b. You can play baseball, can’t you?

In (46b) the proposition is advanced and then the tag asks for confirmationor disconfirmation. In (46a) the proposition is never assumed, although recon-struction allows the system to know what the proposition would be.In other words, by movement out of the Phase, the pragmatic projection toa presupposition of a question, as in (b), does not occur. The trace of Aux-movement allows reconstruction of thematic meaning, but the proposition car-ried by tense is not projected if the tense is not pronounced in the Phase.This requires technical expression that differentiates Phase Head and Phase-

5 See Roeper (2011) and deVilliers & Roeper (to appear) where a detailed minimalist syntacticaccount is provided.6 Note that it can be said in German (M. Zimmerman pc),i. Ob er da ist [whether he here is]but with some presupposition difference, as if the proposition is in the Common Ground, which ishard to pin down.

Sinn und Bedeutung 15 - [PDF Document] (69)

68 Roeper

complement, limited reconstruction, and a relation between pronunciation andpresupposition which we will not articulate here, but which Strict Interfacesallow (see Roeper (2011)):

(47)[CP can [IP you trace play baseball]Phase-Head Phase-complement

Only the Phase Complement is initially transferred to interpretation (Boeckx2008). UG and a Strict Interface delivers, by movement out of the complement,a block on the propositional presupposition. By hypothesis:

(48) Strict Interfaces are a part of UG and therefore immediately availableto a child.

Now the odd fact that inversion undoes a presupposition follows directly fromStrict Interfaces and the seemingly natural whether-question option is automat-ically rendered superfluous.If indeed, this is UG-derived, hence innate, then it should be available for chil-dren very early. In fact, Van Valin (2002), advocating a similar view of the roleof Tense, shows that precisely overtly tensed elements invert first: “Is, are, was,do, does, did, have, has and had occur in inverted questions at the initial stage67% of the time, while can, could, may, might, shall, shoul, will occur only14%.”Moreover, further operations become quickly possible. As mentioned, young3yr olds say things like:

(49) “it IS working” (Danilo Azcarate 3.4 years)

showing an awareness of Verum Focus, requiring Contrastive stress on thetensed element, presupposing the Tense-Proposition connection. In fact, wehave explicit dialogues that reveal the presence of the presupposition when, viacopying, the tense continues to be present in the clause for children:

(50) Father: Do you want to go outside?Child: No.Child (to friend): Do you don’t want to go outside ?

The copying, by maintaining the auxiliary inside the Phase, preserves the pre-supposition and asks a question about it, while the inverted case not only lacksthe presupposition, but acquires a suggestion reading:

(51) Don’t you want to go outside?

Sinn und Bedeutung 15 - [PDF Document] (70)

Acquisition of Propositions 69

Other examples include:

(52) a. “ Where are we ARE?” (Danilo Azcarate 3.8 yrs)b. = Where is it that we are?

which is uttered in a strange forest, with the meaning like a cleft (52) wherethe fact that we are somewhere is presupposed.Now let us ask more carefully what the copying implies. It has been noted anddebated for 40 years in the acquisition.7 It occurs for a brief period of time,before tag-questions are available (Jesney 2007):8

(53) “Is Bill is busy” “Can you can do that”“Is it’s Stan’s radio” “Is this is the powder”“Is that’s a belt” “why do you’re going outside”“why do you’re giving juice” “why do you’re cutting the meat”“what’s he’s doing” “what’s the mouse is doing”“why is there’s big tears” “what is the woman is doing”“why do deze don’t unrase” “why did you didn’t want to go”

Under our hypothesis, the yes/no questions should have the meanings of tag-questions, not open yes/no questions, although the claim is hard to prove, andthe wh-questions are like clefts with presupposed relatives (why is it you’recutting meat).Under the Strict Interface, syntax and semantics converge. One reason that achild would use copying is precisely to preserve the presupposition. A secondreason is that copying transformations can be first learned at phonetic opera-tions with Total Reconstruction, thus the interpretation would involve treatingboth the copy and the trace as if it were the full tensed element.This notion of Total Reconstruction has been articulated by Sauerland & El-bourne (2002) and Miyagawa (2005). If the child first does Total Reconstruc-tion, then we predict that even non-copied inverted SAI questions initially arelike tag questions. Here is a pilot experiment that goes in that direction. Con-sider the contrast between:

(54) a. anyone can lift a ball→ exhaustive = everyoneb. can anyone lift a ball→ Qp-neg→ anyone = Free Choice

(or exhaustive reading)

7 See Guasti (2006); deVilliers, deVilliers & Roeper (2010) for some references.8 Fitzpatrick (2005) shows inversion changes presupposition in wh-questions too. Note the con-trast: why don’t you go outside (no required presupposition) and how come you don’t go outside(necessary presupposition). Conroy (2006) shows that variation in inversion in wh-questions inchildren patterns with this distinction.

Sinn und Bedeutung 15 - [PDF Document] (71)

70 Roeper

This Scene was given to a child and the child was asked:

(55) [insect, dog, fish, child]Can anybody play with a ball?→ adult: yes, the child or dog (yes-bias chooses)

An adult if asked, would give “no, not a insect or a fish” to a tag-question:

(56) Anybody can play with a ball, is that right?

If the child has a yes-bias (and because anybody has a person or personlikebias), we would expect “yes” for “Can anybody play with a ball” instead the4yr old child gave a tag-like response:

(57) “no, not the fish or the insect”

Even for an adult, the copied version seems to prefer this Free Choice, notuniversal reading :

(58) → Can anybody can play with a ball?

To sum up, the experiment supports both the notion of Total Reconstructionand the presupposed proposition reading.What can we conclude from the early evidence:

(59) a. early use of inversionb. verum focusc. copying to create tag-questionsd. disorder linked to Tense

We take this to reflect a mechanical and strict interface between syntax andsemantics.

What is astonishing and critical is that this evolution happens very earlywith children in the 2-3 yr range, while the block of a presupposition underattitude verbs like say, think, tell which are not inherently more complex, is notworked out until several years later, to which we now turn.We argue below that the presence of easy pragmatic inferences that overrulewhat strict syntactic/semantic connections require, like those courtroom con-fusions outlined above, are an important part of why some realizations of StrictInterfaces seem not strict to children and delay acquisition.

Sinn und Bedeutung 15 - [PDF Document] (72)

Acquisition of Propositions 71

4 The Inference toward Default TenselessnessLater stages, nonetheless, show an application of inferences to what may be adefault Tenseless repesentation. Children show:

(60) Event interpreted as proposition.Proposition interpreted as event.

In a pilot experiment conducted by Mary Wilson and associates at LaureateLearning Corporation, 6 children were given scenes and sentences like this:

5/6 answered “Amy” (or comparable for other stories) for “who heard thatJack played the trumpet in his room.” None of the children did the reverse:took heard Jack play the trumpet to be “grandma”. Therefore a tensed clausewas taken to be, or entail, an untensed small clause event, but not the reverse.While the primary direction is: Proposition ⇒ Event, there is some evidenceof the opposite as well.While adults know when an inference from a proposition to an event is war-ranted, a child may not.

4.1 Inferences that Overrule GrammarInference, however, hangs in the air over many situations and is the motive formany statements. Thus the statement above:

Sinn und Bedeutung 15 - [PDF Document] (73)

72 Roeper

(61) “Don’t you want to go outside?”

is usually expressed as a suggestion not a request for a propositional judge-ment of truth or falsity equivalent to: ‘Is it the case that you don’t want to gooutside.” Because such pragmatic goals as suggestion are the purpose of thespeech act, there is an understandable inclination to build this into the gram-mar directly. However, we believe it is of apiece with inferences that overrulethe overt meaning of the grammar.Schulz (2003) explored the contrast in a series of experiments with numeroussimilar stories like this one:

(62) Kermit went shopping and he was supposed to buy eggs.Then in the evening, he got really hungry, but he said “I have nothingto eat in the house.” He didn’t remember the eggs.Did he forget to buy eggs?⇒ noDid he forget that he bought eggs?⇒ yes (Schulz 2003)

A group of 38 children 4-6 yrs made errors on both types. 82% of the childrenmade errors on these sentences and Schulz states that “younger children treatfactive verbs as non-factive.” Consider the younger children who were mostprone to take the non-factive and interpret it as factive: “Did he forget to buyeggs” to mean “did he forget that he bought eggs” and answer “yes”.Unlike the Strict Interface discussed above, verbs like forget are negative im-plicatives which means they do not carry a fixed truth, but rather an implica-tion. The fact that he forget to buy eggs does not entail that he did not buyeggs strictly, just that he forgot the obligation at some point. Inference mustbe applied to know which implication to apply. Consider these more elaboratescenarios.

A boy is supposed to buy eggs to make omelettes. He forgets and bringsno money to the store. When he comes home, his mother says, “Did you forgetto buy eggs?”. He says yes and quickly runs back to the store to get some eggs.His father comes home and says:

(63) “Did you forget to buy eggs?”

The natural, pertinent answer is “no” because he did eventually buy eggs, buthe also had forgotten as well. Most adults would answer such a question withthe most recent and relevant answer: No. Now consider an altered context:Suppose the father comes home, and says “Johnny is always forgetting things– I bet he forgot to buy eggs.” “Did you forget to buy eggs?” Now the answermight be “Yes, even though he got them eventually.” This answer would, in a

Sinn und Bedeutung 15 - [PDF Document] (74)

Acquisition of Propositions 73

way, respond to both interpretations.The conclusion is that the child must decide by a broader situational inference– which should not be collapsed with Conventional implicatures – to decidewhich event should be the basis of an answer. So we see that direct syntac-tic projections, implications, and inferences are all pertinent. It is possible, atearly stages, that a child arrives at a neutral reading, which gives free rein toinferences? Suppose she deleted any tense marking and instead generated agerund:

(64) Did he forget about buying eggs?

This could allow both meanings: forget to buy or forget that he bought. Weconsider this a plausible syntactic move. It is plausible in part because theact of deletion would signal (unconsciously) the fact that the grammar wasincomplete. It is exactly what the children did in the hear him play the trumpetscenario.

4.2 False BeliefA great deal of work has surrounded the fact that children make errors in FalseBelief contexts. In deVilliers et al. (2010); Roeper (2011); deVilliers & Roeper(to appear) we have developed an approach that parallels our discussion ofinversion at the syntactic level. Children allow Total Reconstruction to occurwith respect to a syntactic chain which enables them to obey the Strong Min-imalist Thesis. Children are known to give the wrong answer in this scenario(“birthday cake” instead of “paper towels”):

(65) This mother snuck out one night when her little girl was asleep andbought a surprise birthday cake. The next day the little girl saw thebag from the store and asked, “What did you buy?” The mom wantedto keep the surprise until later so she said, “Just some paper towels.”– What did the mom say she bought?⇒ “a birthday cake”

We argue that when the wh-word is interpreted in the lower clause, the SMTdemands interpretion there, and then they give a factive answer. Thus they putthe what back into the second trace position:

(66) What did she say [ (what) she bought trace.]

That position is inside the first Phase and therefore is interpreted there. Theinterpretation for the adult is ultimately altered by the impact of the higher verballowing the lower CP to inherit a feature in syntax. The Total Reconstruction isexactly like what we saw with auxiliary inversion. In fact several other branches

Sinn und Bedeutung 15 - [PDF Document] (75)

74 Roeper

of adult behavior coincide with this analysis to enhance its plausibility. First inFrench it has been argued that wh-in-situ, unlike the fronted case, entails apresupposed propositon:

(67) Ilhe



vs. Oùwhere

va-t’il.goes he

The first entails that one went somewhere, while the second might be answeredwith “nowhere”. In addition, in German, it has been claimed that (substandard)partial movement where the wh-word is repeated requires factivity for somespeakers (Herberger 2001):

(68) Waswhat








Although this is regarded as substandard and obscure, the effect is exactly whatwe find with children, who are also known, in a variety of languages, to spon-taneously repeat (see deVilliers & Roeper (to appear)):

(69) “What did she say what she wanted for her birthday.”

And finally when a wh-word is in-situ in English the “correct” reading emerges,as this contrast reveals:

(70) What did he guess t the number was t. [=what was his guess]Did he guess what the number was. [=guess correctly or just make aguess]

Thus, looked at carefully, the child’s decision falls within the options of UGand is shown in parallel adult behavior.

But why do children persist in this reconstruction so much longer thanfor yes/no questions? In our initial court cases, we discussed how we developbeliefs from imperfect verbal evidence. In these experiments, a story with twoprongs is developed, both of which in the larger world of inferences is roughlyplausible. The experiment, in effect, seeks to see when children are able to usea precise interpretation of grammar to see which is appropriate. Adults mustlikewise sense that this discrimination is called for. If adults also allow infer-ences to exceed what grammar allows, then why should children not do thesame? If say often takes factive complements and if they are allowed to infer aproposition from a small clause with an event (see John drink), then the mannerin which the child uses inference to expand, and overinterpret, then their ma-neuver is within the larger comprehension strategies available to human beings.

Sinn und Bedeutung 15 - [PDF Document] (76)

Acquisition of Propositions 75

Ultimately, the child must grasp that the momentum of the story is preciselya call for the hearer to contrast what was bought with what was said to havebeen bought, and therefore the factive assumption must not be made.

5 ConclusionsThe reasons, then, that children pass through the misanalysis of SAI years be-fore they reliably handle such False Belief environments involves both the factthat UG favors an SMT analysis of clauses and that they must grasp the role ofcontrast in the story narrative.We have allowed several ideas to circle around the Strict Interface that: Tenseentails a proposition. We have argued that

1. it motivates the first major step to a transformation in the Root infinitivestage.

2. it promotes Total Reconstruction in order to maintain a tag-questionmeaning in early yes/no questions and in later long-distance questions,and

3. it allows deletion in some tensed clauses to fulfill pragmatically invitedinferences.

This article seeks primarily to embed these syntactic options within a largersemantic and pragmatic context by showing that what children do largely re-mains in the repertoire of adult response to language as well.What must the child do to become an adult? She must recognize certain con-texts, particularly those where contrast is part of the rhetoric of a narrative,that calls for, in effect, blocking our usual broad inferences about the possiblemeanings of human utterances.

ReferencesBeaver, David & Bart Geurts. 2010. Presupposition. Ms., Cedric. 2008. On the locus of asymmetry in UG. Harvard Manuscript.Chomsky, Noam. 2005. Three factors in the language design. Linguistic In-

quiry 36(1). 1–22.Chomsky, Noam. 2007. Approaching UG from below. In Uli Sauerland &

Hans-Martin Gärtner (eds.), Interfaces + recursion = language? Chom-sky’s minimalism and the view from syntax-semantics Studies in Gener-ative Grammar, 1–29. Mouton de Gruyter.

Chomsky, Noam. 2008. On Phases. In Robert Freidin, Carlos P. Otero &

Sinn und Bedeutung 15 - [PDF Document] (77)

76 Roeper

Maria Luisa Zubizarreta (eds.), Foundational issues in linguistic theory(Current Studies in Linguistics (CSLing) 45), 133–166. MIT Press.

Conroy, Anastasia. 2006. The semantics of How Come: A look at how fac-tivity does it all. In Nina Kazanina, Utako Minai, Philip J. Monahan& Heather L. Taylor (eds.), University of Maryland Working Papers inLinguistics 14, 1–24. The University of Maryland Department of Lin-guistics.

deVilliers, Jill G., Peter deVilliers & Tom Roeper. 2010. What’s in a phase.Unpublished.

deVilliers, Jill G. & Tom Roeper. to appear. The acquisition of wh-movementand linguistic theory. In Jill G. deVilliers & Thomas Roeper (eds.),Handbook of generative approaches to language acquisition, Springer.

Fitzpatrick, Justin. 2005. The Whys and How comes of presupposition and NPIlicensing. In John Alderete, Chung hye Han & Alexei Kochetov (eds.),Proceedings of the 24th West Coast Conference on Formal Linguistics24, 138–145. Cascadilla Press.

Guasti, María Teresa. 2006. Language acquisition: The growth of grammar.MIT Press.

Herberger, E. 2001. Paper on partial movement. Paper presented at LSA.Higginbotham, James. 1983. The logic of perceptual reports: An extensional

alternative to situation semantics. The Journal of Philosophy 80(2).100–127.

Hinzen, Wolfram. 2007. An essay on names and truth. Oxford UniversityPress.

Jesney, Karen. 2007. The acquisition of neg-cl*ticization and inversion.Manuscript Umass.

Klein, Wolfgang. 2006. On finiteness. In Veerle van Greenhoven (ed.),Semantics meets acquisition Studies in Theoretical Psycholinguistics,chap. 11, 245–272. Springer Verlag.

Lebeaux, David. 2000. Language acquisition and the form of the grammar.John Benjamins.

Legere, C. 2008. The acquisition of two types of factive complements. InAnna Gavarró & M. Joao Freitas (eds.), Language acquisition and de-velopment. Proceedings of GALA 2007, 337–346. Newcastle, UK: Cam-bridge Scholars Publishing.

Miyagawa, Shigeru. 2005. EPP and semantically vacuous scrambling. InJoachim Sabel & Mamoru Saito (eds.), The free word order phe-nomenon: Its syntactic sources and diversity, 181–220. Mouton deGruyter.

Panizza, Daniele, Gennaro Chierchia, Yi-Ting Huang & Jesse Snedeker. (in

Sinn und Bedeutung 15 - [PDF Document] (78)

Acquisition of Propositions 77

press). The relevance of polarity for the online interpretation of scalarterms. To appear in: The Proceedings of Semantics and Linguistic The-ory (SALT) 19.

Potts, Christopher & Tom Roeper. 2005. The narrowing acquisition path: Fromexpressive small clauses to declaratives. In Ljiljana Progovac, Kate Pae-sani & Eugenia Casiellesand Ellen Barton (eds.), The syntax of nonsen-tentials: Multi-disciplinary perspectives, John Benjamins.

Roeper, Tom. 1982. The acquisition of gerunds. In Lila Gleitmann & Eric Wan-ner (eds.), Language acquisition: The state of the art, 3–48. CambridgeUniversity Press.

Roeper, Tom. 2011. Vacate phase and a symmetrical theory of movement.Manuscript Umass.

Sauerland, Uli & Paul Elbourne. 2002. Total reconstruction, PF movement,and derivational order. Linguistic Inquiry 33(2). 283–319.

Schulz, Petra. 2003. Factivity: Its nature and acquisition. In LinguistischeArbeiten, Tübingen: Niemeyer.

Van Valin, Robert D. 2002. The development of subject-auxiliary inversion inenglish wh-questions: An alternative analysis. Journal of Child Lan-guage 29(1). 161–175.

Wexler, Kenneth. (to appear). Grammatical computation in the optional in-finitive stage. In Jill G. deVilliers and Thomas Roeper: Handbook ofGenerative Approaches to Language Acquisition.

Sinn und Bedeutung 15 - [PDF Document] (79)

Sinn und Bedeutung 15 - [PDF Document] (80)

In Reich, Ingo et al. (eds.), Proceedings of Sinn & Bedeutung 15,pp. 79–100. Universaar – Saarland University Press: Saarbrücken, Germany, 2011.

Appositives in Modal Contexts∗

Kjell Johan SæbøUniversity of [emailprotected]

Abstract. It has recently been argued (by Harris and Potts 2009: 523–552) thatso-called non-speaker oriented readings of appositives are a matter of pragmaticperspective shift and thus do not show that appositives contribute to descriptivecontent. In contrast, I argue that appositives are indeed building blocks of propo-sitions and that non-speaker oriented readings are de dicto readings of classicalpresuppositions (definedness conditions on concepts) in modal contexts. Down-ward entailing modal contexts, like surprise contexts, provide the key to thisconclusion: here, appositives provide a useful means to ensure that an individualretains its relevant properties across the modal space.

1 IntroductionThe question whether appositives, like “, a virgin,” as in (1), are building blocksof descriptive meaning is answered in the negative by Harris and Potts (2009)(henceforth: HP) but in the positive by Schlenker (to appear).

(1) . . . she, a virgin, would have a child . . .

In particular, HP hold that ‘non-speaker oriented’ readings, as attested by (3)(from Amaral et al. 2007), as opposed to ‘speaker oriented’ readings as attestedby (2) (from Potts 2005), are a matter of perspective shift: an appositive isusually used to implicate that the speaker is committed to the proposition thatthe referent has the property (in (2), that Chuck is a psychopath), but pragmaticfactors can shift the attribution of that commitment to someone else, typicallythe holder of an attitude (in (3): Sheila).

(2) Sheila believes that Chuck, a psychopath, is fit to watch the kids.

(3) Sheila believes that Chuck, a sweetheart if ever there was one, is fit towatch the kids.

∗ I am greatly indebted to the members of the group Meaning and Understanding across Languagesat the Oslo Center for Advanced Study 2010, particularly to Regine Eckardt, as well as to theaudience at the lecture at Sinn und Bedeutung 15 in Saarbrücken in September 2010 on which thispaper is based.

Sinn und Bedeutung 15 - [PDF Document] (81)

80 Sæbø

In contrast, I will argue that ‘non-speaker oriented’ readings of appositives arede dicto readings and that under such readings, appositives do contribute todescriptive content: they help build propositions in contexts like (4).

(4) Mary could not believe that she, a virgin, would have a child.

Generally, I will propose that appositives introduce ‘classical’ presuppositions,i.e., definedness conditions on functions from evaluation indices, starting withindividual concepts and projecting to propositions. In extensional contexts andon de re readings in intensional contexts, they are “mere” presuppositions, buton de dicto interpretations in intensional contexts, they enter into the argumentpropositions, thus making a genuine contribution to truth conditions.

I thus follow Schlenker (to appear) in pursuing a semantic strategy fordealing with appositives generally and ‘non-speaker oriented’ interpretationsparticularly.1 Our tactics are a bit different, though: while Schlenker uses factsabout the French subjonctif and Sequence of tense to show that non-restrictiverelative clauses can interact with mood and tense operators, concluding thatthey convey a mixture of assertion and presupposition, I will use facts aboutdownward entailing intensional contexts to show that nominal appositives caninteract with attitude predicates, constraining their argument propositions, andconclude that they convey more traditional presuppositions.

HP offer two major arguments in support of their pragmatic strategy. First,they present experimental evidence to show that non-speaker oriented readingsare available outside intensional contexts. I try to counter this in part by arguingthat their extensional contexts are intensional after all and in part by appealingto a mechanism of ‘concealed quotation’.

Second, conducting a corpus study of appositives under attitude verbs,HP report a bias for speaker orientation. However, their attitude predicates areall upward entailing (if monotone). By contrast, in downward entailing attitudecontexts, such as surprise contexts, there is a bias for non-speaker orientation.Thus there is no empirical basis for marginalizing this orientation.

I confront the arguments supporting HP’s pragmatic account in Sect. 2.In Sect. 3, addressing the semantics of surprise predicates, I show how the biasfor non-speaker orientation in such contexts provides indirect evidence thatappositives contribute to propositions, and present direct evidence that they do,in terms of entailment patterns derived from standard analyses.

1 A note on terminology: the term “appositive”, or “apposition”, is sometimes used to cover bothnominal appositives like the indefinite descriptions in (1)–(4) and non-restrictive relative clauses(which form the main concern of Schlenker (to appear)), or even other forms of non-restrictivemodifiers; though most of the points made in this paper are probably relevant for appositives inthat broad sense, I will concentrate on nominal appositives in the form of indefinite descriptions.

Sinn und Bedeutung 15 - [PDF Document] (82)

Appositives in Modal Contexts 81

In Sect. 4 I demonstrate how this contribution can be captured through clas-sical presuppositions, i.e., definedness conditions on functions from worlds.The emerging picture is that when they are read de dicto, appositives providea means to ensure that an individual referred to in a modal context retains itsrelevant properties across the modal space. Sect. 5 brings conclusions.

2 The Arguments for the Pragmatic StrategyHP offer two major arguments in favor of their pragmatic theory of perspectivalorientation for appositives. One is a finding that perspective shift, from speakerto non-speaker, occurs outside intensional contexts. The other is a finding thatin intensional contexts, non-speaker orientation is relatively rare. In this sectionI examine this evidence and try to counter it.

2.1 Perspective Shi� in Extensional Contexts?HP present evidence from experiments eliciting informant judgments to showthat non-speaker oriented interpretations are available even when appositivesare in “matrix clauses” (pp. 530ff.). One of their cases is (5):

(5) I am increasingly worried about my roommate. She seems to be grow-ing paranoid.a. The other day, she told me that we need to watch out for the mail-

man, a possible government spy.b. The other day, she refused to talk with the mailman, a possible

government spy.

The idea is that in (5a), the appositive a possible goverment spy is in a modalcontext while in (5b) it is not. Schlenker (to appear) notes, however, that refusemay well be an attitude verb. To me, it is clear that it is. A test will be if we candetect a de dicto / de re ambiguity in a definite description in a refuse context,and it appears that we can. Consider (6):

(6) Goddess Parvati, wife of God Shiva, lonely during one of her husband’slong absences, molds a son from mud. Ganesh grows up, and one day,Parvati tells him to guard the front door and not let anybody in whileshe bathes.a. Shiva returns home. Ganesh refuses to let his father in, and Shiva

beheads him.b. Shiva returns home. Ganesh refuses to let the intruder in, and Shiva

beheads him.

Sinn und Bedeutung 15 - [PDF Document] (83)

82 Sæbø

(6a) is true on a de re interpretation but false on a de dicto interpretation, whilethe converse holds of (6b). This strongly suggests that one of HP’s key casesdoes not show what it is supposed to show, namely, that a non-speaker orientedinterpretation is possible in an extensional context. But it is not their only case:they also report non-speaker oriented readings in one or two contexts whichare much less clearly intensional, like (7):

(7) My brother Sid hates school.He puts off his homework, a complete waste of time, to the last minute.

I see here two ways to salvage the semantic strategy, on which the orientationof the appositive a complete waste of time toward my brother Sid is a de dicto,intensional phenomenon: either

1. one can argue that put off is an attitude verb in disguise, or2. one can argue that the appositive is a piece of quotation in disguise.2

Way 1 predicts that a non-speaker-orientation reading is unavailable if the verbcannot under any conditions be an attitude verb. (7a) seems to bear this out:

(7) a. ?My brother Sid hates school. His homework, a completewaste of time, earns him bad grades.

Way 1 also predicts that there can be a de dicto / de re ambiguity in a contextunder a transitive verb like put off – and (8) might seem to show that there can;the Senate might or might not subscribe to the characterization of the bill asthe most important one ever presented to it.

(8) The Senate has postponed the most important bill ever presented to it.

Way 2 is to appeal to a mechanism of ‘concealed quotation’ which can affectdescriptions generally, as, for example, in (9):

(9) The ship that could not sink sank on her maiden voyage in April 1912.

According to Geurts and Maier (2005), pieces of quotation create local dicendicontexts with contextually anchored sources (for (9) on a quotation-in-disguiseanalysis of the definite description, this source might be the White Star Line).On a piece-of-quotation-in-disguise analysis of the appositive, (7) will receivemore or less the same interpretation as (7b).

(7) b. My brother Sid hates school. He puts off his homework,

2 At a more general level, this has been suggested by Anand (2007).

Sinn und Bedeutung 15 - [PDF Document] (84)

Appositives in Modal Contexts 83

“a complete waste of time”, to the last minute.

The contextually anchored sources are subject to constraints, and presumably,one constraint would account for the mild infelicity of (7a) above. Anyway, ifthe mechanism of concealed quotation is otherwise freely available, it servesto create local attitude contexts with implicit attitude holders and thus to takethe edge off the evidence adduced by HP against a semantic strategy.

2.2 A Bias for Speaker Orientation in Intensional Contexts?HP report on a corpus study designed to measure the frequency of non-speakeroriented readings of appositives in embedded contexts, based on a sample of 31embedding predicates; attitude predicates and verbs of saying (pp. 540–547).They discover a bias for speaker orientation; non-speaker orientation appearsas an exception.

However, apart from non-monotone verbs like ask, the attitude verbs inHP’s sample are all monotone increasing (upward entailing): believe, claim,say, etc. Once we turn to downward entailing attitude predicates like surprised,the picture actually seems to be reversed: there then appears to be a bias fornon-speaker-oriented readings. (10a) is a case in point:

(10) a. Mary is surprised that John, a Laestadian, wears a necktie.

In Laestadianism, a conservative Lutheran revival movement centered in theextreme north of Scandinavia, men are (often) not supposed to wear neckties.Hence it is surprising if a Laestadian man does wear one.

So the appositive in (10a) seems to contribute to the argument proposition,constraining it; the intuition is that worlds where John is not a Laestadian arenot to be counted when the predicate surprised is evaluated. True, in contrast tothe case (3) in Sect. 1, with the non-factive, upward entailing predicate believe,the appositive is here not controversial: there is (due to the factive predicate) noconflict between the beliefs of the speaker and those of the attitude holder, butit is still evident that the appositive is an ingredient in the object of the attitudeand hence that it is essentially oriented towards the holder of the attitude.

In fact, it is typical of appositives in surprise contexts that there is not thatmuch of a difference between a conjoined predicate and an appositive:

(10) b. Mary is surprised that John is a Laestadian and wears a necktie.

To be sure, there is a difference; there is a partition into background and focusin (10a), corresponding to a distinction between ‘old’ and ‘new’ knowledge:the property expressed in the appositive seems to count as ‘old knowledge’ ofJohn on the part of Mary, whereas the property expressed in the carrier clause

Sinn und Bedeutung 15 - [PDF Document] (85)

84 Sæbø

seems to count as a piece of knowledge just acquired, triggering the surprise.Speaker-oriented readings are possible in surprise contexts, but rare:

(11) Mary is surprised that John, a notorious casanova, betrays her.

That John is a notorious casanova is something the speaker knows, not Mary;that she does not know it is the reason for her surprise.

The asymmetry between upward and downward entailing attitudes as towhat is the ‘normal’ orientation – toward the speaker with the former, towardanother individual with the latter – calls for an explanation. Could there be apragmatic explanation? Well, one might think that the ‘normal’ orientation iswhat provides an explanation for the attitude in discourse relational terms – butnote that while that is indeed the normal case under surprised, cf. (10a), it is theexceptional case under believe, cf. (3); in other words, non-speaker orientationis consistently what would go to explain the attitude.

Other pragmatic explanations may be conceivable, but in the next sectionI will suggest an explanation building on the semantics of surprise predicates,and also supply more concrete evidence that non-speaker oriented appositivesunder such predicates play a semantic role.

3 Appositives and the Semantics of SurpriseBelow, I go into the semantics of surprise predicates and show that in regard tothe inference patterns that standard analyses predict, appositives in embeddedclauses commonly behave as if they were integral parts of those clauses. Theypattern with conjoined predicates in expressing properties that are relevant forevaluating the surprise predicates, thus serving a significant semantic purpose.

3.1 Surprise Semantics and Contextual PerspectivesHP predict widely different inference properties for (10a) and for (10b): whileaccording to their theory of pragmatic perspective shift, (10a) means the sameas (10c) in the dimension of descriptive content and implicates (10d) in thedimension of expressive content, there is no prediction that (10b) entails (10c).

(10) a. Mary is surprised that John, a Laestadian, wears a necktie.b. Mary is surprised that John is a Laestadian and wears a necktie.c. Mary is surprised that John wears a necktie.d. Mary believes that John is a Laestadian.

In fact, scholars who have analyzed predicates like surprise agree that clauseslike (10b) do not entail clauses like (10c) but that clauses like (10c) togetherwith clauses like (10d) do entail clauses like (10b). As acknowledged by those

Sinn und Bedeutung 15 - [PDF Document] (86)

Appositives in Modal Contexts 85

scholars, our intuitions may not be so clear in these regards, and I will return tothe reasons for that below; but note at once that (10a) does not seem to entail(10c) any more than does (10b). That is to say: contrary to what HP predict,in certain contexts, appositives are intuitively not more detachable from thedescriptive content than conjoined predicates.Asher (1987) identifies a class of ‘negative factive’ attitude verbs, obeying, i.a.,the inference rule ‘weakened downward entailment’ (WDE) (a for the agent,α for the attitude predicate at issue, β for believe):

(12) Weakened downward entailment:aα φ ∧ JψK⊆ JφK ∧ aβ ψ ⇒ aα ψ

This class contains a variety of predicates, but the adjective surprised is usuallyconsidered as the paradigmatic case. Thus (10c) and (10d) are to jointly entail(10b), but (10b) is not to entail (10c).3 There is a consensus on this: von Fintel(1999), Sharvit (2002), and van Rooij (2006) concur in predicting (12).

The reason that (weakened) downward entailment is predicted by all isthat all assume the basic truth condition to be that the argument proposition isa subset of something (the ‘unexpectation’ worlds) or has an empty intersectionwith something (the ‘expectation’ worlds); cf., e.g., Sharvit (2002: 103):

(13) Semantics of Surprise according to Sharvit (2002: 103):J surprise K(w)(P)(a) = 1 iff P(w)⊆ NONEXP(a)(w)

And that is of course easier the stronger the argument proposition, P(w), is.Reality is not quite so simple, though. It may often appear as if surprised

fails to entail downward, or even as if it entails upward. (14) would provide acase in point for the former, (10b, c) might go to suggest the latter.

(14) John is surprised that Mary won a medal, but he is not surprisedthat she won a bronze medal.

(10) b. Mary is surprised that John is a Laestadian and wears a necktie.c. Mary is surprised that John wears a necktie.

The customary way to account for the apparent failure of surprise predicatesto entail downward, or their semblance of entailing upward, is to appeal to animplicit restriction, surreptitiously modifying (as the case may be, weakeningor strengthening) the argument proposition in the putative conclusion.

3 Strictly, (10c) and “Mary believes that John is a Laestadian and wears a necktie” are to entail(10b), but since (10c) presupposes that Mary believes that John wears a necktie, (10d) is enough.

Sinn und Bedeutung 15 - [PDF Document] (87)

86 Sæbø

To be specific, when we pass from (10b) to (10c) to test whether the latterfollows from the former, it is very difficult to keep the context constant and notlet the property explicitly ascribed to John in the complement of the premisscarry over, as an implicit restriction, to the complement of the conclusion. Thuspremiss and conclusion are easily judged to be equivalent.

Kadmon and Landman (1993) call this implicit restriction a ‘perspective’and propose to code it as a contextual parameter in the meaning of surprised:

To be surprised that A is always relative to a certain perspective on A, a perspec-tive that determines what it is about A that is surprising and in virtue of what it issurprising. The perspective is a contextually determined parameter in the inter-pretation of surprised, very much in the same way that a ‘modal base’ (Kratzer1981) is a contextually determined parameter in the interpretation of modals.[. . . ] The perspective enters into the semantics of surprised and affects the truthconditions of sentences containing it. (Kadmon and Landman 1993: 381)

Perspectives “can be at least partly specified by explicit linguistic text”, e.g.:

(15) – I can’t believe she’s divorcing HIM.– Yeah, such a rich man.– No, such a KIND man!

von Fintel (1999) elaborates on Kadmon and Landman’s proposal by providing“a semantics for the attitudes that is specifically sensitive to a shifting domainof ordered worlds”. His (main) proposal for surprised is, slightly adjusted:4

(16) Semantics of Surprise according to von Fintel (1999: 122–125):J surprised K f,g

w = λ pλα : Bα(w)⊆ p , Bα(w)⊆ fα(w),fα(w)∩ p 6= /0 , fα(w)−p 6= /0 . ∀w′∈ maxgα (w)( fα(w)) : w′ /∈ p

Here f and g are parameters similar to the modal base and the ordering sourcein Kratzer’s theory of modality, f assigning a set of worlds and g assigning a setof propositions to the subject of surprised and the current index of evaluation.The definedness conditions ( between : and . ) say that the subject of surprisedbelieves the complement p and the set of relevant worlds fα(w) and that therelevant worlds contain p- and non-p-worlds. The content proper is, in words,that in all the most expected of the relevant worlds, p is not true. Note that thelogical structure of this definition is the same as that of Sharvit’s definition (13);the two definiens clauses are contrapositions of each other, and the attitude is

4 von Fintel does not actually define surprised but sorry, intending that definition to carry over tosurprised “with suitable adjustments”. I substitute a ‘denotation equation’ for his truth condition,and I omit his coindex on the attitude predicate and on the functions f and g on the understandingthat f and g are here the functions fitting surprised.

Sinn und Bedeutung 15 - [PDF Document] (88)

Appositives in Modal Contexts 87

downward entailing. The essential novel feature of von Fintel’s definition is therelativization to a set of relevant worlds compatible with the subject’s beliefs.

It is natural to think of fα(w) as the intersection of a set of propositions,e.g., the proposition that Mary is a virgin, or that John is a rich man, or a kindman, or a Laestadian, – or that the topic time is Good Friday:

(17) Bjarne was in Hamburg on Good Friday in 1984 and was outraged thatthe brothels were open.

More generally, if the set fα(w) is the intersection over a set of propositions Φ

and p involves rigid or de re reference to an individual, with a name, a pronoun,or a definite description on a de re interpretation, it is reasonable to expect Φ

to include the ascription of a property to that individual.

3.2 Surprise Semantics and AppositivesThe way appositives function in the cases so far considered suggests that theyprovide a means to ensure that individuals referred to in intensional contextsretain their relevant properties across the possible worlds under consideration,making explicit what would otherwise have to be left implicit and ascertainingthat certain properties are carried along throughout the modal space at issue.In short, they contribute content that can be much at issue. In the next section,I sketch a way to formalize this notion. But first, I will present evidence thatappositives matter for the truth conditions of downward entailing attitudes, andprovide an answer to the question why non-speaker oriented interpretations arenot rare but frequent when appositives are in such contexts.

Consider (18a–c). (18a) contains two instances of complex appositives,appositives consisting of two predicates – Jewish man, Samaritan woman. In(18b), the second member of each pair is deleted, in (18c), the first member is.

(18) In John 4 Jesus spoke with a Samaritan woman and asked for a drink.She had two things against her: she was a woman, and a Samaritan.a. She was surprised that he, a Jewish man, spoke to her, a Samari-

tan woman.b. She was surprised that he, a Jew, spoke to her, a Samaritan.c. She was surprised that he, a man, spoke to her, a woman.

There is a strong intuition here that (18a) entails neither (18b) nor (18c). Thatis to say, surprise contexts are not closed under the weakening of appositives.This is predictable if the appositives help to build the argument propositions,but not – or only with difficulty (see below) – if they do not.

So why does this case present a different picture from a case like (10a, c),

Sinn und Bedeutung 15 - [PDF Document] (89)

88 Sæbø

where the intuition that the first sentence fails to entail the second is weaker?

(10) a. Mary is surprised that John, a Laestadian, wears a necktie.c. Mary is surprised that John wears a necktie.

Here, as with (10b) in relation to (10c) (see Sect. 3.1), it is difficult to blot outthe property explicitly ascribed to the subject of the first complement clause;it easily influences the second as an implicit constraint on the relevant worlds.This source of error is evidently eliminated in (18a/b) and (18a/c), presumablybecause by explicating a property, we signal that implicit ones are not relevant.

Now if an appositive can never contribute to a proposition, the missinginferences in (18) are mysterious: (18a–c) should then be equivalent.

Note, in addition, that the appositives in (18a) represent old information,both in regard to the common ground and to the belief state of the experiencer:the context entails both that he was a Jew and a man and she a Samaritan anda woman and that she knew that. This makes a theory where appositives justdescribe separate, scopeless bits of expressive content not seem very plausible;in fact, they would appear to be superfluous in such cases. More clearly still,noone needs reminding that someone referred to with she or Maria is a female:

(19) She became accustomed to the double-takes from male firefighterswho were surprised that she, a female, could head the station.

(20) Maria tells us that the students are surprised that she, a woman, talksto them about math and natural sciences.

Admittedly, one might try to account for the missing inferences in (18) whilemaintaining that (18a–c) are equivalent by appealing to discourse relations: itcould be argued that experiencer-oriented appositives in attitude contexts arecommonly intended as explanations; if the complex appositives in (18a) givefull explanations, then the simple ones in (18b or c) cannot be expected to doso too. After all, the semantics of surprised refers to contextually determinedrelevant beliefs, and the appositives could be argued to supply those by generalpragmatic principles. Hence it is difficult to actually prove that the appositivesmake a semantic contribution.

But if we do assume that they do, we are able to answer the question whysubject-oriented readings are frequent in downward entailing attitude contextsalthough they are infrequent in upward entailing ones. If under such readings,appositives do restrict argument propositions, then there is a close analogue indeterminer domain restriction, more noticeable, because it is more useful, withdownward than with upward entailing determiners, cf. Heim (to appear):

Sinn und Bedeutung 15 - [PDF Document] (90)

Appositives in Modal Contexts 89

For all we can tell, every determiner may be construed with a covert restrictorin addition to its overt one and thus apply to an effectively narrowed set of con-textually relevant or salient entities. We just don’t see this as clearly with somedeterminers as with others. the, along with every . . . , is not upward monotone,so covert restrictors weaken the presupposition or assertion, giving an otherwisetruth-value-less or false claim a chance to be felicitous and true.

By analog, we can say that decreasing attitude contexts show a bias for subject-oriented readings because here, such readings serve to weaken the statement,giving an otherwise potentially false assertion a better chance to be true, whileincreasing attitude contexts show a bias for speaker-oriented readings becausethere, subject-oriented readings serve to strengthen the statement. If that is so,we would expect negated versions of increasing attitude predicates to show apropensity for subject orientation as well, and this seems to be borne out:

(4) Mary could not believe that she, a virgin, would have a child.

I conclude from the evidence that appositives can help build propositions. Thequestion is now how.

4 Appositives as Definedness ConditionsFaced with a picture of appositives in different roles in different contexts – innonmodal contexts or on speaker-oriented readings in modal contexts, they addinformation on referents in a de re mode, but on non-speaker-oriented readingsin modal contexts, they add information on referents in a de dicto manner – Iwill attempt to unite these two roles by describing appositives as introducingdefinedness conditions on partial individual concepts, or more generally, asthese project, on partial functions from worlds, ultimately partial propositions.This approach necessitates a method of intensional composition, roughly à laBeaver and Krahmer (2001), and I will now specify the necessary machinery.

4.1 From Appositives to Partial PropositionsAt the bottom, I posit a silent appositive functor A whose meaning is:5,6

(21) JA K = λPe(st)λx(se)λw : xw ∈ De , Pw(xw) . xw

Suppose we build the meaning of “surprised that Mary, a virgin, is pregnant”.I consider the indefinite article in the appositive as spurious, i.e., disregard it.

(22) JA K(JvirginK) = λx(se)λw : xw ∈ De , Vw(xw) . xw

5 This is the indefinite case, where the first argument is a function from individuals to propositions;a definite case, where this argument is an individual concept, is also definable.6 Notation: Stuff between : and . are definedness conditions. Convention: fw(g) = f (g)w.

Sinn und Bedeutung 15 - [PDF Document] (91)

90 Sæbø

This is the full appositive meaning, an operation on partial individual concepts.It carries two presuppositions (definedness conditions): one projected from itsargument, the other coincident with its own content. (V = JvirginK.)

The proper name Mary is, standardly, taken to mean a constant, though inprinciple a partial, function from worlds to individuals.

(23) JA K(JvirginK)(JMaryK) = λw : Mw ∈ De , Vw(Mw) . Mw

This is Mary with the definedness condition that she is a virgin.For partial individual concepts like this to combine with properties like

JvirginK or J pregnant K, which are (as in Beaver and Krahmer 2001) functionsfrom individuals to propositions, they are lifted to functions from such e(st)‘properties’ to propositions by the abstract filter function F :

(24) JF K = λ f(se)λPe(st)λw : fw ∈ De . Pw( fw)

(25) JF K(JA K(JvirginK)(JMaryK)) =

λPe(st)λw : Mw ∈ De , Vw(Mw) . Pw(Mw)

This lifted DP meaning can be used directly in subject position, applying toa property like J pregnant K, resulting in (26), or more generally in a positionhosting a DP after QR, maybe from object position. (V = J pregnant K.)

(26) JF K(JA K(JvirginK)(JMaryK))(JpregnantK) =

λw : Mw ∈ De , Vw(Mw) . Pw(Mw)

This is the partial proposition that Mary, a virgin, is pregnant.

4.2 From Partial Propositions to SurpriseNow we need a meaning for surprised which can take partial propositions andproject their definedness conditions appropriately. It is commonly assumed thatsurprised introduces the presupposition that the subject believes the argumentproposition.7 This will now incorporate the presuppositions brought along bythe argument proposition, in particular, the one originating in the appositive,because what the subject must believe is not the partial argument propositionbut its positive extension, the set of worlds where it is defined and not negative.By the same token, what must lie outside the subject’s relevant expectationsfor the attitude predicate to hold of its two arguments is not the first argumentit*elf but the set of worlds where it is true, which means that the definedness

7 van Rooij (2006: 217), following Asher (1987), assume that such predicates introduce two layersof presupposition, the argument proposition itself and the proposition that the subject believes it;however, von Fintel (1999), following Heim (1992), only assumes the latter layer, as I will too.

Sinn und Bedeutung 15 - [PDF Document] (92)

Appositives in Modal Contexts 91

condition (Mary is a virgin) and the content proper (Mary is pregnant) bothcontribute to define the set of worlds at issue.

(27) is an augmentation of (16) in Sect. 3.1, based on von Fintel (1999):

(27) J surprised K f,g = λφλxλw : Bx(w)⊆ λw′φw′=1 .

maxgx(w)( fx(w))∩ [λw′φw′=1] = /0

(maxgx(w)( fx(w)) = those relevant worlds compatible with x’s beliefsthat x most expects)

(28), then, is the meaning of “surprised that Mary, a virgin, is pregnant”.

(28) J surprised K f,g(JF K(JA K(JvirginK)(JMaryK))(JpregnantK)) =

λxλw : Bx(w)⊆ λw′Mw′ ∈ De,Vw′(Mw′),Pw′(Mw′) .

maxgx(w)( fx(w))∩ [λw′Mw′ ∈ De,Vw′(Mw′),Pw′(Mw′)] = /0

Applied to Joseph, this becomes the partial function from worlds to truth valuesthat is defined iff Joseph believes that Mary is a virgin and pregnant and true iffit is defined and the worlds where Mary is a virgin and pregnant all lie outsidethe most expected relevant belief worlds of Joseph.

That may be a fair first attempt at a definition of the definedness and truthconditions of (29a), but note two problematic aspects of it:

1. The definition of the attitude in (27) fails to predict a difference between(29a) and (29b), that is, between appositives and conjoined predicates inembedded positions.

(29) a. Joseph is surprised that Mary, a virgin, is pregnant.b. Joseph is surprised that Mary is a virgin and (is) pregnant.

2. It seems contradictory to say, for (30) for simplicity, on the one hand thatfor the proposition to be defined, the subject must believe that Mary ispregnant, and on the other that for it to be true, the set of worlds whereshe is must lie outside the subject’s most expected relevant belief worlds.

(30) Joseph is surprised that Mary is pregnant.

Now in fact, both problems can be tackled with one tool, by distinguishing inthe definition of surprised between the intersection over x’s ‘old beliefs’ in w,Bx(w), and the intersection over x’s ‘full beliefs’ in w, B+

x (w) (a subset of it).It is the old beliefs that x’s most expected relevant worlds should be based on,before the new belief coming from the argument proposition is formed, and it is

Sinn und Bedeutung 15 - [PDF Document] (93)

92 Sæbø

the old beliefs that the presupposition inherited from the argument proposition,originating in the appositive, should be required to include. So I propose thisamended definition (needless to say, to do a proper job of it, the notions of oldand full beliefs should be defined in terms of eventualities of surprise):

(31) J surprised K f,g = λφλxλw : Bx(w)⊆ λw′φw′ ∈ Dt ,

B+x (w)⊆ λw′φw′ = 1 .

maxgx(w)( fx(w))∩ [λw′φw′=1] = /0

(maxgx(w)( fx(w)) = those relevant worlds from Bx(w) thatx most expects)

This definition leads to the following revised definition of the definedness andtruth conditions of “Joseph is surprised that Mary, a virgin, is pregnant”, (29a):

(32) J surprised K f,g(JF K(JA K(JvirginK)(JmK))(JpregnantK))(J jK) =

λw : B j(w)⊆ λw′Mw′ ∈ De,Vw′(Mw′) ; B+j (w)⊆ λw′Pw′(Mw′) .

maxg j(w)( f j(w))∩ [λw′Mw′ ∈ De,Vw′(Mw′),Pw′(Mw′)] = /0

(maxg j(w)( f j(w)) = those relevant worlds in B j(w) that j expects the most.)There are two layers of definedness conditions here: Joseph must have believedand believe that Mary is a virgin and he must believe now that she is pregnant(to accord with the truth condition, this must be a new belief); by contrast, onthe definition of the definedness and truth conditions of (29b), there is only onesubstantial level of presupposition: Joseph must believe that Mary is a virginand (is) pregnant (and, sloppily, one of the two conjuncts must be a new belief).Thus we have now a reasonable truth condition and a distinction between anappositive and a conjoined predicate at the level of definedness conditions.

4.3 The Speaker Oriented Case: de reHow about speaker-oriented readings? Consider (33), akin to Potts’ (2):

(33) Mary is surprised that John, a notorious casanova, betrays her.

Mary is not aware that John is a notorious casanova. I will treat this reading asa de re reading of the DP “John, a notorious casanova”, employing a designatedvariable for the actual world – v – and an abstract actualizer operator – @:8

8 There are of course various ways of construing de re interpretations (for recent surveys of thestate of the art, see von Fintel and Heim (2009) and Schwager (to appear)); if @ as defined in (34)is considered the general method, it would seem to imply that only definite DPs, with individualconcept, type (se), meanings, can have such readings; for indefinite descriptions, to the extent that

Sinn und Bedeutung 15 - [PDF Document] (94)

Appositives in Modal Contexts 93

(34) J@K = λ f(se)λw . fv

In extensional contexts, this operator is redundant. To see this, consider (35)and its two semantic structures (36) and (37), one with, the other without @:

(35) Professor Duriarti, a notorious criminal, has escaped from prison.

(36) JF K(JA K(Jnotorious criminal K)(JDuriartiK))(JescapedK)

= λw : Dw∈ De , N C w(Dw) . Ew(Dw)

(37) JF K(J@K(JA K(Jnotorious criminal K)(JDuriartiK)))(JescapedK)

= λw : Dv∈ De , N C v(Dv) . Ew(Dv)

The main difference is that the presupposition stemming from the appositive ison the construal in (37) anchored to v, the actual world. This does not make adifference once the sentence is evaluated, though. What has been built above,and all along, are not sentence denotations (truth values) but propositions. Butwhen these are ‘finished’ and are to be used as assertions or more generally asrepresentatives, they are applied to v, yielding a denotation, a truth value. Andthen the difference between (36) and (37) is levelled out.

Returning to (33) and the de re interpretation of the embedded appositive,when J@K applies to JA K(Jnotorious casanovaK)(JJohnK), the result is:

(38) J@K(JA K(Jnotorious casanovaK)(JJohnK)) =

λw [λw : Jw∈ De , N C w(Jw) . Jw](v) =

λw : Jv∈ De , N C v(Jv) . Jv

This is a partial constant function, defined if and only if the name John actuallydenotes an individual and this individual is actually a notorious casanova, andyielding, if defined, that individual no matter what. The partial proposition thatJohn, a notorious casanova, betrays Mary (here the type e(st) property Bm) is:

(39) JF K(J@K(JA K(Jnotorious casanovaK)(JJohnK)))(Jbetrays MaryK)

= λw : Jv∈ De , N C v(Jv) . Bmw (Jv)

So far, this is parallel to (37), and the actualizer is redundant. But when (39) isembedded in an intensional context, this changes. The full partial propositionthat Mary is surprised that John, actually a notorious casanova, betrays her is:

(40) λw : Bm(w)⊆ λw′Jv∈ De,N C v(Jv) ; B+m (w)⊆ λw′Bm

w′(Jv) .

transparent evaluation readings are desired, one could consider a choice function analysis.

Sinn und Bedeutung 15 - [PDF Document] (95)

94 Sæbø

maxgm(w)( fm(w))∩ [λw′Jv∈ De,N C v(Jv),Bmw′(Jv)] = /0

There are two cases: either the name John actually denotes an individual andthis individual is actually a notorious casanova, in which case the proposition isdefined iff Mary believes that this individual betrays her and true iff she is alsosurprised that he does; or not, that is, John is not actually a notorious casanova;then the proposition is only defined if Mary believes the empty set of worlds.That may be thought to be possible; after all, one can have inconsistent beliefs.If we want the proposition to be plainly undefined in case John is not actually anotorious casanova, we could consider including the factive presupposition inthe definedness condition introduced by the attitude: λφλxλw : φw = 1 . . . .

Importantly, though, however the definedness conditions are dealt with,in contrast with the de dicto construal, on the de re construal the content of theappositive plays no role in the content of the clause (the second line in (40)) andso does not affect its truth conditions: if John is indeed a notorious casanova,the conjunct N C v(Jv) is superfluous.

4.4 DiscussionThe analysis developed above has some welcome consequences but also someproblematic features. First, since the actualizer can apply below an appositive,the analysis predicts that there can be de re readings of definite descriptionswith appositives read de dicto, – and this seems to be borne out:

(41) Thoas wants Iphigenia to sacrifice her own brother, a stranger who hastried to steal the statue of Artemis.

Thoas is not yet aware that Orestes is Iphigenia’s brother. But he is well awarethat Orestes has tried to steal the statue of Artemis, and he is convinced thatOrestes is a stranger. If this is a natural reading of (41), it is derivable.

By the same token, however, it is predicted that there cannot be de dictoreadings of definite descriptions with appositives read de re, – and this questionseems more open. To test it, we may consider cases like (42).

(42) ?Gloucester wants his loyal son, a traitor to him, to inherit his title.

The only sensible interpretation is that the son that Gloucester wants to inherithis title is Edmund, loyal in his belief worlds but actually a traitor. To the extentthat this interpretation is unavailable, it serves to confirm the prediction that ifthe appositive is in the scope of @, so must the expression it applies to be. Butit may not be obvious that (42) is dubious. (43) is predicted not to be dubious:the son that Gloucester wants to disinherit is Edgar, a traitor in his belief worldsbut loyal in reality. It is not clear that this derivable reading is more accessible

Sinn und Bedeutung 15 - [PDF Document] (96)

Appositives in Modal Contexts 95

than the intended but not derivable reading of (42).

(43) ?Gloucester wants to disinherit his loyal son, a traitor to him.

If (43) is indeed felt to be no more felicitous than (42), this reflects unfavorablyon the formalism developed in this section.

Another, more general, point of discussion that should be raised concernswhat Schlenker (to appear) terms the epistemic status of appositives in relationto the presupposition-as-definedness-condition that their contribution has beenidentified as. According to Schlenker, nonrestrictive relative clauses introduceconditions similar to presuppositions but special in being easy to accommodate(translucency). Thus their content, while not trivial, should be uncontroversial.The close parallel between (35) and (44) indicates that Schlenker would intendthis characterization of nonrestrictive relatives to carry over to appositives.

(44) Professor Duriarti, who is a notorious criminal, has escaped from prison.

(35) Professor Duriarti, a notorious criminal, has escaped from prison.

The analysis proposed in this paper simply predicts that (35) lacks a truth valueunless Professor Duriarti is actually a notorious criminal, and the question ishow this definedness condition on the world of evaluation can be brought intoaccord with Schlenker’s notion of a ‘translucent’ condition. On the face of it,the two notions seem very different; in particular, it seems too strong to deny(35) a truth value should the professor turn out not to be a notorious criminal.

The relevant notion of presupposition has been referred to as a ‘classical’or ‘traditional’ notion, and my intention has been that this notion is appropriatefor some but not all triggers and coexists with a context oriented conception,apt for triggers like again and also, start and stop, non-emotive factives, andanaphoric definite descriptions. In these and more cases, being entailed by thecontext is the ideal and accommodation is difficult. Appealing to a definednesscondition for appositives or nonrestrictive modifiers more generally can be seenas a way to decrease the demands on the context and to allow for a comparativeease of accommodation, and thus as a way to emulate ‘translucency’.

Thus as long as the information contained in the appositive, though new,does not conflict with the information state of the addressee, she will typicallybe willing to accept it on faith. Conversely, if that information does conflictwith the addressee’s information state, she may well be prepared to disregardit and still assess the information contained in the carrier sentence:

(36) It says here: “Professor Duriarti, a notorious criminal, has escaped

Sinn und Bedeutung 15 - [PDF Document] (97)

96 Sæbø

from prison.” Since he is innocent, I am glad to learn that he is a freeman.

The additional assumption needed is thus that (simplex) sentences containingnonrestrictive modifiers can still have a truth value even if the modifier shouldbe false or undefined – and the need for such a notion of ‘accommodation’,consisting not in adding information but in disregarding it, is shared by otherpresupposition triggers which lend themselves to a definedness analysis, likereferential definite descriptions, as in (45), or gender markers, as in (46).

(45) – Who is the man drinking a martini? (from Donnellan 1966)– It’s not a martini, but anyway, it’s Smith.

(46) – PirminsPirmin’s







– Esit









On the whole, a critical point of discussion is whether and to what extent theanalysis of appositives presented in this paper should be expected to extend tononrestrictive modifiers such as those exemplified by (47)–(49):

(47) The pharmacist was surprised that this old woman was interested incondoms.

(48) My friends are surprised that shy and sweet me is learning self defense.

(49) She was startled to see the devoted detective Mickey Mouse shakehands with his ruthless rival Black Pete.

In all these cases, the italicized NP or AP is redundant in the sense that it doesnot contribute to the identification of the referent of the description it is part of.In (47), this individual would be sufficient for the identification of the referent,in (48), the pronoun I (am) would, and in (49), the name Mickey Mouse would(cf. Matushansky (2008: 595ff.) for a recent analysis of this last construction).All three descriptions are embedded in a downward entailing modal context,and seem to give the same interpretive effect as the above-studied appositives:they may not convey new, or even nontrivial, information, but the informationthat they convey plays a critical role for the evaluation of the modal predicate.What made it surprising to the pharmacist that this individual was interested incondoms was the property of being an old woman, etc. So at least with respectto the contexts in focus in this paper, a parallel treatment of several, if not all,sorts of nonrestrictive modifiers appears as desirable – and to the extent that the

Sinn und Bedeutung 15 - [PDF Document] (98)

Appositives in Modal Contexts 97

different sorts turn out to display a different behaviour in other environments,for instance, in simplex sentences, it is to be hoped that this can eventually bedescribed as a variation within the general frame of analysis developed above:Supplements, including nominal appositives, nonrestrictive relatives, and thetypes of nonrestrictive modifiers exemplified in (47)–(49), can contribute tothe content of a clause because they introduce presupposition-like conditions.

5 ConclusionsThe point of departure for this study was the following double claim made byHarris and Potts (2009) concerning the perspectival orientation of appositives:

1. Non-speaker oriented readings are independent of intensionality; theyare possible even in extensional contexts.

2. However, speaker-oriented readings predominate even in intensional con-texts.

I have disputed both claims, the first by arguing that the relevant extensionalcontexts are intensional after all, in the spirit of Hintikka (1973: 214):

Surprisingly often modal notions are tacitly being considered in apparently non-modal contexts.

The second claim was countered by pointing out that the intensional contextsconsidered by Harris and Potts are not representative; in fact, once downwardentailing attitude contexts are taken into consideration, non-speaker orientedreadings predominate. In turn, this realization paves the way for a recognitionthat appositives are not generally vehicles of expressive meaning, orthogonalto descriptive meaning, but that they do sometimes help to build propositions.The prime piece of evidence for that consists in the observation that weakeningappositives fails to preserve the truth of certain attitude ascriptions, primarilyascriptions of surprise. If a pharmacist is surprised that Mrs. Otis, an old lady,is interested in condoms, it does not follow that she is surprised that Mrs. Otis,a lady, is interested in condoms. Another fact which supports the ‘semanticstrategy’ is that appositives in surprise contexts often represent old information.

The function that appositives fill in this type of context is best understoodagainst the background of the theory of emotive factive attitudes like surprised.When we ascribe surprise to somebody at a certain individual having a certainproperty (like being pregnant or interested in condoms), we tacitly understandsome additional relevant properties (like being a virgin or an old lady), andthese are encoded in a contextual parameter providing a set of relevant (‘old’)belief worlds. If, for fear of being misunderstood because the context mightnot be sufficiently clear in this regard, we want to make a property explicit,

Sinn und Bedeutung 15 - [PDF Document] (99)

98 Sæbø

expressing it in the form of an appositive is a natural move.Thus in modal contexts, appositives act as reminders of properties which

are to follow an individual along through the set of worlds under consideration.They provide a means to ensure that a thing retains its relevant properties acrossthe modal space. This task can be argued to be more important in downwardentailing than in upward entailing modal contexts, and that could be the reasonthat (non-speaker oriented) appositives are more frequent there.

For modelling this behaviour of appositives in a way that also allows forspeaker-oriented interpretations, in extensional as well as intensional contexts,a treatment in terms of partial meanings along the lines of Beaver and Krahmer(2001) commends itself. I hope to have shown how such an approach can yieldcoherent results in a fully compositional fashion: Starting from an abstract so-called appositive functor, appositives add definedness conditions to individualconcepts, conditions that become definedness conditions on higher intensions,finally on propositions. When these propositions are embedded under attitudes,those definedness conditions (on the usual, de dicto construal) become integralparts of them (to be exact, their positive extensions), and thus the intuition thatagent-oriented appositives convey attitude-relevant properties is accounted for.

As far as speaker-oriented interpretations are concerned, it is difficult todiscern a decisive difference between a ‘bidimensional’ view of appositives asbelonging to the dimension of implicature, to do with expressive meaning only,and a ‘unidimensional’ view where appositives give definedness conditions onpartial concepts and propositions. Once we take non-speaker oriented readingsseriously, however, not only in upward entailing but also in downward entailingmodal contexts, it becomes clear that, at the very least, the two ‘dimensions’should be able to communicate: the piece of content expressed in the appositiveshould be allowed to merge with the larger piece expressed in its local context.The present paper has presented one way to produce this effect.


Amaral, Patricia, Craige Roberts and E. Allyn Smith. 2007. Review of TheLogic of Conventional Implicatures by Chris Potts. Linguistics and Phi-losophy 30, 707–749.

Anand, Pranav. 2007. Re-expressing judgment. Theoretical Linguistics 33,199–208.

Asher, Nicholas. 1987. A Typology for Attitude Verbs and their AnaphoricProperties. Linguistics and Philosophy 10, 125–197.

Beaver, David and Emiel Krahmer. 2001. A Partial Account of PresuppositionProjection. Journal of Logic, Language and Information 10, 147–182.

Sinn und Bedeutung 15 - [PDF Document] (100)

Appositives in Modal Contexts 99

Donnellan, Keith. 1966. Reference and definite descriptions. PhilosophicalReview 75, 281–304.

von Fintel, Kai. 1999. NPI Licensing, Strawson Entailment, and Context De-pendency. Journal of Semantics 16, 97–148.

von Fintel, Kai and Irene Heim. 2009. Intensional Semantics. Lecture Notes,MIT.

Geurts, Bart and Emar Maier. 2005. Quotation in Context. Philippe De Bra-banter (ed.) Hybrid Quotations (= Belgian Journal of Linguistics 17).Amsterdam: John Benjamins. 109–128.

Harris, Jesse Aron and Christopher Potts. 2009. Perspective-shifting with ap-positives and expressives. Linguistics and Philosophy 32, 523–552.

Heim, Irene. 1992. Presupposition Projection and the Semantics of AttitudeVerbs. Journal of Semantics 9, 183–221.

Heim, Irene. to appear. Definiteness and Indefiniteness. Claudia Maienborn,Paul Portner and Klaus von Heusinger (eds.) Semantics: an Interna-tional Handbook of Natural Language Meaning. Berlin: Mouton deGruyter.

Hintikka, Jaakko. 1973. Grammar and Logic: Some Borderline Problems.Jaakko Hintikka et al. (eds.) Approaches to Natural Language. Dor-drecht: Reidel. 197–214.

Kadmon, Nirit and Fred Landman. 1993. Any. Linguistics and Philosophy 16,353–422.

Matushansky, Ora. 2008. On the linguistic complexity of proper names. Lin-guistics and Philosophy 21, 573–627.

Potts, Christopher. 2005. The Logic of Conventional Implicatures. OxfordStudies in Theoretical Linguistics. Oxford: Oxford University Press.

Schlenker, Philippe. 2010. Supplements within a Unidimensional SemanticsI: Scope. Maria Aloni, Harald Bastiaanse, Tikitu de Jager and KatrinSchulz (eds.) Logic, Language and Meaning (17th Amsterdam Collo-quium Revised Selected Papers). Lecture Notes in Computer Science6042. Berlin: Springer. 74–83.

Schlenker, Philippe. to appear. Supplements within a Unidimensional Seman-tics II: Epistemic Status and Projection. Patrick Grosz, Irene Heim,Pritty Patel and Igor Yanovich (eds.) Proceedings of NELS 40. GLSAPublications, University of Massachusetts, Amherst.

Schwager, Magdalena. to appear. Speaking of Qualities. Satoshi Ito and EdCormany (eds.) Proceedings of SALT 19. Ithaca, New York: CLC Pub-lications.

Sharvit, Yael. 2002. Embedded Questions and ‘De Dicto’ Readings. NaturalLanguage Semantics 10, 97–123.

Sinn und Bedeutung 15 - [PDF Document] (101)

100 Sæbø

van Rooij, Rob. 2006. Attitudes and Changing Contexts. Berlin: Springer.

Sinn und Bedeutung 15 - [PDF Document] (102)

An Unaccusativity Diagnostic at the Syntax-­‐Semantics Interface: there-­‐Insertion, Indefinites and Restitutive again

Artemis Alexiadou

Universität Stuttgart [emailprotected]

Florian Schäfer Universität Stuttgart


Abstract. This paper argues that the unaccusativity mismatch observed in the literature concerning the availability of there-insertion points to a syntactic difference between two classes of unaccusatives. The paper shows that the theme argument of change-of-location unaccusatives occupies Spec,ResultP, while that of change-of-state unaccusatives occupies Spec,vP. Insertion of there is blocked in the latter case, as the theme and the expletive compete for the same position.

1 Introduction As is well known, there-insertion is possible in the context of unaccusative verbs but impossible with unergative and transitive verbs.1 However, only a subset of unaccusative verbs allows there-insertion (Levin 1993), leading to an unaccusativity mismatch (1a vs. 1d).

(1) a. There arrived a man (in the garden) (unaccusative-1) b. *There walked a man (in the garden) (unergative) c. *There kissed a girl a boy (in the garden) (transitive) d. *There broke a glass (in the kitchen) (unaccusative-2)

In this paper, we argue that this unaccusativity mismatch points to a syntactic difference between the two classes of unaccusatives. Building on the “low-there” hypothesis, recently proposed by Richards & Biberauer (2005), Richards (2007) and Deal (2009), we argue that the theme argument of the two classes of unaccusatives can occupy different structural positions within the vP, namely Spec,vP and Spec,ResultP, see (2a vs. 2b). Insertion of there is blocked, if the theme DP obligatorily occupies Spec,vP, because the two compete for the same position. This is the case for break-type verbs.

1 We only discuss “presentational there”, i.e. expletive there in the context of lexical verbs. We will not be concerned with expletive there in the context of the copula be (i.e. in progressives, passives and existentials); see Deal (2009) for a recent discussion within the “low-there”-hypothesis applied here.

In Reich, Ingo et al. (eds.), Proceedings of Sinn & Bedeutung 15, pp. 101–115. Universaar – Saarland University Press: Saarbrücken, Germany, 2011.

Sinn und Bedeutung 15 - [PDF Document] (103)

102 Alexiadou & Schäfer

(2) a. [vP there [ResultP theme ]] b. [vP *there/theme [ResultP ]]

2 The Standard Account of there-­‐Insertion: there in Spec,TP Chomsky (1981, 1995 and subsequent work) proposes that there is externally merged in the derived subject position Spec,TP to satisfy the Extended Projection Principle (EPP) (i.e. to check the strong D-feature on T). On this logic, (1b, c) are ungrammatical for the following reason: in English, a language lacking Transitive Expletive Constructions (TECs), the subject and the expletive compete for a single specifier position, Spec, TP.2 In TEC-languages such as Dutch in (3), the counterparts of (1b, c) are grammatical because these languages have two specifier positions available for subjects outside the vP.

(3) dat er iemand een appel gegeten heft (TEC) that there someone an apple eaten has

Note that this standard analysis of there-insertion cannot account for the contrast in (1a, d) (cf. also Borer 2005, Deal 2009, Alexiadou 2011).

3 Against the Standard Analysis: there down in Spec,vP The standard analysis of there-insertion has recently been challenged by Richards & Biberauer 2005, and Richards 2007 (see also Deal 2009) as it faces a number of problems which we can touch upon here only super-ficially. In Chomsky (2000, 2001, 2004), there is a head with [uF] and probes from Spec,TP into TP and values T. This proposal faces a technical problem: only root nodes should probe. Since there in Spec,TP is not the root node (which is T), its probing is counter-cyclic. Moreover, it needs a number of extra assumptions to derive Bure’s Generalization, i.e. the observation that TECs are available only in languages with Object Shift(OS)/Scrambling of full DPs. (Why should the availability of a second specifier in the TP-region be related to the availability of a derived object position? (Cf. Alexiadou & Anagnostopoulou 2001, 2007, Richards 2004).

As the aforementioned authors argued, for conceptual reasons, MERGE-Expletive should be a property of phase heads (C, v), i.e. expletives are externally merged either in Spec,CP or in Spec,vP. If an expletive occurs in

2 A crucial assumption is that the subject must leave the vP (see Alexiadou & Anagnostopoulou 2001, 2007).

Sinn und Bedeutung 15 - [PDF Document] (104)

There-Insertion 103

Spec,TP, it must have moved there. The consequence is that the EPP on T is checked only via MOVE.

In agreement with Richards & Biberauer (2005) and Richards (2007), we conclude that i) there is not a probe but a goal (like any other nominal category/DP); ii) it merges in Spec,vP where it is in the probe domain of T; iii) it has the interpretable but incomplete φ-feature set [person], rendered active via an unvalued Case feature.3 It is probed by T and gets its case valued. T’s φ-feature set remains unvalued as there is φ-incomplete. T there-fore remains active for Agree with the associate DP. Afterwards, there moves to Spec,TP.

As expletives are dummies (they do not have reference and cannot bear a theta-role), they can merge (externally) only in non-thematic specifiers, i.e. as a) the specifier of a defective head vpassive; b) the specifier of a defective head vunaccusative or c) the outer specifier of thematic v/Voice (the OS-position).

The third option determines the availability of TECs; English lacks both TECs and OS as it has no outer Spec,vP/outer Spec,VoiceP (the complementarity between Expl and external arguments is due to the mutual exclusivity of thematic and non-thematic v in English).4 It also explains the complementarity between expletives and raised internal arguments of unaccusatives in (4)-(5); both target the non-thematic specifier of vPunaccusative. Note that under the traditional Expl-in-TP-approach (4c)/(5b) should be fine. It also explains why in languages that have both OS and TECs the former bleeds the latter (6): again, we have competition for the same position.

(4) a. *There seems [TP a man to be ta man in the garden] b. There seems [TP to be a man in the garden] c. *[TP There [VP a man [VP arrived ta man]]] d. There arrived a man

(5) a. dat *(daar) gister ’n skip gesink het (Afrikaans) that (there) yesterday a ship sunk has b. dat (*daar) ’n skip gister gesink het

3 But see Deal (2009) for the claim that there must have uninterpretable φ-features and that it locally probes the associate DP. This, she claims, is necessary in order to avoid the “too-many-theres” problem (*There seemed there to arrive a train in the station). We do not discuss the feature content of there but concentrate on its configurational, i.e. external-merge properties. 4 Something in addition has to be said about cyclic A’-movement of vP-internal elements which is, of course, possible in English.

Sinn und Bedeutung 15 - [PDF Document] (105)

104 Alexiadou & Schäfer

(6) a. *dat er veel mensen dat boek gisteren gekocht hebbe that EXPL many people the book yesterday bought have (Dutch) b. dat daar baie mense baie/*die bier gedrink het (Afrikaans) that EXPL many people many/the beer drunk have

The conclusion we can draw for English is that there is blocked if i) an external argument occupies the specifier of v/Voice or ii) an object raises to Spec,vdefective in passive or unaccusative structures.

Now, recall our mismatch within the class of unaccusatives in (1a vs. 1d) replicated in (7):

(7) a. There appeared a man in the garden. b. *There melted a lot of snow on the streets of Chicago.

The verb’s influence on there-insertion holds also in raising constructions (Deal 2009).

(8) a. There seemed to appear a dagger in front of Macbeth. b. *There seemed to melt a lot of snow on the streets of Chicago.

Ideally, we should be able to explain the contrast between the two classes of unaccusatives along the same lines as the contrast between e.g. transitives and passives. More concretely, appear/arrive-verbs should make available an empty Spec,vP where there can merge, while melt/break-verbs should not make available such an empty Spec,vP; it follows then that Spec,vP of break-unaccusatives must be occupied. The question then is: what is located in Spec,vP of melt/break-unaccusatives?

4 What Does there Correlate with? Two Classes of Unaccusatives

4.1 A Classification of Verbs Allowing there-­‐Insertion

Levin (1993) characterizes the verbs allowing there-insertion roughly as verbs of existence or appearance. They can be broken down into the following subclasses (a-f) of unaccusatives. Verbs of change of state (g) do not permit there although they are unaccusatives too:5 5 Levin (1993) points out that verbs of manner of motion also allow for there in the context of directional PPs, but they differ in that the subject must follow this PP. (i) a. There arrived three gentlemen from Verona. b. ??There arrived from Verona three gentlemen. (ii) a. *There ran a raggedy looking cat into the room. b. There ran into the room a raggedy looking cat.

Sinn und Bedeutung 15 - [PDF Document] (106)

There-Insertion 105

(9) a. Verbs of Existence: blaze, bubble, cling, coexist, … b. Verbs of Spatial Configuration: crouch, dangle, hang, kneel, … c. Meander Verbs: cascade, climb, crawl, cut, weave, wind, … d. Verbs of Appearance: accumulate, appear, arise, … e. ?Verbs of disappearance: die, disappear, vanish, … f. Verbs of Inherently Directed Motion: arrive, ascend, come, … g. *Verbs of Change of State: bend, break, chip, rip, shatter, split, …

4.2 Is there a Causative Event in Spec,vP (Deal 2009)?

Deal (2009) offers an account for the contrast between the two classes of unaccusatives concerning there-insertion that strongly influenced our analysis. Specifically, she proposes that there is inserted at the edge of a vP that lacks an external argument, i.e. into a non-thematic Spec,vP position. Unaccusatives rejecting there have Spec,vP already occupied by a causative event.

While we are sympathetic with her blocking account, her proposal faces a number of theoretical and empirical problems which we will not discuss here for reasons of space. Below, we make an alternative proposal about what blocks there-insertion at the edge of vP which, in turn, strongly builds on the work by Dobler (2008a, b).

4.3 Hypothesis: There Is an Internal Argument in Spec,vP

Hale & Keyser (2000) assume two different lexical syntactic representations for unaccusatives. With verbs such as arrive, occur, …, the theme is introduced within the complement of the verb, in the specifier of a small-clause headed by a (potentially covert) P-projection (10a). With verbs such as break, open, … the theme is introduced in the specifier of the verb that takes an adjective as its complement (10b). ((10b) is a composite dyadic lexical projection, also called a complex predicate; see e.g. Beck & Johnson 2004, Embick 2004, McIntyre 2004).

(10) a. [vP arrive [PP many guests [ Pcovert/in the garden ]]] b. [vP the sky [ v Aclear ]]

Hale & Keyser do not actually propose this solution, but with the background of the “low-there”-hypothesis discussed above, these structures

(iii) Suddenly there flew through the window [that shoe on the table] Cases such as (iib) are called “outside verbals” in Deal (2009). Outside verbals do not obey the definiteness restriction (iii) and allow “a bewildery variety of verbs” (Milsark 1974). See Deal (2009) for an analysis of these cases. We concentrate here on “inside verbals”.

Sinn und Bedeutung 15 - [PDF Document] (107)

106 Alexiadou & Schäfer

could, in principle, explain the distribution of there in the context of unaccusatives. With clear-type predicates, Spec,vP is already occupied by the theme argument; with arrive-type predicates, Spec,vP is available. In the next section, we investigate whether this is the correct explanation for the unaccusativity mismatch observed with there.

5 Tracing the Position of Internal Arguments Both structures in (10) above are bi-eventive/resultative. They differ con-cerning the position where the theme argument is merged; either it is merged as the argument of the lower-event small-clause or as the argument of the higher-event verb. Over the years, there has been a lot of discussion about the correct analysis of resultative structures. Some authors argued that the small-clause analysis is generally correct (e.g. Hoeckstra 1988), some claimed that the complex-predicate analysis is generally correct (e.g. Beck & Johnson 2004).

Dobler (2008a, b) discusses transitive, resultative constructions and con-cludes that both structures co-exist. The small-clause analysis is correct for transitive resultatives referring to a change of location, i.e. the position of the object in (11a) is similar to that of the theme argument in the unaccusative structure in (10a). The complex-predicate analysis is correct for transitive resultatives referring to a change of state, i.e. the position of the object in (11b) is similar to that of the theme argument in the unaccusative structure in (10b).

(11) a. Thilo sent the plane to Yubara . b. He wiped the floor clean.

To determine this, she investigated whether an existential operator in object position can be part of the presupposition of restitutive again. In what follows, we summarize her argumentation.

5.1 The Interaction of wieder/again and Existential Operators in Object Position

Von Stechow (1996) argued in detail that the repetitive vs. restitutive interpretation associated with the adverb again is the result of a structural ambiguity. Evidence for this is provided by word order facts such as the ones in the German examples in (12), where the syntax disambiguates between the two interpretations. Von Stechow took this as evidence for the syntactic decomposition of the VP into a vP and a ResultP component.

(12) a. Thilo öffnet die Tür wieder

Sinn und Bedeutung 15 - [PDF Document] (108)

There-Insertion 107

Thilo opened the door again i) He had opened the door before (repetitive & ii) The door used to be open restitutive) b. Thilo öffnet wieder die Tür (only repetitive)

As is well known, German definite objects always leave the vP, cf. (13) (von Stechow 1996, Dobler 2008a, b modifying Webelhuth 1992). The examples in (14) show that if the adverb again precedes the theme, it has necessarily wide scope over vP and ResultP, leading to a repetitive reading. If the adverb follows the theme, the adverb might either outscope just the ResultP, leading to a restitutive reading, or once again both the vP and the ResultP leading to the repetitive reading.

(13) a. weil er (wohl) das Buch (wohl) gelesen hat as he particle the book particle read has b. weil er (wohl) [vP das Buch [vP(wohl) [vP tsubj tobj lesen]]

(14) a. weil er wieder die Tür geöffnet hat as he again the door opened has a’. wiederrepetitive [ die Tür [vP tsubject v [AP tobj offen b. weil er die Tür wieder geöffnet hat as he the door again opened has b’. [ die Tür [(wiederrepetitive) [vP tsubject v [AP (wiederrestitutive) tobject offen

(15) shows that indefinite objects remain inside the vP (unless they get a strong interpretation). As shown in (16), this is compatible with both the small-clause analysis as well as the complex predicate analysis of re-sultatives, if we assume that the subject is introduced by an extra projection (VoiceP):

(15) weil er (wohl) ein Buch (*wohl) gelesen hat as he particle a book particle read has

(16) [VocieP Subject Voice [vP (Object) v [ResultP (Object) state]]]

Von Stechow (1996) only discusses the interaction of definite DPs and again. Nissenbaum (2006) investigates scope-interactions between again and in-definites. In (17), we get different readings, depending on where the indefinite is interpreted, within the vP or in the IP.

(17) Someone is sneezing again a. again [∃x.x is sneezing] (different person) b. ∃x [x is sneezing] (same person) c. [IP Someonei is [ [vP ti sneezing] again]

Sinn und Bedeutung 15 - [PDF Document] (109)

108 Alexiadou & Schäfer

In German, the readings are determined once again by the surface order:

(18) a. weil [wieder [VoiceP jemand [vP nießt (different person) b. weil [VoiceP jemand [vP wieder nießt (same person) as (again) someone (again) sneezes

Dobler (2008a, b) uses the scope-interaction between restitutive again and an indefinite object to investigate the position of the internal argument in transitive resultative constructions. The main goal of her investigation is to determine whether the internal argument is an argument of the result state (small-clause (SC) analysis) or of the verb (complex-predicate (CP) analysis). Importantly, only the small-clause analysis predicts that the existential operator can be interpreted inside the result-state clause, i.e. inside the presupposition triggered by restitutive again (cf. 19).6

(19) [VoiceP Subject Voice [vP (Objectindef) v [againrest [RP (Objectindef) state ]]]] | | same potentially different

Dobler argues that the following interpretative picture emerges (in both, English and German):

(20) Change of state: Pandora scrubbed a donkey clean again a. #again [∃x.x is a donkey and x is clean] (SC-analysis) b. ∃x.x is a donkey and again [x is clean] (CP-analysis)

(21) Change of location: Pandora put a donkey in her stable again a. again [∃x.x is a donkey and x is in Pandora’s stable] (SC-analysis) b. ∃x.x is a donkey and again [x is in Pandora’s stable] (CP-analysis)

Dobler concludes that the theme is (syntactically) the argument of the verb (vP) in change-of-state resultatives, while it is the argument of the secondary predicate (ResultP) in change-of-location resultatives. In the latter case, it can, of course, move out of the scope of again yielding reading (21b).

Below we list some further examples provided by Dobler (2008a) which test whether the relevant reading (restitutive again outscopes the indefinite theme) is available or not in English and German. (22)-(23) illustrate the situation with change-of-state predicates, (24)-(25) illustrate it with change-

6 The #-sign indicates that a reading ‘restitutive again > indefinite theme’ is not available.

Sinn und Bedeutung 15 - [PDF Document] (110)

There-Insertion 109

of-location predicates. The contexts given before the test sentences are meant to exclude an irrelevant repetitive reading and force a restitutive reading:7

(22) a. Context: Sally owns a brown mouse and a great number of white mice. While she is gone, Harry takes care of them and the brown mouse dies. Harry is freaked out and wants to cover up the loss…

b. #Er färbt wieder eine Maus braun. c. #He dyes a mouse brown again.

(23) a. Context: Yesterday, Sally visited a popsicle factory. There she had the opportunity to taste the popsicle mixture before it was frozen. She really loved it.

b. #Daheim angekommen hat Sally wieder ein Eis am Stiel geschmolzen.

c. #Once she was home, Sally melted a popsicle again.

(24) a. Context: Until about 200 years ago, bears used to live in the Alps. b. Gestern haben Biologen wieder Bären in den Alpen angesiedelt. c. Yesterday, scientists put bears in the Alps again.

(25) a. Context: The island had a mountain that practically disappeared in the course of an earthquake.

b. Die Bewohner der Insel haben wieder einen Berg errichtet. c. The inhabitants constructed a mountain again.

5.2 Conclusion

To conclude, Dobler (2008a, b) shows that there are two classes of transitive bi-eventive verbs that differ in whether the indefinite/existential object can be in the scope of restitutive again or not. Below, we list some further verbs of these two classes:

Group A: #restitutive again > existential operator

7 The repetitive reading (repetitive again > indef) is available in English and German but it is irrelevant for the present argumentation. The sentences in (22b, c) have therefore the following interpretative properties: (i) a. again [∃x.x is a mouse and x is brown] impossible reading (restitutive) There is a brown mouse and there was a (different) brown mouse. b. again [∃x.x is a mouse and x is dyed brown] possible reading (repetitive) A mouse is (being) dyed brown and at a previous time, there was a (different) mouse that was (being) dyed brown.

Sinn und Bedeutung 15 - [PDF Document] (111)

110 Alexiadou & Schäfer

melt, freeze, cool, warm, empty, fill, open, close, paint (in) pink, dye brown, …

Group B: restitutive again > existential operator put, place, donate, construct, build, …

At first sight, it seems that group A contains verbs undergoing the causative alternation. However, this does not seem to be the correct generalization, as group B contains such verbs too (e.g. German (sich) ansiedeln and its English counterpart settle). Group A contains de-adjectival verbs, but we get the same result if we replace for example “paint pink” with “paint in pink”. The correct generalization is a division into change of state verbs and change of location verbs (as well as creation verbs ≈ cause to be in a location (see Dobler 2008a, b for detailed discussion)).

To explain these differences, we conclude with Dobler (2008a, b) that the direct object of change-of-state predicates is necessarily located outside of the Result phrase (when scope is computed). With change-of-location predicates, we note the reverse situation; the direct object can be located inside of the Result phrase (when scope is computed). Following Hale & Keyser (2000), we assume the structures in (26) for these two types of (transitive) verbs/predicates. (For structural variants of (26a, b) which are, in principle, compatible with the above findings, see Beck & Johnson (2004), von Stechow (2007), Dobler (2008a, b) or Ramchand (2008)).

(26) a. [VoiceP subject Voice [vP object v [RP Result ]]] (change-of-state) b. [VoiceP subject Voice [vP v [RP object Result ]]] (change-of-location)

6 On the Position of the Subjects of Unaccusatives; Are They Blocking there-­‐Insertion in Spec,vP?

Dobler (2008a, b) investigated transitive constructions while we are interest-ed in unaccusatives. Many of the verbs in Group A discussed in Dobler’s work express a change of state and have an unaccusative counterpart. If the widely held assumption is correct that the object of transitives has the same syntactic base position as the sole argument of unaccusatives, we expect that the unaccusative counterparts of these verbs should behave alike in terms of scope interaction between an indefinite theme argument and restitutive again. The transitives in Group B are change-of-location verbs. As noted in 4.1, the unaccusatives allowing there-insertion also express a change of location (come into existence ~ come to be in a location). We thus predict that these verbs should behave like transitive change-of-location verbs as far as scope interaction between an indefinite theme argument and restitutive again is

Sinn und Bedeutung 15 - [PDF Document] (112)

There-Insertion 111

concerned. In other words, if the argument of change-of-state unaccusatives is responsible for the blocking of there-insertion, it should necessarily be located outside the ResultP in the specifier of the un-accusative vP and it should never occur inside the scope of restitutive again. On the other hand, the argument of change-of-location un-accusatives should be located inside the ResultP and thus inside the scope of restitutive again; if it stays there, it does not block there-insertion in Spec,vP. These predictions are indeed borne out, as shown in the following two sub-sections.8

6.1 Verbs of Change of Location

Unaccusative verbs of appearance (27) and unaccusative verbs expressing an inherently directed motion (28) both allow, as predicted, the relevant reading where the indefinite/existential theme argument is in the scope of restitutive again.9

(27) a. Context: Until about 200 years ago, bears used to live in Bavaria, but they were completely wiped out by the inhabitants in the 19th century.

b. Letzten Sommer ist wieder ein Bär in Bayern aufgetaucht/erschie-nen.

c. Last summer, a bear appeared in Bavaria again.

(28) a. Context: Until about 200 years ago, bears used to live in Bavaria, but they were completely wiped out by the inhabitants in the 19th century.

b. Letzten Sommer ist wieder ein Bär nach Bayern gekommen. c. Last summer, a bear/bears came to Bavaria again.

6.2 Verbs of Change of State10 Unaccusative verbs of change of state, on the other hand, do not allow the relevant reading; their indefinite/existential theme argument cannot be

8 We would like to thank Eva Dobler (German), Andrew McIntyre, Walter Pederson, Marc Richards and Mike Putnam for their judgements. 9 Levin (1993) notes that verbs of disappearance allow there-insertion marginally. Deal (2009) argues that these verbs do not allow there-insertion. We do not discuss this class here, as it is hard to test (see also footnote 10). 10 There is a general complication with change-of-state verbs. Many of these verbs express “the disruption of material integrity” (Levin 1993). Since we are interested in a restitutive reading, these verbs are complicated to test; how can something start out broken, become united and break again?

Sinn und Bedeutung 15 - [PDF Document] (113)

112 Alexiadou & Schäfer

interpreted as being in the scope of restitutive again. (Repetitive again can outscope the indefinite theme but this reading is irrelevant for the argument.)

(29) a. Context: Yesterday, Sally visited a popsicle factory. There she had the opportunity to taste the popsicle mixture before it was frozen. She really loved it.

b. #Daheim angekommen ließ sie wieder ein Eis am Stiel schmelzen. c. #Once she was at home she made/let a popsicle melt again.

(30) a. Context: Many years ago, a type of squirrel existed which was yellow. Unfortunately, they all died due to a mysterious infection.

b. #Forscher haben es geschafft, dass sich in einem Labor wieder ein Eichhörnchen gelb gefärbt hat. c. #Scientists working in a Swiss laboratory managed to bring it about

that a squirrel turned yellow again.

6.3 ‘Verbs of Change of State’ under a ‘Come into Existence’ Reading

In addition to its use as a verb of change of state, the verb break also has a use as a verb of coming into existence, as in “The war broke (out)”. Similarly, the verb open has an appearance sense which can be paraphrased as ‘become visible’ or ‘come into existence’ in addition to its change-of-state sense. The question then is whether this difference is relevant for the availability of there-insertion. The judgements of our four informants shown in (31a, b) vs. (31c, d, e) suggest that such an effect indeed exists at least as a general tendency (1[(very good] - 5[very bad]).

(31) a. There broke a vase in the living room 5 4 4 4 b. There opened a window in the living room 5 4 3 5 c. During the spring, there suddenly broke (out) a war in west India 5 2 2.5 2 d. Suddenly, there opened a cavity underneath their feet 1 2 2 3 e. Suddenly, there opened a gap in the middle of the street 3 1 4 3

Crucially, and in accordance with our overall proposal, the ‘come into existence’ reading of these basically change-of-state verbs makes available the scope againrestitutive > indefinite:

(32) a. Context: For hundreds of years, people could get into the mountain through a small hole/crack. After a strong earthquake, this entrance was blocked. But after a long period of rain,

b. A hole opened in the rock again which allowed people to enter.

Sinn und Bedeutung 15 - [PDF Document] (114)

There-Insertion 113

c. Im Laufe der Zeit hat sich aber wieder eine Lücke geöffnet.

(33) a. Context: When we started here, all the walls were covered with numerous gaps and holes which we closed with great effort.

b. But during the storm, a huge gap opened again. c. Durch den Sturm hat sich plötzlich wieder ein Spalt in der Wand


This suggests that the relevant parameter is not strictly syntac-tic/categorial (adjectival vs. prepositional), but semantic/conceptual (change-of-state vs. change of location/existence). However, this semantic parameter is syntactically reflected in the position available for the theme.

We thus conclude that the theme of change-of-location verbs originates inside the Result phrase where it can stay in principle. The theme of change-of-state verbs is obligatorily located in Spec,vP, not in the Result phrase. There-insertion is blocked in the latter context as it competes with the theme argument, see (34).

(34) a. [vP there [ResultP theme]] vs. b. [vP theme/*there [ResultP ]]

7 Conclusion In this paper, we argued that the unaccusativity mismatch observed in the literature concerning the availability of there-insertion points to a syntactic difference between two classes of unaccusatives. We showed that the theme argument of change-of-location unaccusatives occupies Spec, ResultP, while that of change-of-state unaccusatives occupies Spec,vP. Insertion of there is blocked in the latter case, as the theme and the expletive compete for the same position.

References Alexiadou, Artemis. 2011. Post-verbal nominatives: an unaccusativity

diagnostic under scrutiny. In R. Folli & C. Ulbrich (eds.), Interfaces in Linguistics: New Research Perspectives. Oxford: Oxford University Press, 56–77.

Alexiadou, Artemis & Elena Anagnostopoulou. 2001. The subject in situ generalization, and the role of Case in driving computations. Linguistic Inquiry 32, 193–231.

Alexiadou, Artemis & Elena Anagnostopoulou. 2007. The subject in situlenageneralization revisited. In H.-M. Gärtner & U. Sauerland (eds.), Interfaces + Recursion = Language. Berlin: Mouton de Gruyter, 31–60.

Sinn und Bedeutung 15 - [PDF Document] (115)

114 Alexiadou & Schäfer

Beck, Sigrid & Kyle Johnson. 2004. Double objects again. Linguistic Inquiry 35:1, 97–124.

Borer, Hagit. 2005. Structuring sense, Vol. II. The normal course of events. Oxford University Press.

Chomsky, Noam. 1981. Lectures on government and binding. Dordrecht: Foris.

Chomsky, Noam. 1995. The Minimalist program. Cambridge, MA: MIT Press.

Chomsky, Noam. 2000. Minimalist inquiries: The framework. In R. Martin, D. Michaels & J. Uriagereka (eds.), Step by step: Essays on minimalist syntax in honor of Howard Lasnik. Cambridge: MIT Press, 89–155.

Chomsky, Noam. 2001. Derivation by Phase. In M. Kenstowicz (ed.), Ken Hale: A Life in Language. Cambridge, MA: The MIT Press, 1–52.

Chomsky, Noam. 2004. Beyond Explanatory Adequacy. In A. Belletti (Ed.), Structures and Beyond: The Cartography of Syntactic Structures, Vol. 3. Oxford, Oxford University Press, 104–131.

Deal, Amy Rose. 2009. The origin and content of expletives: evidence from "selection". Syntax 12:4, 285–323.

Dobler, Eva. 2008a. ‘Again’ and the structure of result states. In Proceedings of ConSOLE XV.

Dobler, Eva. 2008b. Creating as putting something into the world. In Proceedings of ConSOLE XVI.

Embick, David. 2004. On the structure of resultative participles in English. Linguistic Inquiry 35.3, 355–392.

Hale, Ken & Samuel Jay Keyser. 2000. There-insertion unaccusatives. Ms., MIT.

Hoekstra, Toen. 1988. Small clause results. Lingua 74:101–139. Levin, Beth. 1993. English verb classes and alternations: a preliminary

investigation. Chicago: University of Chicago Press. McIntyre, Andrew. 2004. Event Paths, Conflation, Argument Structure and

VP Shells. Linguistics 42:3, 523–571. Milsark, Gary. 1974. Existential sentences in English. Ph.D. dissertation,

MIT. Nissenbaum, Jon. 2006. Decomposing resultatives: two kinds of restitutive

readings with ‘again’. Poster presented at NELS 37, University of Illinois, Urbana-Champaign, October 2006.

Ramchand, Gillian. 2008. Verb meaning and the lexicon: a first phase syntax. Cambridge University Press.

Sinn und Bedeutung 15 - [PDF Document] (116)

There-Insertion 115

Richards, Marc. 2007. On object shift, phases, and transitive expletive constructions in Germanic. In P. Pica et al. (eds), Linguistic Variation Yearbook 6. Amsterdam, John Benjamins, 139–159.

Richards, Marc & Theresa Biberauer. 2005. Explaining Expl. In M. den Dikken & C. Tortora (eds.), The function of function Words and functional categories. Amsterdam: John Benjamins, 115–153.

Richards, Norvin. 2004. A distinctness condition on linearization. Ms, MIT. von Stechow, Arnim. 1996. The different readings of wieder ‘again’: a

structural account. Journal of Semantics 13, 87–138. von Stechow, Arnim. 2007. Syntactic and lexical causativization: BECOME

and CAUSE again. Talk presented at ConSOLE XV, Brussels. Webelhuth, Gert. 1992. Principles and parameters of syntactic saturation.

Oxford University Press.

Sinn und Bedeutung 15 - [PDF Document] (117)

Sinn und Bedeutung 15 - [PDF Document] (118)

In Reich, Ingo et al. (eds.), Proceedings of Sinn & Bedeutung 15,pp. 117–132. Universaar – Saarland University Press: Saarbrücken, Germany, 2011.

Optatives: Deriving Desirability from Scalar Alternatives∗

María BiezmaUMass Amherst


Abstract. A compositional analysis of optative sentences is challenging for atleast two reasons: they encode desirability without having any overt marker, andthey are if -clauses with or without consequents, raising the question of whetherthey are actually conditionals. In this paper I argue that optatives are condition-als even when they do not have overt consequents. With respect to desirability, Iargue that in optatives modality is pragmatically derived. The investigation of op-tatives sheds light on the interaction between syntax, pragmatics and discourse.

1 Introduction: Where Does Desirability Come from?Structures like (1) are known as optatives in the literature, and they presentchallenges in several respects.

(1) If only I had been taller, I would have played in the NBA.

The utterance of an optative like (1) signals the speaker’s desires, and yet thereis no lexical item encoding desirability. Notice that what is desired when aconditional optative is uttered is not the antecedent proposition, i.e. that thespeaker were taller. What the speaker desires is the consequent, (2).

(2) A: If only I had been taller, I would have played in the NBA.B: That would not have been necessary, you were such a great player!

What would have made a difference was if you had been in a bettercollege team.

A: Yeah. . . !, you are right. . . , If only I had played for UCLA, I wouldhave played in the NBA.

The dialogue in (2) illustrates that what is really desired is not being taller orhaving played for UCLA. What the speaker really desires at the time of utter-ance is to have played in the NBA. The antecedent proposition is not desiredper se but just as means to bring about the consequent, i.e to have played in theNBA.∗ I would like to thank Rajesh Bhatt, Lyn Frazier and Chris Potts for comments and help. Alsothanks to the audience of Sinn und Bedeutung 15, especially Cleo Condoravdi, Sven Lauer andMalte Zimmermann.

Sinn und Bedeutung 15 - [PDF Document] (119)

118 Biezma

The main question this paper addresses is where desirability comes from. Iwill propose that the modal meaning in optatives comes about pragmatically.It arises from the discourse assumptions leading to the utterance of the opta-tive, and is revealed by the topic-focus structure in optatives and the semanticsof conditionals. Optatives illustrate the importance of investigating meaningwithin the larger context provided by discourse and paying attention to prag-matic meanings derived from different components.

Overview: in §2 I investigate the syntax of optatives, in §3 I argue that alloptatives are conditionals, in §4 I show that optative conditionals differ withrespect to topicality, in §5 I argue that the reversal of topicality brings aboutdesirability by constraining the questions that license optatives in the discourse.

2 Scope and StructureI will adopt the view that if -clauses restrict the domain of quantification of amodal (Lewis-inspired proposal by Kratzer 1977). For a conditional to bringabout optativity, there must be a focus adverb in the antecedent that obeyscertain distributional restrictions. Let us consider the contrast between (3) and(4).

(3) OptativesIf only I had left earlier/ If only he didn’t have a gun/ If I had only leftearlier/ If he had only always acted honorably/ If he only didn’t havea gun/ If a hurricane only had razed the city/ If he had only not had agun/ Had I only read a letter

(4) Not Optatives (ungrammatical or not optative meaning)a. If he had always only acted honorably.b. *If he did only not have a gun.c. If he didn’t only have a gun.d. If he hadn’t only had a gun.e. *Had only I read a letter. (Rifkin 2000)

Only is an adverb, and can attach at any level in the structure that is seman-tically permitted. The data in (4) shows that in order to obtain an optativemeaning, the adverb must adjoin higher than vP. In (4a) there is no optativemeaning and only has attached either at the vP level or at the VP level (alwaysis adjoined at the vP level and only adjoins below it). The same is illustratedby (4c) and (4d), in which negation is constituent negation at the vP level. Thestructure below offers a summary of the positions where only may show upwith an optative interpretation (see Biezma in progress for details regarding

Sinn und Bedeutung 15 - [PDF Document] (120)

Optatives 119

the structures).

(5) [CP[TP only [TP[T only [T [PerfP only [PerfP [vP. . . ]]]]]]]]

To sum up so far, we have seen that the syntax of conditional optatives in-volves a focus adverb scoping over a clause denoting a proposition (vP or TP).In what follows I argue that it also needs to associate with the entire proposi-tion.

If only does not associate with the proposition there is not optativity, (6b).

(6) a. If only mom invited grandpa, he wouldn’t come.b. If only [F MOM] invited grandpa, he wouldn’t come.

The sentence in (6a) has an optative meaning, i.e. it is inferred that the speakerwants grandpa not to come. The contrast in (7) illustrates the meaning of (6a):

(7) a. I wish so much to see grandpa, # If only mom invited grandpa, hewouldn’t come.

b. I do not want to see grandpa this weekend,√

If only mom invitedgrandpa, he wouldn’t come.

I assume that the intonation in (6a) is just neutral intonation and as we seein (7), with neutral intonation there is optative meaning. (6b), however, is notan optative, i.e. it is not understood that the speaker does not want grandpa tocome. In (6b) the small caps on mom indicate emphatic intonation. In this case,only associates with mom, and the optative meaning disappears.

(8) A: Grandpa is getting old. He only travels when the whole family triesto convince him to get together.

B: Well, there is a possibility that mom ask him to visit us next week.A: Don’t be stupid! If only [F MOM] invited him, he wouldn’t come.

In (8) the conditional does not carry an optative meaning even though there is afocus adverb in the antecedent. The meaning of A’s utterance is that were momto be the only person inviting, he would not feel compelled to visit at all.

A second related argument to claim that optativity arises only when thefocus adverb associates with the entire proposition comes from the possibilityof having silent consequents.1 When the focus adverb does not associate withthe entire antecedent propositions, the absence of the consequent is ungram-matical.

1 In §3 I argue that optatives with and without spelled-out consequents are indeed conditionals.

Sinn und Bedeutung 15 - [PDF Document] (121)

120 Biezma

(9) a. If only mom invited grandpa.b. *If only MOM invited grandpa.

If we use the strategy of placing emphasis on a constituent to mark focus, (9b),thus forcing the association of only with something that is not a proposition,the result is a regular conditional and the consequent needs to be spelled out.If there is no special intonation, (9a), and only can associate with the entireproposition, it’s understood as an optative and the consequent can be silent.

In this section we have seen that optativity only arises when there is afocus adverb c-commanding a proposition level constituent, and associatingwith the entire proposition. In the next section we will see that, despite theabsence of consequents in some cases, optatives are always conditionals.

3 Optatives Are ConditionalsOne of the main characteristics of optatives is that they are fine with a conse-quent that is not spelled out. Indeed, this is even preferred and has been takento cast doubt on their characterization as conditionals. In what follows I reviewRifkin’s (2000) arguments against the view that optatives without consequents(if only!2) are conditionals, showing that a closer look at the data underminesRifkin’s conclusions. The main claim made in this section is that in spite of ap-pearances, if only! constructions have a conditional structure (cf. Rifkin 2000).We do not see a conditional because the structures denote properties of proposi-tions with a variable ranging over propositions that have been abstracted over.3

(10) Proposal: if only! constructions are abstractions over propositionsλ p.q⇒ p (where ‘⇒’ stands for the semantics of the modal)

In order to support the proposal in (10) I offer arguments to show that ifonly! constructions do not denote propositions, but denote instead propertiesof propositions, and to show that if only! constructions are conditionals. Argu-ments of the first kind are presented in §3.1 and §3.2, of the second kind in§3.3.

3.1 EmbeddingRifkin (2000) argues against if only! optatives being conditionals by showingthat they cannot be embedded, whereas regular conditionals can.

(11) a. Avi thinks that if it would snow, things would be good.b. *Avi thinks that if only it would snow.

2 I adopt Rifkin’s (2000) label for optatives without spelled-out consequents in the rest of the paper.3 See Biezma (in progress) for details.

Sinn und Bedeutung 15 - [PDF Document] (122)

Optatives 121

c. Avi thinks that if only it would snow, things would be good.

The data in (11a) illustrates a regular embedded conditional. (11b) shows thatif only! constructions cannot be embedded. The example in (11b) contrastswith (11c), in which an optative spelling out the consequent can be embed-ded. According to Rifkin, if if only! constructions were conditionals withoutconsequents they should behave like regular conditionals, but they don’t.

Rifkin’s (2000) observations regarding embeddability actually lend sup-port for the view presented above according to which if only! constructions donot denote propositions. This is the reason why they cannot be embedded in thesame way as optatives in which the consequent is spelled out, which do denotepropositions. The predicate think takes a proposition as argument, and if only!constructions are not of the right type to instantiate this argument.

3.2 ConjunctionRifkin (2000) claims that if if only! constructions were conditionals, they shouldbehave as conditionals across the board, and signals (12) as a counterexample.

(12) a. *If only Sue had money and if she had time, she would ski Mt.McKinley

b. *If Sue had money, she would ski Mt. McKinley, and if only shehad money

c. If Sue had money, things would be good, and if she had time, shecould ski Mt. McKinley (Rifkin 2000: ex. (31), (33) and (32))

Rifkin (2000) uses the data in (12) to argue that if only! constructions do notbehave like regular antecedents of conditionals with respect to coordination. Inprinciple we could conjoin two conditionals without only, (12c), but we cannotconjoin one with only and one without only, (12a) and (12b).

However, Rifkin himself points out that it is possible to conjoin two an-tecedents with only.

(13) I can’t believe the picnic went so poorly!a. If only Meg had brought a corkscrew and if only Jim had made a

decent saladb. If Meg had only brought a corkscrew and if Jim had only made a

decent salad (Rifkin 2000: footnote 5, ex. (iv))

Rifkin’s (2000) observations regarding coordination also provide support forthe view according to which if only! constructions are properties of proposi-tions. The contrast between (12a), (12b) and (13) is perfectly explained once

Sinn und Bedeutung 15 - [PDF Document] (123)

122 Biezma

we consider that if only! constructions are properties of propositions and notpropositions. The ungrammaticality of (12a) and (12b) is explained by the gen-eral impossibility of conjoining two objects of different semantic types (propo-sitions, in the case of regular conditionals, and properties of propositions in thecase of if only! constructions). This problem does not arise in (12c), since thetwo conjuncts are regular conditionals (and hence of the same type), and doesnot arise either in (13), where we have two if only! optatives conjoined.

3.3 Recovering the ConsequentEven though if only! constructions denote properties of propositions, they areused in contexts in which it is possible to recover a consequent, thus supportingthe claim that they are conditionals. Example (14), where B’s response showsthat B has worked out the silent constituent in A’s statement, illustrates this:

(14) A: If only I were taller.B: Then your desires wouldn’t have become true either.4

(14) illustrates that we process A’s statement as giving sufficient conditionsfor a desired consequence to be brought about. After the utterance of an ifonly!, we accommodate a consequent. In the most general case, as in (14),such consequent is merely that the consequences of the antecedent being trueare desired.

The fact that we can take B to be contradicting A’s claim is importantbecause B’s claim is itself an overt conditional. The proform in B’s statementprovides the antecedent for the modal ‘would’. In this context, it picks out thesame antecedent as the one in A’s statement. What follows in B’s claim is thenegation of the implicit consequent in A’s claim, and thus we understand thatB is disagreeing with A.5 The shape of B’s disagreement provides support forthe view that upon hearing A’s utterance, we process a conditionalized claim.

3.4 Summary and Further DataIn this section we have seen arguments that support the view that if only! con-structions are conditionals and we have proposed that in these cases the conse-quent is a silent pronoun that is abstracted over to generate a property of propo-sitions. With these ingredients we have been able to review Rifkin’s (2000)original arguments and show that the data does not actually argue against aview of if only! constructions as conditionals. There are further arguments thatcan be provided to support the view that conditionals and optatives have thesame underlying logical form (contra Rifkin 2000). These include the fact that

4 I thank a Sinn und Bedeutung 15 anonymous reviewer for this data.5 B’s utterance form is very telling since it is if α⇒¬β , the negation of the conditional statement.

Sinn und Bedeutung 15 - [PDF Document] (124)

Optatives 123

counterfactuality is not obligatory in optatives, the fact that the same ques-tions follow up conditionals and optatives, and the behavior of stacked an-tecedents. This discussion cannot be included for reasons of space (see Biezmain progress).

4 Reversed TopicalityIn this section, we turn to the issue of why the consequent can remain silent inif only! constructions and take the first steps towards explaining desirability. Ithas been argued in the literature that the antecedents of conditionals are topics(Haiman 1978).6 In the kind of regular conditionals that interest us here, theantecedent is an aboutness topic (Reinhart 1981).7 When the conditionals areoptatives, however, topicality is reversed. Since in optatives the focus adverbscopes and associates with the antecedent proposition, α , it is the antecedentproposition that is the focus. In these structures, the consequent, β , is nowthe topic. Recall that it is crucial for optativity that the focus adverb scopesand associates with the entire antecedent proposition. It is this that allows the(sentence level) information structure to be reversed in this type of conditional(we can also have focused elements in topic constituents, as in (6b) above,while the constituent itself remains the sentence topic).

The fact that the consequent in optatives is the topic, thus treated as dis-course old, explains why it can remain silent. The possibility of not spellingout the consequent in optative conditionals is the result of topic drop (and thisalso explains why speakers actually prefer not to spell out the consequent).8

The presence of focus adverbs in optatives plays a crucial role in ex-plaining the reversal in topicality. So far we have only considered optativescontaining only in the antecedent, but optativity can arise with other adverbstoo:9

(15) a. EnglishIf at least I had been taller, I would have played in the NBA.

b. SpanishSiif



/ tan sóloas only





habríawould have

6 Indeed, antecedents of conditionals can constitute topics of different kinds. See Ebert, Endriss &Hinterwimmer (2008) a.o.7 See Biezma (in progress) for arguments on this respect.8 Notice that in regular conditionals, in which the antecedent is the topic, the antecedent can remainsilent (see Kasper 1992). This is the opposite of what we find in optatives, since in optativesinformation structure is reversed.9 Below there is data from English, Spanish and German. My account is meant to explain the caseof English and Spanish. Further research would be needed to discuss the German data.

Sinn und Bedeutung 15 - [PDF Document] (125)

124 Biezma





c. German (optatives are preferred without a spelled-out consequent)? Wenn

















What about desirability? Where does this come from? The data in (15) showsthat there is a range of focus adverbs whose presence in the conditional an-tecedent brings about desirability, (with the constraints in §2). Given that onthe surface, optatives differ from conditionals only regarding the presence of afocus adverb in the antecedent, but do not depend on the semantics of that par-ticular focus adverb (there are several that do the trick), desirability needs to bederived from the mere presence of a focus adverb, not from its truth conditions.In what follows I argue that desirability arises from the interaction between thetypes of interpretations associated with focus adverbs and the Immediate Ques-tion Under Discussion.

5 Deriving DesirabilityIn this section we will finally tackle the issue of how desirability arises in op-tatives. We have reached the following important conclusions: (i) in optativesa focus adverb scopes over and associates with a proposition, §2; (ii) optativesare conditionals that spell out the antecedent, §3; (iii) information structure inoptative conditionals is reversed with respect to regular conditionals, §4. Wewill now bring these ingredients together to argue that desirability in optativesarises because the focus adverb appeals to a scale setting up discourse licensingconditions such that the question under discussion can only be a goal orientedquestion. Desirability is analyzed as an implicature arising from the discoursegiven an (implicit) goal oriented question.10

To reach this conclusion I proceed by first giving a brief overview ofRoberts’s (1996) discourse model, §5.1. Then I discuss the questions underdiscussion that license optatives, §5.2. Afterwards I establish a link with thescales in optatives, §5.3. Finally I show how desirability is derived, §5.4.

10 I am using the term goal oriented in a very broad sense. Goal oriented is meant to indicatethat the question inquires about how to bring about the desired state of affairs, without implyingagentivity.

Sinn und Bedeutung 15 - [PDF Document] (126)

Optatives 125

5.1 Topicality and the IQuDRoberts’ (1996) theory of discourse is devoted to the recognition of the in-terlocutor’s intentions in understanding the meaning of the utterance. Robertsprovides a framework for discourse as a sequence of intentional actions struc-tured with a given goal. Following Stalnaker, Roberts considers that the maingoal of a discourse is the communal inquiry to discover what the actual worldis like. During discourse, the participants’ goal is to reduce the context set (aset of possible worlds) characterized by the Common Ground (CG).

Roberts takes questions to be the obvious counterpart of an inquiry anduses them as the formal objects reflecting interlocutor’s goals. In Roberts’ sys-tem, we can track the speaker’s intentions by assuming that every utterance iseither an answer (pay-off move) to an (implicit) question that the speaker ac-cepts to address (the immediate question under discussion (IQuD)), or a ques-tion itself (set-up move). Assertions are pay-off moves because they choosebetween the alternatives proffered by a set up move. In this system the in-terpretation of every move involves two aspects: (i) the presupposed content,which constrains the contexts in which an utterance can be made, and (ii) theproffered content, which corresponds with what is asserted (in assertions) andthe non-presupposed content of questions and commands.

Besides recognizing that the primary goal of every discourse is a com-munal inquiry, Roberts also recognizes the existence of more particular goals,domain goals. These particular goals are ultimately what lies behind the type ofconversational inquiry conducted by the speaker. In the next section I explorewhat are the domain goals behind the utterance of an optative (i.e. the IQuD).

5.2 Mention-Some and the IQuDs in OptativesIn this section we will discuss the role of optatives in the discourse. Our goal isto identify the IQuDs that can be answered (paid off ) with an optative. This isimportant because my objective is to link desirability in optatives to the IQuD.Let us start by noting that, in general, in answers we find focus on the elementsthat are under question, (16).

(16) A: What did Lauren buy?B: Lauren bought [F BANANAS]

Even if B’s utterance is not preceded by an explicit question, we can assume,given the structural characteristics of B’s utterance (syntax and intonation),that the utterance is answering the question what did Lauren buy?. This fol-lows from the fact that the question under discussion has to be congruent withthe utterance. So, in order to find out the IQuD in optatives, we first need to

Sinn und Bedeutung 15 - [PDF Document] (127)

126 Biezma

understand the semantics of the conditional, since the implicit question has tobe congruent with this too. A conditional if α , β claims (roughly) that in themost similar worlds to the actual world in which α is true, β is true (à laLewis-Stalnaker). With this semantics in hand, and considering the informa-tion structure of regular conditionals as discussed above (α is the topic), theIQuD when a regular conditional is uttered would be as in (17).

(17) What does α bring about? or What would α have brought about?

The conditional uttered as answer to the question in (17) provides the answervia the consequent, β , which bears focus.

(18) A: What would happen after the fall of the dictatorial Government?B: If the Government fell, a democratic system would be estab-


The consequent proposition, a democratic system is established, is the answer.As argued above, however, in optatives (if only α , β ) topicality is reversed

and β does not bear focus. The sentence focus is α , the antecedent, whereas β ,the consequent, is now the topic. Given this, and considering the semantics ofconditionals, I claim that the (implicit) IQuD for optatives is (19).

(19) How do we bring β about? or How would we have brought β about?

The IQuD when an optative is uttered asks what are sufficient conditionsto bring about the consequent (the topic).

Notice that the questions in (19) are a special kind of question. They aregoal oriented questions. We understand that the speaker wants to know aboutthe best way to bring about β . In the case of goal oriented questions, we donot ask about all the alternatives that bring about the truth of the embeddedproposition (β ), but about the best alternative that the addressee is aware of.

The questions in (19) have another important characteristic, they implythat the proposition embedded in the question is desired by the speaker. To seethat this is so, let us consider the questions in (20).

(20) a. How do I get to the supermarket?b. How do I get to play in the NBA?c. How do I get to die?

The question in (20a) implies that the questioner wants to get to the supermar-ket and asks about the best way to get there that the addressee is aware of,(21).

Sinn und Bedeutung 15 - [PDF Document] (128)

Optatives 127

(21) John is walking on the street and John asks a passer-by a question.John: How would I get to the supermarket?Bill: Walk south and turn right on the next street.

If after Bill’s directions John starts walking north, Bill would be perplexed,since he would wonder why he is going in the opposite direction to that of thesupermarket. He would be even tempted to call him out and indicate that he isjust walking opposite to what he indicated, south is in the other direction. Thisis because John’s utterance implied that he wants to go to the supermarket. Thesame strategy would show that in (20b) it is implied that the speaker wants toplay in the NBA. And (20c) is odd in most contexts because it implies that thespeaker wants to die and that is an odd desire to have.

If the IQuD when an optative is uttered is a goal oriented question, thiswould provide us with an explanation for why we understand that optativesconvey the desirability of the consequent: the IQuD asks how to bring the con-sequent about and implicates that the consequent is desired. However, we stillneed to provide arguments to support the claim that the IQuD addressed by anoptative is of the kind in (19), i.e. goal oriented.

Notice that the claim that it is goal oriented questions that license opta-tives is not trivial. If we just consider the semantics of conditionals and theirinformation structure, other types of questions may be expected to serve asIQuDs leading to optatives. Paying attention only to the semantics of condi-tionals and the reversal in information structure, one could also argue that (22)could serve as an IQuD licensing an optative answer.

(22) What are the circ*mstances that would bring about β?

Given (22), a conditional in which the antecedent is the sentence focus wouldbe an appropriate answer, and this is exactly what we find in optatives giventhe presence of a focus adverb. But, of course, if the IQuD were something like(22), we would not explain desirability in optatives. (22) is not a goal orientedquestion and we do not understand that the embedded proposition is desired.

Why can’t (22) be the IQuD for an optative? The important differencebetween the questions in (19) (goal oriented) and questions like (22) is thatthey privilege different readings. The questions in (19) privilege a mention-some reading, whereas in (22) a mention-all reading is prominent.11 In themention-some readings, the answerhood conditions for a question require thatthe answer meet the questioner’s goals. The relevant answer is then the one

11 How to account for these two readings in a theory of questions is a debate far from settled andbeyond the scope of this paper.

Sinn und Bedeutung 15 - [PDF Document] (129)

128 Biezma

indicating the best alternative for achieving the goal (mention-some questionsare goal oriented questions). It has been argued that some questions are spe-cialized for mention-some readings (e.g. Asher & Lascarides 1998) argue thathow and where questions give rise to a mention-some reading in most of thecases). In what follows I argue that the IQuD in optatives has to be a questionwith a mention-some reading. Since such questions are typically goal orientedquestions this explains the desirability effects in optatives. In my explanation Iwill appeal to the semantics of focus particles present in optatives. In the nextsections I will argue that certain aspects of the semantics of the focus adverbsin optatives are crucial in establishing the IQuD addressed by an optative. Inparticular, I will appeal to the fact that these adverbs are scalar.

5.3 The Scale in OptativesIn this paper I adopt Beaver & Clark (2008) analysis of conventionally focussensitive expressions (like only and at least). This analysis argues that such ex-pressions encode a dependence on the IQuD. As these authors point out, theirproposal is not the first proposal claiming that there is a relation between focussensitive expressions and the IQuD. Other authors already established such linkwith the discourse topic or the IQuD (von Fintel 1994; Roberts 1996). How-ever, Beaver & Clark (2008) go a step further and claim that this relationship isencoded in the meaning of the expressions and that these must comment on theIQuD. In what follows I focus on only and conditional optatives containing thisadverb. According to Beaver & Clark (2008), “the function of exclusives likeonly is to say that the strongest true answer to the IQuD is weaker than someexpected answer.” Thus, utterances containing only trigger a partial rank of al-ternatives (the possible answers) ordered according to a contextually providedscale (see Beaver & Clark 2008 for details).12 According to these authors, ut-terances containing only carry the presupposition that “the strongest true al-ternatives in the IQuD are at least as strong as the prejacent”,13 and that thedescriptive content of utterances with exclusives indicates that “the strongesttrue alternatives in the IQuD are at most as strong as the prejacent”. With theprevious background in hand, let us see now how only works in optatives. Con-sider the optative in (23).

(23) John had a job interview this morning. He drove there but his car brokedown. John called Tom, a mechanic friend, but by the time he got thecar running it was too late for John to make it.

12 The ordered alternatives do not need to logically entail alternatives lower in the scale.13 The prejacent of an utterance containing only is the proposition denotated by the sentence inwhich the exclusive is not present.

Sinn und Bedeutung 15 - [PDF Document] (130)

Optatives 129

Tom:If only I had arrived earlier

In order to make use of Beaver & Clark’s (2008) proposal, we need to adaptit to the case of conditionals. The optative uttered by Tom is “If only I hadarrived earlier, John would have gotten to his interview on time”. With theassumption that the antecedent proposition is focused, the prejacent itself is X.In the context of the optative conditional, we obtain (24):

(24) If X, John would have gotten to his interview on time.(Where X= Tom arrives on time and repairs John’s car)

The alternative values for X are presented in (25). The alternatives in (25) areordered according to a scale provided by what could intuitively be thoughtof as likelihood (factors like the degree of deviation from the history of theactual world, the effort required to bring about the truth of the proposition, andplausibility can all play a role here). The strongest alternatives are the mostlikely ones, while the weakest alternatives are the ones that require more effort,are more implausible given the history of the world, etc.

(25) +(likely) John drove his car more carefullyTom arrived earlier and fixed the carTom fixed the car faster

−(likely) John went out and bought a new car

In (25) we find a variety of alternatives. The order is provided by likelihoodand the amount of effort required to bring each about. Suppose that John isactually a careful driver and Tom is habitually late. It would have been morelikely/easier for John to drive even more carefully than he actually did thanfor Tom to arrive on time. John is actually rather poor, so the amount of effortit would have taken for him to buy a new car, and the unlikelihood of thathappening, is much greater than for the alternative of Tom arriving on time.

I will follow Beaver & Clark (2008) with respect to the presuppositionsand descriptive content associated with only. Since we are dealing with alterna-tives that are antecedents of (counterfactual) conditionals, we cannot ask for thestrongest true alternative. Instead, in the context of a conditional, we will lookfor the strongest sufficient alternative. When Tom utters the optative in (23), hepresupposes that the strongest sufficient alternatives are at least as strong as theantecedent proposition. The descriptive content associated with Tom’s claim isthat the strongest sufficient alternatives are at most as strong as the antecedent.

Let us examine the predictions made by this proposal with respect to (23).Tom’s utterance carries the presupposition that the sufficient alternatives are atleast as strong as the chosen alternative. This is true, since the only other suffi-

Sinn und Bedeutung 15 - [PDF Document] (131)

130 Biezma

cient alternative (that Tom fixed the car more quickly) is as strong as the chosenalternative (the other sufficient alternatives are weaker). The descriptive con-tent associated with Tom’s utterance is that the strongest sufficient alternativesare at most as strong as the chosen alternative. This is true given our scale, sincethe stronger alternatives are not sufficient (the car breaking down had nothingto do with John’s driving style).

The proposal above makes correct predictions regarding unacceptable op-tatives in this context. Imagine that in the scenario above, Tom had uttered Ifonly you had driven more carefully!. This would have been deviant in the con-text, since driving more carefully would not have had any useful consequences.We would be surprised by Tom’s utterance. The deviancy is predicted. The de-scriptive content associated with such a claim would have been false. This isnot the strongest sufficient alternative. Indeed, this is not a sufficient alternativeat all. With the assumption (following Beaver & Clark 2008) that only marksthe strongest sufficient condition, this optative is predicted to be deviant.

Let us turn now to another deviant optative. Suppose that in the scenarioabove, Tom had uttered If only you had bought another car. This optativewould also have been deviant. John would have felt that Tom’s utterance wasa bit exaggerated. This is also predicted by the proposal above. The presup-positions associated with Tom’s utterance would not be respected. There aresufficient alternatives that are stronger than the chosen alternative. Again, theproposal predicts that this optative is deviant.

The role of only in an optative is to signal the position that the antecedentproposition occupies on a scale. We have followed Beaver & Clark (2008) withrespect to the presuppositions and descriptive content associated with only.Given that our interest lies in only in the antecedent of conditionals, we havenot relativized the scale to truth, but to the sufficiency of the proposition tobring about the consequent. The scales we have adopted order the alternativesin terms of likelihood, with the most likely being considered stronger. This hasthe result that propositions that are harder to bring about, or wildly implausible,are characterized as weaker. This may appear rather unintuitive, but, as we haveseen, this scale fits our intuitions regarding the acceptability of optatives.

We have not discussed Beaver & Clark’s (2008) claim that only weakenssalient or natural expectations. A discussion of this point remains for futurework. It is worth noting that the case of conditionals is different from the caseof assertions discussed in Beaver & Clark (2008). It is unclear how expec-tations would work in the antecedent of (counterfactual) conditionals. Noticethat in Beaver & Clark’s (2008) example Brad only got a Soames, getting aSoames is understood as being ‘less’ than was expected/hoped for. However,in the context of a conditional If only Brady had gotten a Soames! the judgment

Sinn und Bedeutung 15 - [PDF Document] (132)

Optatives 131

disappears. Expectations seem to work differently in the case of conditionals,but this discussion lies outside the scope of the current work.

An optative provides the best/strongest alternative that a speaker knowswould bring about the desired consequent. If an optative is considered a pay-offmove, it requires an IQuD that asks for the best strongest alternative that bringsabout the consequent. These are goal oriented /mention-some questions.

5.4 Desirability Derived!When uttering an optative the speaker indicates that he is answering a mention-some/goal oriented question. This is because of the congruence requirementbetween the optative and the IQuD. Optatives require a IQuD that asks aboutthe best alternative amongst the set, and mention-some/goal oriented questionsdo exactly that. Since only mention-some/goal oriented questions can licenseoptatives and these questions imply that the embedded proposition14 is desired,we understand that the consequent in optatives is desired.

6 ConclusionIn this paper I have proposed an analysis of optatives that draws heavily on theinteraction between syntax, semantics, pragmatics and discourse to explain themeaning of the construction. The focus of the paper has been the expressionof desirability in optatives. I have shown that the modal meanings associatedwith desires can be derived pragmatically. There isn’t a “desirability modal” inoptatives. There is, however, a focus adverb that appeals to ordered alternativesand invokes a question under discussion with desirability implicatures.

ReferencesAsher, Nicholas & Alex Lascarides. 1998. Questions in dialogue. Linguistics

and Philosophy 23(3). 237–309.Beaver, David I. & Brady Z. Clark. 2008. Sense and Sensitivity: How focus

determines meaning. Malden, MA: Wiley-Blackwell.Biezma, María. in progress. Anchoring pragmatics in syntax and semantics.

Amherst, MA: University of Massachusetts Amherst dissertation. Ms.Ebert, Christian, Cornelia Endriss & Stefan Hinterwimmer. 2008. A unified

analysis of indicative and biscuit conditionals as topics. In T Fried-man & S Ito (eds.), Proceedings of semantics and linguistic theory 18,Ithaca: Cornell University.

von Fintel, Kai. 1994. Restrictions on quantifier domains: University of Mas-

14 The embedded proposition in the question is also the (implicit) consequent in the optative con-ditional.

Sinn und Bedeutung 15 - [PDF Document] (133)

132 Biezma

sachusetts Amherst dissertation.Haiman, John. 1978. Conditionals are topics. Language 54(3). 564–589.Kasper, Walter. 1992. Presuppositions, composition, and simple subjunctives.

Journal of Semantics 307–331.Kratzer, Angelika. 1977. What ‘must’ and ‘can’ must and can mean. Linguis-

tics and Philosophy 1. 337–355.Reinhart, Tanya. 1981. Pragmatics and linguistics: An analysis of sentence

topics. Philosophica 27. 53–94.Rifkin, Jay I. 2000. If only if only were if plus only. In Cls 36, 369–383.

Chicago, Il: Chicago Linguistics Society.Roberts, Craige. 1996. Information Structure in Discourse: Towards an Inte-

grated Formal Theory of Pragmatics. In OSU Working papers in lin-guistics 49: papers in semantics, 91–136.

Sinn und Bedeutung 15 - [PDF Document] (134)

On French un même and Antispecificity

Isabelle Charnavel University of California, Los Angeles/Institut Jean Nicod, Paris


Abstract. In this paper, I examine the definiteness problem raised by sentence-internal même 'same' in French, as in this language (vs. English), même does not only combine with the definite determiner (le même), but also with the indefinite article (un même). Even if le même, like the same, does not behave like typical definite descriptions do, it contrasts with un même with respect to definiteness and distribution: un même is more indefinite than le même in that it does not trigger any presupposition at all; moreover, un même is more restricted in distribution than le même in that it only exhibits sentence-internal readings in antispecific contexts. I hypothesize that both le même and un même are quantifiers over a plural event that has been distributed, but also contain a domain variable: that of un même has to be quantificationally bound while that of le même can also be identified by the context.

1 Introduction Sentence-internal same (like different) poses a problem of compositionality that aroused the interest of several linguists (Carlson 1987, Moltmann 1992, Barker 2007, Brasoveanu 2009, ...): due to its meaning involving comparison, the interpretation of same relies on the presence of a licenser (underlined in 1) that does not directly combine with the DP containing same. Under the sentence-internal reading, (1) means that there exists some book x such that Mike read x and Sue read x (while the sentence-external reading depends on identifying some contextually salient book).

(1) Mike and Sue read the same book.

But in this paper, I will focus on another related issue raised by sentence-internal same: even if DPs with sentence-internal same do not behave the way typical definite descriptions do, 'a same' is ungrammatical in English; C. Barker (2007: 428) formulates this definiteness puzzle as follows:

Why does same require the definite determiner? (Baker 2007: 428)

However, French equivalent of 'same' même interestingly combines either with the definite (le/la/les même(s)) or the indefinite (un/une/de même(s))

In Reich, Ingo et al. (eds.), Proceedings of Sinn & Bedeutung 15, pp. 133–147. Universaar – Saarland University Press: Saarbrücken, Germany, 2011.

Sinn und Bedeutung 15 - [PDF Document] (135)

134 Charnavel

article. Moreover, le même behaves like the same in that it also presents indefiniteness effects. The same can always be translated by le même, but also by un même in certain cases; in other cases, un même cannot be translated by the same without changing the meaning. The main issue is thus to understand in which cases un même is licensed if le même already behaves like an indefinite. My goal is to describe the semantics and the distribution of un même and examine the theoretical implications with respect to definiteness and specificity. To this end, I will first compare le même and un même with respect to definiteness, then with respect to distribution, and I will finally suggest an hypothesis about the antispecificity of un même: I will propose that un même and le même are quantifiers containing a variable that gets interpreted in different ways.

2 Le même, un même and Definiteness 2.1 Le même and Indefiniteness Effects Sentence-internal même differs from other terms expressing identity and difference with respect to determiner use.

(2) a. Julie et Paul ont lu le même livre/#le livre différent/similaire. ‘Julie and Paul read the same/#the different/#the similar book.’ b. Julie et Paul ont lu ??un même livre/un livre différent/similaire. ‘Julie and Paul read ??a same/a different/a similar book.’

Moreover, le même exhibits indefiniteness effects. First, it does not trigger a presupposition of unique existence as typical definite descriptions do (cf. Barker 2007: 428). This is shown in (3) presenting basic tests for presuppositions: even if the sentence is negated or questioned, the existence of a unique book that Luc and Flore read is not presupposed, but this is precisely what is at issue here.

(3) a. Luc et Flore ont lu le même livre. ‘Luc and Flore read the same book.’ b. Est-ce que Luc et Flore ont lu le même livre? ‘Did Luc and Flore read the same book?’ c. Luc et Flore n'ont pas lu le même livre. ‘Luc and Flore did not read the same book.’

Also, le même can introduce a new discourse referent like indefinites (cf. Novelty Condition) as opposed to standard definites.

Sinn und Bedeutung 15 - [PDF Document] (136)

French un même and Antispecificity 135

(4) A: Pourquoi tu penses que Luc et Flore ont des goûts similaires? B: Pendant les vacances, ils ont lu le même livre. ‘A: Why do you think Luc and Flore have similar tastes? B: They read the same book during the holidays.’

Furthermore, le même can occur in existential constructions unlike typical definites.

(5) a. Il existe le même problème dans ces trois pays. ‘There exists the same problem in these three countries.’ b. * Il existe le/ce problème dans ces trois pays. ‘*There exists the/this problem in these three countries.’

Finally, le même can be non specific as exemplified in (6). This is not predicted based on Enç (1991) who argues that specificity corresponds to an inclusion relation in a contextually determined set; since the linking relevant for definite DPs is the identity relation and identity of referents entails inclusion, all definites are expected to be specific.

(6) Claire et Anne ont acheté la même robe. ‘Claire and Anne bought the same dress.’ a. Specific: there is a particular dress that the speaker has in mind that

Claire and Anne each bought. b. Non specific: there exists a unique dress – whatever it is, the speaker

does not know which one – that Claire and Anne each bought.1

So le même behaves like an indefinite in several respects.

2.2 Un même more Indefinite than le même? If le même has the properties of an indefinite, how can un même contrast with le même? First, le même unlike un même triggers what I call a global presupposition of unique existence. As seen above, (3) does not presuppose the existence of a unique book that Luc and Flore read (usual presupposition of existence), but it presupposes that Luc and Flore each read a unique book (global presupposition of existence), as shown by Solomon (2009) for same. This idea is corroborated by the following examples: the infelicitousness of (7) points to a global existence presupposition, while that of (8) indicates the existence of a global uniqueness presupposition.

1 Note that specificity cannot be determined by scope when same is involved: because of its meaning, narrow scope of same does not entail covariation.

Sinn und Bedeutung 15 - [PDF Document] (137)

136 Charnavel

(7) Est-ce que Luc et Flore ont lu le même livre? #Non, Luc n'a pas lu de livre. ‘Did Luc and Flore read the same book? #No, Luc didn't read any book.’

(8) ??Luc et Flore ont lu le même livre pendant les vacances, et Luc a également lu les Misérables et Madame Bovary. ‘??Luc and Flore read the same book during the holidays, and Luc also read les Misérables and Madame Bovary.’

On the other hand, un même does not trigger any presuppositions at all: when un 'a' is used, (9) does not presuppose that the children eat round a table, and (10) does not presuppose that each country only has one enemy.

(9) Dans la plupart des familles nombreuses, les enfants ne mangent pas autour de la/une même table. ‘In most large families, the children do not eat round the/UN same table.’

(10) Quand deux pays ont le/un même ennemi, ils s'allient. ‘When two countries have the/UN same enemy, they form an alliance.’

Furthermore, un même has to be non specific (what I call antispecificity) while le même can be either specific or non specific. (11) illustrates that un même unlike le même cannot be used when the referent can be identified by the context.

(11) a. Dans chaque système planétaire, toutes les planètes tournent autour de la/une même étoile.

‘In each planetary system, every planet revolves around the/UN same star.’

b. Dans le système solaire, toutes les planètes tournent autour de la/*une même étoile. ‘In the solar system, every planet revolves around the/*UN same star.’

So the semantic difference between un même and le même pertains to pre-suppositions and specificity.

3 Le même, un même and Distribution It appears that un même can always be replaced by le même, but the reverse does not hold. What is then the contrast between un même and le même with respect to distribution?

Sinn und Bedeutung 15 - [PDF Document] (138)

French un même and Antispecificity 137

3.1 DP-­‐internal Distribution of même First, let's note that même does not exhibit the DP-internal distribution of standard adjectives, which supports the hypothesis that le même and un même are actually complex determiners. In fact, même cannot be used predicatively.

(12) a. Ces livres sont *(les) mêmes. ‘These books are *(the) same.’ b. *Je ne le trouve pas même aujourd'hui. ‘*I don't find him same today.’

Secondly, the only determiners compatible with sentence-internal même are the definite determiner le/la/les, and the indefinite one un/une/de.

(13) a. lire *quelques/*divers/*certains/*plusieurs/*trois/de/les même livres. ‘to read *some/*various/*certain/*several /*three/ø/the same books.’ b. lire un/*leur/#ce/le même livre. ‘to read a/*their/#this/the same book.’2

Thirdly, même cannot be modified by adverbs.

(14) Luc et Flore ont lu le (*vraiment/*très/*presque/*tout) même livre. ‘Luc and Flore read the (?really/very/?almost/very) same book.’

Finally, même cannot be coordinated with any adjective.

(15) Luc et Flore ont acheté le (*petit et/*premier et/*seul et) même livre. ‘Luc and Flore bought the (*small and/*first and/*only and) same book.’

Based on these data, I hypothesize that le même and un même are complex determiners. This is supported by crosslinguistic evidence: Braseovanu (2009) shows that in Romanian, while singular and plural 'different' are adjectival in nature (alt and diferit, respectively), 'same' is a determiner – the so-called demonstrative article (or pronoun) of identity acelaşi (agreeing in gender and number), which is the counterpart of the English 'the+same' rather than just 'same'.

3.2 Only one Reading for un même But there are several distributional differences between un même and le même. First, un même unlike le même only has sentence-internal readings.

2 # indicates that the sentence-external reading is possible, but not the sentence-internal reading that concerns us here.

Sinn und Bedeutung 15 - [PDF Document] (139)

138 Charnavel

In fact, le même appears in three kinds of contexts: it can have sentence-internal readings (cf. 3); but it can also exhibit sentence-external-readings that depend on identifying some contextually salient book, whether deictically (16) or anaphorically (17); and it can occur in comparative constructions (18).

(16) Regarde! Luc et Flore ont lu le même livre. ‘Look! Luc and Flore read the same book.’

(17) J'ai lu Germinal pendant les vacances. Luc et Flore ont lu le même livre. ‘I read Germinal during the holidays. Luc and Flore read the same book.’

(18) Luc et Flore ont lu le même livre que toi/l'année dernière/celui que tu as emprunté à la bibliothèque. ‘Luc and Flore read the same book as you/last year/the one you borrowed from the library.’

On the other hand, un même only presents sentence-internal readings (19): it cannot appear in comparative constructions (20) and cannot have sentence-external readings whether anaphorically (21) or deictically (22) constructed.

(19) Une même expression peut avoir plusieurs sens. ‘UN same phrase may have several senses.’

(20) On ne peut jamais employer *une/la même expression que Paul. ‘One can never use *UN/the same phrase as Paul.’

(21) Paul a choqué l'assistance en employant une expression très familière. Ses collègues ne pourraient pas employer *une/la même expression. ‘Paul shocked the audience by using a very colloquial phrase. His colleagues could never use *UN/the same phrase.’

(22) Ecoute ça! On employait *une/la même expression il y a dix ans. ‘Listen to that! One used *UN/the same phrase ten years ago.’

So un même is more constrained than le même in that it only presents sentence-internal readings.

3.3 The Distributional Constraints on un même Furthermore, the sentence-internal reading of un même is itself more constrained. To realize that, we need to first identify the distributional constraints common to le même and un même.

Sinn und Bedeutung 15 - [PDF Document] (140)

French un même and Antispecificity 139

Le/un même requires distributive licensers, whether obligatorily distributive (some quantifiers, cf. 23a) or optionally distributive (plurals and conjunctions, cf. 23b).

(23) a. Chaque enfant doit lire le/un même livre. ‘Each child has to read the/UN same book.’ b. Ces enfants/Luc and Flore doivent lire le/un même livre. ‘These children/Luc and Flore have to read the/UN same book.’

Conversely, elements that cannot be distributive (singulars or collectives cf. 24a, mass nouns cf. 24b) cannot license le/un même.

(24) a. Luc/la classe doit lire #le/*un même livre. ‘Luc/the class has to read #the/*UN same book.’ b. Le riz coûte #le/*un même prix. ‘Rice costs #the/*UN same prize.’

Moreover, the relation between the licenser and le/un même resembles the relation between two scope-taking quantifiers. Thus, le/un même does not need to be c-commanded by its licenser as opposed to anaphors.

(25) Le/un même joueur peut remporter tous les tournois. ‘The/UN same player may win every tournament.’

Also, le/un même is sensitive to island constraints (adjunct constraint (26), coordination constraint (27), extraction constraints related to non-bridge verbs (28), wh-islands (29), subject islands (30)).

(26) a. Aucune région n'est en colère parce que #le/*un même nombre de députés a démissionné.

‘No region is angry because #the/*UN same number of deputies resigned.’

b. Aucune région ne peut élire le/un même nombre de députés. ‘No region can elect the/UN same number of deputies.’

(27) a. Chaque électeur peut voter pour ce président et #le/*un même trésorier. ‘Each voter can vote for this president and #the/*UN same treasurer.’

b. Chaque électeur peut voter pour le/un même trésorier. ‘Each voter can vote for the/UN same treasurer.’

(28) a. Si tous les habitants chuchotent que #la/*une même personne a commis le crime, il n'y a pas d'espoir.

Sinn und Bedeutung 15 - [PDF Document] (141)

140 Charnavel

‘If every inhabitant whispers that #the/*UN same person committed the crime, there is no hope.'

b. Si tous les habitants accusent la/une même personne, il n'y a pas d'espoir.

‘If every inhabitant accuses the/UN same person, there is no hope.’

(29) a. Quand Anne et Flore se demandent où #la/*une même personne ira, elles finissent par le savoir.

‘When Anne and Flore wonder where #the/*UN same person will go, they end up finding it out.’

b. Quand Anne et Flore critiquent la/une même personne, elles n'ont pas de pitié. ‘When Anne and Flore criticize the/UN same person, they have no pity.’

(30) a. Qu'un individu commette #le/*un même acte peut constituer un crime contre l'humanité et un crime de guerre.

‘That an individual commits #the/*UN same act can constitute a crime against humanity and a war crime.’

b. Le/un même acte peut constituer un crime contre l'humanité et un crime de guerre. ‘The/UN same act can constitute a crime against humanity and a war crime.’

Furthermore, le/un même is not only licensed by distributive DPs, but also by conjoined PPs, conjoined Ps, conjoined VPs, conjoined APs and possibly conjoined Advs, as observed for same and different by Carlson (1987) and Moltmann (1992), who based on such data proposed an analysis of same in terms of events.

(31) a. Le/un même homme peut composer des opéras et jouer au football. ‘The/UN same man may compose operas and play soccer.’ b. Le/un même homme peut aimer peindre dans son atelier et à l'exté-

rieur. ‘The/UN same man may like painting in his studio and outside.’ c. Le/un même homme peut voter pour et contre un projet de loi. ‘The/UN same man voted for and against the bill.’ d. On peut peindre le/un même jouet en rouge et en bleu. ‘One may paint the/UN same toy red and blue.’ e. On peut cuisiner le/un même plat joyeusem*nt et tristement. ‘Luc cooked the/UN same meal joyfully and sadly.’

Sinn und Bedeutung 15 - [PDF Document] (142)

French un même and Antispecificity 141

I showed in Charnavel (2011) that moreover, le/un même is also licensed by several possible aspectual notions, such as frequentativity, iterativity (32, 33) or continuativity, durativity (34, 35), which can be expressed by several categories (verbs, adverbs, preverbs, nouns, adjectives).

(32) Lire le/un même livre plusieurs fois est instructif. ‘Reading the/UN same book several times is instructive.’

(33) La répétition de la/une même erreur n’est pas acceptable. ‘The repetition of the/UN same mistake is not acceptable.’

(34) Il est difficile de continuer à travailler dans la/une même entreprise quand on aime le changement. ‘It is hard to keep working in the/UN same company when one likes changes.’

(35) Une habitation prolongée dans le/un même logement peut poser problème.

‘Extended habitation in the/UN same housing may pose problems.’

In all these cases, a plural event is involved, which is distributed over the overall running time in several ways depending on the aspect that is expressed. So I hypothesize that le/un même is an existential quantifier over a plural event (see Charnavel 2011 for more details); this event needs to have independently been distributed through participants or times as formalized below: le/un même takes two arguments, its restriction Y and the event predicate Z, and says that for every event en part of this (obligatorily) plural event, there is a corresponding individual xn in en part of the restriction set, and all these individuals xn are identical:

(36) [[ le/un même]] =λY<et>.λZ<e,vt>. ∃x1, x2… xn ≤ x (en-1 ≠ en; n is a positive integer and n ≥ 2) such that Y(xn)=1 and Z(xn)(en)=1, and xn-1=xn

This is illustrated in (37). In (a), the event has been distributed through participants (possibly through a silent distributive operator): for every subevent e1 (Luc reading) and e2 (Flore reading), there is a book x1 and a book x2, such that x1 is identical to x2. In (b), the event has been distributed through times by the adverbial quantifier: for every subevent e1 (Flore reading at time t1) and e2 (Flore reading at time t2), there is a book x1 and a book x2, such that x1 is identical to x2.

(37) a. Luc et Flore ont lu le même livre. ‘Luc and Flore read the same book.’ b. Flore lit toujours le même livre.

Sinn und Bedeutung 15 - [PDF Document] (143)

142 Charnavel

‘Flore always reads the same book.’

Besides, un même has further distributional constraints (note that in the previous examples, these additional constraints were fulfilled so that the constraints common to le même and un même could be independently examined). The presence of a plural event is a necessary, but not sufficient condition for licensing sentence-internal un même, which is only licensed by the following contexts: inside a DP (38), in the context of modality and genericity (39), in the context of arbitrary PRO (40), or in the context of hypotheses (41).

(38) a. Quatre générations sous un/le même toit ‘Four generations under UN/the same roof.’ b. Ces quatre générations ont vécu sous ??un/le même toit. ‘These four generations lived under ??UN/the same roof.’

(39) a. Un/le même mot peut avoir plusieurs sens. ‘UN/the same word may have several senses.’ b. Un/le même mot a (généralement) plusieurs sens. ‘UN/the same word (generally) has several senses.’ c. Dans ce texte, ??un/le même mot a plusieurs sens. ‘In this text, ??UN/the same word has several senses.’

(40) a. Utiliser un/le même mot de passe pour différents services, ordi-nateurs et sites internet augmente les risques de se faire voler des informations personnelles. ‘Using UN/the same password for different services, computers and websites increases the risks of having personal informations stolen.’

b. J'ai utilisé ??un/le même mot de passe pour différents services, ordi-nateurs et sites internet. ‘I used ??UN/the same password for different services, computers and websites.’

(41) a. Si un/le même joueur fait plus de 5 fautes, il est disqualifié. ‘If UN/the same player makes more than 5 mistakes, he gets dis-qualified.’

b. ??Un/Le même joueur a fait plus de 5 fautes. ‘??UN/The same player made more than 5 mistakes.’

So the intuition is that un même is licensed in contexts presenting a flavor of generality, i.e. when multiple situations are involved: it is unfelicitous as soon as a particular situation is at stake (antispecificity).

Sinn und Bedeutung 15 - [PDF Document] (144)

French un même and Antispecificity 143

To sum up, I have proposed several hypotheses concerning both le même and un même: based on the fact that même does not exhibit the standard distribution of an adjective, I have hypothesized that un même and le même are complex determiners. Based on the observation that un même and le même require distributive licensers (individuals or times) and that they are sensitive to island constraints, I have proposed that un même and le même are quantifiers over a plural event. Furthermore, I have observed that un même has a more constrained distribution than le même: un même only exhibits sentence-internal readings in antispecific contexts. The question is now to know how to analyse the antispecificity of un même.

3 Un même and Antispecificity Let's review in which sense un même is antispecific as opposed to le même. First, un même cannot be used when the speaker has an individual in mind as its referent.

(42) Dans le système solaire, toutes les planètes tournent autour de la/*une même étoile.

‘In the solar system, every planet revolves around the/*UN same star.’

Secondly, un même is not licensed if the referent of its DP has a linking relation with an antecedent, as shown by the absence of sentence-external readings with un même. According to Enç (1991), nonspecifics require that their discourse referents not be linked to previously established discourse referents, while specifics exhibit an inclusion relation or any other association with an antecedent.

(43) Paul a choqué l'assistance en employant une expression très familière. Ses collègues ne pourraient pas employer *une/la même expression. ‘Paul shocked the audience by using a very colloquial phrase. His col-leagues could never use *UN/the same phrase.’

Thirdly, un même does not trigger any presupposition of existence.

(44) Dans la plupart des familles nombreuses, les enfants ne mangent pas autour de la/une même table. ‘In most large families, the children do not eat round the/UN same table.’

Finally, un même requires variability of situations.

(45) Utiliser un/le même mot de passe pour différents services, ordinateurs et

Sinn und Bedeutung 15 - [PDF Document] (145)

144 Charnavel

sites Internet augmente les risques de se faire voler des informations personnelles. ‘Using UN/the same password for different services, computers and websites increases the risks of having personal informations stolen.’

(46) Quatre générations sous un/le même toit ‘Four generations under UN/the same roof.’

(47) Un/le même mot peut avoir plusieurs sens. ‘UN/the same word may have several senses.’

(48) Si un/le même joueur fait plus de 5 fautes, il est disqualifié. ‘If UN/the same player makes more than 5 mistakes, he gets dis-qualified.’

(49) a. Dans chaque système planétaire, toutes les planètes tournent autour de la/une même étoile.

‘In each planetary system, every planet revolves around the/UN same star.’

b. Dans le système solaire, toutes les planètes tournent autour de la/*une même étoile. ‘In the solar system, every planet revolves around the/*UN same star.’

Thus in (49), the plurality of planets illustrates the requirement for même that there be several events: même expresses uniqueness across multiple events; furthermore, the plurality of planetary systems in (a) illustrates the requirement for the indefinite article combined with même that there be several situations: un même requires a covarying interpretation and expresses uniqueness relativized to a situation.

Based on Von Fintel (2004)'s idea that quantifiers have a hidden domain argument, I propose that un même is not only a quantifier over a plural event, but also contains a domain variable (resource situation pronoun) whose value has to be quantificationally bound; it cannot be identified by a contextually supplied situation. That's why un même is licensed by contexts which contain a quantifier over situations or worlds.

(50) [Dans chaque système planétaire]i, toutes les planètes tournent autour d'unei même étoile. ‘In each planetary system, every planet revolves around UN same star.’

On the other hand, the variable contained in le même can be either bound by the same operators (non specific reading) or identified by the context

Sinn und Bedeutung 15 - [PDF Document] (146)

French un même and Antispecificity 145

(specific reading). This is formalized below. In the case of un même expressing uniqueness relativized to a situation, s needs to be quantificationally bound, but since le même expresses uniqueness either relativized to a situation or in a particular context, s is either quantificationally bound or contextually supplied in this case.

(51) [[le/un même]] =λs.λY<et>.λZ<e,vt>. ∃x1, x2…xn ≤ x (en-1 ≠ en; n is a positive integer and n ≥ 2) such that Y(xn)(s)=1 and Z(xn)(en)(s)=1, and xn-1=xn in s.

The contrast between un même and le même is parallel to other phenomena. First, it is reminiscent of Florian Schwarz's dissertation (2009) concerned with the description and analysis of two semantically different types of definite articles in German (weak and strong).

(52) a. Hans ging zum Haus. (Schwarz 2009: 12) Hans went to-theweak house ‘Hans went to the house.’ b. Hans ging zu dem Haus. Hans went to-thestrong house ‘Hans went to the house.’

The weak article encodes uniqueness (relativized to a situation); the strong article is anaphoric in nature (dependent on a antecedent). The interpretation of the weak article definite depends on the interpretation of its situation pronoun, which can stand for the topic situation or a contextually supplied situation, or be quantificationally bound.

So, le même contrasts with un même like the strong article with the weak article in that it has an anaphoric capacity that un même lacks as shown in sentence-external readings. Like the weak article, un même expresses uniqueness relativized to a situation, which can be analyzed by use of a resource situation pronoun.

Nevertheless, the empirical divisions between the weak/strong articles and un/le même are different: the resource situation pronoun of the weak article can also stand for the topic situation or a contextually supplied situation while that of un même cannot: it has to be quantificationally bound. Moreover, the contrast between le même and un même is the reverse of the contrast observed by Beghelli and Stowell (1995) between each and every: the set variable of each must be identified by the context, while the set variable of every can also be bound by operators.

Sinn und Bedeutung 15 - [PDF Document] (147)

146 Charnavel

(53) a. Every dog has a tail. (Beghelli and Stowell 1995: 32) b. Each dog has a tail.

(53a) can be construed as a claim about dogs in general, whereas (53b) must be construed as a claim about a particular set of dogs previously mentioned in the discourse. Thus, the sentence with each means that there is a particular situation s, a set X of all dogs in s, such that all the members of X have a tail, while the sentence with every means that in the default situation s where X is the set of all dogs in s, all members of X have a tail. When every-DPs occur in generic contexts, they are interpreted as though they were universal-generic quantifiers because they contain restricted variables (ranging over sets) bound by a silent generic quantifier. When every occurs in a context associated with reference to a single situation time, it acquires its contextualized universal-distributive reading because it is bound by a silent definite quantifier (existential quantifier ranging over situation-time (existential counterpart of GEN)).

Note that Beghelli and Stowell distinguish quantifiers ranging over situation-times from quantifiers ranging over events. This fits the present analysis since un même and le même, being quantifiers over events, could not be bound by quantifiers over events themselves, but can only be bound by hierarchically higher operators like quantifiers over situations.

Domain variable bound by operators Domain variable contextually supplied

le même x x un même x

every x x each x

4 Conclusion To summarize, même presents two main puzzles: a compositionality problem and a definiteness issue. Concerning the first one, I hypothesized that le même and un même are existential quantifiers over an event that has been distributed over participants or times. As for the second problem which was my main concern here, I observed that French un même documents the availability of the indefinite determiner with same, even if le même already presents indefiniteness effects. The distribution of un même shows its antispecificity and therefore questions the following generalization: definites are assumed to be specific while indefinites are specific or non specific; but

Sinn und Bedeutung 15 - [PDF Document] (148)

French un même and Antispecificity 147

actually, le même is specific or non specific while un même is non specific (antispecificity). To account for that, I proposed that un même introduces a variable that has to be bound by operators over situations, while the variable introduced by le même can also be contextually supplied. I believe it would be worth further investigating this phenomenon and similar ones, as it may be fruitful to relate definiteness and specificity with the interpretation of domain variables in quantifiers.

References Barker, Chris. 2007. Parasitic Scope. Linguistic and Philosophy 30. 407–444. Beghelli, Filippo & Tim Stowell. 1997. Distributivity and Negation: the

Syntax of each and every. In Szabolcsi (ed.), Ways of Scope Taking. 71–109. Kluwer.

Brasoveanu, Adrian. 2009. Sentence-internal different as Quantifier-internal Anaphora. In Proceedings of the 27th West Coast Conference on Formal Linguistics. 72–80.

Carlson, Greg. 1987. Same and Different: some Consequences for Syntax and Semantics. Linguistics and Philosophy 10.

Charnavel, Isabelle. 2011. On the Sentence-internal Reading of French le même. In Proceedings of the 20th Semantics and Linguistic Theory Conference.

Enç, Mürvet. 1991. The Semantics of Specificity. Linguistics Inquiry 22 (1). 1–25.

Fintel, von, Kai. 2004. A Minimal Theory of Adverbial Quantification. In Barbara Partee & Hans Kamp (eds), Context Dependence in the Analysis of Linguistic Meaning, Volume 11 of Current Research in the Semantics/Pragmatics Interface. 137–175. Elsevier.

Moltmann, Friederike. 1992. Reciprocals and same/different: towards a Semantic Analysis. Linguistics and Philosophy 15. 411–462.

Solomon, Mike. 2009. Partitives and the Semantics of same. Sinn und Bedeutung 14 handout.

Schwarz, Florian. 2009. Two Types of Definites in Natural Language: Umass PhD. Thesis.

Sinn und Bedeutung 15 - [PDF Document] (149)

Sinn und Bedeutung 15 - [PDF Document] (150)

In Reich, Ingo et al. (eds.), Proceedings of Sinn & Bedeutung 15,pp. 149–164. Universaar – Saarland University Press: Saarbrücken, Germany, 2011.

Performative Verbs and Performative Acts∗

Cleo CondoravdiPalo Alto Research Center &

Stanford [emailprotected]

Sven LauerStanford University


Abstract. Searle (1989) posits a set of adequacy criteria for any account ofthe meaning and use of performative verbs, such as order or promise. Centralamong them are: (a) performative utterances are performances of the act namedby the performative verb; (b) performative utterances are self-verifying; (c) per-formative utterances achieve (a) and (b) in virtue of their literal meaning. He thenargues that the fundamental problem with assertoric accounts of performatives isthat they fail (b), and hence (a), because being committed to having an intentiondoes not guarantee having that intention. Relying on a uniform meaning for verbson their reportative and performative uses, we propose an assertoric analysis ofperformative utterances that does not require an actual intention for deriving (b),and hence can meet (a) and (c).

Explicit performative utterances are those whose illocutionary force is madeexplicit by the verbs appearing in them (Austin 1962):

(1) I (hereby) promise you to be there at five. (is a promise)

(2) I (hereby) order you to be there at five. (is an order)

(3) You are (hereby) ordered to report to jury duty. (is an order)

(1)–(3) look and behave syntactically like declarative sentences in every way.Hence there is no grammatical basis for the once popular claim that I promise/order spells out a ‘performative prefix’ that is silent in all other declaratives.Such an analysis, in any case, leaves unanswered the question of how illocu-tionary force is related to compositional meaning and, consequently, does notexplain how the first person and present tense are special, so that first-personpresent tense forms can spell out performative prefixes, while others cannot.Minimal variations in person or tense remove the ‘performative effect’:

(4) I promised you to be there at five. (is not a promise)

(5) He promises to be there at five. (is not a promise)

An attractive idea is that utterances of sentences like those in (1)–(3) are asser-∗ The names of the authors appear in alphabetical order.

Sinn und Bedeutung 15 - [PDF Document] (151)

150 Condoravdi & Lauer

tions, just like utterances of other declaratives, whose truth is somehow guar-anteed. In one form or another, this basic strategy has been pursued by a largenumber of authors ever since Austin (1962) (Lemmon 1962; Hedenius 1963;Bach & Harnish 1979; Ginet 1979; Bierwisch 1980; Leech 1983; among oth-ers). One type of account attributes self-verification to meaning proper. An-other type, most prominently exemplified by Bach & Harnish (1979), tries toderive the performative effect by means of an implicature-like inference thatthe hearer may draw based on the utterance of the explicit performative.

Searle’s (1989) ChallengeSearle (1989) mounts an argument against analyses of explicit performativeutterances as self-verifying assertions. He takes the argument to show that anassertoric account is impossible. Instead, we take it to pose a challenge that canbe met, provided one supplies the right semantics for the verbs involved.

Searle’s argument is based on the following desiderata he posits for anytheory of explicit performatives:

(a) performative utterances are performances of the act named by the per-formative verb;

(b) performative utterances are self-guaranteeing;(c) performative utterances achieve (a) and (b) in virtue of their literal mean-

ing, which, in turn, ought to be based on a uniform lexical meaning ofthe verb across performative and reportative uses.

According to Searle’s speech act theory, making a promise requires that thepromiser intend to do so, and similarly for other performative verbs (the sincer-ity condition). It follows that no assertoric account can meet (a-c): An assertioncannot ensure that the speaker has the necessary intention.

“Such an assertion does indeed commit the speaker to the exis-tence of the intention, but the commitment to having the intentiondoesn’t guarantee the actual presence of the intention.”

Searle (1989: 546)

Hence assertoric accounts must fail on (b), and, a forteriori, on (a) and (c).1

Although Searle’s argument is valid, his premise that for truth to be guar-anteed the speaker must have a particular intention is questionable. In the fol-lowing, we give an assertoric account that delivers on (a-c). We aim for an

1 It should be immediately clear that inference-based accounts cannot meet (a-c) above. If theoccurrence of the performative effect depends on the hearer drawing an inference, then such sen-tences could not be self-verifying, for the hearer may well fail to draw the inference.

Sinn und Bedeutung 15 - [PDF Document] (152)

Performative Verbs and Performative Acts 151

account on which the assertion of the explicit performative is the performanceof the act named by the performative verb. No hearer inferences are necessary.

1 Reportative and Performative UsesWhat is the meaning of the word order, then, so that it can have both reporta-tive uses – as in (6) – and performative uses – as in (7)?

(6) A ordered B to sign the report.

(7) [A to B] I order you to sign the report now.

The general strategy in this paper will be to ask what the truth conditions ofreportative uses of performative verbs are, and then see what happens if theseverbs are put in the first person singular present tense. The reason to start withthe reportative uses is that speakers have intuitions about their truth conditions.This is not true for performative uses, because these are always true when ut-tered, obscuring the truth-conditional content of the declarative sentence.2

An assertion of (6) takes for granted that A presumed to have authorityover B and implies that there was a communicative act from A to B. But whatkind of communicative act? (7) or, in the right context, (8a-c) would suffice.

(8) a. Sign the report now!b. You must sign the report now!c. I want you to sign the report now!

What do these sentences have in common? We claim it is this: In the rightcontext they commit A to a particular kind of preference for B signing thereport immediately.

If B accepts the utterance, he takes on a commitment to act as though he,too, prefers signing the report. If the report is co-present with A and B, he willsign it, if the report is in his office, he will leave to go there immediately, andso on. To comply with an order to p is to act as though one prefers p. One neednot actually prefer it, but one has to act as if one did. The authority mentionedabove amounts to this acceptance being socially or institutionally mandated.

Of course, B has the option to refuse to take on this commitment, in eitherof two ways: (i) he can deny A’s authority, (ii) while accepting the authority, hecan refuse to abide by it, thereby violating the institutional or social mandate.Crucially, in either case, (6) will still be true, as witnessed by the felicity of:

2 Szabolcsi (1982), in one of the earliest proposals for a compositional semantics of performativeutterances, already pointed out the importance of reportative uses.

Sinn und Bedeutung 15 - [PDF Document] (153)

152 Condoravdi & Lauer

(9) a. (6), but B refused to do it.b. (6), but B questioned his authority.

Not even uptake by the addressee is necessary for order to be appropriate, asseen in (10) and the naturally occurring (11):3

(10) (6), but B did not hear him.

(11) He ordered Kornilov to desist but either the message failed to reachthe general or he ignored it.4

What is necessary is that the speaker expected uptake to happen, arguably aminimal requirement for an act to count as a communicative event.

To sum up, all that is needed for (6) to be true and appropriate is that (i)there is a communicative act from A to B which commits A to a preference forB signing the report immediately and (ii) A presumes to have authority over B.The performative effect arises precisely when the utterance itself is a witnessfor the existential claim in (i).

There are two main ingredients in the meaning of order informally out-lined above: the notion of a preference, in particular a special kind of preferencethat guides action, and the notion of a commitment. The next two sections laysome conceptual groundwork before we spell out our analysis in section 4.

2 Representing PreferencesTo represent preferences that guide action, we need a way to represent prefer-ences of different strength. Kratzer’s (1981) theory of modality is not suitablefor this purpose. Suppose, for instance, that Sven desires to finish his paperand that he also wants to lie around all day, doing nothing. Modeling his pref-erences in the style of Kratzer, the propositions expressed by (12) and (13)would have to be part of Sven’s bouletic ordering source assigned to the actualworld:

(12) Sven finishes his paper.

(13) Sven lies around all day, doing nothing.

But then, Sven should be equally happy if he does nothing as he is if he finisheshis paper. We want to be able to explain why, given his knowledge that (12)and (13) are incompatible, he works on his paper. Intuitively, it is because thepreference expressed by (12) is more important than that expressed by (13).

3 We owe this observation to Lauri Karttunen.4

Sinn und Bedeutung 15 - [PDF Document] (154)

Performative Verbs and Performative Acts 153

Preference StructuresDefinition 1. A preference structure relative to an information state W is apair 〈P,≤〉, where P⊆℘(W ) and ≤ is a (weak) partial order on P.

We can now define a notion of consistency that is weaker than requiring thatall propositions in the preference structure be compatible:

Definition 2. A preference structure 〈P,≤〉 is consistent iff for any p,q ∈ Psuch that p∩q = /0, either p < q or q < p.

Since preference structures are defined relative to an information state W , con-sistency will require not only logically but also contextually incompatible propo-sitions to be strictly ranked. For example, if W is Sven’s doxastic state, and heknows that (12) and (13) are incompatible, for a bouletic preference structureof his to be consistent it must strictly rank the two propositions.

In general, bouletic preference structures need not be consistent, and theyoften will not be. We assume that the desires, preferences, and obligations ofvarious kinds of an agent A are represented by a set Pw(A) of preference struc-tures, some of which may be inconsistent, internally or with each other.

A consistent preference structure will give rise to a partial order≺ amongworlds. There are various ways to define this partial order, but for the presentpaper, we leave it open which definition is most appropriate. Nothing in whatfollows hinges on the choice. The basic intuition is that ≺ should be ‘lexi-cographic’: lower-ranked propositions in the preference structure should onlymake a difference for the ranking of two worlds w and v if they are on equalfooting with respect to all the higher-ranked propositions.

Consolidated PreferencesGiven the multitude of preference structures influencing an agent’s decisions,if an agent wants to act, he has to integrate these structures into a global one,resolving any conflict. Thus, a rational agent A in world w has a distinguished,consistent preference structure


⟩. We call this A’s effective pref-

erence structure in w.We require that Pw(A) ⊆

⋃Pw(A) and also that if p,q ∈ Pw(A) such that

there is 〈P,≤P〉 ∈ Pw(A) and p <P q and there is no 〈P′,<P′〉 ∈ Pw(A) such thatq≤ p, then p <Pw(A) q, ensuring that no spurious goals are introduced into theeffective preference structure and rankings that are consistent are retained.

In w, A’s induced preference order �Pw(A) will partially5determine theagent’s behavior: If A has the choice between w1 and w2 (as continuations of wdiffering in what action, if any, A performs), and w1 ≺ w2, then A will choose5 Only partially, as an agent may be genuinely indifferent between two possible courses of affairs.

Sinn und Bedeutung 15 - [PDF Document] (155)

154 Condoravdi & Lauer

w2. That is, the definition of � is a first step to defining a non-probabilistickind of Decision Theory,6 with preference structures corresponding to utilityfunctions in classical decision theory, while information states correspond tosubjective probability distributions.

We propose the following desiderata for a more developed version of sucha theory:7

Positive introspection for effective preferences If an agent a effectivelyprefers p, he believes that he does.

Negative introspection for effective preferences If an agent a believesthat he effectively prefers p, he does effectively prefer p.

3 CommitmentsThe idea that a main effect of utterances is to modify the commitments of theinterlocutors is an old one, going back at least to Hamblin (1971). More re-cently, it has been fruitfully developed by Gunlogson (2008) and Davis (2009),who take utterances to update commitment states, typically modeled as sets ofpropositions.

Commitments as Restricting Future States of the WorldHamblin and Gunlogson only model discourse commitments, that is, commit-ments that constrain the linguistic actions of the interlocutors in the future ofthe present discourse. This enables them to characterize commitments simplyas a set of ‘legal’ (Hamblin) or ‘expected’ (Gunlogson) future discourse states:If the discourse ends up in a state that is not in this set, something is off.

While such a model may be sufficient for what these authors were after,it is not quite enough in general. Commitments arising by linguistic meansalso constrain non-linguistic actions and actions that are performed after thediscourse has ended. Promises and orders are particularly obvious examples.

In order to capture this more general notion of commitment, we can thinkof the taking on of a commitment as excluding possible future states of theworld, thereby making certain future states of the world impossible. Given thisconception, we cannot just specify a set of ‘good’ futures (in which all com-mitments are honored), for, of course, taking on a commitment does not ex-clude the possibility of violating it. However, we can think of commitments asexcluding those futures in which the agent does not act according to the com-

6 By ‘decision theory,’ we mean any theory that models how agents choose actions on the basisof their beliefs and preferences. We use the term ‘classical decision theory’ for what is called‘decision theory’ in mathematics and economics.7 It should be kept in mind that what we want to model are conscious preferences. Thus, thesedesiderata are appropriate even though an agent may be influenced by factors he is not aware of.

Sinn und Bedeutung 15 - [PDF Document] (156)

Performative Verbs and Performative Acts 155

mitment, yet is not at fault. Before the commitment was taken on, this kind offuture was possible, afterwards it is impossible.

Consider the simple case of an agent committing himself to raise his handwhen prompted the next time. There are three kinds of possible futures:

(i) futures in which the agent is prompted and raises his hand(ii) futures in which the commitment is voided, either by being rescinded be-

fore the agent is prompted, or because it becomes evident that the agentwill not be prompted

(iii) futures in which the agent is prompted and does not raise his hand, butthe commitment was not voided before he was prompted.

Taking on the commitment excludes those futures of type (iii) in which theagent does not count as having violated a commitment.

Keeping a CommitmentIn the (somewhat contrived) example above, it is clear what ‘acting in accor-dance with the commitment’ amounts to (raising the hand when prompted),and also at which time the commitment has to be voided so as to not count asviolated (before the prompting).

In general, matters are more complicated. If I promise to meet you at theairport at noon tomorrow, what is required of me is not only to be at the airportat noon. Rather, what is required is a complex ensemble of actions that resultin me being at the airport at noon. Suppose the trip to the airport takes an hour,and I sleep in until 11:30. I am at fault, even if you call at 11:35 to tell me thatyour flight has been delayed by several hours and so I do not have to meet youat noon. You may never know that I violated my commitment, but I did violateit. On this conception, there is not only a time when the commitment was kept,there is also a time when the commitment was (first) violated: The (first) time Ifail to act in a way that would ensure my being at the airport at noon.

Commitments are always commitments to act. When we say ‘an agent iscommitted to believing the proposition p’, this is short for ‘the agent is com-mitted to act as though he believes p’. Similarly, ‘an agent is committed to an(effective) preference for p’ is short for ‘an agent is committed to act as thoughhe (effectively) prefers p to be actualized.’ This is exactly the right notion ofcommitment for promises and the like. In the example above, what I am com-mitted to is to act as though I effectively prefer to be at the airport at noon.Some of the required actions have to happen quite sometime before noon.

Sinn und Bedeutung 15 - [PDF Document] (157)

156 Condoravdi & Lauer

So we can characterize the notion of ‘taking on a commitment’ as follows:

(14) If an agent a takes on a commitment, he thereby excludes possiblefuture states in whicha. the agent does not act according to the commitment AND

b. the commitment is not voided before the agent fails to act accord-ing to the commitment AND

c. the commitment does not count as violated.

Construing commitments as commitments to act means that the features of thedecision theory from the last section get ‘lifted’ to the respective commitments:

Positive introspection for preference commitment If an agent is com-mitted to an effective preference for p, he is also committed to act as thoughhe believes he is committed to an effective preference for p.

Doxastic reduction for preference commitment If an agent is commit-ted to act as though he believes that he is committed to an effective preferencefor p, he is also committed to act as though he effectively prefers p.

Positive introspection for doxastic commitment If an agent is commit-ted to act as though he believes that he is committed to act as though he believesthat p, he is committed to act as though he believes that p.

We end this section by introducing the following bit of notation (omitting,for simplicity, a necessary temporal parameter, introduced later):

Definition 3. We let

PEPa(p) :={

w ∈W∣∣∣∣ p is a maximal element of a’s public

effective preference structure in w

}(Where p is a maximal element of a’s public effective preference structure iff ais committed to act as though p is a maximal element of his effective preferencestructure.)

Assertions and Public CommitmentsWe use a deliberately weak notion of assertion: All that it takes to assert is to(sincerely) utter a declarative sentence. We characterize assertions in terms oftheir minimal effect in the sense of Zeevat (2003). With Gunlogson (2003) andDavis (2009), we take this effect to be the coming about of a doxastic commit-ment on the part of the speaker. Additional properties of assertions can arguablybe explained as pragmatic inferences on the basis of this speaker commitment.8

8 A prominent example of such a secondary effect is that it becomes common ground that p. Wefollow Gunlogson and Davis in assuming that an assertion becomes part of the common ground

Sinn und Bedeutung 15 - [PDF Document] (158)

Performative Verbs and Performative Acts 157

(15) An assertion of a declarative φ in context C adds JφKC to the publicbeliefs of the speaker, thereby publicly committing the speaker to actas though he believes JφKC.

Assertions are, of course, communicative events. Let the totality of the doxasticcommitments of a speaker S resulting from a communicative event u be desig-nated as PBS[u] and PBt

S stand for the set of beliefs of S that become publiclymanifest at time t. We do not identify PBS[u] strictly with the truth-conditionalcontent of u. Rather, the commitment can come about as a result of the meaningof the utterance plus information available in the context in which it is made.9

Part of what it means to say that a commitment results from an eventis that the commitment comes about at the very end of the event. We henceassume the following principle, where tu is the final instant of the runtime of u:

(16) p ∈ PBS[u]⇔ (p ∈ PBtuS ) ∈ PBS[u]

Analogously, we let PEPS[u] and PEPtS refer to the set of preferences resulting

from u and that become publicly manifest at t, respectively, and assume

(17) p ∈ PEPS[u]⇔ (p ∈ PEPtuS ) ∈ PBS[u]

4 Explicit Performatives as Self-Verifying AssertionsIn this section, we present our assertoric analysis of explicit performatives us-ing the three verbs claim, promise and order, which are representative, inSearle’s (1975) classification, of ASSERTIVES, COMMISSIVES and DIREC-TIVES, respectively. What performative verbs have in common is that they allreport communicative events. In the following, we conceive of these events asconcrete particulars, and hence take every communicative event u to be associ-ated with a unique context c(u) whose speaker is the agent of u and whose timeis the runtime of u. The shape of the argument that the utterance ensures theperformative effect will be the same in all three cases, but the lexical seman-tics for the verb will get progressively more complex. What we have to showin each case is that an utterance of a sentence S with an explicit performativeverb is self-verifying, i.e., for any world w, if u is an utterance of S in w, thenw ∈ JSKc(u).

only as a secondary effect, after the hearer has accepted the assertion.9 We leave it open here whether the commitments a speaker takes on with an utterance can beidentified with Gricean speaker meaning.

Sinn und Bedeutung 15 - [PDF Document] (159)

158 Condoravdi & Lauer

Commitment to a Belief: ‘I claim that p’The problem posed by assertive performative verbs like claim and assert isnicely illustrated by what has come to be known as Cohen’s problem (Lycan(1999), based on Cohen (1964)). On the one hand, claim is ‘truth-conditionallytransparent’: the speaker of (18) cannot react to the continued absence of rainby saying ‘Well, I only said I CLAIMED that it was going to rain’. On the otherhand, claim obviously contributes to truth conditions: For example, (18) entailsthat somebody claims that it is going to rain.

(18) I claim that it is going to rain.

Cohen’s problem can be solved by analyzing claim as a performative verb. Thecontent of (18) is just a statement about what the speaker claims, but there isalso a performative effect, through which the speaker also becomes committedto the complement of claim in the way we demonstrate below.

The Reportative UseWhat has to be the case for (19) to be true?

(19) Peter claimed that it was going to rain.

There must have been a communicative event u from Peter (to someone). Whatkind of sentence out of Peter’s mouth could verify (19)? (18) would do, but sowould any utterance that, in its context, commits the speaker to the belief thatit is going to rain.

(20) w � claim(u,a, p) iffa. u is a communicative event from a: w � CEa(u)b. in c(u), u commits a to the belief that p: w � p ∈ PBa[u].

(18) and the plain assertion of (21) will bring about the required commitmentin any context in which they are sincerely uttered.

(21) It is going to rain.

However, recall that the commitments resulting from an utterance can go be-yond its truth-conditional content, hence, (19) can be supported by utterancesof sentences that have (21) as a contextual implication.

The Performative UseThe goal is to explain why, by virtue of uttering (18), a speaker is doxasticallycommitted to (21). Let u∗ be an utterance of (18) in context C∗ and world w∗.The truth-conditional content of (18) is given in (22), where the identification

Sinn und Bedeutung 15 - [PDF Document] (160)

Performative Verbs and Performative Acts 159

of the run time τ of the two utterance events is contributed by the simple presenttense. Given the semantics of claim in (20), (22) is equivalent to (23).

(22) {w |w � ∃u : τ(u) = τ(u∗)∧ claim(u,S,Rain)},where Rain = Jit is going to rainKC∗

(23) {w |w � ∃u : τ(u) = τ(u∗)∧CES(u)∧Rain ∈ PBS[u]}

u∗, as an assertion, commits the speaker to the belief in (22)/(23). The speakeris therefore committed to the belief in the existence of a communicative eventthat commits him to the belief that it is going to rain, i.e.

(24) w∗ � (23) ∈ PBS[u∗]

Therefore, at the final instant t∗ of τ(u∗), we have:

(25) w∗ �{

w∣∣w � Rain ∈ PBt∗


}∈ PBt∗


Given positive introspection for doxastic commitment, (25) reduces to (26):

(26) w∗ � Rain ∈ PBt∗S

(24) and (26) together imply (27), which by postulate (16) reduces to (28).

(27) w∗ � (Rain ∈ PBt∗S ) ∈ PBS[u∗]

(28) w∗ � Rain ∈ PBS[u∗]

This means that u∗ satisfies the conditions in (23) and hence w∗ ∈ (23), in otherwords, an utterance of (18) is necessarily self-verifying.

Commitment to an Effective Preference: ‘I promise to p’Moving to commissives, what has to be the case for (29) to be true?

(29) Peter promised (Mary) to get the tickets.

Once again, there has to have been a communicative event from Peter (to Mary)that creates a particular kind of commitment. And again a number of sentencescould have been uttered in order to make (29) true:

(30) a. I promise you to get the tickets.b. I will get the tickets.c. You will have the tickets tomorrow.

Sinn und Bedeutung 15 - [PDF Document] (161)

160 Condoravdi & Lauer

We propose the following semantics for promise:10

(31) w � promise(u,a,b, p) iffa. u is a communicative event from a to b: w � CEa→b(u)b. in c(u), u commits a to PEPa(p): w � p ∈ PEPS[u]

Thus, any utterance that verifies (29) publicly commits its speaker Peter (toMary) to effectively prefer to get the tickets. As before, (30b,c) will bring aboutthe requisite commitment only if the context is right, while the explicit perfor-mative (30a) will create it in any context in which it is sincerely uttered.

The Performative UseAn utterance u∗ of (30a) to addressee A in context C∗ and world w∗ commitsthe speaker S to acting as if he believes the proposition in (32):

(32) {w |w � ∃u : τ(u) = τ(u∗)∧CES→A(u)∧Tickets ∈ PEPS[u]},where Tickets = JS will get the ticketsKC∗

The derivation of the performative effect is as follows:

(33) w∗ � (32) ∈ PBS[u∗]

(34) w∗ �{

w∣∣w � Tickets ∈ PEPt∗


}∈ PBt∗


Given doxastic reduction for preference commitment, (34) reduces to (35):

(35) w∗ � Tickets ∈ PEPt∗S

(33) and (35) together imply (36), which by postulate (17) reduces to (37).

(36) w∗ � (Tickets ∈ PEPt∗S ) ∈ PBS[u∗]

(37) w∗ � Tickets ∈ PEPS[u∗]

We have thus derived that the assertion of (30a) is a witness for its own truth—and hence, an utterance of (30a) is necessarily self-verifying.

Commitment to an Effective Preference for an Effective Preference:‘I order you to p’Finally, what has to be the case for (38) to be true?

(38) Mary ordered Peter to sign the report immediately.

10 The semantics we give only spells out the truth-conditional part of the meaning of promise.There is a presuppositional part, as well. The presupposition, roughly, is that a presumed that b hasa stake in p.

Sinn und Bedeutung 15 - [PDF Document] (162)

Performative Verbs and Performative Acts 161

As before, there must have been a certain kind of communicative event fromMary to Peter. In the right context, an utterance of (7) or any of the sentencesin (8) will suffice.

(7) I order you to sign the report immediately!

(8) a. Sign the report immediately!b. I want you to sign the report immediately!c. You have to sign the report immediately!

In section 1, we said that order requires that the event commit the speaker to acertain kind of preference. We can now refine this claim. The event in questionmust commit the speaker to effectively prefer that the hearer commit himselfto effectively prefer that he signs the report immediately.11

(39) w � order(u,a,b, p) iffa. u is a communicative event from a to b: w � CEa→b(u)b. in c(u), u commits a to PEPa(PEPb(p)): w � Pb(p) ∈ PEPa[u],

where Pb(p) ={

w∣∣w � ∃t > τ(u) : p ∈ PEPt


}The Performative UseAn utterance u∗ of (7) to addressee A in context C∗ and world w∗ commits thespeaker S to believe the proposition in (40):

(40) {w |w � ∃u : τ(u) = τ(u∗)∧order(u,S,A,Sign)},where Sign = JA signs the report immediatelyKC∗

The derivation of the performative effect is like that for promise except that,given the lexical semantics we propose for order, the equivalent of (33) is (42):

(41) {w |w � ∃u : τ(u) = τ(u∗)∧CES(u)∧PA(Sign) ∈ PEPS[u]}(42) w∗ � (41) ∈ PBS[u∗]

From this, we can derive

(43) w∗ � PA(Sign) ∈ PEPS[u∗]

As before, this means that w∗ ∈ J(7)K, i.e. (7) is self-verifying.

11 Again, order also carries a presupposition, namely that a presumes to have authority over b withrespect to p, i.e. that b is socially or institutionally obligated to take on the commitment effectivelypreferred by a.

Sinn und Bedeutung 15 - [PDF Document] (163)

162 Condoravdi & Lauer

Features of the AnalysisWhich verbs give rise to explicit performative utterances? Our analysis pre-dicts that it is those verbs that denote communicative events and whose truth-conditional content is fully specified in terms of speaker commitments. Whilethis is the case for verbs like claim, promise, or order, it is not the case forverbs like insult, annoy, or frighten.

Our account, unlike some of its assertoric predecessors, derives the self-verification of explicit performative utterances without assuming that they areself-referential. They can be made self-referential, though, by the use of hereby,which on the present analysis is best seen as an adverbial modifier that requiresthe identification of the described event with the utterance event.

Another central issue about explicit performatives that our analysis canexplain is their interaction with the progressive. A well-known generalizationis that utterances in the progressive cannot (usually) be used performatively.Our account plus the assumption that performative verbs are accomplishmentsimplies that the utterance of a performative progressive sentence does not com-mit the speaker to the existence of a commitment. This is so because progres-sive sentences describing accomplishments do not entail the culmination of thedescribed event.

Our proposal is similar in several respects to two recent, independentlydeveloped accounts by Eckardt (2009) and Truckenbrodt (2009). We cannotundertake a detailed comparison here but we note that it differs in (a) how itderives the self-verifying property of performative utterances, (b) the lexicalmeaning it assumes for assertives, commissives and directives, (c) in the ex-planation of how performative utterances restrict possible future states of theworld.

5 Concluding RemarksSearle’s argument against assertoric accounts relies on the assumption that anintention is required for a speech act to happen. We circumvent the problem byrequiring only that the speaker be committed to having a belief or an intention(in our terms, an effective preference). On our view, what matters for speechacts, or at least the truth conditions of performative verbs, is public facts.

Our analysis can also readily meet a challenge brought up by Jary (2007).He argues that explicit performatives cannot be assertions because their contentgets added to the common ground automatically, rather than being conditionedon the acceptance of the addressee, as is the case for run-of-the-mill assertions.However, as Jary himself notes, the fact that the assertion happened alwaysautomatically enters the common ground. Our account then predicts Jary’s ob-

Sinn und Bedeutung 15 - [PDF Document] (164)

Performative Verbs and Performative Acts 163

servation. Since the utterance itself is a witness for its own truth, the contentof the assertion is entailed by the fact that the assertion happened, and so thiscontent will become part of the common ground automatically.

ReferencesAustin, J. L. 1962. How to do things with words. Cambridge, MA: Harvard

University Press.Bach, K. & R. M. Harnish. 1979. Linguistic communication and speech acts.

Cambridge, MA: MIT Press.Bierwisch, Manfred. 1980. Semantic structure and illocutionary force. In

John R. Searle, Ferenc Kiefer & Manfred Bierwisch (eds.), Speech acttheory and pragmatics, 1–35. Dordrecht: Reidel.

Cohen, L. Jonathan. 1964. Do illocutionary forces exist? The PhilosophicalQuarterly 14(55). 118–137.

Davis, Christopher. 2009. Decisions, dynamics, and the Japanese particle yo.Journal of Semantics 26(4). 329–366.

Eckardt, Regine. 2009. Integrated speech acts. Manuscript, University of Göt-tingen.

Ginet, Carl. 1979. Performativity. Linguistics and Philosophy 3(2). 245–265.Gunlogson, Christine. 2003. True to form: Rising and falling declaratives in

English. New York: Routledge.Gunlogson, Christine. 2008. A question of commitment. Belgian Journal of

Linguistics 22(1). 101–136.Hamblin, C. L. 1971. Mathematical models of dialogue. Theoria 37(2). 130–

155.Hedenius, I. 1963. Performatives. Theoria 29. 115–136.Jary, Mark. 2007. Are explicit performatives assertions? Linguistics and Phi-

losophy 30(2). 207–234.Kratzer, Angelika. 1981. The notional category of modality. In H. J. Eikmeyer

& H. Rieser (eds.), Words, worlds, and contexts. New approaches inword semantics, 38–74. Berlin: de Gruyter.

Leech, Geoffrey N. 1983. Principles of pragmatics, vol. 30 Longman linguis-tics library. London and New York: Longman.

Lemmon, E. J. 1962. On sentences verifiable by their use. Analysis 22(4).86–89.

Lycan, William G. 1999. Philosophy of language: A contemporary introduc-tion. Routledge.

Searle, John R. 1975. A taxonomy of illocutionary acts. In K. Gunderson (ed.),Language, mind, and knowledge, Minneapolis: University of Minnesota

Sinn und Bedeutung 15 - [PDF Document] (165)

164 Condoravdi & Lauer

Press.Searle, John R. 1989. How performatives work. Linguistics and Philosophy

12(5). 535–558.Szabolcsi, Anna. 1982. Model theoretic semantics of performatives. In Hun-

garian and general linguistics, 515–535. John Benjamins.Truckenbrodt, Hubert. 2009. Performatives and agreement. Manuscript, ZAS,

Berlin.Zeevat, Henk. 2003. The syntax semantics interface of speech act markers.

In Proceedings Diabruck, 7th workshop on the semantics and the prag-matics of dialogue, Wallerfangen.

Sinn und Bedeutung 15 - [PDF Document] (166)

Identity and Definiteness in Chinese wh-­‐Conditionals

Stephen Crain Macquarie University


Qiong-Peng Luo Macquarie University


Abstract. Previous studies of Chinese wh-conditionals leave several issues unresolved, including (i) definiteness effects; (ii) apparent violations of the novelty condition and (iii) accounting for the range of readings of Chinese wh-conditionals. We attempt to resolve some of these issues by analysing wh-indefinites as unique indefinites and wh-conditionals as special instances of topic-comment structures (i.e. wh-conditionals are topic-comment structures with an identity relation). Chinese wh-conditionals can refer to either a single situation or multiple situations, leading to either a definite interpretation or a generic interpretation respectively.

1 Introduction For several decades, the semantics of wh-indefinites and wh-conditionals in Chinese has been a topic of debate for Chinese linguistics. In this study, we offer a somewhat novel analysis for wh-conditionals. A typical wh-conditional always contains a pair of matching wh-phrases, one in the antecedent clause and the other in the consequent clause. The wh-phrases in the antecedent and consequent clauses must be identical in number, form and reference. We add one more observation: Chinese wh-conditionals sometimes have an additional flavour of definiteness, semantically akin to free relatives in English.

Several accounts of Chinese wh-conditionals have been advanced in the literature. The most frequently cited account, by Cheng and Huang (1996), treats wh-indefinites as recurring indefinite expressions, but this appears to violate the novelty condition, which requires indefinites to introduce novel entities into the domain of discourse. To circumvent this problem, Chierchia (2000) proposes that wh-indefinites in Chinese are indefinite pronouns (i.e. pronominals), thus they can appear in the consequent of wh-conditionals without violating the novelty condition. However, Chierchia’s account does not explain why ordinary wh-indefinites display Principle C effects, a finding that seems to indicate that wh-indefinites are R-expressions rather than pronominals. We propose to reconcile the tension inherited from previous research by analyzing wh-conditionals as identity statements, which are not

In Reich, Ingo et al. (eds.), Proceedings of Sinn & Bedeutung 15, pp. 165–179. Universaar – Saarland University Press: Saarbrücken, Germany, 2011.

Sinn und Bedeutung 15 - [PDF Document] (167)

166 Crain & Luo

subject to the novelty condition. On the present account, Chinese wh-con-ditionals are ambiguous between being correlatives and conditionals. On one hand, when definiteness is added into the equation, wh-conditionals can be seen to share properties with free relatives in that they refer to a particular (unique) individual in a particular situation. When the context/antecedent establishes a plurality of situations, the unique individuals picked up by wh-phrases get relativised to situations, and the identity of the referent is not known, or not relevant. This reading is semantically akin to–ever free re- latives in English. This reading involves universal quantification over situations. Therefore, the intuitive insight in Cheng & Huang (1996) is intact on the current account; wh-conditionals are donkey conditionals and have a generic interpretation.

2 Chinese wh-­‐Conditionals: Definiteness Effect The seminal Cheng & Huang (1996) summarize the typical properties of wh-conditionals as follows (Cheng & Huang 1996:132):

(1) Properties of wh-conditionals a. The (donkey) anaphor must take the form of a wh-word b. The (donkey) wh-word must be identical to the wh-word in the

antecedent clause c. There must be an element in the consequent clause referring back the

wh-word in the antecedent clause

What appears mysterious here is that unlike donkey conditionals in English, where the anaphors always take the form of a pronominal, Chinese wh-conditionals take an identical wh-word as the donkey anaphor. This is the notorious ‘matching effect’: wh-phrases in the antecedent and consequent clauses of wh-conditionals must be identical in number, form and reference. Even minor variations are unacceptable. Example (2) below illustrates a typical wh-conditional in Chinese, while (3) illustrates the matching effect:

(2) Shei xian lai, shei xian chi. who first come who first eat Lit.: ‘If X comes first, X eats first’

(3) *Shei xian lai, shenme ren /tongyang de ren xian chi. who first come what person the-same DE person first eat

Cheng & Huang analyse the wh-conditionals as a case of ‘unselective binding’ a la Heim (1982) and Kamp (1981). They treat wh-phrases in wh-conditionals as indefinites (i.e. variables) that are unselectively bound by a

Sinn und Bedeutung 15 - [PDF Document] (168)

Identity and Definiteness in Chinese wh-Conditionals 167

default universal necessity operator. The implicit operator provides universal quantificational force for wh-conditionals. Their semantic representation for (2) are provided as (4) below:

(4) ∀x (x come first → x eats first) (Cheng & Huang 1996:132)

According to Cheng & Huang, (2) means everybody who comes first eats first. This semantics has a plurality commitment. It is committed to multiple comers and eaters. However, intuitively, (2) is true if for one particular situation, say, Ann’s birthday party tonight, there turns out to be exactly one individual who comes first and eats first. It is semantically odd to say everybody in the room studies the kangaroo if there is exactly one man in the room. If this is the case, it shows (2) has a unique reading in the sense of Kadmon (1990).

The unique reading of wh-conditionals correlates with the definiteness effect of wh-conditionals, an observation missed in Cheng & Huang (1996). The definiteness effect of wh-conditionals can be illustrated by example (5), which shows that the wh-indefinite in the antecedent clause can be referentially linked to a partitive expression in the consequent clause:

(5) Shenme ban biaoxian hao, what class perform well shenme ban de sanfenzhiyi jiu keyi dedao jiangli. what class DE one third then can get reward ‘One third of whatever class that perform(s) well will get a reward’

Example (5) casts doubt on Cheng & Huang’s claim that wh-indefinites in Chinese wh-conditionals are genuine indefinites, because the wh-phrase is used as the complement of a partitive with the form ‘NP of wh-NP’. An ordinary indefinite cannot be used as the complement DP in a partitive. It is well-known that the partitives with the form ‘NP of DP’ are subject to the Partitive Constraint (Jackendoff 1972, Barwise & Cooper 1981).

If we take a stand that wh-indefinites in wh-conditionals are definite description-like expressions, we may be able to capture both the uniqueness and the definiteness effect.

3 The Novelty Condition and Principle C Cheng & Huang treat wh-indefinites as Heimian indefinites (e.g. a farmer, a donkey, etc.), but this runs into a problem with the novelty condition. Ordinary indefinites are subject to the novelty condition (cf. Heim 1982, 365f):

Sinn und Bedeutung 15 - [PDF Document] (169)

168 Crain & Luo

(6) * If a mani comes first, a mani eats first.

As example (6) indicates, ordinary indefinites are required to introduce novel entities into the discourse. If wh-phrases are like indefinites, they should each introduce a novel entity to the discourse. This prediction hasn’t been borne out, because in Chinese wh-conditionals, the wh-phrase in the antecedent and the one in the consequent are identical in reference. In other words, the wh-phrase in the consequent of a conditional introduces a familiar referent rather than a novel one. As Chierchia (2000:17) puts, this represents a very bizarre picture:

(7) a. wh-words must introduce a novel variable in the antecedent of a con-ditional

b. wh-words must introduce a non-novel variable in the consequent of a conditional

Chierchia has convincingly shown that if (7) is right, then we no longer have a predictive theory of indefinites. The question is why Chinese wh-conditionals bluntly violate this novelty condition, which is supposed to be obeyed by indefinites generally.

To solve this problem, Chierchia proposes that wh-indefinites in Chinese are indefinite pronouns (i.e. pronominals). This explains why wh-indefinites can appear in the consequent clause of wh-conditionals without violating the novelty condition. A pronominal can be used as a discourse anaphor. A simple example would illustrate this idea:

(8) If a mani comes first, hei eats first.

At first glance, this seems to be a reasonable solution. Some issues need to be addressed, however. First, if wh-phrases in Chinese are indeed indefinite pronouns (i.e. pronominals), we expect they should always introduce a familiar discourse referent in the antecedent of a conditional, as pronominals (and definite descriptions) always do. But a wh-phrase in the antecedent of a conditional, however, doesn’t require a linguistic antecedent. One might wonder why the familiarity condition doesn’t apply here. The second problem is more severe. On Chierchia’s account, wh-phrases are expected to be subject to Principle B (because they are pronominals) and pattern with ordinary pronouns. However, wh-phrases in Chinese display Principle C effect, a fact unexpected on Chierchia’s analysis. Consider the following examples:

(9) a. Sheii shuo tai xihuan wo? who said he like me

Sinn und Bedeutung 15 - [PDF Document] (170)

Identity and Definiteness in Chinese wh-Conditionals 169

‘Who said he likes me?’ {Johni said hei likes me, Peterj said hej like me, …} b. *Tai shuo sheii xihuan wo? he said who like me ‘Who did he say likes me?’ {hei said Johni likes me, hej said Peterj likes me, …}

(9b) is a strong crossover case. The contrast between (9a) and (9b) indicates Chinese wh-phrases are not like pronominals but R-expressions. The following examples adopted from Tran & Bruening (2006) constitute another supporting observation:

(10) a. *Tai shuo sheii xihuan wo meimei? he said who like my sister ‘Who did he say like my sister?’ b. Tai zongshi shuo *sheii / tai xihuan wo meimei. he always said who he/she like my sister ‘Hei (always) says *whoi/ hei likes my sister’ c. Sheii (yaoshi) shuo tai /*sheii xihuan wo meimei, wo jiu zou ta. who if say he who like my sister I then hit he ‘If somebodyi says hei/*whoi likes my sister, I will hit him’

As all the examples under (10) clearly indicates, Chinese wh-phrases stand with R-expressions rather than pronouns. We face a paradoxical dilemma here. On one hand, if Chierchia’s proposal is indeed right, then we have to explain why wh-phrases display Principle C effect everywhere else. On the other hand, if wh-phrases are not pronominals, why can they appear in the consequent clause and remain anaphorically linked to the wh-phrase in the antecedent in wh-conditionals?

4 Indefinites and Uniqueness The definite reading of wh-conditionals is most ready when a unique referent is being established. To consider:

(11) A: (Zai zheci xuanju zhong), Zhang San bu xihuan shei? (in this election,) Zhang San NEG like who ‘Whom doesn’t Zhang San like (in this election)?’ B: Shei bu tou Wang Wu de piao, who neg vote Wang Wu DE vote Zhang San jiu bu xihuan shei. Zhang San then NEG like who

Sinn und Bedeutung 15 - [PDF Document] (171)

170 Crain & Luo

‘Who doesn’t vote for Wang Wu, Zhang San then doesn’t like who.’

When uttered out of the blue (i.e. in a neutral context), (11B) can mean Zhang San hates whoever that doesn’t vote for Wang Wu. If there is more than one event involving voting for or against Wang Wu, then Zhang San would have correspondingly more than one (possibly different) persons to dislike. This reading is akin to general statement about Zhang San’s personal disposition. However, when the antecedent establishes some unique referent, like in (11A) above, (11B) is true in the situation that there is exactly one person, say, Li Si, who doesn’t vote for Wang Wu, and Zhang San dislikes Li Si. However, (11B) also allows a multiple-individual reading. Suppose there are three persons, Li Si, Ma Liu and Zhang Qi, who decide not to vote for Wang Wu, then (11B) is true only when Zhang San dislikes all the persons that don’t vote for Wang Wu (in this case, they are Li Si, Ma Liu and Zhang Qi).

The exactly one reading is the uniqueness reading. But what is uniqueness? How can an indefinite generate a unique interpretation? In the literature, it has been reported that an indefinite under certain circ*mstances can have a unique reading (cf. Evans (1980), Kadmon (1990), Heim (1990), among others). There are multiple ways to encode the uniqueness into semantic representations. We follow Brasoveanu (2007, 2008), who adopts a Russellian treatment. The Russellian semantics of definites consists in existence, maximality and singleton presuppositions. This can be demon-strated below: When the wh-phrase denotes a singleton:

(12) Shei xian lai ∃X [X≠∅ & X= {y: person (y) & first_come (y)} & #X=1] existence maximality singleton uniqueness

When the wh-phrase denotes a plurality:

(13) Na-xie ren xian lai which-cl (pl.) person first come ∃X[X≠∅ & X= {y: person (y) & first_come (y)} & #X>1] existence maximality plural

When the wh-phrases denote a plurality, we assume there is a maximality operation in the sense of Link (1983) and Grosu & Landman (1998) that turns the plurality into a maximalized individual. In lattice-theoretic terms, if a and b are individuals, then the sum of a and b (written as a ⊕ b) is also an

Sinn und Bedeutung 15 - [PDF Document] (172)

Identity and Definiteness in Chinese wh-Conditionals 171

individual. The technical details of this lattice-theoretic approach to plurality are immaterial here. We entertain here the maximal individual of a poset <X, ≤> is the least upper bound of X.

The MAX operation always returns a plural individual. In this sense it is still singular, and the wh-phrases remain unique. Kadmon (1990) has another example that shows uniqueness is related to maximal collections. In the following (14), they /three of them refer to the maximal collection of the chairs that Leif owns.

(14) Leif has four chairs. They / Three of them are in the kitchen. (ex. 24)

This uniqueness-based account offers a straightforward explanation for the definiteness effect, for both the singular-individual reading and the multiple-individual reading. Our analysis predicts that the following sentence is ambiguous between distributive and collective interpretations:

(15) Shenme ban biaoxian hao, what class perform well shenme ban de sanfenzhiyi jiu keyi dedao jiangli. what class DE one-third then can get reward ‘One third of whatever class(es) that perform well get(s) a reward’

The sentence allows both the distributive reading and collective reading. On the distributive reading, it means for each class that performs well, one third of its members will be rewarded. The other reading, i.e. collective reading is compatible with the situation that for some class, none of its members get rewarded, while for some other classes, all of the members get rewarded. While this ambiguity can be attributed to a lack of number specification in nominal quantification in Chinese, the definiteness/uniqueness plays an essential role here.

5 Wh-­‐Conditionals as Identity Statements We have shown that the definiteness effect that remains elusive on the previous accounts can be captured by assuming wh-indefinites encode uniqueness. The uniqueness effect shows up when anaphora is attempted. However, there is a notable difficulty with this claim. While a unique indefinite is always referred back by a pronoun (e.g. Leif has a chair. It is in the kitchen), in wh-conditionals, the anaphor is an identical wh-phrase rather than a pronoun. How to account for this matching requirement in wh-conditionals?

Sinn und Bedeutung 15 - [PDF Document] (173)

172 Crain & Luo

A straightforward solution to this problem is to treat wh-conditionals in Chinese as identity statements. If wh-conditionals are treated along with identity statement, we will also be able to reconcile the tension between the novelty condition and Principle C. It is well-known that overt identity statements are immune to the novelty condition:

(16) A man who drinks alcopops is a man who gets a hangover.

In (16), the indefinite a man doesn’t c-command the other one. However, the novelty condition doesn’t apply here, and the indefinite expressions are happy to remain identical in reference. We assume in identity statements like (20), the novelty condition is being overridden here. It is being overridden because there is an overt identity operation that forces the indefinite expressions to pick up the same referent. In another word, the novelty condition is an Elsewhere Condition (EC) which applies only when it can. If Chinese wh-conditionals are subject to a similar identity operation, then we find a way to reconcile the tension between the novelty condition and Principle C. But how could this be achieved?

On the present account, wh-indefinites are subject to a σ-operation, where σ should be understood to stand for uniqueness:

(17) Shei xian lai, shei xian chi. The antecedent: [[ shei xian lai ]] = σx. person(x) & first_come (x)

The issue here is how the wh-indefinite in the consequent clause is being interpreted. We assume there is a covert identity operation:

(18) Shei xian lai, shei xian chi [who first come]x λx [first eat [σy [person (y) & y=x ]]]

On this account, the antecedent wh-indefinite shei xian lai ‘who comes first’ binds the variables x by λ-abstraction. The wh-anaphor is interpreted as a definite description, introducing a variable that is identical to the one previously introduced, which is x in the antecedent.

It is been proposed, since Cooper (1979), that donkey anaphors should be interpreted as generalized D-type pronouns (cf. also Heim & Kratzer 1998, Elbourne 2005, among others). The D-type pronouns contain both a definite description and a free relation variable R which helps fix the referent of the definite description. Cooper assumes R is provided by pragmatic saliency. This idea has been challenged by Heim (1990), who notices that donkey anaphora is subject to a condition which she dubs as Formal Link Condition, that is, the donkey anaphor requires an explicit linguistic antecedent (e.g. every man who has a wife is sitting next to her vs. */??every married man is

Sinn und Bedeutung 15 - [PDF Document] (174)

Identity and Definiteness in Chinese wh-Conditionals 173

sitting next to her). Chinese wh-conditionals may constitute another supporting evidence for the D-type pronoun analysis for donkey anaphors. Instead of looking for any linguistic antecedent, the wh-anaphor looks for an identical antecedent to fix its referent. R in this case is always provided by an explicit linguistic antecedent. And ‘identity’ is to be understood in Leibniz’s way (i.e. ‘x =y’ is true iff for any predicate P, P(x) if and only if P(y)). This treatment yields the correct semantics for wh-conditionals:

(19) a. the antecedent: [[ shei xian lai ]] = σx. [person (x) & first_come (x)] b. the wh-anaphor: [[ shei (xian chi) ]] = σy.[person (y) & y=x & R(y))] c. R → λx. first_come (x) d. the wh-anaphor: [[ shei (xian chi) ]] = σx. [person (x) & first_come

(x)] e. the consequent:λz. first_eat (z) (σx. [person (x) & first_come (x)]) = first_eat (σx. (person (x) & first_come (x))) g. [[ shei xian lai, shei xian chi]] =1 iff the individual who comes first is

the individual who eats first.

On this account, wh-conditionals are semantically akin to free relatives in English. Despite the structural differences, it is easy to see Chinese wh-conditionals and English free relatives may share a common semantics, since all English free relatives can be translated as identity statements (cf. Moltmann (2010)):1

(20) Whoever comes first eats first = the first comer is the first eater I don’t like whatever you bought = the thing(s) you bought is(are) the thing(s) I don’t like

The matching requirement provides another independent evidence for this analysis. We assume without the copula to mark identity in Chinese wh-conditionals, identity of form is a prerequisite to identity of reference (see (3)). Not surprisingly, we find the same form-matching restriction is also operative in English identity statements. Consider the following examples:

(21) */? A man who drinks alcopops is someone / the same person / the man

1 The structural differences between Chinese wh-conditionals and English free relatives may turn out to be superficial. Citko (2001) proposes that in a simple free relative like John ate what Mary cooked, the single instance of what is an argument of both ate and cooked. However, due to Deletion under Identity, the lower copy what gets deleted at PF. The only difference between Chinese and English, viewed in this light, is unlike English, the two copies of the wh-indefinite must stay at PF in Chinese.

Sinn und Bedeutung 15 - [PDF Document] (175)

174 Crain & Luo

who gets a hangover.

To summarize, it is the identity relation that is responsible for the identity in reference between the wh-indefinites and the inapplicability of the novelty condition in wh-conditionals. This identity reading renders wh-conditionals semantically akin to identity statement.

The above discussion results in a novel syntactic analysis for wh-conditionals. That is, wh-conditionals are topic-comment structures like correlatives (cf. Bittner 2001, Dayal 1997, among others), in which the antecedent wh-indefinite is topical, which is commented by the consequent wh-clause. We believe this analysis is on the right track, for several reasons. First, to treat wh-conditionals as topic-comment structures, we can derive the identity in reference between the wh-indefinites effortlessly. All we need to do is to assume the relationship between the topic and comment is that of identity. Second, the analysis suggests a more motivated explanation for the matching requirement in wh-conditionals. On this analysis, the wh-indefinite antecedents are topics, and we cannot mark an indefinite as topical (by means of wh-morphology) and not comment about it: ‘the intuitive idea is that topic-comment sequencing presupposes that the comment is about the topic. It requires … every topical discourse referent introduced in the topic updated to be picked up by an anaphoric element in the comment update’ (Bittner 2001). We believe this move (i.e. to treat wh-conditionals as topic-comment structures) is welcome. Recently, it has been frequently proposed that conditionals are topic-comment constructions (cf. Lewis 1973, Bittner 2001, Schlenker 2004, Ebert, Endriss & Hinterwimmer 2008, among others).

6 On the Generic Interpretation of wh-­‐Conditionals In addition to the definite interpretation, wh-conditionals can also be used as general statements and are open to a generic interpretation. This interpretation has ignorance, indifference and free choice implications.

6.1 The Ignorance Implication

Wh-conditionals have some ignorance implications (i.e. the speaker/agent’s epistemic uncertainty about identity of the referent denoted by the wh-phrase, or more plainly, the speaker/agent doesn’t know who has the property P). So (22a) has some implication as (22b):

(22) a. Shei xian lai, shei xian chi. who first come who first eat b. The person who comes first eats first, but I don’t know who will be

the one that comes first

Sinn und Bedeutung 15 - [PDF Document] (176)

Identity and Definiteness in Chinese wh-Conditionals 175

6.2 The Indifference Implication

Wh-conditionals also have indifference implications (i.e. the speaker/agent’s intentional or unintentional indiscriminateness with respect to the identity of the referent denoted by the wh-phrase, or more plainly, the speaker/agent doesn’t care who has the property P). (23a) has some implication as shown by (23b):

(23) a. Shei zuihou lai wanhui, shei xi wan who last come party who wash dish b. the person who arrived last for the party washes the dishes Counterfactual implication: it could be anyone else that washed the dishes if he was the last person for the party

6.3 Free Choice Implication

Wh-conditionals also have some free choice implication under certain circ*mstance. To consider:

Context: the university requires 50 credits for a bachelor’s degree, and Mary has already got 47 credits. To fulfill the university’s requirement, Mary has to get 3 more credits. There are three courses Mary can register for this purpose. Each course has 3 credits. The following sentence is felicitous:

(24) Ni xuan na-men kecheng, na-men kecheng jiu keyi you choose which-CL course which-CL course then can rang ni biye. let you graduate ‘Whichever course you take can let you graduate’

6.4 Deriving the Generic Interpretation

These observations bring wh-conditionals semantically closer to –ever FRs in English. Dayal (1997) argues that –ever FRs in English always involve some universal quantification over identity alternatives to the worlds of evaluation. Following Dayal, we assume the generic reading of Chinese wh-conditionals are derived in a similar way. The wh-conditionals contain a null adverbial quantifier GEN over world variables. And wh-phrases are concepts, i.e. from possible worlds to individuals:

(25) [[ shei xian lai ]] = λi. σx[first-come (x)](i)

(26) GEN ⇒ λPλQ.∀i-Alt∈f(w)(s)){P(i),Q(i)}, where (i) f(w)(s) is the set of worlds the speaker’s belief hold and (ii) a world w’∈f(w)(s) is an i-alternative iff there exists some w’’ such that σx[P(w’)(x)]≠σx[P(w’’)(x)]

Sinn und Bedeutung 15 - [PDF Document] (177)

176 Crain & Luo

This semantics captures the modal implications (i.e. the ignorance, indifference and free choice implications) by treating them as presuppositional content of wh-clauses and attributes the generic reading to a generic context.2 The unique referent denoted by the wh-phrase is being relativised to worlds, i.e. for each world, there is a unique individual involved in it. And quantification is over the worlds rather than individuals. We arrive at the following truth conditions for the generic reading:

(27) a. [[ shei xian lai, shei xian chi]] = 1 iff ∀i∈f(w)(s)){first_eat (i) (σx (person (x) & first_come (x) (i))} b. As far as the speaker’s belief is concerned, the first comer is the first


7 Are Chinese wh-­‐Conditionals Ambiguous? The previous discussion unambiguously leads to an ambiguous end, namely, that Chinese wh-conditionals are ambiguous. Semantically, Chinese wh-conditionals are akin to English FRs, which have two varieties: plain FRs and –ever FRs. English plain FRs are argued to have a prima facie definite/unique interpretation, while –ever FRs have some universal quantification interpretation (cf. Jacobson (1995), among others). Dayal (1997) proposes the universal quantificational force of –ever FRs is contributed by ever, which adds some modality to the semantic representation and renders FRs to be interpreted attributively.3 A plausible assumption extending to Chinese wh-conditionals is that Chinese wh-conditionals conflate this distinction (between plain FRs and –ever FRs) and are always open to two interpretations. Chinese lacks a lexical item like ever for the generic interpretation, and sometimes only the context/pragmatics can tell which reading is the most salient one.

At this moment, we should give some credit to Cheng & Huang (1996), who analyse wh-conditionals on a par with donkey conditionals in English. The ambiguity between definite and generic readings of wh-conditionals is also present in English donkey conditionals. Kadmon (1990) observes that donkey conditionals have both an absolute unique (definite) reading and a universal reading. She distinguishes one-case conditionals from multi-case conditionals (e.g. one-case conditionals: If there is a doctor in London and

2 For a slightly different version about the modal flavour of –ever FRs please see von Fintel (2000). Limit of space prohibits a fuller comparison and implementation of those ideas. 3 Donnellan (1966) distinguishes two uses of definites: referential vs. attributive. According to Dayal, the primary semantic function of ever is to force the FRs to be read attributively. Otherwise, FRs always receive a referential /absolute unique reading.

Sinn und Bedeutung 15 - [PDF Document] (178)

Identity and Definiteness in Chinese wh-Conditionals 177

he is Welsh, then we are all set vs. multi-case conditionals: If a semanticist hears of a good job, she applies for it).

We propose the choice between the two interpretations is regulated by context in Chinese. When the context is unspecified about the fixation of the referent, it has a generic reading, whereas when the context imposes some absolute uniqueness requirement of the referent, it has a definite reading:

(28) Wo wangji ni jie-le ji-ben shu gei wo le I forget you lend-ASP how-many-CL book to me ASP Danshi, ni jie-le shenme gei wo, wo jiu huan-gei ni But you lend-ASP what to me I then return-to you shenme le. what ASP I don’t remember how many books you lent me, but I’ve returned to you whatever books you lent me I’ve returned to you the books you lent me I’ve returned to you all the books you lent me

In (28), when the antecedent specifies a particular case/situation (i.e. book-lending by you to me), the generic reading is no longer the preferred one. And the ignorance and indifference implications also disappear. (28) simply expresses the speaker has returned all the books the addressee lent to him.

Semantically, the difference between the definite vs. universal readings of Chinese wh-conditionals boils down to a difference in granularity level of the quantification (see Brasoveanu 2007). The quantification can be coarse-grained, i.e. we ‘collectively’ quantify over topical cases/situations, which boils down to quantifying over topical individuals – and the consequent clause is predicated about these individuals. This yields the definite /unique reading. Alternatively, the quantification can be fine-grained, i.e. we ‘distributively’ quantify over the topical cases/situations introduced by the antecedent – and the consequent clause is predicated of each of such cases/situations. This yields the universal interpretation.

References Bhatt, Rajesh & Roumyana, Pancheva. 2006. Conditionals. In Martin

Evearert et al. (eds.), The Blackwell Companion to Syntax I, 638–687. Oxford: Blackwell.

Bittner, Maria. 2001. Topical referents for individuals and possibilities. In R. Hastings et al. (eds.), Proceedings of SALT 11, 36–55.

Sinn und Bedeutung 15 - [PDF Document] (179)

178 Crain & Luo

Bruening, Benjamin & Tran, Thuan. 2006. Wh-conditionals in Vietnames and Chinese: Against unselective binding. Paper presented at Berkeley Linguistic Society 32. University of California at Berkeley.

Cheng, Lisa & Huang, James. 1996. Two types of donkey sentences. Natural Language Semantics 4. 121–163.

Chierchia, Gennaro. 2000. Chinese conditionals and the theory of conditionals. Journal of East Asian Linguistics 9. 1–54.

Citko, Barbara. 2001. Deletion under identity in relative clauses. In M. Kim & U. Straus (eds.), Proceedings of NELS 31, 131–145. GLSA: Univer-sity of Massachusetts, Amherst.

Cooper, Robin. 1979. The interpretation of pronouns. In F. Heny and H. Schnelle (eds.), Syntax and Semantics 10, 61–92. New York: Academic Press.

Dayal, Veneeta. 1997. Free relatives and –ever: identity and free choice reading. In A. Lawson & E. Cho (eds.), Proceedings of SALT 7, 99–116. CLC Publications.

Donnallan, K. 1966. Reference and Definite Descriptions. Philosophical Review 75. 281–304.

Ebert, Christian, Endriss, Cornelia and Hinterwimmer, Stefan. 2008. Topics as speech acts: an analysis of conditionals. In Natasha Abner and Jason Bishop (eds.), Proceedings of the 27th West Coast Conference on Formal Linguistics, 132–140. Somerville, MA: Cascadilla Proceedings Project.

Elbourne, Paul. 2005. Situations and Individuals. Cambridge, Mass.: The MIT Press.

Evans, G.. 1980. Pronouns. Linguistic Inquiry 11. 337–362. von Fintel, Kai. 2000. Whatever. In Proceedings of Semantic and Linguistic

Theory 10, 27–39. Grosu, Alexander & Landman, Fred. 1998. Strange relatives of the third kind.

Natural Language Semantics 6. 125–170. Gu, Chloe. 2008. Maximalization and the definite reading in Mandarin wh-

conditionals. Paper presented at NELS 40. MIT. Heim, Irene. 1982. The semantics of definite and indefinite noun phrases:

Umass at Amherst Phd. Thesis. Heim, Irene. 1990. E-type Pronouns and donkey anaphora. Linguistics and

Philosophy 13. 137–177. Jackendoff, Ray. 1972. Semantic interpretation in generative grammar. Cam,

Mass.: The MIT Press.

Sinn und Bedeutung 15 - [PDF Document] (180)

Identity and Definiteness in Chinese wh-Conditionals 179

Jacobson, Pauline. 1995. On the quantificational force of English free relatives. in E. Bach, E. Jelink, A. Kratzer and B. Partee (eds.), uantification in natural language, 451–486. Dordrecht: Kluwer.

Kamp, Hans. 1981. A theory of truth and semantic representation. In J. Goenendijk et al (eds.), Formal Methods in the Study of Language, 277–322. Amsterdam: Mathematical Center Tracts.

Kadmon, N.. 1990. Uniqueness. Linguistics and Philosophy 13. 274–324. Kratzer, Angelika & Heim, Irene. 1998. Semantics in generative grammar.

Cam, Mass.: The MIT Press. Lewis, David. 1973. Counterfactuals. Harvard University Press. Li, Y.-H. Audrey. 1992. Indefinite wh in Mandarin Chinese. Journal of East

Asian Linguistics 1. 125–156. Link, Godehard. 1983. The logical analysis of plurals and mass terms: A

lattice-theoretical approach. In R. Bäuerle et al (eds.), Meaning, Use and Interpretation of Language, 302–323. Berlin: Walter de Gruyter.

Moltmann, Friedericke. 2010. Identificational sentences and the objects of direct perception. Ms. IHPST, Paris.

Neale, Stephen. 1990. Descriptions. Cam., Mass.: The MIT Press. Schlenker, Philippe. 2004. Conditionals as definite descriptions (a referential

analysis). Research on Language and Computation 2. 417–462.

Sinn und Bedeutung 15 - [PDF Document] (181)

Sinn und Bedeutung 15 - [PDF Document] (182)

In Reich, Ingo et al. (eds.), Proceedings of Sinn & Bedeutung 15,pp. 181–192. Universaar – Saarland University Press: Saarbrücken, Germany, 2011.

Total vs. Partial Adjectives:Evidence from Reduplication∗

Mojmír DocekalMasaryk [emailprotected]

Ivona KucerováMcMaster [emailprotected]

Abstract. This paper provides evidence for a structural difference betweentwo classes of antonym adjectives, namely, total and partial adjectives, for ex-ample, clean and dirty (Yoon 1996; Rotstein & Winter 2004). Based on datafrom morpho-phonological processes in Czech we argue that only total adjec-tives have their standard value represented in the derivation. In contrast, the stan-dard value of the partial adjectives is determined pragmatically. Furthermore, weargue that antonym adjectives must be at least sometimes represented by over-lapping scales. A consequence of the proposed analysis is that an empiricallyadequate account of antonym adjectives must supply a part of the denotationfrom lexical semantics and part from the context.

1 IntroductionCzech, West Slavic, has a productive system of a semantically driven morpho-phonological reduplication (Marantz 1982; Inkelas & Zoll 2005). One suchexample comes from the morphological marking of aspect. The imperfectiveverbal morpheme -va- is often called habitual since it may encode iterativityif reduplicated, as in (1). The effect of reduplication is indeed semantic and assuch has truth-conditional effects: the reduplicated form may be used in habit-ual or generic sentences, as seen in (2a), but it is incompatible with episodicsentences, as can be seen in (2b).

(1) a. praco-va-lwork-IMPERF-PP.M.SG.‘he worked’ Imperfective/generic

b. praco-vá-va-lwork-IMPERF-IMPERF-PP.M.SG.‘he used to work’ iterative

c. praco-vá-vá-va-lwork-IMPERF-IMPERF-IMPERF-PP.M.SG.

∗ We would like to thank Chris Kennedy, Henk Zeevat and the audiences at the Szklarska Poreba2010 worhshop and the SuB conference. The authors are happy to acknowledge that MD wasfinancially supported by GACR (grant 405/09/0677).

Sinn und Bedeutung 15 - [PDF Document] (183)

182 Docekal & Kucerová

‘he used to work’ iterative (emphatic)

(2) a. PetrPetr





‘Peter used to clean the window every morning.’ Xhabitual/ge-neric

b. *PetrPetr





‘Peter used to clean the window yesterday in the morning.’ *episo-dic

This paper focuses on another type of semantically driven reduplication, na-mely, reduplication in antonym adjectives. Czech gradable antonym adjectivesmay contain a degree morpheme which meaning roughly corresponds to En-glish very. If the degree morpheme undergoes a process of reduplication, the re-sulting meaning of the adjective may be paraphrased as ‘very, very. . . (clean)’,i.e., emphasizing the standard value of the adjective. Native speakers charac-terize the resulting interpretation as that of reaching the absolute degree ofadjectiveness (for example, of cleanness).

The fact that interests us here is that not every gradable antonym adjectivemay undergo the reduplication process. Even though any gradable antonymadjective may contain a degree morpheme, the morpheme may be reduplicatedonly in so-called total adjectives, never in their partial counterparts, followingthe terminology of Yoon 1996. The contrast is shown in (3) and (4). Here,the adjectives cistý ‘clean’ and zavrený ‘closed’ provide an example of totaladjectives and the adjectives špinavý ‘dirty’ and otevrený ‘open’ provide anexample of partial adjectives.

(3) cistý ‘clean’ vs. špinavý ‘dirty’a. cistý→ cist’ounký→ cist’oulinký→ cist’oulilinký. . . Xredupli-

cationb. špinavý→ špinavoulinký→ *špinavoulilinký. . . *reduplication

(4) zavrený ‘closed’ vs. otevrený ‘open’a. zavrený→ zavrenoulinký→ zavrenoulilinký Xreduplicationb. otevrený→ otevrenoulinký→ *otevrenoulilinký. . . *reduplication

For presentational purposes we demonstrate the reduplication process in stages.First, we observe that for the degree morpheme to be inserted the stem of theadjective need to be modified. The change of the stem is independently moti-vated by phonotactic constraints on this type of morphological formation anddoes not directly concern us here. Once the morpheme – in our case, an in-

Sinn und Bedeutung 15 - [PDF Document] (184)

Total vs. Partial Adjectives 183

fix -li-, meaning roughly ‘very’ – is inserted, the morpheme may be redupli-cated without any further phonological or morphological change of the stem. Incontrast, as the (b) examples show, even though the partial adjectives špinavý‘dirty’ and otevrený ‘open’ may be modified by the same degree morpheme,reduplication of this morpheme is impossible. Further examples demonstratingthe contrast are given in (5).

(5) Some further examples (source: the Czech National Corpus):

cistý (clean) cist’oulilinkýzavrený (closed) zavrenoulilinkýzdravý (healthy) zdravoulilinkýrovný (straight) rovnoulilinkýtenký (thin) tenoulilinkýjemný (slight) jemnoulilinkýchabý (faint) chaboulilinkýkrehký (fragile) krehoulilinkýšpinavý (dirty) *špinavoulilinkýotevrený (open) *otevrenoulilinkýnemocný (ill) *nemocnoulilinkýzahnutý (bent) *zahnutoulilinkýtlustý (thick) *tlust’oulilinkýhrubý (rough) *hruboulilinkýpevný (solid) *pevnoulilinkýnerozbitný (unbreakable) ???

The observed restriction on reduplication is rather puzzling since it does nothold for its semantically closest variant, i.e., adverbial modification by velmi‘very’, as can be seen in (6). Similarly, the closest English paraphrase (therepetition of ‘very’) is compatible with both total and partial adjectives as well,as in (7).

(6) No restriction on adverbial modification:a. velmi


b. velmivery


(7) No restriction on English adverbial modification:a. very very very cleanb. very very very dirty

Sinn und Bedeutung 15 - [PDF Document] (185)

184 Docekal & Kucerová

Crucially, the restriction on reduplication cannot be explained in phonologi-cal or morphological terms either because neither semantic class of the ad-jectives forms a phonological or morphological natural class. Furthermore, ifthere were any phonotactic restrictions they should be equally resolved withinthe morpho-phonological changes accompanying the primary modification bythe degree morpheme.

The rest of the paper provides evidence that the restriction on reduplica-tion stems from distinct semantic properties of the two classes of adjectives.Concretely, we will argue that the reduplication process refers to a meaningcomponent that is structurally present only in total adjectives. The correspond-ing meaning component in partial adjectives is never present in the derivation.Instead, it is supplied by the context. Consequently, it cannot serve as an anchorfor morpho-phonological processes. We will outline our semantic assumptionsand present the actual proposal in section 2. Section 3 investigates Englishadverbial modification by almost and compares it with the Czech reduplica-tion case. As we will see, there is a sharp contrast between English and Czech:while the English type of modification is sensitive to the context, this pragmaticstrategy fails to rescue reduplication of Czech partial adjectives thus providingfurther evidence for the present proposal. Section 4 concludes.

2 ProposalWe assume scalar semantics for adjectives, i.e., the positive form of an adjec-tive denotes a subinterval of the scale SA where the subinterval depends on astandard value dA in the scale and where the scale is ordered by a relation RAdefined with respect to the standard value dA ∈ SA (Cresswell 1977; von Ste-chow 1984; Bierwisch 1989; Klein 1991). Furthermore, we assume that thestandard value variable dA is context dependent.1 The denotation of the posi-tive form of an adjective can be formalized as in (8) (after Rotstein & Winter2004: ex. (18)):2

(8) JAKde f= {x ∈ SA : RA(dA,x)}

The complete lexical semantics of the adjective like long can be then formal-ized using λ -abstraction as follows:

(9) JlongK = λdAλx.long(x)≥ dA

1 For example, the standard value for big is set differently in a big house than in a big mouse.2 Notice the denotation of an adjective in (8) must be mapped on the set of entities for the degreeof A-ness to be included in JAK otherwise the intersection interpretation of the AP within an NPyields a type-mismatch.

Sinn und Bedeutung 15 - [PDF Document] (186)

Total vs. Partial Adjectives 185

In the function talk – the denotation of an adjective like long is a functionfrom a context set degree dA and an individual x which yields truth value 1 ifand only if the degree of the length of the individual x exceeds the degree dA.3

Degree dA can be given explicitly (in most cases by a noun which is modified bythe adjective) or implicitly. The relation ≥ is supplied by an invisible operatorpos which operates on the adjective and assigns truth value 1 only to thoseindividuals (when the adjective is used predicatively) which exceed the averagedegree for the comparison class. Von Stechow (1984: R6) defines pos in thefollowing way:

(10) PositiveLet A0 be any adjective meaning, C be any appropriate property, x beany appropriate individual and w be any world. Then w ∈ JposK iff(∃d) [d is an A0-degree & d > average [A0,C] & x has d in w & w ∈C(x)].

A sentence like Ferda is a big cat is true in a world w iff Ferda has a degree ofbigness which exceeds the average degree of bigness for cats in the world ofevaluation.

Here we are concerned with two basic types of antonym adjectives: par-tial and total adjectives.4 We semantically represent total and partial adjectivesby a scale and a standard value. A partial adjective indicates some amount ofthe relevant property (moisture, dirt, sickness etc.), while a total adjective in-dicates no amount of such property (e.g., a dirty object has some degree ofdirtiness, but it is not necessarily free of cleanliness; in contrast, a clean objectis free of dirtiness). As for their semantic denotation, we follow Rotstein &Winter (2004) in formalizing total v. partial adjectives as overlapping scales,schematized in Figure 1.

Figure 1: Total and Partial adjective scales

3 Notice that the type of the adjective is not a predicate, 〈e, t〉, but a function from degrees into afunction from individuals to truth values: 〈d,〈e, t〉〉.4 We put aside so-called relative adjectives. In relative adjectives, no member of an antonym pairhas it standard value set independently of the context (Kennedy & McNally 2005; Kennedy 2007).

Sinn und Bedeutung 15 - [PDF Document] (187)

186 Docekal & Kucerová

In Figure 1, SP is a partial scale, without a fixed standard value (context-dependent). ST is a total scale; its standard value is fixed as the lower boundof its partial counterpart (a healthy man is a man that lacks any illness). Cru-cially, ST and SP are ordered inversely and SP may partially overlap with ST .What this means is that some amount of the relevant partial property does notexclude some amount of the complementary total property. For instance, if acoat is dirty it can mean that it is stained on sleeves but the rest of the coat isclean.

Following von Stechow (2007); Heim (2008), among others, we representantonym adjectives as complements/negations of each other, the denotation ofadjectives like clean and dirty are related by the operation of being a com-plement of each other’s respective scales. We thus follow a linguistic traditionwhich treats antonyms as lexically related instead of being syntactically de-composed in syntax.5

More formally, we define the relation between partial and total adjec-tives with respect to the standard value of the total adjective represented as thelower bound of its partial counterpart (following Rotstein & Winter 2004 contraKennedy & McNally 2005). Crucially, the scales may partially overlap and theimpression of their antonymous interpretation (not clean dirty) comes froman interaction of their interval boundaries and their standard values. As for thedenotation of partial adjectives, their standard value is determined contextu-ally. Consequently, the standard value of a partial adjective has no structuralrepresentation:

(11) dP ∈ SP; SP . . . closure of the partial scale

In contrast, the denotation of a total adjective defines the standard value of thetotal member of an adjectival pair as the lower bound of its partial counterpart:

(12) dT = Pmin ∈ ST ; ST . . . closure of the total scale

With the formal semantics of antonym adjectives in place we can ap-proach the question of the denotation of reduplication. The intuition is thatreduplication corresponds to semantic modification, i.e., adjectives with redu-plicated morphemes denote some interval close to the standard value. Sincethis is semantic modification, it depends on the type of the scale in the denota-tion of the adjective with which it combines. More formally, we argue that thedenotation of reduplication corresponds to a limit function where the limit isdefined as the standard value of the total adjective. Thus, our first step to the

5 Heim (2008) provides an argument that antonyms are not decomposed in syntax contra Büring(2007a,b).

Sinn und Bedeutung 15 - [PDF Document] (188)

Total vs. Partial Adjectives 187

formalization can be captured as follows:

(13) J-li-li-K≈ dT

The resulting denotation corresponds to approaching the standard value of theadjective. If we apply the denotation of the reduplication to a total adjective,for example, to cistý ‘clean’, we get the following outcome:

(14) J cist’ou-li-li-. . . nký K = λx∃d[d ≈ dT ∧ clean(d)(x)]

The formalization captures the fact that the denotation of a total adjective isnot a function from degrees and individuals to truth values, but instead it is afunction from individuals to truth values. This stems from the fact that the de-gree dA is not a context dependent variable,6 but instead the variable is existen-tially closed and its value approaches the standard value of its total counterpart.Consequently, the denotation of a reduplicated adjective is not dependent on acontext fixation of the standard value dA anymore.

After we further abstract over the adjective, we end up with the followingdenotation:

(15) J -li-li-. . .K = λGλx∃d[d ≈ dT ∧G(d)(x)]

We speculate that the reduplication takes place instead of the invisible oper-ator pos which can be found in relative adjectives. There are two differencesbetween pos and reduplication morpheme: First, since the reduplication mor-pheme replaces pos, it does not need any comparison class from the linguisticor extralinguistic context, consequently, a reduplicated adjective is not depen-dent on the context. Second, the relation between the degree and an individualis a limit function, instead of ≥.7 Even if both pos and the reduplication mor-pheme existentially close the degree variable, they do it in a different way – theoperator pos makes the denotation of the adjective dependent on a comparisonclass C but the reduplication is not dependent on any comparison class at all.8

The proposed formalization makes certain predictions about interactionsbetween different types of adjectives and reduplication. First of all, since redu-

6 As it is the case in the denotation of relative adjectives.7 We put aside any relativization to possible worlds because we use a purely extensional frameworkin this paper.8 We assume that different speaker standards for total adjectives come from extralinguistic factorsand not from the semantics itself – e.g. if we consider the sentence The dishes are clean, which canbe true for one speaker and false for another one depending on their personal standard, one mightthink that the interspeaker disagreement comes from a semantic context dependency. However, webelieve this type of difference in speakers judgements come solely from extralinguistic factors anddoes not need to be represented in semantics proper.

Sinn und Bedeutung 15 - [PDF Document] (189)

188 Docekal & Kucerová

plication is a morpho-phonological process, it may only apply to a materialpresent in the derivation.9 What this means with respect to our data is that redu-plication is possible only if the standard value is structurally represented. Con-sequently, we predict that reduplication applies only to total adjectives sinceonly total adjectives have their standard value structurally represented as somevalue in the closure of the total scale. In contrast, the standard value of a partialadjective is determined contextually and may fall anywhere within the inter-val. What this means is that there is no structural representation of the standardvalue. Consequently, there is no material that could be used for reduplication.Thus, we have successfully derived the contrast between partial and total ad-jectives with respect to reduplication.

3 Further Evidence: Context and the Standard ValueInterestingly, English adverbial modification by almost shows similar proper-ties to the Czech adjectival reduplication. As can be seen in (16), almost usuallycombines with total adjectives but not with partial adjectives. This restrictionis parallel to the restriction observed for the Czech reduplication paradigm andas such invites the question of whether we deal with the same phenomenon.

(16) (from Rotstein & Winter 2004: ex. (9))a. The work is almost complete/*incomplete.b. The patient is almost dead/*alive.c. The explanation is almost clear/*unclear.

The basic observation about almost is that almost cross-categorically denotesnegation of the denotation of the constituent it modifies:

(17) a. John almost passed the exam John didn’t pass the examb. Almost every student passed the exam Not every student passed

the examc. John is almost healthy John isn’t healthy

In order to account for the semantics of English almost, Rotstein & Winter(2004) proposed that the interval associated with the phrase almost A denotesdegrees that are adjacent to the standard value of A and are in the oppositedirection from the ordering of the scale associated with the adjective A. If weapply this denotation to our semantics of total and partial adjectives, the in-

9 We believe this claim is fairly theory neutral. At least, we are not aware of any generative modelof morpho-phonology where this reasoning wouldn’t apply. In fact, this type of dependency can beeasily reformulated in representational terms as well, yielding the same result.

Sinn und Bedeutung 15 - [PDF Document] (190)

Total vs. Partial Adjectives 189

compatibility of almost and partial adjectives can be derived in the followingway. First, recall that under our semantics for total and partial adjectives, ifthe standard value of a partial adjective equals the standard value of a totaladjective (just on the opposite scale), then the adjectives are complementary.Consequently, the partial adjective cannot be modified by almost because thereis no complement interval between dP and 0.

There is an interesting prediction stemming from this formalization. Oneof the crucial distinctions between total and partial adjectives lies in the waytheir standard value is represented. While the standard value of a total adjectiveis structurally fixed as the lower bound of its partial counterpart, the standardvalue of a partial adjective is not fixed in the structure but instead it is con-textually dependent. What this amounts to is that in a neutral context Englishspeakers tend to fix the standard value of a partial adjective as the minimum.However, this is not necessary. If we create an appropriate context, the standardvalue can be shifted further up the scale. If this happen, we create a non-emptyinterval between dP and 0. Such a shift is schematically shown in Figure 2. Ifsuch an interval exists, then it should be able to feed into the denotation of al-most. Consequently, if such modification is possible, a partial adjective shouldbecome modifiable by almost. This prediction is indeed borne out as observedby Kennedy (2007). Examples in (18) and (19) demonstrate the shift.

Figure 2: Modification by almost

(18) (from Kennedy 2007)a. We need a rod that is bent in an angle of 90 degrees. Let’s pick

up that rod over there and bend it a little: it should be easy, as it’salmost bent already.

b. We consider a glass dirty and wash it as soon as there are fivespots on it. This glass is now almost dirty – it has four spots onit.

(19) (from Kennedy 2007)a. We need a TALL basketball player – one whose height is at least

1.95 meters. But we cannot take John, who is 1.90 meters – he’s

Sinn und Bedeutung 15 - [PDF Document] (191)

190 Docekal & Kucerová

just almost tall.b. The publisher considers a book long if it’s 300 pages or more.

This book is almost long – it’s 298 pages.

In the previous section we have proposed that the impossibility of reduplicationof partial adjectives is a direct consequence of their standard value not beingstructurally present. We have argued that the restriction has nothing to do withthe actual position of the standard value on the scale but instead it results fromthe lack of a structural material available for the morpho-phonological pro-cess of reduplication. Could it be the case that in fact the denotation of theprocess of reduplication should be stated in terms of a complement intervalexactly as the denotation of English almost? These two hypotheses make dis-tinct predictions. If we follow the denotation for English almost, we predictthat exactly as in English, reduplication of partial adjectives should improvein a context that pushes the standard value of a partial adjectives further awayfrom the minimum. In contrast, our structural hypothesis predicts that the dis-tance between the standard value and the minimum should not make any dif-ference: if reduplication depends on the standard value being structurally fixed,it should not matter whether or not the standard value is distinct from the mini-mum value. The reason is that the contextual fixation happens only later in thederivation (in the semantics/pragmatics component) and as such it cannot affectthe morpho-phonological process that necessarily takes place before the prag-matic component sets the standard value. Thus, the prediction of our proposalis that even if we modify the context, reduplication of Czech partial adjectivesshould still fail. Interestingly, this prediction is indeed borne out, as examplesin (20), modelled after Rotstein & Winter (2004), show. No matter how hardwe try to modify the context, what we see is that reduplication in Czech partialadjectives, unlike almost-modification in English, cannot be improved.

(20) a. This glass is certainly not clean, since it has several big spots onit and I am not willing to drink from it even if you insist. Theglass is simply. . .*špinavoulilinká ‘very very dirty’

b. This glass is certainly not dirty, since it has absolutely no dirtyspots on it. The glass is simply. . .Xcist’oulilinká ‘very very clean’

4 ConclusionWe have examined a surprising contrast between partial and total adjectivesthat emerges in a semantically driven morpho-phonological process of redupli-

Sinn und Bedeutung 15 - [PDF Document] (192)

Total vs. Partial Adjectives 191

cation in Czech. While total adjectives can be reduplicated, partial adjectivescannot. This is unexpected because a formally distinct but semantically parallelprocess of adverbial modification does not show any such restriction. We haveargued that the pattern can be explained if we adopt Yoon’s and Rotstein &Winter’s account of total and partial adjectives. According to them, there is astructural difference between partial and total adjectives: Only total adjectiveshave their standard value represented in the derivation, the standard value ofpartial adjectives is derived from the context. Furthermore, we have argued thatantonym adjectives cannot be represented by adjacent scales but instead theymust be allowed to partially overlap (in agreement with Rotstein & Winter 2004and contra Kennedy & McNally 2005; Kennedy 2007). Crucially, we have ar-gued that the semantics of antonym adjectives must be formalized as a com-bination of grammatically encoded (semantics) and contextually-determined(pragmatics) meanings. Thus, in our model, the proper formalization of thesetwo types of adjective must be represented in two components of the grammar.As we have seen in our case study, the different representations are empiricallytestable.

ReferencesBierwisch, Manfred. 1989. The semantics of gradation. In Manfred Bier-

wisch & Ewald Lang (eds.), Dimensional adjectives, 71–262. Berlin:Springer-Verlag.

Büring, Daniel. 2007a. Cross-polar nomalies. Paper presented at Semanticsand Linguistics Theory 17.

Büring, Daniel. 2007b. More or less. Paper presented at the Chicago LinguisticSociety Annual Meeting.

Cresswell, Maxwell J. 1977. The semantics of degree. In Barbara Partee (ed.),Montague grammar, 261–292. New York: Academic Press.

Heim, Irene. 2008. Decomposing antonyms? In Atle Grønn (ed.), Proceedingsof SuB12, Oslo, 212–225. Oslo: ILOS.

Inkelas, Sharon & Cheryl Zoll. 2005. Reduplication: Doubling in morphology.New York: Cambridge Univ Press.

Kennedy, Christopher. 2007. Vagueness and grammar: the semantics of relativeand absolute gradable adjectives. Linguistics and Philosophy 30. 1–45.

Kennedy, Christopher & Louise McNally. 2005. Scale structure, degree mod-ification, and the semantics of gradable predicates. Language 81(2).345–381.

Klein, Ewan. 1991. Semantik: Ein internationales Handbuch der Zeitgenössis-chen Forschung chap. Comparatives, 673–691. Walter de Gruyter.

Sinn und Bedeutung 15 - [PDF Document] (193)

192 Docekal & Kucerová

Marantz, Alec. 1982. Re reduplication. Linguistic inquiry 13(3). 435–482.Rotstein, Carmen & Yoad Winter. 2004. Total adjectives vs. partial adjectives:

Scale structure and higher-order modifiers. Natural Language Seman-tics 12(3). 259–288.

von Stechow, Arnim. 1984. Comparing semantic theories of comparison. Jour-nal of Semantics 3. 1–77.

von Stechow, Arnim. 2007. The temporal degree adjectives früh(er)/spät(er)’early(er)’/’late(r)’ and the semantics of the positive. In A. Giannakidou& M. Rathert (eds.), Quantification, definiteness and nominalization,214–233. Oxford: Oxford University Press. 2009.

Yoon, Youngeun. 1996. Total and partial predicates and the weak and stronginterpretations. Natural Language Semantics 4(3). 217–236.

Sinn und Bedeutung 15 - [PDF Document] (194)

Focus Marking via Gestures*

Cornelia Ebert University of Stuttgart


Stefan Evert University of Osnabrück


Katharina Wilmes University of Amsterdam


Abstract. This paper contributes to the recent investigations of speech-accompanying gestures under a formal semantic view. We show that gestures can serve to disambiguate a sentence with respect to its possible focus domains. We provide a statistical evaluation of data gained from a corpus annotated with gestures and information structure. The language under investigation is German. We argue that a sentence that, in isolation, is ambiguous concerning the extension of its focus domain is disambiguated via speech-accompanying gestures. Gesture thus is a means to mark information structure next to intonation and word order.

1 Introduction It is widely known that gestures are temporally aligned with the speech signal, in particular it has often been claimed that the stroke, i.e. the main part of a gesture where the actual gesture movement takes place, falls together with the main accent of the gesture-accompanying sentence (McNeill 1992 among many others). The relationship of complete gestures or gesture phrases and foci, however, has not been investigated systematically yet. We want to fill this gap by showing that the possible focus projection of a focus exponent is restricted by the point of time at which a speech-accompanying gesture starts. Gesture thus serves as a means to mark focus domains. Consider the following example for illustration (the main accent is indicated by capital letters):

(1) I ate baNAnas.

* First and foremost, we would like to thank Hannes Rieser and Florian Hahn for giving us access to the gesturally annotated SAGA-corpus of the University of Bielefeld. This work would not have been possible without the possibility to access and make use of the accurate and fine-grained gestural annotations of the SAGA-corpus. We would also like to thank Hannes Rieser and Florian Hahn for their constant help with technical and other questions of all sorts as well as for numerous valuable discussions about gestures and information structure.

In Reich, Ingo et al. (eds.), Proceedings of Sinn & Bedeutung 15, pp. 193–208. Universaar – Saarland University Press: Saarbrücken, Germany, 2011.

Sinn und Bedeutung 15 - [PDF Document] (195)

194 Ebert, Evert & Wilmes

The sentence in (1) with the given intonation pattern can be read as an answer to the two questions in (2), each inducing a different focus-background structure.

(2) a. What did you do? b. What did you eat?

(2a) is a VP-focus invoking question, while (2b) requires narrow focus on the direct complement. Following (2a), (1) allows for the focus pattern in (3a); if (1) follows (2b) on the other hand, the focus pattern is the one of (3b).

(3) a. I [ate baNAnas]F. b. I ate [baNAnas]F.

In the following we will defend the hypothesis in (4).

(4) Hypothesis (Focus-gesture alignment): How far a focus projects is determined by the onset of the ac-companying gesture (if one exists).

In other words, the onset of a speech-accompanying gesture indicates the left border of the focus phrase (independent of the type of gesture – be it a beat, a deictic or an iconic gesture or any other kind of gesture). A speech-accompanying gesture can thus serve to disambiguate an information-structural ambiguity in a sentence towards a certain focus-background pattern. Simplifying matters for now, we expect the patterns in (5). (|G marks the hypothesized onset of the speech-accompanying gesture.)

(5) a. I |G[ate baNAnas]F. b. I ate |G[baNAnas]F.

Although (1) is ambiguous with respect to the underlying information structure, |G disambiguates the sentence towards one of the focus-background patterns in (3). In order to test the hypothesis in (4), we looked at the temporal occurrences of gestures and foci. We therefore annotated the multimodal Bielefeld Speech-And-Gesture-Alignment (SAGA) corpus with focus features – in addition to the existing gestural annotation – and marked the nuclear accents of certain intonation units. A subsequent statistical analysis confirmed our hypothesis that the onsets of focus and gesture align indeed – with a systematic shift, however: on average gestures start about 0.3 seconds earlier than the corresponding focus phrases. That is, there is a certain time lag between the onset of a gesture and its associated focus.

Sinn und Bedeutung 15 - [PDF Document] (196)

Focus Marking via Gestures 195

In this paper, we mostly present material that has also been discussed in Wilmes (2009). We re-evaluate some of the results of Wilmes (2009) and further elaborate on various aspects. The remainder of this paper is structured as follows: Section 2 sets the stage and discusses the relevant findings from the gesture literature that will be needed in the remainder of the paper. Section 3 presents the methodology underlying our investigations. Here, we explain what the data set that our study is based on looks like, how we annotated these data and how we finally investigated the temporal interdependence of gestures and foci. Section 4 then presents the results of a statistical investigation of the temporal occurrences of gestures and foci. In section 5, we evaluate and discuss these results. Section 6 discusses some controversial issues and loose ends. And finally, section 7 concludes the paper.

2 Speech-­‐Accompanying Gestures It is a widely held view that gesture is a distinct mode of expression and that the study of gestures can tell us more about language than one might think at first sight (see e.g. Kendon 1972, 1980 and Loehr 2004 and references therein). We subscribe to this view and we will argue in particular that for a comprehensive view of focus phenomena it is inevitable to take speech-accompanying gestures into account.

To set the stage, we will have a look at some important findings concerning the interpretation of speech-accompanying gestures. First of all, one has to define what a gesture phrase is, i.e. where it starts and where it ends. In order to determine which movements can be considered to contribute to a particular gesture, Kendon (1972, 1980) identified a certain structure that can be found for gestures quite generally. The smallest unit of a gesture is its main element, i.e. the minimally required element for being reckoned as a proper gesture: the stroke. The stroke can be identified with the strongest movement within the gesture. A stroke is usually preceded by a preparation phase and followed by a retraction phase, for the hands must be brought into an appropriate position for the stroke to be executed and back into the resting position. Taken together, these three phases constitute the gesture phrase1. Preparation and retraction are optional, so a gesture phrase may consist of nothing but a stroke. Between preparation and stroke and stroke and

1 This notion of a gesture phrase cannot be applied to all kinds of gestures. So-called beats are only biphasal, i.e. they consist of two movement phases, constituting a repeated movement pattern, like up and down or in and out.

Sinn und Bedeutung 15 - [PDF Document] (197)

196 Ebert, Evert & Wilmes

retraction holds may occur, which are termed pre- or postholds, respectively. These are considered to enhance timing between speech and gesture (cf. McNeill 1992, Lascarides and Stone 2009).

Importantly, it has been argued that gesture and speech can work together to convey one single thought (McNeill 1992, Kendon 1980) and hence that the semantic content of speech-accompanying gestures is intertwined with the semantic content of the speech signal. What is especially important for our purposes is that speech-accompanying gestures are known to be temporally aligned with the speech signal. It has been argued that speech and gesture synchronise in that the stroke of the gesture falls together with the main accent of the gesture-accompanying utterance (see among others: Pittenger, Hockett, & Daheny 1960; Kendon 1980; McNeill 1992; Loehr 2004; Jannedy & Mendoza-Denton 2005). The general claim is that the stroke occurs just before or at the same time as (but not later than) the nuclear accent. Although there are very few empirical studies that back this claim (see Loehr 2004 for a recent study), this is a fairly established finding in gesture theory.

What has been far less investigated is the interaction of entire gesture phrases and speech. In the literature one can find only a few hints and claims concerning their interdependence and there seems to be no general agreement. Kendon (1972: 184) suggests that gesture phrases align with so-called ‘tone units’ (i.e. ‘the smallest grouping of syllables over which a completed intonation tune occurs’, cf. Loehr 2004). Loehr (2004) on the other hand argues that gesture phrases and ‘intermediate phrases’ in the sense of Pierrehumbert (1980) align. We want to add to this list and argue that it is actually focus phrases that gesture phrases align with. Hence, while Loehr (2004) and Kendon (1972) argue that the temporal occurrence of gesture phrases is mainly triggered by intonational aspects, we think that gesture phrases rather synchronise with focus phrases, which means that their temporal appearance is determined by information structure. While there is, of course, a clear connection between intonation and focus, we still believe that the alleged interdependence between gesture phrases and whichever kind of intonationally motivated category is – at best – an epiphenomenon of the gesture-focus alignment for which we argue.

3 Methodology To verify our hypothesis in (4) that (the onsets of) gesture phrases align with (the left border of) focus domains, we investigated the temporal interdependence of gesture phrases and focus domains. In addition, we also

Sinn und Bedeutung 15 - [PDF Document] (198)

Focus Marking via Gestures 197

looked at the timing of stroke and nuclear accent. Our study is one of the very few empirical studies about the interplay between gesture and intonation; to the best of our knowledge, it is the first empirical study of the interplay between gesture and focus. We analysed a 20-minute video sequence with 275 gestures, which makes this study the most extensive empirical study on gesture and speech (cf. Loehr 2004: Condon & Ogston 1966: 5 sec; Kendon 1972: 90 sec; McClave 1991: 125 gestures; Loehr 2004: 164 sec and 147 gestures).

3.1 Data For our study, we worked with one sequence of the Bielefeld SAGA-corpus (Lücking et al. 2010), which is a multimodal corpus (video and audio) that collects dialogues from an experiment where one subject (the router) gives directions to another subject (the follower) for navigation through a dynamic virtual world (see Lücking et. al 2010 for details). While talking, the movements of the subjects’ hands were recorded by sensors attached to the hands and fingers. Three video cameras recorded the scene from different angles. Sound was also recorded.

From this corpus we selected a 20-minute sequence with two male participants. Gestures were already annotated, including gesture type (e.g. iconic or deictic) and duration of gesture phases (i.e. preparation, stroke, holds and retraction).

3.2 Annotation For our purposes, it was necessary to add information-structural annotation (accent and focus) to the existing gestural annotation of the selected video. Our annotation was entirely based on the audio material, which had already been transcribed (but not annotated with parts of speech or other morpho-syntactic information). The information-structural annotation was carried out without reference to the video and its gesture annotations in order to exclude a possible bias. We annotated nuclear accents and distinguished two types of foci: new-information and contrastive. All annotations were based on the recommendations of Dipper et al. (2007) (in particular Chapters Phonology and Intonation (Féry et al. 2007) and Information Structure (Götze et al. 2007)). We treated as new-information focus those cases where information is provided which is new and/or carrying the discourse forward. Here, we predominantly found rather broad focus domains: whole sentences (all-focus sentences), e.g. if these sentences were text-initial or answers to polar questions, and VP-foci. However, our data also contain narrow foci such as

Sinn und Bedeutung 15 - [PDF Document] (199)

198 Ebert, Evert & Wilmes

DP- or AdjP-foci. An expression was tagged as contrastive focus if it overtly contrasted with other elements in nearby utterances. We kept track of all pitch accents in the data, i.e. the points of highest or sometimes lowest pitch that make syllables intonationally salient (X* in the ToBI framework2) and filtered out the nuclear pitch accents among them. There was always one unique nuclear accent for each new-information focus domain. For reasons of space, we cannot go into the details of the annotation procedure and refer to (Wilmes 2009: 26-31) for further information.

3.3 Data Extraction To verify hypothesis (4), i.e. to show that gesture phrases and focus phrases align in fact, we investigated the temporal interdependence of focus phases (FocPs) and gesture phrases (GPs). This left us with the following task:

(6) Verification Task (Focus-gesture alignment): For each gesture phrase, find the corresponding focus phrase and compare the temporal position of the two.

For each gesture, we had a look at the associated speech (not the other way round).3 Making use of the result from the literature that nuclear accents and strokes align, we associated a gesture phrase with a focus phrase if the nuclear accent of the focus phrase overlapped with the gesture phrase’s stroke (see Figure 1 for an example). In the few cases where there was no main accent coinciding with the gestural stroke, we considered a focus phrase overlapping with at least the stroke phase to be associated with the gesture, unless the overlap was very small and a close investigation of the gesture-focus pair made an association implausible (because there was another focus that was more likely to associate with the gesture). This was the case for only two gestures. Moreover, there were eight cases of strokes that did not overlap with any focus. In one case, an entire gesture did not coincide with any focus at all and for seven gestures, though they overlapped with a focus in some parts, it was not the stroke that overlapped with the focus. We excluded these ten gestures and strokes from our statistical evaluation.

Figure 1 illustrates an example that shows how gesture time and focus time can be compared. Time differences are assessed by subtracting focus

2 TOBI stands for Tone and Break Indices. The system is based on work by Pierrehumbert (1980). In our study, we did not distinguish between different kinds of pitch accents like high (H*), low (L*) or rising (L+H*). 3 Thus, if there is no gesture there is also no need to identify a focus to verify our hypothesis. However, in most cases we found a one-to-one mapping of focus phrase and gesture phrase.

Sinn und Bedeutung 15 - [PDF Document] (200)

Focus Marking via Gestures 199

times from gesture times (e.g. start difference = gesture start – focus start). The corresponding sentence from the corpus is given in (7):

(7) Ja, also die Busfahrt, die hat äh fünf Stationen, die auf jeden Yes so the bus tour RP has eh five stops that on every Fall angefahren werden müssen. case approached will must

‘Yes, so on the bus tour there are five stops that have to be approached in any case.’

The onset time of the focus phrase (StFoc) is subtracted from the onset time of the associated gesture phrase (StGest), i.e. the onset of the preparation phase (or the stroke if there is none). The time when the focus phrase ends (EFoc) is subtracted from the time when the stroke ends (representing the end of the gesture phrase, hence EGest). We treat the end of the stroke and not the end of the retraction phase as the end of a gesture for two reasons: First, according to McNeill (1992: 29) the retraction phase is ‘semantically neutral’ and second, Loehr (2004) discusses the possibility to disregard retractions and post-holds in his statistical evaluation as well, because they seem to have a different status as the other phases of a gesture phrase.4

4 Cf. Loehr (2004: 117): ‘Typically, an entire g-phrase [CE/SE/KW: gesture phrase] aligned with an intermediate phrase. Occasionally, however, it was clear that a g-phrase aligned with an intermediate phrase only when disregarding post-stroke holds, [or] retractions [...] within the g-phrase. These internal components are included within g-phrases by definition, following

Figure 1: Comparison of focus and gesture times

StGest EGest

StFoc EFoc

Sinn und Bedeutung 15 - [PDF Document] (201)

200 Ebert, Evert & Wilmes

As a base for comparison, we also studied the temporal occurrences of nuclear accents (NAcc) and strokes in order to verify the by now well-established claim from the literature that nuclear accents and strokes align (cf. section 2). For each stroke, we considered a nuclear accent that overlapped with the stroke as associated with the stroke. If there was no such accent, we took the nearest nuclear accent. Time differences were again calculated by subtracting accent time from stroke time (e.g. start difference = stroke start – accent start).

4 Results In the following we present our results on the hypothesized gesture-focus alignment and our reassessment of the question whether stroke and main accent align, as has been claimed in the literature. Statistical analysis was carried out with the R environment for statistical computing (R Development Core Team 2005).

4.1 Alignment of Main Accent and Stroke In total, we analysed 275 stroke-accent pairs. In the majority of cases (209 pairs) the stroke began earlier than the main accent (versus 66 pairs where accent began earlier). Similarly the stroke ended later than the main accent for 183 pairs (versus 92 pairs where the accent ended later). In 124 cases, the stroke encompassed the main accent, in 100 cases stroke and main accent overlapped in some other way, and in 51 cases they did not overlap at all. Figure 2 shows a histogram for the time difference between the onsets of nuclear accents and the corresponding strokes. As can be seen, the distribution is ap-proximately Gaussian (the solid line shows the empirical distribution, the dashed line a Gaussian approximation). On average, the stroke starts 0.36s earlier than the corresponding nuclear accent. The standard deviation is about 0.55s. We interpret this as a tendency for gestures to precede the corresponding accent (though there are a

Kendon’s hierarchical packaging. However, there may be some different quality about these post-stroke components. Occurring after the heart of the gesture, they may have a less important status in terms of timing with speech.’

Figure 2

Sinn und Bedeutung 15 - [PDF Document] (202)

Focus Marking via Gestures 201

considerable number of cases where the gesture starts later). For comparison, the offset differences have a mean of 0.53s (i.e. stroke usually ends later than the accent) and a standard deviation of 1.25s (Figure 3). It is obvious that the onsets align much better than the offsets: their standard deviation is considerably smaller. On the whole, we take our results to show that there is indeed an alignment between the beginning of the stroke and the beginning of the main accent, as claimed in the literature.

4.2 Alignment of Focus and Gesture Having obtained experimental confirmation for the alignment of nuclear accents and strokes, we now turn to our hypothesis that gesture phrases and focus phrases are also synchronised. We found that contrastive foci and new-information foci behave somewhat differently with respect to their accompanying gestures, so we evaluated the two types of foci separately. We analysed 260 new-information focus–gesture pairs and 56 contrastive focus-gesture pairs. As pointed out above in Section 3.3, ten gestures were excluded from the analysis because no focus could be associated with them.

4.2.1 New-­‐Information Focus and Gesture Figure 4 shows the distribution of the onset differences of gesture and new-information focus (we refer to new-information focus simply as focus in the following), which corresponds almost perfectly to a Gaussian distribution. With 0.41s, the standard deviation is rather small. Again we find a systematic shift: gestures start on average about 0.31s earlier than foci, and there are only few cases where focus precedes gesture. While there is thus a certain time lag, most gesture-focus pairs are within less than one second of each other and can be considered to be aligned. A one-sample t-test shows that the time lag effect is genuine (t=12.41, df=259, p < .001; H0: mean time lag = 0). The corresponding Figure 4

Figure 3

Sinn und Bedeutung 15 - [PDF Document] (203)

202 Ebert, Evert & Wilmes

95%-confidence interval places the true mean time lag between gesture and focus in the range from 0.264s to 0.363s. We consider these results as a confirmation of our hypothesis (4) that gestures and foci align in their onsets. For the offsets, the situation is not as clear. Figure 5 shows the distribution of the time differences between the end of a gesture (i.e. the end of the stroke) and the end of the corresponding new-information focus. With a mean of –0.15s, there is no evidence for a systematic shift. The standard deviation of 1.24s, however, is comparatively huge, and some gestures end several seconds after the corresponding focus phrase. On the basis of our data, offsets of gestures and foci thus do not seem to synchronise.

4.2.2 Contrastive Focus and Gesture For contrastive foci and the accompanying gestures, the alignment was not as neat as for the new-information foci. Figure 6 shows a histogram of the onset differences between gestures and contrastive foci. With 0.70s the standard deviation is rather high. The mean is –0.77s, so gestures have a clear tendency to start earlier than the corresponding foci. We interpret these data to show that there is no tight alignment between the onsets of contrastive foci and those of the associated ges- tures. We also tested whether contrastive foci align with the stroke rather than the entire gesture. The histogram for the onset differences of contrastive foci and strokes is given in Figure 7. Again, the standard deviation is quite large (0.75s), but in this case there is no evidence of a systematic shift (mean lag = –0.11s). With such high variability, it is impossible to inter-pret these results as evidence for an alignment of contrastive foci and strokes.

Figure 5

Figure 7

Figure 6

Sinn und Bedeutung 15 - [PDF Document] (204)

Focus Marking via Gestures 203

To conclude, we have not found any focus-gesture or focus-stroke alignment effects for contrastive foci. One has to keep in mind, though, that our data set of contrastive foci is rather small. We therefore leave a detailed investigation of contrastive foci and their accompanying gestures for future research, which will need to build on larger amounts of empirical data in order to draw any reliable conclusions.

5 Discussion In the following we will briefly discuss and evaluate the results that we presented in Section 4. Since our data set for contrastive foci is too small to draw reliable conclusions, we limit our discussion to the comparison of new-information foci and gestures as well as nuclear accents and strokes.

5.1 Shift Effect As indicated above, our observation that strokes usually start 0.36s earlier than the corresponding nuclear accents is entirely in line with the claims from the literature, where it has been noted that a stroke usually coincides with or starts earlier than its corresponding nuclear accent, but in general does not start later than the accent (Kendon 1980, McNeill 1992). We found the same type of shift for gesture phrases and focus phrases, too. Gestures usually start 0.31s earlier than the corresponding focus domains. We believe that this significant time shift may have its roots in the fact that it allows the hearer to draw attention to the upcoming focus phrase, as its occurrence is made predictable by the preceding gesture. Moreover, it is plausible to assume that gesture production is faster than speech production and that the time lag between the onsets of speech and gesture is due to this difference in generation complexity (cf. also Loehr 2004: 29).

5.2 Alignment We interpret our results above as support for hypothesis (4), i.e. they show that gesture phrases and (new-information) foci align (with a certain time lag). We still need to clarify what exactly counts as ‘alignment’, though. Our main arguments supporting the gesture-focus alignment hypothesis are as follows. First and foremost, we take the stroke-accent alignment, which is a well-established effect from the literature, as a point of reference. The onset differences between nuclear accents and strokes have a mean of –0.36s and a standard deviation of 0.55s. Our results show a considerably better gesture-focus alignment, with a similar shift of –0.31s and smaller standard deviation (0.41s). Compare the corresponding histograms in Figures 2 and 4: the better alignment of gesture and focus is immediately obvious.

Sinn und Bedeutung 15 - [PDF Document] (205)

204 Ebert, Evert & Wilmes

There is a second argument to support the interpretation of our results in favour of hypothesis (4). As to our knowledge, there is one empirical survey that our study can directly be compared with (Loehr 2004). When interpreting his results, Loehr (2004) was confronted with the same problem, i.e. to define what exactly can be considered as an alignment. He found that the so-called apex (the peak of a stroke) and the main accent coincide with a standard deviation of 0.27s (and without any significant shift). He interpreted this as showing that there is a tight alignment of apex and nuclear accent. Furthermore, he also suggested that there is an interdependence of Pierrehumbert’s (1980) intermediate phrases and gesture phrases. Similar to our results for gesture phrases and focus phrases, he found that gesture phrases usually start before the corresponding intermediate phrases. The standard deviation for the onset differences between intermediate phrases and gesture phrases was 0.55s. As Loehr (2004) interpreted his results as evidence for a genuine alignment, we think that our study (with standard deviation of only 0.41s) can safely be interpreted to show an alignment of gesture and focus, too.

We did not find evidence for a corresponding alignment of the offsets of gestures and focus phrases. With 1.24s, the standard deviation was very large (recall that the end of a gesture is defined as the end of the stroke). Looking at the histogram in Figure 5, however, it seems that for some gestures there is a good alignment (the main peak of the histogram), while for others the stroke is held much longer (the long right-hand tail of the histogram). This suggests that there may be two different types of gestures – one that aligns well with the focus of the accompanying speech signal and another type that does not. We have not investigated this possibility in depth yet, but it would be worthwhile for future research to examine whether there are certain types of gestures (e.g. beats, deictics and iconic gestures) whose purpose it is to structure information and which thus align better with the speech signal than others (e.g. discourse gestures) that might serve a different purpose.

Finally, let us briefly point out once again that we did not reach a conclusion with respect to contrastive foci. We would need more data in order to see how they relate to the accompanying gestures (see Section 4.2.2 for a discussion) and we hope that future research will shed light on this question.

6 Further Issues Some issues are still open for discussion and call for further research. In the following, we address some of these topics. In particular, we want to point

Sinn und Bedeutung 15 - [PDF Document] (206)

Focus Marking via Gestures 205

out that the alignment of focus phrases and gesture phrases is ‘real’ and not merely an epiphenomenon of some underlying alignment effect of a different nature.

6.1 A Qualitative Argument It has been proposed in the literature that gesture phrases align with ‘tone groups’ (Kendon 1972) or ‘intermediate phrases’ (Loehr 2004), cf. section 2. We have now added another suggestion: gestures align with focus phrases. However, it is possible that none of these claims are true, and that gestures are simply synchronised with certain syntactic categories, e.g. entire sentences or VPs. As our corpus predominantly consists of all-foci sentences and VP-foci, this possibility cannot be excluded without further inspection. Unfortunately, the SAGA corpus is not syntactically annotated, so a quantitative evaluation of how well different kinds of syntactic categories align with gestures cannot easily be carried out without time-consuming manual work. However, we attempted a qualitative assessment of this question. We took a closer look at narrow foci and foci that begin a considerable time later than the corresponding utterance and checked how well they align with an accompanying gesture. We found that if a focus does not begin at the start of the utterance, the corresponding gesture also begins at some later point in nearly all cases. In (8) we give some examples in point:

(8) a. genau äh also [e|Grst Kreisverkehr]F exactly eh so first roundabout ‘exactly, eh, first the roundabout’ b. die haben beide |G[dieselben Türen und dieselben Fenster]F they have both the same doors and the same windows ‘they have both the same doors and the same windows’ c. rechts von dieser Kap|Gelle [ist ein großer Laubbaum]F right of this chapel is a big broadleaf tree ‘to the right of this chapel there is a big broadleaf tree’

In all three example cases, the gesture starts near the start of the focus phrase and not at the beginning of the utterance. The gesture phrase thus seems to be aligned with the focus phrase and not with the entire utterance. Furthermore, we found no evidence for a general alignment of gesture phrases with any syntactic categories such as sentences or VPs (see Wilmes 2009 for details).

6.2 A Quantitative Argument Here, we attempt to show that the alignment of gesture phrase and focus phrase cannot be a secondary effect of the well-known stroke-accent

Sinn und Bedeutung 15 - [PDF Document] (207)

206 Ebert, Evert & Wilmes

alignment and the fact that the initial part of the focus phrase (up to the main accent) and the preparation phase have similar lengths. Note that the time difference ΔtF between onset of gesture and focus phrase is the sum of the time difference ΔtA between onset of nuclear accent and stroke and the length difference Δl between preparation phase of the gesture and focus phrase up to the main accent. Assuming that ΔtA and Δl are independent alignment effects, we would expect the standard deviation of the resulting gesture-focus alignment ΔtF to be greater than the standard deviations of ΔtA and Δl. This is not the case: the standard deviation of ΔtF was only 0.41s in our study, whereas the expected standard deviation would be 0.82s (see Wilmes 2009 for details on this calculation). Moreover, we would then expect a strong correlation between the time differences ΔtF and ΔtA as well as ΔtF and Δl, while ΔtA and Δl themselves should be independent or weakly correlated. Our data show an opposite effect: there is only a weak correlation between ΔtF and ΔtA (Pearson’s r ≤ 0.219), but a very strong correlation between ΔtA and Δl (Pearson’s r = 0.759). From these results and the pairwise correlation plots (omitted for lack of space), we conclude that the length differences arise from two independent alignment effects for stroke and main accent, and for gesture and focus phrase.

7 Conclusion In our study, we were able to verify claims from the literature that gestural strokes and nuclear accents align (albeit with a systematic shift). We also found a clear, but shifted alignment for the onsets of gesture phrases and (new-information) foci. We interpret these results to show that gestures are a means of marking information structure next to intonational and syntactic means, i.e. speech-accompanying gestures can indicate focus domains.

Furthermore, we were able to show that gestures can serve to disambiguate. A sentence that is information-structurally ambiguous in iso-lation can be disambiguated by its accompanying gestures. This is yet another observation suggesting that ambiguity might be less of a problem for natural language than was originally thought. While many sentences (e.g. simple SVO sentences with two quantifiers) that seem ambiguous at first sight are disambiguated via intonation in natural speech, we showed that sentences that seem ambiguous even when intonation is taken into account are in fact disambiguated by accompanying gestures.

We hence support the view of Lascarides and Stone (2009) that a formal semantic model should represent not only the usual semantics of linguistic

Sinn und Bedeutung 15 - [PDF Document] (208)

Focus Marking via Gestures 207

expressions, but also take care of the semantic contribution of their accompanying gestures.

References Condon, William & W. Ogston. 1966. Soundfilm analysis of normal and

pathological behavior patterns. Journal of Nervous and Mental Disease 143. 338–347.

Dipper, Stefanie, Michael Götze & Stavros Skopeteas. 2007. Information structure in cross-linguistic corpora: Annotation guidelines for phonology, morphology, syntax, semantics, and information structure. ISIS 7. 147–187.

Jannedy, Stefanie & Norma Mendoza-Denton. 2005. Structuring information through gesture and intonation. Interdisciplinary Studies on Information Structure, 3. 199–244.

Kendon, Adam. 1972. Some relationships between body motion and speech: An analysis of an example. In A. Siegman & B. Pope (Eds.), Studies in dyadic communication. New York: Pergamon Press.

Kendon, Adam. 1980. Gesticulation and speech: Two aspects of the process of utterance. In Mary Ritchie Key (Ed.), The Relationship of Verbal and Nonverbal Communication. The Hague: Mouton.

Lascarides, Alex & Matthew Stone. 2009. A formal semantic analysis of gesture. Journal of Semantics. 1–57.

Loehr, Daniel. 2004. Gesture and Intonation. PhD thesis, Georgetown University, Washington, DC.

Lücking, Andy, Kirsten Bergman, Florian Hahn, Stefan Kopp & Hannes Rieser. 2010. The Bielefeld Speech and Gesture Alignment Corpus (SaGA). In M. Kipp, J.-C. Martin, P. Paggio & D. Heylen (eds.), LREC 2010 Workshop: Multimodal Corpora–Advances in Capturing, Coding and Analyzing Multimodality.

McClave, Evelyn. 1991. Intonation and Gesture. Doctoral Dissertation, Georgetown University, Washington DC.

McNeill, David. 1992. Hand and Mind: What gestures reveal about thought. The University of Chicago Press, Chicago and London.

Pierrehumbert, Janet B. 1980. The Phonology and Phonetics of English Intonation. PhD dissertation, MIT. [IULC edition, 1987].

Pittenger, R., Hockett, C. & Danehy, D. 1960. The first five minutes: A sample of microscopic interview analysis. Ithaca, NY: Paul Martineau.

Sinn und Bedeutung 15 - [PDF Document] (209)

208 Ebert, Evert & Wilmes

R Development Core Team. 2005. R: A language and environment for statistical computing, reference index version 2.x.x. R Foundation for Statistical Computing, Vienna, Austria. URL

Wilmes, Katharina A. 2009. Focus marking by speech-accompanying gestures. Bachelor Thesis. University of Osnabrück. Germany.

Sinn und Bedeutung 15 - [PDF Document] (210)

Imperatives as Future Plans

Regine Eckardt University of Göttingen


Abstract. Disjoint imperative sentences like (Nimm die) Hände hoch, oder ich schiesse!, literally (take your) hands up, or I’ll shoot! intuitively present the addressee with all her alternatives for action. The speaker informs that all future worlds, as far as the speaker can forsee, are such that the addressee raises her hands or gets killed. I propose a semantic/pragmatic analysis for sentences in the imperative mood that adopts this exhausitve description of future alternatives as a semantic backbone. Different contextual instantiations of alter-natives capture a wide range of uses of sentences in imperative mood, as well as coordinations of imperative and declarative sentences, in a uniform way.

1 Some Observations about Imperatives 1.1 Variety It has frequently been noted that sentences in imperative mood (Simp) can express a wide variety of speech acts, some directive, some not. I will take my starting point from the following range of examples.

(1) Leave my garden! (command/request)

(2) Lend me your bike, please! (plea)

(3) Take a cookie! (offer, invitation)

(4) Take an umbrella with you! (advice)

(5) Ok. Go kill yourself. Smoke! (concession, „giving in“)

(6) Get well soon! (well-wish)

(7) Come and take the ball (if you dare)! (dare)

These are part of the agenda set by Condoravdi & Lauer (C&L, 2010a, b) in a recent series of talks, drawing on earlier literature (e.g. Schwager 2006a,

In Reich, Ingo et al. (eds.), Proceedings of Sinn & Bedeutung 15, pp. 209–223. Universaar – Saarland University Press: Saarbrücken, Germany, 2011.


Sinn und Bedeutung 15 - [PDF Document] (211)

210 Eckardt

2008, Donhauser 1986, Bybee, Pagliuca & Perkins 1994).1 It would be desirable to derive the different types of act compositionally from the literal content of the sentence, the semantics of mood, and knowledge in context that pertains to the interpretation of imperatives. Such an approach would certainly be preferable over stipulating a range of speech act operators and leave the choice of the correct operator subject to a holistic, noncompositional evaluation of the overall utterance situation (Searle 1969). An operator approach would, for instance, force us to postulate categorical distinctions between different act types where in practice, we find a gradual continuum between e.g. command and request, request and plea, request and advice and so on. Schwager (2006) and Portner (2007) have led the way in demonstrating how compositional semantics for imperative mood in speech acts can look like.

1.2 Conjoining Simp and Sdecl Simp can be conjoined with sentences in declarative mood Sdecl. The result are speech acts of different natures, including anti-directive acts such as threats, like in (10).

(8) Clean your room, and I will take you to the movies. (request + incentive) (9) Open the newspaper, and you will find the king’s picture on page 2. (conditional) (10) Touch this glass, and I will kill you. (threat + sanction)

Sometimes, the speaker wants the addressee to act as required by the imperative (Do!) but sometimes he aims to avoid exactly that, practically intending to say Don’t! (Schwager 2006a, Russell 2007, van Rooij & Franke 2010, Bolinger). A commonality of examples like (8) to (10) seems to be that they all can equivalently be expressed by a conditional (‘If you clean your room, then I will take you to the movies’ etc.). This is why scholars have proposed to class Simp as pseudo-imperatives here and propose a common conditional meaning for the construction. It would be attractive to have an analysis that relates the meaning of (8) to (10) to the interpretation of “normal” imperative sentences in a transparent manner.

1.3 Disjunctions Simp or Sdecl Simp can likewise enter disjunctions with a “face the consequences” clause, like in the following example. 1 Two more types of act that they include, namely WISH and ILL-WISH, will only be touched later in this paper.

Sinn und Bedeutung 15 - [PDF Document] (212)

Hands up! 211

(11) Freeze! or I’ll shoot you.

These intuitively present the addressee with all her alternatives for action. The speaker informs that all future worlds are such that the addressee raises her hands or gets killed. It is not possible to add a disjunct that describes more promising alternative prospects, in order to prohibit the addresse from acting as specified by Simp (Russell 2006, van Rooij & Franke 2010 a.o.).

(12) Go on fighting, or you’ll get chocolate. ≠ ‘If you stop fighting, you’ll get chocolate’ (ironically?)

Whenever the speaker seriously attempts to motivate the addressee to act according to Simp, it is standardly possible to spell out his underlying in-centives by an or-clause. Hence, while conjunctions like in 1.2 might be viewed as a deviant case, the use of disjunctions frequently just explicates the reasoning behind a typical directive uses of Simp.2 Again, it would be appealing to read that off directly from the semantics of imperative mood.

2 Modal Theories for Imperatives 2.1 Earlier Theories I agree with earlier authors on imperatives who assume that literal meaning and speech act should be captured in one integral overarching theory. This leads naturally to analyses of Simp that play on their semantic closeness to deontic necessity. A recent prominent example is Schwager (2006a, b and subsequent). We will generally assume that the sentential root [[ Simp ]] denotes a property which gets instantiated by the addressee A to yield [[ Simp ]] (A).

[[ Freeze! ]] Schwager = ∀w [ “BEST-WORLDS(w, wo)” → FREEZE(A, w)) ]

I use BEST-WORLDS as a cover term for factors that determine the domain of quantification. These include the choice of a modal base (FUTURES which are CIRc*msTANTIALLY POSSIBLE) and a partial ordering of the worlds which, among other criteria, refers to what ACCORDS.WITH.SPEAKERS.DESIRES(wo,w). Schwager proposes that this is further specified by context (“In what sense does the speaker want this to happen?”) which leads to different flavours of imperatives. The modal quantification is contributed by an imperative mood operator, and finally the scope of this quantification is contributed by the content Simp(A) (following Kratzer 1981, 1991). Schwager’s analysis is attractive because it shows tight fit with necessity modals, it can be naturally 2 See Schwager 2006a, 2008 for a very lucid discussion of the relation between sentence mood and typical associated speech acts.

Sinn und Bedeutung 15 - [PDF Document] (213)

212 Eckardt

extended to conditionals, and it has a smooth semantics-speech act interface. The theory doesn’t need extra components or ontology like TO-DO-lists, plans, action schemes or the like.

Another, more recent proposal in a similar line was issued by Condo-ravdi and Lauer (C&L, 2010a, b). They suggest that Simp expresses the desire of the SPEAKER that the ADDRESSEE commits herself to act as if she (= A) preferred [[ Simp(A) ]] (Condoravdi & Lauer 2010a: 10). Like Schwager, they assume that the information content of imperatives alone is sufficient to predict its speech act qualities, and envisage a smooth semantics – speech act interface. Portner’s (2007) analysis will be disregarded here because it stipulates the use of an extra list of propositions called the TO-DO LIST.3

2.2 Coverage of Observations Variety is accounted for by both approaches, where both fit more naturally for some cases than for others. Specifically, as both analyses rest on speaker desires, they will need to ascribe the speaker strangely desinterested and altruistic desires in some cases. Schwager captures flavours by different specific ACCORDS.WITH.SPEAKERS.DESIRES properties (capturing offer, warning, advice, wish, and several “deviant” uses). C & L (this volume) likewise attempt to derive known examples from their intricate mix of speaker and hearer preference, which I will discuss in section 4. The coordination cases can not be captured easily by a naive extension of the modal analyses (Schwager, 2006a); perhaps to the exception of the Simp and Sdecl,good cases. In view of the obvious problems that arise, Schwager (2006a) proposes very different, and much more sophisticated ways to interpret the respective conjunctions and disjunctions. The coordination Simp and Sdecl is simply interpreted as conditional. The imperative operator will contribute the modal quantification scheme:

∀w [ FUTURE(wo,w)∧CIRC(wo,w)∧ … P(w) … → Q(w) ]

In non-coordinate imperatives, the syntax-semantics interface instantiates P(w) with speaker-desire, and Q(w) is instantiated by Simp(A). In the conjunctive case, however, P(w) gets instantiated by the speaker’s desires plus the content of imperative (if you do Simp…) whereas Q(w) instantiated by and-clause (…then Sdecl will happen). The result is descriptively adequate, but the semantic derivation of dubitable legitimation. (It is claimed that a topical status of the imperative leads to its analysis in the restrictor of some

3 While the components of Portner might be reconstructed in terms of the other two competing theories, a full comparison is beyond the scope of the present paper.

Sinn und Bedeutung 15 - [PDF Document] (214)

Hands up! 213

quantifier; the topical status is attributed to the imperative on basis of prosodic cues that are inconclusive.) According to Schwager on Simp or Sdecl, the idea that ‘or’ could mean Boolean disjunction has to be radically denied. Her treatment of Simp or Sdecl rests on Geurts (2005) who proposes that ‘or’ denotes a conjunction of modal quantifications where background and propositional slots get instantiated by the sentence to be interpreted. Ci are contextually given sets of worlds; Mi ∈ {¸,¨} and Pi = disjuncts.

(14) C1 M1 P1 ∧ C2 M2 P2

Schwager makes use of this scheme in a sophisticated way, assuming that C1 = CG (common ground); M1 P1 = ¸Simp(A) ∧ [[ Simp ! ]] , second context C2 = CG \ Simp(A) and finally M2 P2 = ¨ [[ Sdecl ]] . The result can be spelled out as “It is possible that Simp; and in all speaker-desirable worlds, Simp actually happens; and in all worlds where it does not happen, Sdecl will necessarily be true.” This leads to a descriptively adequate semantic representation. How-ever, Geurts’ background theory and the cases at hand do not yet match perfectly. The first conjunct doesn’t unify well with Geurts’ scheme (14), likewise Geurts does not discuss changes between modal bases extensively (e.g. from epistemic to buletic to future-no-matter-what).4 Condoravdi & Lauer do not address coordinate constructions with imperatives. I will come back to their proposal and undertake a more detailed comparison once the Hands-Up theory has been presented.

3 Hands-­‐Up Theory for Imperatives 3.1 The Backbone I propose two kinds of imperative construction operators [ ! ] and [ ¡ ], each with syntactic requirements, denotation and presupposition. Given that I will not deal with conflicting desires or obligations explicitly, I will notate modal quantification in an entailment format. FUTURE, CIRC, DEONT etc. are intended to deliver the future, circ*mstantial etc. alternatives of wo and LEWIS-SIM is used to remind us of the fact that we want to exclude the more obscure of all logical possibilities sometimes. The notation should be reversible to one based on modal base and ordering source.

[ ! ]: Syntax: one obligatory argument: finite sentence in imperative mood Simp

4 Schwager herself comments on the analysis in much the same spirit. It should be kept in mind that all simpler mappings from syntax to semantics were inevitably bound to yield wrong results, so this analysis constitutes true progress.

Sinn und Bedeutung 15 - [PDF Document] (215)

214 Eckardt

one optional argument: or-phrase with or-P → ‘or’ Sdecl Semantics: λpλq∀w[FUTURE(wo,w)∧CIRC(wo,w)∧LEWIS-SIM(wo,w) → p(w)∨q(w)] Presupposition: the speaker believes that the addressee, taking a choice in all life future options λw.Future(wo,w)∧Circ(wo,w)∧Lewis-Similar(wo,w) prefers p-worlds to q-worlds.

[ ¡ ]: Syntax: first obligatory argument: finite sentence in imperative mood Simp second obligatory argument: and-phrase with and-P → ‘and’ Sdecl Semantics: λpλq∀w [FUTURE(wo,w)∧CIRC(wo,w)∧LEWIS-SIM(wo,w) → p(w); q(w) ∨ C(w)] Pragmatics: C propositional variable to be instantiated in context C ⊆ ¬p and C ∩ ¬q ≠ Ø (hence C-worlds might avoid the consequences presented in second conjunct5)

The coordinations and and or do not enter semantic composition but are interpreted syncategorematically. Coordinating and is reflected as narrative sequencing; as in DRT. Usually, Simp and Sdecl are tightly linked anaphorically under conjunction. I will not go into the details of [[ Simp;Sdecl ]] which would require dynamic lambda logic. The disjunction is strengthened to exclusive disjunction in most cases. I will in one case below refer to this strengthening. Finally note that the approach once again stipulates a semantic difference between and-coordinations and or-disjunctions. Unlike other analyses, the one defended here treats the two cases maximally parallel, differing only in how the same semantic parameters get instantiated in either case, and in the presence or absence of one presupposition.

3.2 Examples of [ ! ]-­‐Imperatives I will now survey how the analysis can treat various kinds of uses of the imperative. Different types of propositional OR arguments yield different flavours of imperatives. I assume throughout that if the second argument of [ ! ] is not overtly realised, it will be instantiated in context. Let us start with Command, the most prototypical use of imperative mood.

(16) Remove your car! 5 Thanks to Sven Lauer who suggested this specific version of restriction.

Sinn und Bedeutung 15 - [PDF Document] (216)

Hands up! 215

The first argument of the [ ! ] operator λw[REMOVE(A, CAR-OF-A, w)] is provided by the imperative sentence. The hearer understands the presupposition that the speaker believes that the addressee will prefer REMOVE-CAR-worlds to q-worlds. Depending on the situation at hand, the hearer might guess that leaving the car will cause trouble with the police, e.g. she will get a ticket. Hence, the overall proposition conveyed is this:

∀w[FUTURE(wo,w)∧CIRC(wo,w)∧LEWIS-SIM(wo,w) → [REMOVE(A, CAR-OF-A, w) ] ∨ TICKET(w) ]

The utterance presupposes that REMOVE-worlds are better than TICKET-worlds. The “force” of the command derives from the threatening nature of the alternatives. The more likely the speaker holds the TICKET case, and the less she is inclined to loose money, the more likely will she comply to the command. — Next, consider Warning/Advice, like in (17).

(17) Wear a raincoat!

The speaker in (17) need not have a personal desire for the addressee to comply. (Theories that rephrase imperative sentences as reports about the speaker’s desires will find such examples worrisome.) The present analysis predicts that [ ! ] will take λw[WEAR(A, RAINCOAT, w)] as its first argument. The second derives from common knowledge about the current weather, the health state of the addressee; let us assume a simple q = ‘you will get wet’. The speaker conveys, and the hearer accepts the presupposition: RAINCOAT-worlds are better for the hearer than WET-worlds.

∀w[FUTURE(wo,w)∧CIRC(wo,w)∧LEWIS-SIM(wo,w) → [WEAR( A, RAINCOAT, w)] ∨ WET(A, w)]

In giving desinterested advice, the speaker points out certain facts and leaves it to A to act in the most reasonable way. We’d expect, however, that the speaker does not mind if A reacts in the indicated manner—or else, the speaker would not have pointed out these facts in the first place. We will come back to this fact. The analysis can nicely reflect speaker’s Authority and, more interestingly, the Lack of Authority. Consider the unspecific request in (18).

(18) Be quiet! (or ... ?) ∀w[FUTURE(wo,w)∧CIRC(wo,w)∧LEWIS-SIM(wo,w) → [QUIET( A, w) ] ∨ PUNISHMENTS(A, w) ]

The hearer could draw on knowledge about speaker like “wow, this speaker is a fierce guy who could earlier think of nasty PUNISHMENTS”. The speaker

Sinn und Bedeutung 15 - [PDF Document] (217)

216 Eckardt

conveys the presupposition: “I, the speaker, believe that you’ll like the QUIET-worlds better than PUNISHMENT-worlds”. Drawing on earlier knowledge, the hearer will believe this presupposition and accordingly hold her mouth. Speakers with little authoritative force lead to a different kind of hearer knowledge, e.g. “this speaker’s PUNISHMENTS are not severe”. Again, the speaker conveys the presupposition that ‘he believes that the hearer finds QUIET-worlds are better than PUNISHMENT-worlds’. The addressee, however, might disagree with the speaker and therefore opt for the worlds where λw.¬[QUIET(A, w)]. We see that Authority can be based on the experience that S was able to think about drastic measures in the “or”-case on earlier occasions. I will next address Permissions like the cookie invitation.

(19) Take a cookie! λq∀w[FUTURE(wo,w)∧CIRC(wo,w)∧LEWIS-SIM(wo,w)→ [TAKE( A, COOKIE, w) ] ∨ q(w) ]

(19) suggests that—contrary to earlier belief of A—it is not forbidden to take a cookie. Why? The speaker conveys the presuppositions that the addressee will prefer COOKIE-worlds to NO-COOKIE-worlds. If the speaker can be trusted, this includes a commitment to not punish Addressee if she takes a cookie. (19) is typically understood as a permission in contexts where the only disadvantage of NO-COOKIE-worlds for A is that she does not get a cookie (which the speaker should believe a desirable thing to have). Consequently, the addressee can decide to decline this offer—for instance if she is on a diet or does not like cookies very much. Hence, permissions arise as one possible instantiation of q. — Let us finally look at Concessives. I will use an example in a naturally sounding prediscourse.

(20) a. Don’t smoke (, or you’ll die young)! b. (nag nag nag) — Well, then do smoke! Kill yourself!

Intuitively, (20) shows that speaker and addressee disagree in certain respects. This is reflected in the presuppositional discourse record. (20a) entails that the Speaker believes that Addressee prefers NON-SMOKE-worlds (= LIVE-LONG-worlds) to DIE-YOUNG-worlds (= SMOKE-worlds). In (b), [ ! ] takes a first argument λw[SMOKE( A, w) ] with the second argument missing. With the presupposition conveyed in (b), the Speaker acknowledges that Addressee prefers SMOKE-worlds (= DIE-YOUNG-worlds) to NON-SMOKE-worlds (= LIVE-LONG-worlds). As part of the discourse record, however, the speaker has made it clear that she does not share this preference and does not think it reasonable. This also leads to an ironic undertone.

Sinn und Bedeutung 15 - [PDF Document] (218)

Hands up! 217

I will leave the remaining cases to the reader. PLEAS are characterized by the moral pressure of the ‘or’-alternative. The speaker hopes that the addressee will prefer making her happy to making her miserable. WELL-WISHES straightforwardly acknowledge addressee’s preferences without that either addressee or speaker can do anything to drive the course of events towards such happier alternatives. DARE! cases, finally, convey an intricate conditional preference of the addressee: If A ‘dares’, i.e. overcomes her fear of bad consequences, then she will prefer worlds where she takes the ball (ex. 7) to worlds where she doesn’t take it. ‘Daring’ is tantamount to ‘countering the situation with enough strength and energy so as to overcome the obviously threatening dangers’. We find a continuum of attitudes between the encouraging “come, take the ball if you dare” by the provocative coach and evident threats as Dare! (and you will see what happens). As the present analysis assumes that the flavour of imperatives derives from contextual instantiation of the ‘or’-cases, we’d expect such a continuum.

3.3 Examples of [ ¡ ]-­‐Imperatives I will now turn to the conjunction Simp-and-Sdecl which are analysed with [ ¡ ]. Recall that the second argument is obligatorily instantiated (i.e. we overtly see the and clause) and there is no presupposition as to what is good or bad for the addressee. The content of the second argument alone determines whether the worlds where Simp(A) is true are better or worse for A.

(21) Come in, and you will get coffee.

[ ¡ ] = λpλq∀w[FUTURE(wo,w)∧CIRC(wo,w)∧LEWIS-SIM(wo,w) → p(w);q(w) ∨ C(w)] will apply to the first argument λz.COME-IN(A, z) and the second argument by dynamic update: λz.GET(A, COFFEE, z). Pragmatics requires that C is a proposition to be instantiated in context where C ⊆ [[ NOT Simp ]] = λz.¬ COME-IN(A, z) and moreover C ∩ [[ NOT Sdecl ]] = C ∩ λz.¬GET(H, COFFEE, z) ≠ Ø. The elsewhere-case C describes a missed occasion: Speaker believes that Addressee prefers COFFEE-worlds to NO-COFFEE-worlds.6 As in the cookie example, the “force” of the offer depends on the addressee’s eagerness not to miss an occasion to get coffee. Note that it is incoherent to combine motivational conjuncts and threatening disjuncts.

(21) a. #Come in, and you will get coffee, or I won’t talk to you for days.

6 In a richer account, the not-getting coffee needs to be tied to a limited interval of time; the time that would correspond to the time after the non-occurring entry.

Sinn und Bedeutung 15 - [PDF Document] (219)

218 Eckardt

This is captured by the syntactic (raw) analysis. Given that the coordinate clause Simpand/orSdecl as a whole does not count as Simp, the structure is not recursive. More interestingly, perhaps, is what happens in the Threat case.

(22) Touch this cookie, and I will kill you. (22) a. first argument of [ ¡ ]: λz.TOUCH(A, COOKIE, z) second argument of [ ¡ ], dynamic update: λz.KILL(S, A, z) Psp. for alternatives C: (i) C ⊆ λz.¬TOUCH(A, COOKIE, z) (ii.) C ∩ λz.¬KILL(S, A, z) ≠ Ø

The speaker assumes that Addressee prefers not being killed to being killed. The imperative informs her that if she avoids touching the cookie there is a chance to stay alive. Earlier theories interpret Simp-and-Sdecl as conditionals. The present analysis treats Simp-and-Sdecl maximally similar to other imperative clauses but it predicts that Simp-and-Sdecl entail conditional statements. The example in (23) is a typical conditional case, but the reasoning holds for all examples.

(23) Open the newspaper, and you’ll see the king on page 2. first argument of [ ¡ ]: λz.∃x(NEWSPAPER(x) ∧ OPEN(A, x, z)) second argument of [ ¡ ]: λz.SEE(A, KING, PAGETWOOF(x), z) Presupposition: C in context, (i) C ⊆ λz.¬∃x(NEWSPAPER(x) ∧ OPEN(A, x, z)) (ii.) C ∩ λz.¬∃x(NEWSPAPER(x)∧SEE(A, KING, PAGETWOOF(x), z)) ≠ Ø i.e. there is a chance for A to see the photo of the king.

The instantiation of C is restricted to sets of worlds where A doesn’t open a newspaper. The overall modal quantification states that all future courses where newspapers get opened by A are such that the king’s picture is on p.2. This entails the conditional “If you open the newspaper, you’ll see the king”. Admittedly, the entailment is again hard-wired in the interpretation of [ ¡ ] and maybe therefore no less stipulative than in competing analyses. However, the stipulation here echoes the strengthening of disjunction in the plain imperative case. [ ! ]-imperatives typically inform the addressee what happens if, and what happens if she does not engage in certain actions (e.g. freezes). In the simple case, this dichotomy can be modeled by exclusive disjunction. In the [ ¡ ] case, exclusive disjunction will not be sufficient to maintain this division of worlds into cases. The condition that C ⊆ ¬p therefore simply transfers exclusivity of cases to the [ ¡ ] denotation. This concludes the discussion of examples. What is missing so far are ill-wishes like “Die!”, “Eat sh*t!”. These obviously rest exclusively on what is

Sinn und Bedeutung 15 - [PDF Document] (220)

Hands up! 219

desirable for the speaker. They do not fit into the basic version of the analysis and will be touched in section 4.2.

4 More Ties to Earlier Literature 4.1 Van Rooij and Franke, 2010 In making the assumption that there are two imperative operators [ ! ] and [ ¡ ], I stipulate a fundamental difference between and and or in imperatives. Of course, it would be desirable to derive the different behaviours from more basic facts about imperatives and coordination. In a recent paper, van Rooij & Franke propose that it can be predicted on a game theoretic basis. They address the fact that only and can be used to “reverse” the intention of an imperative, as illustrated again in (24). Only (24b) conveys a serious invitation to eat spinach.

(24) a. *Don’t eat your spinach, or I will give you a dollar. b. Leave your spinach, and I will beat you.

R&F’s idea is simply this: Both imperatives in (24a) and (24b) state what the speaker wants not be done. Both erroneously prime the listener to not eat spinach. (24a) counteracts by promising a reward for the elsewhere case; (24b) counteracts by promising a punishment in the imp! case. Now the reward case competes with other ways to call out similar rewards, e.g. (25).

(25) If you eat your spinach, I will give you a dollar.

Van Rooij and Franke argue that the reward in (25) can be somewhat lower than the one in (24a) because in (25), it only needs to overcome the addressees reservations against spinach whereas (24a) has to overcome these plus the additional linguistic priming to not eat spinach, caused by the imperative. Therefore (25) systematically wins over (24a). In principle, the dual threat in (24b) faces a similar competition. The speaker likewise could decide to say ‘If you eat your spinach, I will not beat you.’ or such. And again, cheaper threats are required here because priming of the unwanted action has been avoided. However, van Rooij and Franke say, a costly punishment is not as binding a social commitment as a costly reward. Society will sanction those who promise big rewards and do not pay. In contrast, society rather rewards those who lower punishment. Therefore, false priming is not equally uneconomic when it only raises punishment costs: You can always lower your costs again by simply not punishing so badly. This argument would certainly be appealing, but there are parallel examples where speakers indeed offer promising vs. unpromising alternatives in those

Sinn und Bedeutung 15 - [PDF Document] (221)

220 Eckardt

costly ways that R&F want to exclude. This is possible both with disjoint declaratives and with disjoint imperatives. (26) demonstrates the strategy in a common parent-child interaction (the parent wants actually to get home).

(26) You can either stay on the playground longer, or we’ll have time to have an ice cream on our way home.7

(27) is to be understood in a context which advertises cosmetic surgery. (Of course, I do not submit to the argument.)

(27) It’s your decision: Remain an unremarkable average person for the rest of your life, or make an appointment with Dr. Knock’s cosmetic surgery clinics today!

Such examples show that speakers are indeed willing to make suggestions in ways where priming has to be countered with higher rewards, even sugges-tions that are worded in the imperative mood. The pattern is just conven-tionally not available for the Simp or Sdecl coordination. I therefore conclude that the asymmetric behaviour of and/or coordination is a conventional part of the pragmatics of Simp coord Sdecl and needs to be coded in grammar.

4.2 Condoravdi and Lauer, again C&L (2010a, b, 2011) argue in favour of a general model for speech acts in terms of public beliefs and commitments. Specifically, they propose that “the utterance of an imperative p! commits the speaker to act as if he had a preference for the hearer committing himself to act as if he preferred p” (C&L, 2010b). They assume that commitments are part of public beliefs in common ground update (Stalnaker, 2002). Public commitment to p will add p to a (public) list of the agent that reveals his preferences that drive his decisions for action. Preference lists feed modal quantification and offer a natural link to statements like ‘I must p’ that are entailed by imperatives. General public will watch whether the agent’s behaviour accords with his public commitments. If discrepancies get too large, the general public can decide on sanctions, thereby taking responsibility for the ‘elsewhere’ worlds that are part of the imperative’s meaning in the Hands Up! approach.

C&L’s analysis is a sophisticated variant of a speaker-buletic modal. It is therefore ideally suited to analyse imperative uses for wishes, including ill-wishes. These are hard for my own proposal, according to which the speaker basically asserts that it would be in the hearer’s own interest to take a certain action. Arguably, this does not fit the ‘drop dead’ example.

7 Thanks to Manfred Sailer who brought up this type of example.

Sinn und Bedeutung 15 - [PDF Document] (222)

Hands up! 221

(28) Please, be blond! (wish in absence of addressee) Drop dead! (ill-wish)

For the same reason, however, C&L’s analysis has problems with imperatives used for desinterested ADVISE, cookie INVITATIONS, CONCESSIVES and DARE! imperatives. They do not discuss the use of imperatives in threats of the Simp and Sdecl form, but given that the speaker will not have an interest for the listener to follow Simp! in these cases either, these coordinations should be extremely problematic in that approach. The appealing vision in C&L’s approach is that ordering sources in modal semantics can be reconciled with preference lists in action planning. In accepting an imperative p!, the addressee is assumed to rank p high on his list of preferences. The actual ranking of preferences will not be fully determined by the imperative utterance, because the hearer could have other aims that he pursues with even higher priority. This underspecification is certainly adequate.

The Hands Up! analysis, in contrast, contributes in a more local, but also more explicit way to the facts that determine the addressees actions. Take the drastic initial Freeze, or I will kill you! Before hearing and believing the content of this imperative, the addressee A might have planned (= pre-ferred with high priority) to not freeze but have a coffee. By learning that his next future options are either to freeze or to get killed, A does not simply demote his earlier plan ‘I will now have a coffee’ to a somewhat lower rank. What A indeed faces is a quite drastic belief revision: He learns that the coffee plan is not part of any possible future at all, and that his choice is a quite different one.

4.3 Ross’ Paradox Let me finally show how Ross’ paradox can be avoided. Ross (1944) is quoted as the first to observe that simple-minded modal analyses of imperatives carry the danger of falsely predicting that (29) entails (30).

(29) Come! (30) Come, or stay!

The problem arises due to the fact that any world that has property p also has the weaker property p∨q. The Hands-Up! analysis does not predict this false entailment: In the following, I will use the proposition SANCTIONS as a cover predicate for contextually given sanctions that the hearer could understand.

(29’) [[ [ ! ] Come! ]] = [ ! ] (λw.COME(A,w)) (λw.SANCTIONS(A,w)) = λz.∀w[FUTURE(z,w) ∧ CIRC(z,w) ∧ LEWIS-SIM(z,w) → Come(A,w) ∨ Sanctions(A,w) ]

Sinn und Bedeutung 15 - [PDF Document] (223)

222 Eckardt

Presupposition: Speaker believes that A will like worlds in λw.COME(A, w) better than worlds in λw.SANCTIONS(A, w).

(30’) [[ [ ! ] Come, or stay! ]] = [ ! ] (λw.COME(A, w) ∨ STAY(A, w)) (λw. SANCTIONS (A, w)) = λz. ∀w(FUTURE(z, w) ∧ CIRC(z, w) ∧ LEWIS-SIM(z, w) → (Come(A, w) ∨ Stay(A, w)) ∨ Sanctions(A, w) )

Presupposition: Speaker believes that A will like worlds in λw[COME(A, w) ∨ STAY(A, w)] better than worlds in λw.SANCTIONS(A, w). There are two ways to avoid Ross’ paradox. The first way is to assume, as we did in earlier places, that the or which separates the sanction case from the Imp! cases is an exclusive or ∨e. With this assumption, (29’) no longer entails (30’): Assume that there is a world which is both a STAY(A)-world and a SANCTION-world. Then (29’) can be true but (30’) will be false. Hence, (30’) is not entailed by (29’).

Exclusive disjunction: ( φ(x) ∨e ξ(x) ) –×→ ( (φ(x) ∨ ψ(x)) ∨ ξ(x) )

The second way to block the inference from (29) to (30) will leave us the option for inclusive ‘or’ in the representation of imperatives in the Hands Up! format. It argues via presuppositions. (29) presupposes that the speaker believes that the addressee prefers worlds in λw.COME(A, w) over worlds in λw.SANCTIONS(A, w). (30) presupposes that the speaker believes that the addressee prefers worlds in λw(COME(A, w) ∨ STAY(A, w)) over worlds in λw.SANCTIONS(A, w). We can model these preferences by universal statements of the following kind: All worlds in λw(COME(A, w) ∨ STAY(A, w)) are better than any world in λw.SANCTIONS(A, w). With this explication of preferences, the person who utters (29) will not be committed to the content of the presupposition of (30) because s/he believes that some STAY worlds are also SANCTION worlds and therefore not any better than other SANCTION worlds. The details of weighing worlds against worlds would need to be worked out in detail, but the approach opens up another way to avoid Ross’ paradox in the Hands Up! theory.

References Austin, J. 1972. How to Do Things With Words. Edited by J.O. Urmson, M.

Sbisà. Cambridge, Harvard University Press. Bach, K. & R. M. Harnish. 1979. Linguistic Communication and Speech Acts.

Boston: MIT Press.

Sinn und Bedeutung 15 - [PDF Document] (224)

Hands up! 223

Bybee, J, W. Pagliuca & R. Perkins. 1994. The Evolution of Grammar: Tense, Aspect and Modality in the Languages of the World. Chicago: Chicago University Press.

Condoravdi, C. & S. Lauer. 2010a. Imperatives and Public Commitments. Handout, 11th Annual Semantics Fest, Stanford, March 2010.

Condoravdi, C. & S. Lauer. 2010b. Performative Verbs and Performative Acts. Handout, Sinn und Bedeutung 15, Saarbrücken, September 2010.

Donhauser, K. 1986. Der Imperativ im Deutschen. Hamburg, Buske. Geurts, B. 2005. Entertaining alternatives: disjunctions as modals. Natural

Language Semantics 13. 383–410. Jary, M. 2007. Are explicit performatives assertions? Linguistics and

Philosophy 30. 207–234. Kratzer, A. 1981. The Notional Category of Modality. In Words, Worlds, and

Contexts. New Approaches in Word Semantics, edited by H. J. Eikmeyer & H. Rieser. 38–74. Berlin: de Gruyter.

Portner, P. 2007. Imperatives and Modals. Natural Language Semantics 15. 351–383.

van Rooij, R. & M. Franke. 2010. Promises and Threats with Conditionals and Disjunctions. Ms., University of Amsterdam.

Ross, A. 1944. Imperatives and logic. Philosophy of Science 11: 30–46. Russell, B. 2007. Imperatives in Conditional Conjunction. Natural Language

Semantics 15. 131–166. Schwager, M. 2006a. Interpreting Imperatives. Dissertation, UFrankfurt. Schwager, M. 2006b. Conditionalized Imperatives. In: C. Tancredi et al.

(eds.) Proceedings of SALT 16. CLC Publications. Schwager, M. 2008. Optimizing the Future. Course material at the 20th

ESSLLI summer school, Bordeaux. Searle, J. 1969. Speech Acts. Cambridge: Cambridge University Press. Searle, J. 1989. How Performatives Work. Linguistics and Philosophy 12.

535–558 Stalnaker, R. 2002. Common Ground. Linguistics and Philosophy 25. 701–

721 Truckenbrodt, H. 2009. Speech Acts and Agreement. Draft, HU Berlin

Sinn und Bedeutung 15 - [PDF Document] (225)

Sinn und Bedeutung 15 - [PDF Document] (226)

In Reich, Ingo et al. (eds.), Proceedings of Sinn & Bedeutung 15,pp. 225–239. Universaar – Saarland University Press: Saarbrücken, Germany, 2011.

Focus Influences the Presence of Conditional Perfection:Experimental Evidence

Marie-Christine FarrGoethe University Frankfurt


Abstract. This contribution tests whether focus has a determining influenceon the occurrence of Conditional Perfection. Two off-line experiments supportthis hypothesis. Conditional Perfection occurs significantly more often if the an-tecedent of the conditional is focused compared to the non-focused case. Ad-ditionally, in contrast to the scalar implicature associated with or (Zondervan2009), Conditional Perfection occurs only infrequent if the antecedent is not fo-cused. The second experiment suggests that this distinct behavior is due to dif-ferent properties of the scalar implicature and the implicature associated withConditional Perfection.

1 IntroductionConditional Perfection (CP) describes the phenomenon that speakers interpretconditional sentences, under certain conditions, as biconditionals. The phe-nomenon was given its name by Geis & Zwicky (1971), who observed thattheir students extended or “perfected” the meaning of conditionals. ConsiderGeis & Zwicky’s original example:

(1) a. If you mow the lawn, I’ll give you five dollars.b. If you don’t mow the lawn, I won’t give you five dollars.c. Only if you mow the lawn, I’ll give you five dollars.

As McCawley (1993) points out, “Only if p, q” can be paraphrased as “If notp, not q”. So examples (1b) and (1c) should be regarded equivalents. CP thusdescribed the inference from (1a) to (1b) or (1c). Geis & Zwicky note that per-fection of conditionals is clearly wrong from a logical viewpoint. The utteranceof a sentence like (1a) does not exclude the possibility that the hearer gets a re-ward for some other effort, for example, for cleaning the living room. Still theyaffirm that many speakers interpret conditionals in exactly this way and thatCP is “highly regular” (Geis & Zwicky 1971: 564). This paper addresses theconditions that influence the occurrence of CP.

Sinn und Bedeutung 15 - [PDF Document] (227)

226 Farr

2 Conditional Perfection Is a Pragmatic PhenomenonStandard approaches analyze conditionals within the framework of possibleworlds semantics, as in (2):

(2) “If p, q” is true in w if and only if:∀ w′ such that w′ ∈ C(w) & p(w′) : q(w′)with w being the actual world and C(w) being the set of possible worldsaccessible from w.

A conditional “If p, q” is considered to be true in w, if for all possible worldsthat are accessible from w and in which p is true q is also true. No informationis incorporated in the conditional meaning about what will be the case if pdoes not hold. So if we want to keep the semantics for conditionals unchanged,we must assume that CP is pragmatic. To affirm this claim, conditionals thatdo not allow perfection offer convincing support. If CP was contingent uponthe semantics it must arise with all conditionals. Biscuit Conditionals, as thestandard example taken from Horn (2000), belong to this group:

(3) a. If you’re thirsty, there’s some beer in the fridge.b. If you’re not thirsty, there’s no beer in the fridge.

The conditional in (3a) clearly does not invite the inferences in (3b). The rea-son for this is that the two propositions are conditionally unrelated. Furtherevidence in favor of a pragmatic analysis of CP provides the fact that the infer-ence is usually cancelable. That is, it can be “taken back” by adding additionalinformation. The following example illustrates this (van Canegem-Ardijns &van Belle 2008):

(4) If you mow the lawn, I’ll give you five dollars.But also if you paint the garage.

Through adding the additional condition under which five dollars will be paid,the CP inference that mowing the lawn is the only possible way to earn themoney is canceled.1 Another feature suggesting a pragmatic account of CP isexpressed by Boër & Lycan (1973). They support the idea that not all condi-tionals, in all situations are perfected and give the following counterexample:

(5) If John quits, he will be replaced.

1 As discussed in van Canegem-Ardijns & van Belle (2008: 371 ff.), this cancels just the “Onlyif p, q” inference. Canceling the “If not p, not q” inference is often not possible for speech actslike promises, threats or warnings. This might indicate that the two inferences “If not p, not q” and“Only if p, q” are indeed not equivalent. Still, we will stick to this assumption.

Sinn und Bedeutung 15 - [PDF Document] (228)

Focus Influences Conditional Perfection 227

Here, according to Boër & Lycan, CP does not arise, because common senseinhibits the implication that if John does not quit his job, he won’t be replaced.It is easily imaginable that John will be replaced, e.g. because his work lacksquality or because he does not get along with his colleagues. This constitutesthe main argument of Boër & Lycan. CP does not arise because of the specialform or intrinsic features of conditionals but simply because of additional in-formation like world or common-sense knowledge.All this provides convincing evidence that CP is a pragmatic rather than a se-mantic inference.

3 Conditional Perfection Is not a Scalar ImplicatureOriginally, Geis & Zwicky claimed that it is difficult to explain CP in terms ofimplicatures (cf. Grice 1989) and that the inference is clearly not a conversa-tional implicature. They argue that Grice, when characterizing conversationalimplicatures

looks for general principles governing the effects that utterances have, principlesassociated with the nature of the speech itself. CP is, in some sense, a principlegoverning the effects that utterances have (...) but it is in no way that we cansee derivable from considerations having to do with the nature of the speech act.(Geis & Zwicky 1971: 565)

Later in their squib, Geis & Zwicky relativize this claim and state that an ex-planation of CP in terms of conversational implicatures is not easily establish-able. Unfortunately they conclude without further elaboration or clarification.Despite this rejection, it is the most popular assumption today to explain CPwith respect to conversational implicatures. One particularly favored approachis to analyze CP in terms of Scalar Implicatures (SI). The general idea of a SIis that an inference can be drawn based on the amount of information that isexpressed. As Grice’s first maxim of Quantity advises, speakers should maketheir contribution as informative as required. Thus, very simplified, if we knowthat more informative statements than the one actually made exist (and somefurther assumptions hold (cf. e.g. Geurts 2010)) we can argue that the speakerdoes not believe that the more informative statements are true. For this reason,Horn Scales are created to order terms according to their information content.In the following, three scales will be introduced that were suggested to accountfor CP.

Sinn und Bedeutung 15 - [PDF Document] (229)

228 Farr

3.1 Attempt 1: Atlas & Levinson (1981)Atlas & Levinson (1981) discuss the following scale:

i f and only i f p, q↑ i f p, q

There are two problems with this scale: First and most important, this scaleis inadequate to explain CP. As mentioned above, by uttering a weaker state-ment the speaker expresses that he does not believe that the stronger statementholds. Thus an effect opposing CP arises. Uttering a conditional would alwaysimply that p is only one among more conditions which will bring about q (Horn2000). Atlas & Levinson assumed that this scale does not elicit SIs, hence theydid not see this problem. They argued that the scale is not well-formed becausethe two terms are not lexicalized to the same degree. This is why the unwantedSI does not arise. However, this restriction does not hold, as will be discussedbelow. The second deficiency of the above scale is pointed out by van der Auw-era (1997,b). The element at the top of the scale is just the literal meaning plusthe SI which is expected. It is clear that this combination will always be moreinformative than the literal meaning on its own. Since a construction of thistype is excluded with respect to other scalar terms like <some, some but notall>, it should also be excluded for CP. A related weakness is the complexity ofthe statement at the top of the scale. A restriction which is often proposed forpotential alternatives is that they must not be considerably longer or more com-plex than the statement made. In these cases, the speaker could just choose theshorter statement in order to be brief. However, this restriction does indeed nothold. As Matsumoto (1995) points out, more informative statements need to beasserted if they contain relevant information and even if they are of a greatercomplexity. Therefore, the above scale does elicit the unwanted SI. However,the scale can be ruled out due to other reasons such as the above-mentioneddeficiencies and additional constraints on monotonicity.2

3.2 Attempt 2: van der Auwera (1997b)Van der Auwera (1997; 1997b) assumes the following scale for his approach:

...i f p, q and i f r, q and i f s, qi f p, q and i f r, q

↑ i f p, q

If someone utters “If p, q” the comparably stronger statements such as, e.g., “Ifp, q and If r, q” are automatically denied and hence the speaker expresses that p

2 Horn scales must not include items of different monotonicity behavior. (cf. e.g. Matsumoto 1995).

Sinn und Bedeutung 15 - [PDF Document] (230)

Focus Influences Conditional Perfection 229

is the only condition that will bring about q. The weakness of van der Auwera’sscale is that he has to assume an infinitely large expression at the very top ofthe scale that incorporates all possible antecedents. Only if the statement “If p,q” negates for all other antecedents r that “If r, q” is true, CP arises. The prob-lem of this account is the reference to particular antecedents. Someone whohears a conditional like (1a) would have to reason for all possible antecedentsthat they will not bring about q. This however, does not seem appropriate torepresent the reasoning involved in CP. It is implausible and probably impossi-ble that someone who hears a conditional has an infinitely long list of possibleantecedents in mind (Horn 2000; von Fintel 2001).

3.3 Attempt 3: Horn (2000)Horn (2000) suggests yet another approach. He believes that the CP effect isdue to pragmatic strengthening and suggests the following scale:

q /Whatever the case, q↑ i f p, q

Unfortunately, a proposal like Horn’s is also not sufficient to derive CP, asvon Fintel (2001) notes. Uttering “If p, q”, and thereby negating the simplestatement q/whatever the case, q, does not mean that p is the only neces-sary and sufficient condition under which q will occur. It only elicits the muchweaker implicature that q is not unconditionally true. Whether only one or sev-eral conditions exist that render q true cannot be decided at that point. ThusHorn’s account is missing an important step towards CP. Still, following Horn,CP is derivable under his account (Horn 2000, 2004). Horn performs a reduc-tion of Grice’s maxims which is motivated by the desire to be in accordancewith the idea of a dualistic functional model which guides conversation. Thismodel assumes that utterances are subject to two forces. All of Grice’s maximsand submaxims, except for the maxim of Quality which remains unchanged,can be reduced to express just these two forces. Horn ends up with the follow-ing two principles (Horn 1993):

• Q Principle: Make your contribution sufficient. Say as much as you can,given quality and the R Principle.

• R Principle: Make your contribution necessary. Say no more than youmust, given the Q Principle.

The Q Principle embodies the first maxim of Quantity and the first two sub-maxims of Manner. It corresponds to a lower bound on information content.Within the R Principle the maxim of Relation, the second maxim of Quantityand the third and fourth submaxims of Manner are collected. This principle

Sinn und Bedeutung 15 - [PDF Document] (231)

230 Farr

constitutes an upper limit on the form of the utterance.Under these prerequisites uttering “If p, q” implies that the speaker does notbelieve that q is unconditionally true. Given the R Principle, this contributionmust be more relevant to the conversation than uttering q alone. This is why CParises. As Horn puts it: “what could make such a condition [i.e. “If p”] morerelevant than its necessity?” (Horn 2000: 310). Thus, CP occurs naturally andregularly due to systematic strengthening which is evoked by the R Principle.A serious problem of Horn’s account is that it would never be relevant to claimthat p is just a sufficient condition. This is however, wrong. With mention-somequestions it is typically sufficient to provide only the most relevant out of sev-eral conditions (cf. von Fintel 2001).The attempts to explain CP as a SI have not been satisfactory. The next sectiondiscusses the possibility to analyze CP still as a Quantity Implicature (QI) butnot as a SI.

4 Conditional Perfection as Quantity ImplicatureThe basic idea, following von Fintel (2001), is that CP is a QI3 which arisesas a by-product of an exhaustive interpretation. The following examples, takenfrom von Fintel (2001) and Groenendijk & Stokhof (1984) illustrate exhaustiveinterpretations.

(6) a. Q: Who left the party early?A: Robin and Hilary left the party early.

b. Robin and Hilary but no one else left the party early.

(7) a. Q: Who walks?A: John and Mary walk.

b. John and Mary but no one else walk.

If the answers in (6a) and (7a) are interpreted exhaustively, they correspondsto (6b) and (7b). These inferences are thought to be QIs. However, they can-not be derived by the “standard procedure” introduced for SIs, i.e. by negatingstronger statements. The problem is the set of relevant alternatives. In orderto infer “John and Mary but no one else walk” from “John and Mary walk”an infinitely large set of more informative statements needed to be rejected, asillustrated below:

• John, Mary and Peter walk.

3 Although QIs entail SIs, I use QI in the following to refer to all kinds of QIs except for SIs.

Sinn und Bedeutung 15 - [PDF Document] (232)

Focus Influences Conditional Perfection 231

• John, Mary, Ann and Peter walk.• . . .

This is again very implausible.Van Rooij & Schulz (2004) and Schulz & van Rooij (2006) provide an accountof exhaustive interpretation which overcomes these problems. Their theory ofexhaustification captures the intuition that exhaustive reasoning is based onthe closed world assumption. Details of their theory are not important for ourpurpose; what is important is that they explain exhaustification with the use ofminimal models. An exhaustive interpretation corresponds to a minimal model.Consider the following models for example (7a).

M1 M2

Individuals = Individuals ={Robin, Ben, John, Mary} {Robin, Ben, John, Mary}JwalkKM1 ={Robin} JwalkKM2 ={Ben}

M3 M4

Individuals = Individuals ={Robin, Ben, John, Mary} {Robin, Ben, John, Mary}JwalkKM3 ={John, Mary} JwalkKM4 ={John, Mary, Ben}

Table 1: The table presents four potential models for example (7a)

Exhaustification arises with a model that makes the answer true and in whichthe extension of the predicate in question is minimal. From the above givenexamples M3 is the model leading to an exhaustive interpretation. The otherseither make the answer false (M1 and M2) or the extension of walk is not min-imal (M4). An important feature of this account is that the focus-backgroundpartitioning determines the predicate in question that gets minimalized. There-fore, minimalization and hence exhaustification applies only to focused terms.In this regard Rooth (1996) makes an interesting observation with respect tofocus and the nature of question-answer pairs. When we consider question-answer pairs, the position of focus in the answer corresponds to the wh-constitu-ent in the question. Examples (8a) and (8b) illustrate this.4

4 Boldness equals focus in this and further examples.

Sinn und Bedeutung 15 - [PDF Document] (233)

232 Farr

(8) a. Q: Who walks?A: John and Mary walk.

b. Q: What do John and Mary do?A: John and Mary walk.

Due to the different focus-background partitioning, different models will beminimal. For (8a) a model in which only John and Mary walk is minimal, butfor (8b) a model where John and Mary do nothing else but walk is minimal.

4.1 Predictions for Conditional PerfectionBased on what we have seen so far, CP can be explained in terms of exhaus-tification which in turn can be explained by the selection of a minimal model.Since exhaustification is sensitive to focus, the same conditional can be in-terpreted exhaustively and non-exhaustively dependent on the question of thecontexts and the predicate it specifies. Consider the following examples:

(9) A: What happens if I sell an eel?B: If you sell an eel, you get 2.50 euros.

(10) A: When/Under which conditions do I get 2.50 euros?B: If you sell an eel, you get 2.50 euros.

Again, depending on the question different parts of the conditional are focused.In the first case, where the consequent is focused, a minimal model is one inwhich selling an eel results in nothing else than getting 2.50 euros. Hence, CPis not expected. With focus on the antecedent, as in dialogue (10), a minimalmodel is one where the only condition under which 2.50 euros are received isif an eel is sold. CP is expected to arises.

5 Experimental InvestigationsThe purpose of the experimental investigations was to test whether the predic-tions with respect to the influence of focus on CP are right. To investigate thissix minimal context pairs were created that differed only in the question asked.The question was either of the type what-if-p or when-q. The answer to eitherquestion was a conditional of the form “If p, q”. Thus, questions of the typewhat-if-p put focus on the consequent of the conditional. Questions of the formwhen-q on the contrary, put the antecedent in a focus position. An alternativeformulation for when-q would be under-which-conditions-q (cf. (10)). How-ever, it was argued in the literature that questions with the Dutch equivalentare necessarily understood to ask for an exhaustive answer (cf. van Canegem-Ardijns & van Belle 2008: footnote 12). If this was also the case for the German

Sinn und Bedeutung 15 - [PDF Document] (234)

Focus Influences Conditional Perfection 233

counterpart (the experiment was conducted in German) the questions would beuseless to investigate the influence of focus. Thus we choose the when-q formu-lation. It is also known that for when-sentences a difference between a condi-tional and a temporal reading exists. In this study, we examined the conditionalreading. To promote a conditional reading, we did not include terms referringto temporal aspects within the contexts of the test items. This was meant to en-sure a conditional reading and with that bring about the effect that the answer“If p, q” rather than “When p, q” was not surprising or unintuitive. The task ofthe participants was then to judge in either case, whether the answer providedby the conditional was sufficient. Hence, felicity judgements rather than truthvalue judgements were collected. It is expected that a felicity judgement task(FJT) is more adequate to investigate whether implicatures were calculated (cf.Papafragou & Musolino 2003). The test items always specify two conditionsthat lead to a particular consequent (cf. 2). Thus participants are expected torate the conditional answer as insufficient when CP occurred. If focus influ-ences CP we expect that subjects rate the conditional answer more often asinsufficient in the when-q condition than in the what-if-p condition.

Monika sells seafood on the market. She gets 1 euro for a crab, 2.50 euros foran eel, 15 euros for a lobster and 2.50 euros for a pike.Kerstin, an employee of Monika, cannot remember the prices. Since she doesnot want to ask Monika again, she asks Sahra, who also works for Monika.Sahra knows the prices exactly.

what-if-p when-qKerstin: What happens if I sell an eel? Kerstin: When do I get 2.50 euros?

Sahra: If you sell an eel, you get 2,50 euros.Did Sahra answer Kerstin’s question sufficiently? [Yes] [No]

Table 2: The table presents a sample item with both what-if-p and when-q ques-tions.

For elicitation of the data a repeated-measures design was used with the ques-tion type as independent variable. Thus subjects were confronted with bothtypes of questions, but never for the same context. The investigations wereconducted in paper-and-pencil form and the questionnaire consisted of six testitems and six fillers. Three out of the six test items contained what-if-p ques-tions and the other three when-q questions. Each questionnaire contained threefillers that tested whether participants were in general able to understand con-ditionals as answers to questions. These conditional fillers were very similar

Sinn und Bedeutung 15 - [PDF Document] (235)

234 Farr

to the test items, but they specified only one antecedent for each consequent.Within a true conditional filler, the conditional answer corresponds to whathas been described in the context. False conditional fillers provide the wrongantecedent for a consequence. Furthermore the questionnaires contained threefillers which tested whether subjects were sensitive for exhaustification. Likethe test items, these fillers specified two conditions which will lead to the sameconsequent. Contrary to the test items, the answer in the dialogue was not givenin conditional form, but was a simple statement starting with only. Answers oftrue exhaustification fillers provided both conditions that were specified by thecontext. The false exhaustification filler gave only one of these. Thus, if par-ticipants are in fact sensitive to exhaustification they should rate the responsesufficient in the true condition and insufficient in the false condition. The par-titioning was balanced over all six fillers, so that in total three were of the truecondition and three of the false condition. Two versions of the questionnairewere used and items and fillers were presented in a pseudo-random order. Nomore than two test items were presented in a row. The same was ensured forthe fillers. The second version of the questionnaire contained in each case theother condition of items and fillers and in the inverse order as in version one.

5.1 ProcedureThe experiment was conducted in German. Participants received a paper copyof the questionnaire with written instructions on the first page. Before the ques-tionnaires were handed out, participants also received oral instructions thatsummarized the written ones. Within the questionnaire, three items or fillerswere printed on one page. At the end of the questionnaire participants wereasked to provide some information about their background knowledge of logicand pragmatics. Additionally, space for comments was provided. The partici-pants took about 15 minutes to fill out the questionnaire.

5.2 ParticipantsParticipants were 50 students from an introductory linguistics class of the Uni-versity of Frankfurt. Two subjects were excluded, prior to analysis, becausethey did not provide correct answers to three or more fillers. Hence, the data of48 participants was evaluated.

5.3 ResultsNegative answers, signaling CP, occurred in 89.2 % of the when-q contextsand in 16.3 % of the what-if-p contexts. An analysis of variance showed thatover items as well as over subjects there was a main effect of question type(when-q vs. what-if-p) on the occurrence of CP, F(1,5) = 145.93, p < 0.001and F(1,47)= 309.93, p < 0.001 respectively. Most participants also showed

Sinn und Bedeutung 15 - [PDF Document] (236)

Focus Influences Conditional Perfection 235

a consistent behaviour over all test items. 61 % of the subjects labeled theanswer as insufficient only with when-q and never with what-if-p questions.33 % labeled the answers more often as insufficient in when-q contexts thanin what-if-p contexts. Only 3 participants said that the answers are equallyfrequent insufficient in both question conditions. Moreover, nobody’s answerssignaled that CP occurred more often or even exclusively in what-if-p contexts.

5.4 DiscussionThe results clearly indicate that focus has an effect on the occurrence of CP.This effect was highly significant over subjects as well as over items. We foundthat the percentage of no-answers, signaling CP, was high in when-q contextsand comparably low in what-if-p contexts. These results indicate that von Fin-tel’s theory, together with Schulz & van Rooij’s account on exhaustive inter-pretation, is adequate to account for CP. Focus seems to be the decisive factorfor the occurrence of this inference.

However, the results are important in another respect. Zondervan (2009)investigated the effect of focus on the SI associated with or (the inference from“A or B” to “A or B but not both”). He also found more SIs when or was ina focus-position but the effect, though significant, was much smaller than ex-pected. Zondervan found that the SI occurred in 77 % of the focused cases,versus 51 % of the non-focused cases. In contrast, the present study reflectsa partitioning closer to the expected one. The main problem of Zondervan’sresults was the high amount of implicatures in the non-focus condition. Thiscould be due to the fact that a different paradigm was used. As described, thepresent study used a FJT while Zondervan used a truth-value judgement task(TVJT) in his study. Within the TVJT, participants had to label the target sen-tences true or false. The underlying assumption is that a sentence like “A or B”is considered false in the case that the subjects calculated the SI (“A or B but notboth”) and indeed A and B holds. However, this appears to be rather a strongclaim. The statement is clearly inappropriate if the SI is calculated but does thatlead to falsehood? To judge this could be a problematic task for participants, sothat the responses might not reflect natural understanding. Judging whether ananswer is suitable or sufficient might be more natural. Additionally, labelingan utterance as false does not automatically mean that an implicature arose.As Zondervan (2006) discusses, a false answer can be interpreted in two ways:Either the subject calculated the implicature and thus labels the target sentencefalse in a situation where both A and B were the case. Or alternatively, thesubject did not calculate an implicature but noticed that in the situation whereboth A and B are the case and would be more suitable than or and thus labelsthe target sentence false. So in addition to the problem that labeling sentences

Sinn und Bedeutung 15 - [PDF Document] (237)

236 Farr

true or false might be unnatural for participants, interpreting the results is dif-ficult. Furthermore, the different results for the non-focus condition in this andZondervan’s study could be due to different properties of or and if. It could bethe case that for if the occurrence of QIs is more dependent on focus, whileor also elicits the inference regularly when not located in a focus position (forwhatever reason). To investigate which of these two factors was decisive for thedifferent results in Zondervan’s and the present study a follow-up experimentwas conducted.

6 Follow-up ExperimentThe follow-up experiment replicated the first experiment with the exceptionthat a TVJT was used. If this experiment yields the same results as the firstexperiment we can conclude that if and or behave differently when located infocus. If however, Zondervan’s results are replicated we can conclude that thedifferent paradigm was responsible for the response pattern.

6.1 MaterialTest and control items were the same as in the previous experiment, with theonly difference that the question “Did X answer Y’s question sufficiently?” waschanged to “Is the answer of X true?” This was done for test as well as controlitems.

6.2 ProcedureThe procedure, the instructions, oral and written, and the design were identicalto the first experiment.

6.3 Participants36 students from an introductory linguistics class of the University of Frankfurtparticipated in the experiment. Two participants were excluded prior to anal-ysis, because they stated having substantial knowledge of implicature theory.Additionally, one participant was excluded because she did not provide correctanswers to three or more fillers. Thus, the data of 33 participants was evaluated.

6.4 ResultsWithin when-q context negative answers, signaling CP, occurred in 56.6 % ofthe cases and they occurred in 11.1 % of the what-if-p contexts. The effect wasagain significant over subjects, F(1,32) = 58.175, p < 0.001, as well as overitems, F(1,5) = 50.845, p < 0.001. 52 % of the participants labeled the answeronly in when-q contexts as insufficient. One half of the remaining participantslabeled more answers insufficient in when-q contexts. The other half said thatthe answers are equally often insufficient in when-q and what-if-p contexts.

Sinn und Bedeutung 15 - [PDF Document] (238)

Focus Influences Conditional Perfection 237

Nobody’s answers signaled that CP occurred more often or even exclusively inwhat-if-p contexts.

6.5 DiscussionThe results indicate once more that focus influences the emergence of CP andfurthermore that this is independent of the paradigm used. The amount of CPwas again significantly higher in the when-q condition than in the what-if-pcondition. This holds over subjects as well as over items. The percentage ofno-answers, indicating CP, is overall lower when a TVJT is used comparedto the FJT. This may either indicate that felicity judgements are indeed moreadequate to detect implicatures or that the FJT overestimates the amount ofimplicatures. However, what is interesting with respect to Zondervan is thatusing a TVJT also lowers the percentage of CP in the non-focus condition.This signals that the high percentage of implicatures which Zondervan foundin the non-focus condition was due to properties of or rather than due to theTVJT.

7 ConclusionThis contribution provides evidence that von Fintel’s (2001) theory togetherwith the account of Schulz & van Rooij makes right predictions for the occur-rence of CP. The experiments showed that focus influences whether an answeris interpreted exhaustively, i.e. whether CP arise. The amount of CP was sig-nificantly higher when the antecendent was in a focus position. We also foundthat the high amount of implicatures which Zondervan found in his study onor in non-focus conditions was probably not due to using a TVJT rather thana FJT. For the case of if the amount of CP was even lower in the non-focuscondition when a TVJT was used.

As mentioned in the literature, not all implicatures behave the same (cf.Papafragou & Musolino (2003) on numerals and other scalar terms, Geurts& Pouscoulous (2009) on different embeddings and Chemla (2009) on scalarterms and free choice inferences). Geurts (2010: 122 ff.) makes an interest-ing observation with respect to potential differences of SIs and QIs that seemsrelevant for the different behavior of or and if. Even though Geurts proposesone account to derive both inferences he clearly distinguishes between the twotypes of implicatures. While we have a well-defined, closed set of potential al-ternatives for SIs such as, e.g., all, most and many for the scalar term some, thisis not the case for QIs as those involved in exhaustive interpretation. For an an-swer like “Robin and Hilary left the party early” (example (6a)) no such clearlydefined set of alternatives exists. We rather have to deal with an open-ended,possibly infinitely large set of alternatives. This core difference seems to play

Sinn und Bedeutung 15 - [PDF Document] (239)

238 Farr

an important role for the different experimental results found by Zondervan andmy study. Zondervan investigated the SI associated with or, so a well-definedset of alternatives was present, namely and. Since CP as analyzed here is not aSI but a QI, an open set of alternatives exists which cannot be clearly defined.This difference appears to be crucial for the reasoning of the participants. AsZondervan had argued, labeling a target sentence false might not mean thatthe SI was calculated but only that the stronger lexical item would be moreadequate. So maybe the amount of SIs is overestimated in the non-focus con-dition and possibly also in the focus condition. It is reasonable that with a finiteand reasonably small set of alternatives, participants think about whether oneof the alternative statements might have been more suitable. This could lead tothe high amount of false ratings in the non-focus conditions. Within the presentexperiments on CP there are no concrete alternatives which could detract theparticipants and lead to a large amount of no answers in the non-focus condi-tion. So answering no seems to be more clearly traceable to the emergence ofthe QI.If this difference was indeed crucial for the different results for or and if, wecould conclude that for the case of or the SI is also not elicited regularly inthe non-focus condition. Rather the clear awareness of the better alternativestatement and interferes and leads the subjects to conclude that the sentenceis false. Furthermore, this predicts that items that are assumed to produce SIs(e.g., some) reproduce Zondervan’s results, whereas items associated with QIs(e.g., the exhaustive interpretation in (6a)) should reproduce the results of thecurrent study.

ReferencesAtlas, Jay David & Stephen C. Levinson. 1981. If-clefts, informativeness, and

logical form. In Peter Cole (ed.), Radical pragmatics, 1–61. AcademicPress.

van der Auwera, Johan. 1997. Pragmatics in the last quarter century: The caseof conditional perfection. Journal of Pragmatics 27. 261–274.

van der Auwera, Johan. 1997b. Conditional perfection. In Angeliki Athanasi-adou & Rene Dirven (eds.), On conditionals again, 169–190. John Ben-jamins Publishing Company.

Boër, Steven E. & William G. Lycan. 1973. Invited inferences and other un-welcome guests. Papers in Linguistics 6. 483–505.

van Canegem-Ardijns, Ingrid & William van Belle. 2008. Conditionals andtypes of conditional perfection. Journal of Pragmatics 40. 349–376.

Chemla, Emmanuel. 2009. Universal implicatures and free choice effects: Ex-

Sinn und Bedeutung 15 - [PDF Document] (240)

Focus Influences Conditional Perfection 239

perimental data. Semantics and Pragmatics 2. 1–33.von Fintel, Kai. 2001. Conditional strengthening.Geis, Michael L. & Arnold M. Zwicky. 1971. On invited inferences. Linguistic

Inquiry 2. 561–566.Geurts, Bart. 2010. Quantity implicatures. Cambridge University Press.Geurts, Bart & Nausicaa Pouscoulous. 2009. Embedded implicatures?!? Se-

mantics and Pragmatics 2. 1–34.Grice, Paul. 1989. Studies in the way of words. Harvard University Press.Groenendijk, Jeroen A.G. & Martin J.B. Stokhof. 1984. Studies on the seman-

tics of questions and the pragmatics of answers: University of Amster-dam dissertation.

Horn, Laurence R. 1993. Economy and redundancy in a dualistic model ofnatural language. In Susanna Shore & Vilkuna Maria (eds.), Sky 1993:Yearbook of the linguistic association of Finland. 33–73.

Horn, Laurence R. 2000. From if to iff: Conditional perfection as pragmaticstrengthening. Journal of Pragmatics 32. 289–326.

Horn, Laurence R. 2004. Implicature. In Laurence R. Horn & Gregory Ward(eds.), The handbook of pragmatics, Blackwell.

Matsumoto, Yo. 1995. The conversational condition on horn scales. Linguisticsand Philosophy 18. 21–60.

McCawley, James D. 1993. Everything that linguists have always wanted toknow about logic ... but were ashamed to ask. University of ChicagoPress 2nd edn.

Papafragou, Anna & Julien Musolino. 2003. Scalar implicatures: Experimentsat the semantics-pragmatics interface. Cognition 86. 253–282.

van Rooij, Robert & Katrin Schulz. 2004. Exhaustive interpretation of complexsentences. Journal of Logic, Language and Information 13. 491–519.

Rooth, Mats. 1996. Focus. mats/.Schulz, Katrin & Robert van Rooij. 2006. Pragmatic meaning and non-

monotonic reasoning: The case of exhaustive interpretation. Linguisticsand Philosophy 29. 205–250.

Zondervan, Arjen. 2006. The question under discussion focus condition forscalar implicatures. Universiteit Utrecht MA thesis.

Zondervan, Arjen. 2009. Experiments on QUD and focus as a contextual con-straint on scalar implicature calculation. In Ulrich Sauerland & KazukoYatsushiro (eds.), Semantics and pragmatics: From experiment to the-ory, Palgrave Macmillan.

Sinn und Bedeutung 15 - [PDF Document] (241)

Sinn und Bedeutung 15 - [PDF Document] (242)

In Reich, Ingo et al. (eds.), Proceedings of Sinn & Bedeutung 15,pp. 241–257. Universaar – Saarland University Press: Saarbrücken, Germany, 2011.

Stative Passives and Event Kinds∗

Berit GehrkeUniversitat Pompeu Fabra


Abstract. Motivated by particular restrictions on event-related modificationwith German stative passives, this paper proposes that stative passives instantiatea consequent state kind of an event kind. The participle in such constructionshas to be derived from a verb whose event structure contains a consequent state,represented by an event-semantically interpreted BECOME component. Event-related modifiers with BE-passives modify either the event kind argument or thestate itself, and are therefore semantically licensed.

1 Introduction

German morphologically distinguishes between so-called eventive (or verbal)and stative (or adjectival) passives (Kratzer 1994, 2000; Rapp 1996; Maienborn2007a: among others). In particular, a past passive participle combines withwerden ‘become’ in eventive passives (1a) and with sein ‘be’ in stative passives(1b) (examples after Kratzer 2000).

(1) a. Diethe




‘The tires are (being) inflated.’b. Die




‘The tires are inflated.’

The semantics of sentences like (1b) is the main topic of this paper, and through-out, I will employ the descriptive labels BECOME- and BE-passives to distin-guish between these two constructions in German.

According to the traditional view, going back to at least Wasow (1977),stative passives are copula-adjective constructions, eventive passives periphras-tic verb forms. Nevertheless, an underlying event is still accessible in BE-∗ This paper benefitted greatly from comments at various stages and locations by many people,in particular Sascha Alexejenko, Boban Arsenijevic, Helga Gese, Claudia Maienborn, Louise Mc-Nally, and Carla Umbach. All errors are mine. This work was supported by the grant HUM2007-60599 from the Spanish Ministry of Education and Science and by an award from the FundacióICREA to Louise McNally.

Sinn und Bedeutung 15 - [PDF Document] (243)

242 Gehrke

passives, given the (albeit restricted) availability of event-related modifiers (2).

(2) a. Derthe







b. Dasthe





Kratzer (1994, 2000) and Rapp (1996) therefore propose that it is possible toadjectivise not just verbs but also verb phrases (VPs). Schlücker (2005) andMaienborn (2007a), in contrast, argue that such modifiers are merely pragmat-ically licensed. In this paper, I will argue, contra the latter, that the particularmodifiers available with BE-passives are semantically licensed.1

The paper is organised as follows. In section 2, I will outline the generalsemantics associated with BE-passives, its input requirements, as well as themore restricted availability of event-related modification with this construction.Section 3 proposes a semantic account of BE-passives based on the differencebetween event kinds and tokens. Section 4 addresses the availability of one typeof event-related modifier, by-phrases, and shows how the facts are accountedfor by the proposal. Finally, section 5 concludes.

2 German BE-Passives

In the literature on German, the view prevails that the participle in BE-passivesis adjectival (Kratzer 1994, 2000; Rapp 1996; von Stechow 1998; Maienborn2007a), and that it expresses the result or outcome of an event. If BE-passivesare copula-adjective constructions, their semantics has to be the one commonlyassumed for such constructions: A stative property is ascribed to an individual.

However, BE-passives co-exist with ‘true’ copula-adjective constructionsthat employ primary adjectives (3) (examples from Maienborn 2009).

(3) a. Diethe





b. Diethe






This suggests that there has to be some difference between the two, and it isnatural to assume that the difference is to be found in the nature of the under-lying verb in BE-passives. The following sections discuss the contribution of

1 For reasons of space, this paper will leave aside issues concerning the syntax-semantics inter-face, including whether or not phrasal adjectivisation of VPs exists. It can also not provide a deepcomparison to existing approaches, but see Gehrke (to appear).

Sinn und Bedeutung 15 - [PDF Document] (244)

Stative Passives and Event Kinds 243

the verb in BE-passives, their input requirements and semantic peculiarities, aswell as restrictions on the availability of event-related modification.

2.1 The Role of the Underlying Verb in BE-Passive Constructions

The stative property, as a rule, is ascribed to the internal (theme) argument ofthe underlying verb. The external argument, on the other hand, is completelyabsent, as illustrated by the unavailability of purpose clauses and depictives(4).2

(4) a. ??? Derthe




umin order


Fahrtjourney continue

b. ??? Dasthe








A first attempt at stating the input requirements for a BE-passive is based onthe assumption that the stative property has to be recovered from the eventstructure licensed by the underlying verb. This means that only verbs whichlicense an event structure with a stative component should be able to deriveBE-passives. This is basically the hypothesis defended in Rapp (1996).

Indeed, the data show that BE-passives are fully acceptable with transitiveverbs that have a lexically specified consequent state (in the sense of Moens &Steedman 1988) ((1b), (5)), i.e. with accomplishments and achievements.

(5) a. Diethe






b. Derthe




c. Diethe




With other verbs, BE-passives are acceptable only in certain contexts ((6)-(8),b. examples from Maienborn 2009) (see also Kratzer 2000). With activities((6), (7)), this is to be expected: The event structure does not contain a state.

(6) a. #Diethe




2 This contrasts with BECOME-passives, where the external argument is syntactically active, evenwhen it remains implicit (see also Gehrke & Grillo 2009: and literature cited therein).

Sinn und Bedeutung 15 - [PDF Document] (245)

244 Gehrke

b. AnnaAnna


















‘Anna has done her neighbourly duties: the mailbox is emptied,the flowers are watered and the cat is petted.’

The BE-passive of a semelfactive verb3 like streicheln ‘stroke’ in (6a) is ratherbad out of context. However, if we create a context under which someonepromised his or her neighbour to take care of things while the neighbour ison holiday, and one of the chores is to stroke the cat, the construction becomesacceptable (6b).

Similarly, the performative verb zitieren ‘cite’ out of context, as in (7a), isnot a good candidate for the construction because it does not license an eventstructure with a stative component. In the right context, in this case adding theby-phrase by Chomsky (7b)4, the sentence becomes acceptable again.

(7) a. #Dasthe




b. Dasthe






What is not expected if all we needed were a stative component to license theconstruction, is the fact that there are also restrictions on deriving BE-passivesfrom some stative predicates ((8), though see (10), below).

(8) a. #Diethe




b. Istis






The BE-passive of wissen ‘know’ out of context is rather bad but gets betterwhen embedded under the question in (8b). It is important to note, however,

3 The terms activity, accomplishment, and achievement are used in the sense of Rothstein (2004).It could be debated whether semelfactives (or performatives, as in (7b)) are activities, but there isgeneral agreement that semelfactives and performatives do not lexically specify a consequent state.4 A German PP headed by von ‘of, from’ in these contexts is commonly translated into English witha by-phrase. However, since it is generally claimed for English that by-phrases are not possible withstative passives, it is not fully clear whether (a) this claim is simply wrong (exceptions for Englishexist; German data are discussed in more detail in section 4); or (b) whether German von-phrasesare not fully equivalent to English by-phrases. Given the facts in (4) (which extend to combinationswith acceptable by-phrases), we have to assume that these by-phrases are still different from theby-phrases with eventive passives, which introduce true external arguments.

Sinn und Bedeutung 15 - [PDF Document] (246)

Stative Passives and Event Kinds 245

that many speakers do not even accept (7b) and (8b), whereas (6b) is acceptedby everyone. A more restrictive hypothesis, then, is the one in (9).

(9) Only verbs that lexically specify a consequent state derive BE-passives.

It follows from (9) that accomplishment and achievement verbs, whose internalargument undergoes a change of state and as a result is the bearer of a conse-quent state, are the perfect candidates for this construction. In addition, stateverbs that allow an inchoative (re-?)interpretation (see also Gehrke & Grillo2009) are also fine, as evidenced by the acceptable BE-passives of the psychpredicates in (10), in contrast to (8a).

(10) MarieMarie







This hypothesis also makes sense of the fact that the subject is always thetheme argument of the underlying verb: Changes of state, as a rule, affect themearguments.

2.2 BE-Passives Involve Event Kinds

We saw in (2) and (7b) that the underlying event can be modified by event-related modifiers, such as instrumentals, manner modifiers, and by-phrases.However, it has often been observed that such modification is only possibleif it pertains to the consequent state; cf. the contrast between (11) and (12)(examples after Rapp 1996).

(11) Derthe

Mülleimerrubbish bin


(* vonby





) geleert.emptied

(12) a. Dasthe





b. Erhe






The modifiers in (11) are out because they refer to an event participant or themanner of the event that (could have) brought about the particular state de-scribed by the sentence without having an impact on or being ‘visible’ duringthe consequent state. In contrast, the event participants described by the by-phrases in (12) clearly belong to the state described (see section 4).

Similarly, the modifiers in (13) have an impact on the underlying eventthat is still visible during the consequent state.

Sinn und Bedeutung 15 - [PDF Document] (247)

246 Gehrke

(13) a. Diethe







b. Derthe







c. Dasthe






In section 3, I will propose that BE-passives only allow modifiers that eithermodify an event kind (as in (13)) or that modify the state directly (as in (12)).

Another important restriction on the modification of BE-passives is thatthe underlying event cannot be temporally or spatially modified.5 For example,a modifier like recently can only modify the state (14).

(14) Diethe





‘The door was in the opened state recently, but probably is no longer.’(NOT: The door is in the opened state, the opening took place re-cently.)

BE-passives are also incompatible with temporal frame adverbials (15) (exam-ples from von Stechow 1998) (see also Rapp 1996, 1997).

(15) a. *Derthe







(‘The computer is repaired three days ago.’)b. Der







This has to do with the fact that BE-passives with a present tense copula arestatements about the present (in contrast to present perfect BECOME-passives,which - at least in German - are statements about the past6).

Furthermore, spatial modifiers that pick out the location of the event thatbrought about the consequent state are also generally bad (16).

(16) a. ??? Diethe







5 The incompatibility of spatial and temporal modifiers with (many or most) stative predicates isdiscussed extensively in Maienborn (2007b) and literature cited therein.6 See also Rapp (1996); Kratzer (2000); Maienborn (2007a) and literature cited therein for ar-guments against treating BE-passives as an ellipsis of an eventive passive perfect construction.(German werden ‘become’ forms the perfect with the auxiliary be.)

Sinn und Bedeutung 15 - [PDF Document] (248)

Stative Passives and Event Kinds 247

b. ??? Dasthe



imin the



From these data I conclude that the event associated with the underlying verblacks spatiotemporal location. To capture this fact, the account outlined in thenext section crucially builds on the assumption that BE-passives involve eventkinds, not event tokens.

3 The Proposal

Based on the discussion in the previous section, I propose that a BE-passiverefers to the instantiation of a consequent state kind of an event kind (17).

(17) a. Diethe




b. ∃ek,sk,s [BECOME(ek,sk) ∧ THEME(ek,door) ∧ closed(s)∧ THEME(s,door) ∧ R(s,sk)]

R is Carlson’s (1977) realisation relation whereas BECOME should be under-stood as an event semantic version of Dowty’s (1979) BECOME-operator, asso-ciated with accomplishment and achievement predicates; e.g. (18).

(18) Informal event semantics of BECOME (von Stechow 1996)[[BECOME]] (P)(e) = 1 iff e is the smallest event such that P is not trueof the prestate of e but P is true of the target state of e.

The use of BECOME is motivated by the hypothesis in (9).7

The idea that BE-passives involve event kinds, in turn, is motivated bythe restricted availability of event-related modifiers discussed in the previoussection.8 The unavailability of spatial and temporal modifiers shows that theevent in BE-passives has no spatiotemporal manifestation. Instrumental, man-ner modifiers and by-phrases, on the other hand, are only available if they canbe interpreted as event kind modifiers, or if they modify the state (token) di-rectly. Event kind modification will be discussed in this section, whereas state

7 Something like BECOME is also employed in Embick’s (2004) account of one type of stativepassive participles he identifies, namely the resultative one.8 Event kinds are natural to expect if we assume that events form a subsort in our ontology of(token) individuals (Davidson 1967), kinds form another subsort in that ontology (Carlson 1977),and as a rule, any token in the ontology should be the realisation of some kind in that ontology.Event kinds have an analog in e.g. the Situation Semantics notion of event type (Barwise & Perry1983), though the formal details are quite different. Under a Neo-Davidsonian view (e.g. Parsons1990), events can be decomposed into subevent, which motivates the additional assumption aboutthe existence of subevent kinds.

Sinn und Bedeutung 15 - [PDF Document] (249)

248 Gehrke

modifiers are addressed in section 4.Empirical arguments for event kinds as an ontological category have been

brought forward by e.g. Landman & Morzycki (2003); Ginzburg (2005); Gehr-ke & McNally (to appear). For example, Landman & Morzycki (2003) proposeto model manner modification in terms of kinds. Since their line of argumen-tation provides additional independent support for my analysis of BE-passives,the following subsection will briefly recapitulate the relevant points.

3.1 Modeling Manner in Terms of Kinds (Landman & Morzycki 2003)

Landman & Morzycki observe semantic and syntactic parallels with so-anapho-ra in the nominal and verbal domain across various languages. Their examplesfrom German are given in (19).

(19) a. soso





‘such a dog like this one’b. Er






‘He danced like Mary.’

In (19) there is a direct semantic parallel in the adnominal and adverbial usesof so, which refers back to a particular kind of entity (a kind of dog or a kind ofdancing event). There is furthermore a syntactic parallel in that both can occurwith an additional clause of comparison introduced by wie ‘like’.

Given that elements like so under the adnominal use (19a), in partic-ular English such, are commonly treated as kind anaphors, following Carl-son (1977), Landman & Morzycki (2003) treat adverbial so analogously, asanaphor to event kinds. In particular, they propose that (adverbial) so denotesa property of events that realise a (particular contextually supplied) kind (20).

(20) [[soi]] = λe.e realises ki

An additional argument that kinds are involved comes from the fact that tempo-ral and locative adverbials generally cannot antecede adverbial so (21), unlessthey can be seen as creating a new (or sub-)kind (22) (examples from Landman& Morzycki 2003).

(21) a. *MariaMary











Sinn und Bedeutung 15 - [PDF Document] (250)

Stative Passives and Event Kinds 249

b. *MariaMary











(22) MariaMary




Schlafsack,sleeping bag





‘Mary sleeps in a sleeping bag and John does so, too.’

For example, the locative modifier in (22) does not serve to specify the locationof a particular sleeping event, but rather serves to create a new sub-kind ofsleeping event, namely the kind of sleeping in sleeping bags. Hence, it is not aproper spatial modifier but rather used as a kind of manner modifier.

Given facts like these, Landmann & Morzycki suggest to treat mannermodifiers as event kind modifiers in general.

3.2 Relevance for this Paper

Returning to the topic of this paper, it is striking to see that the same kindof modifiers that are acceptable antecedents for so are also possible with BE-passives, namely and foremost manner modifiers, which modify an event kind.Spatial and temporal modifiers, on the other hand, modify an event token andare neither good antecedents of so nor acceptable with BE-passives, as observedin the previous section.

If manner modification is taken to be kind modification, one type of re-striction on event-related modification with BE-passives is straightforwardlyaccounted for under the current proposal (abstracting away from state mod-ifiers for the time being). Since BE-passives involve event kinds, only kind-related event modification is possible, including modifiers that serve to createa new or a subkind. It should also be clear, then, that the particular modifiersare semantically and not just pragmatically licensed, contra Maienborn (2007a,2009).

To illustrate how the kind-based approach captures the restrictions onevent-related modification with BE-passives, let us come back to the contrastbetween (11) and (13) (I will return to the stative examples in (12) in section4). The example in (11) without the modifier describes the state the rubbish binis in as a result of an emptying event kind. Combining event-related modifierswith the BE-passive should only be allowed either if these modify the (conse-quent) state (token) or if they create a new subkind, by narrowing down theevent kind of emptying rubbish bins. The particular modifiers in (11), however,do not do either, since they do not relate to the consequent state itself and sincethere are also no common or established subkinds of rubbish-bin-emptying by

Sinn und Bedeutung 15 - [PDF Document] (251)

250 Gehrke

my niece or slowly.9 The modifiers in (13), in turn, pick out particular subkindsof the events in question, namely childish drawings, pencil-writings or sloppycombings.

In the following section, I will take a closer look at restrictions on theavailability of by-phrases and show that some by-phrases modify a state whereasothers should be treated as event kind modifiers.

4 Different Types of by-Phrases with BE-PassivesSchlücker (2005) observes that there are two types of by-phrases that can com-bine with BE-passives. She argues that one type constitutes VP-adjuncts (23),which do not form a prosodic unit with the participle.

(23) a. weilbecause







b. weilbecause







Neutral stress with these phrases is on the participle, secondary stress on themodifier (the latter point is not noted in Schlücker, but see Hoekstra 1999;Gehrke 2008 for similar facts from Dutch).

The second type of by-phrases is argued to be V-adjuncts, which form aprosodic unit with the participle, with neutral stress on the modifier (24), (25).

(24) a. weilbecause






sindare NEUTRAL

b. weilbecause








(25) a. weilbecause





SANGesmusemuse of singing




b. weilbecause





Sangesmusemuse of singing




9 The question remains how to determine whether a subkind is common or established. Ultimately,a pragmatic account should answer such questions, and this is where the current proposal meetspragmatic accounts like Schlücker (2005) and Maienborn (2007a).

Sinn und Bedeutung 15 - [PDF Document] (252)

Stative Passives and Event Kinds 251

Schlücker observes that the intonation facts with her ‘V-adjuncts’ match thosewith other event-related modifiers that are allowed with BE-passives (26), (27).

(26) a. weilbecause




ROTweinred wine


sindare NEUTRAL

b. weilbecause




Rotweinred wine



(27) a. weilbecause







b. weilbecause







She concludes that these latter event-related modifiers, as well as by-phraseswhich behave like V-adjuncts, are pragmatically licensed in line with the ac-count proposed by Maienborn, by forming a complex ad hoc property with thestative property denoted by the verbal participle.10

In addition, Schlücker notes that the two types of by-phrases further dif-fer with respect to the nature of their complements. With her VP-adjuncts, theby-phrase is stated to denote the agent or direct cause of the underlying event.Animate entities are commonly expressed by proper names or members of agroup denoted by a collective noun, e.g. Polizist ‘police-man’. Inanimate enti-ties are referred to by definite uses of mass nouns, e.g. vom Feuer ‘by the fire’,or by appellatives used definitely, e.g. von der Bombe ‘by the bomb’.

With her V-adjuncts, on the other hand, a by-phrase is argued to denotethe theme of the underlying event or an indirect cause. It is supposed to havean instrumental character and to provide information about the manner or rea-son of the event. Animate entities are referred to by collective nouns, e.g. vonder Polizei ‘by the police’, inanimate entities by generic uses of mass nouns(von Feuer ‘by fire’) or indefinite uses of appellatives (von einer Bombe, vonBomben ‘by a bomb, by bombs’).

In the following, I will make some qualifications with respect to the datadiscussed in Schlücker and relate the facts to the current proposal.

4.1 Some Qualifications

A first observation is that the V-adjuncts discussed by Schlücker are parts offixed expressions and idioms. For example, there is no literal ‘verbal’ mean-ing in (25), in the sense that the muse of singing actually kisses or kissed the

10 This is basically what I called a common or established subkind above. Her account of the syntaxof ‘VP-adjuncts’ remains unclear, given that she rejects the possibility of phrasal adjectivisation.

Sinn und Bedeutung 15 - [PDF Document] (253)

252 Gehrke

daughters.11 A second type of by-phrases behaving like V-adjuncts is foundin examples already discussed in previous sections, such as (7b) and (13a),repeated in (28) with the additional neutral stress pattern identifying them asV-adjuncts.

(28) a. Dasthe






b. Diethe







The ‘VP-adjuncts’ discussed by Schlücker, on the other hand, are fully accept-able only with stative predicates. An example from previous sections, whichdisplays this intonation pattern, is the stative one in (12b), repeated in (29)with the relevant intonation pattern.

(29) Erhe






For the other alleged ‘VP-adjuncts’, i.e. those that do not combine with stativepredicates, I do not find them very good and I do not share the judgments aboutthe intonation identifying them as VP-adjuncts. If acceptable at all, they ratherbehave like V-adjuncts, e.g. there is no secondary stress ((30), Schlücker’s ex-amples, my judgments about stress).12

(30) a. ?? weilbecause








b. ?? weilbecause









11 Similarly, the combination ‘von Feuer geschwärzt’ in (24) appears rather fixed. A preliminarygoogle-search revealed very few instances of ‘geschwärzt’ in combination with a by-phrase. Thesewere limited to von Feuer, von Rauch ‘by smoke’, and von Ruß ‘by carbon black’.12 Other native speakers agreed with my judgments. Further syntactic tests to distinguish betweenV- and VP-adjuncts, mentioned by Schlücker (2005), such as the relative placement (with respectto modifier and participle) of sentence negation, sentence adverbials and floating quantifiers, yieldthe same results.

Sinn und Bedeutung 15 - [PDF Document] (254)

Stative Passives and Event Kinds 253

I conclude from these facts, then, that by-phrases that behave like VP-adjunctsare fully acceptable only with states.13 Given observations in previous sections,this makes sense, since only with these predicates, the entity denoted by the by-phrase modifies the state (which is interpreted as an inchoative state).

4.2 Rapp (1996)

A similar modification restriction to stative predicates with BE-passives is al-ready observed in Rapp (1996). Rapp notes that by-phrases that relate to theaction or the process itself (as opposed to a stative component) are generallyincompatible with un-prefixation (31), while this combination is more accept-able with stative verbs (32) (examples due to Lenz 1993).

(31) a. Diethe






b. Derthe






(32) a. Diethe







unbeeindruckt.unimpressed‘The citizens of Dresden are not concerned with such problems.’

b. ... weilbecause







‘... because they are not satisfied by their work’

She concludes that the by-phrases with these verbs do not relate to an activity oraction but express arguments of the adjective (i.e. of the state): The constructionexpresses the attitude of an experiencer with respect to his stimulus.

She furthermore observes that there are word order differences betweennon-action-related by-phrases (33) and other event-related modifiers in the BE-passive (34).

(33) a. Diethe








13 The marginal acceptability of such by-phrases with BE-passives could be explained along thelines of Welke (2007), who assumes that there are a few instances where a BE-passive constructionhas to be interpreted as an elliptical BECOME-passive perfect construction.

Sinn und Bedeutung 15 - [PDF Document] (255)

254 Gehrke

b. ... weilbecause







(34) *... weilbecause












Only the former can be extraposed, whereas the latter have to remain within theVP (or the verbal cluster). From these facts she concludes that the modifiers in(33) modify the adjective, while those in (34) modify a VP.14

4.3 Taking Stock: The Licensing of by-PhrasesTo take stock, we have different kinds of by-phrases with BE-passives that arelicensed semantically (and possibly syntactically) in different ways. First, wehave by-phrases that behave like V-adjuncts, diagnosed by intonation and wordorder possibilities. Such by-phrases are only possible with idioms or whenthey serve to create a new (sub-)kind (e.g. (13a)). Furthermore, they behavelike other event-related modifiers of BE-passives with respect to intonation andword order, as the data discussed by Rapp and Schlücker show.

In contrast, there are by-phrases that behave like VP-adjuncts with respectto intonation and word order possibilities, and thus contrast with other event-related modifiers with this construction. Such by-phrases are fully acceptableonly with stative predicates, in which case they modify a state token. It is possi-ble that such phrases are really to be treated as arguments of the AP (along thelines of Rapp), rather than arguments of the underlying VP. By-phrases withstative predicates commonly do not refer to agents, i.e. they are not true ex-ternal arguments. For example, with psych predicates, they rather refer to thestimulus of the state expressed.

Schlücker’s observation, then, that the complements of V-adjunct by-phraseshave more of a generic character fits these conclusions and the overall pro-posal. If the by-phrase modifies an event kind rather than an event token, thepotential agent of such a kind naturally has a more generic character. With theVP-adjuncts, on the other hand, we have by-phrases modifying an actual statetoken, so they are prone to be less generic.15

14 Rapp takes the latter facts as an argument in favour of phrasal adjectivisation (along the lines ofKratzer 1994).15 Recall that Schlücker notes that VP-adjunct by-phrases can also refer to agents or direct causers.I assume that these must be the by-phrases in the non-stative examples, which are not very good tobegin with.

Sinn und Bedeutung 15 - [PDF Document] (256)

Stative Passives and Event Kinds 255

5 Conclusion

In this paper, I proposed an account of German stative passives (BE-passives),which employed the concept of kinds in the domain of eventualities. On thebasis of the restricted availability of event-related modifiers, it was argued thatBE-passives instantiate a consequent state kind of an event kind. The input re-quirements for this construction therefore involve a participle derived from averb whose event structure contains a consequent state, which was representedby an event-semantically interpreted BECOME component. Event-related mod-ifiers with BE-passives, in turn, were argued to be semantically licensed, sincethey modify either the event kind argument or the state itself.

It was briefly noted that in some cases (for some speakers), BE-passivescan be derived from verbs which do not license an event structure with a con-sequent state component, and it was suggested that these cases have to becontextually (pragmatically) licensed. The precise mechanisms were not dis-cussed, however (see Maienborn 2007a, 2009; Gese 2010: for issues concern-ing the pragmatics of BE-passives); we could also assume that such cases in-volve coercion of the underlying event type. Given that the event kind is notspatiotemporally located, it might also be possible to interpret the scale under-lying BECOME in a non-temporal way. This could explain different readingsascribed to BE-passives that have been discussed in the literature under differ-ent labels, such as consequent state vs. characterisation readings (Brandt 1982;Rapp 1996), resultant state vs. target state readings (Kratzer 2000) or temporalvs. qualitative readings (Maienborn 2009; Gese 2010).16 This remains to beworked out in future research.


Barwise, Jon & John Perry. 1983. Situations and Attitudes. Cambridge, MA:MIT Press.

Brandt, Margareta. 1982. Das Zustandspassiv aus kontrastiver Sicht. Deutschals Fremdsprache 19. 28–34.

Carlson, Gregory Norman. 1977. Reference to Kinds in English: University ofMassachusetts at Amherst dissertation.

Davidson, Donald. 1967. The logical form of action sentences. In NicholasResher (ed.), The Logic of Decision and Action, 81–95. Pittsburgh: Uni-versity of Pittsburgh Press.

Dowty, David. 1979. Word Meaning and Montague Grammar: The Semantics

16 A somewhat different division is found in Embick (2004), who differentiates between stativesand resulatives in English (see also Gehrke to appear).

Sinn und Bedeutung 15 - [PDF Document] (257)

256 Gehrke

of Verbs and Times in Generative Semantics and in Montague’s PTQ.Dordrecht: Reidel.

Embick, David. 2004. On the structure of resultative participles in English.Linguistic Inquiry 35.3. 355–392.

Gehrke, Berit. 2008. Ps in Motion: On the Syntax and Semantics of P Elementsand Motion Events: Utrecht University dissertation. LOT DissertationSeries 184.

Gehrke, Berit. to appear. Passive states. In Violeta Demonte & Louise McNally(eds.), Telicity, Change, and State: A Cross-Categorial View of EventStructure, Oxford University Press: Oxford.

Gehrke, Berit & Nino Grillo. 2009. How to become passive. In KleanthesGrohmann (ed.), Explorations of Phase Theory: Features, Arguments,and Interpretation at the Interfaces Interface Explorations 17, 231–268.Berlin: de Gruyter.

Gehrke, Berit & Louise McNally. to appear. Frequency adjectives and asser-tions about event types. In Proceedings of SALT 19.

Gese, Helga. 2010. Implizite Ereignisse beim Zustandspassiv. Paper presentedat the workshop ‘Zugänglichkeit impliziter Ereignisse’, University ofTübingen, July 2010.

Ginzburg, Jonathan. 2005. Situation Semantics: The ontological balance sheet.Research on Language and Computation 3.4. 363–389.

Hoekstra, Teun. 1999. Auxiliary selection in Dutch. Natural Language andLinguistic Theory 17. 67–84.

Kratzer, Angelika. 1994. The Event Argument and the Semantics of Voice.Ms. University of Massachusetts at Amherst.

Kratzer, Angelika. 2000. Building statives. Ms. University of Massachusettsat Amherst.

Landman, Meredith & Marcin Morzycki. 2003. Event-kinds and manner mod-ification. In Nancy Mae Antrim, Grant Goodall, Martha Schulte-Nafeh& Vida Samiian (eds.), Proceedings of the Western Conference in Lin-guistics (WECOL) 2002, California State University, Fresno.

Lenz, Barbara. 1993. Probleme der Kategorisierung deutscher Partizipien.Zeitschrift für Sprachwissenschaft 12. 39–76.

Maienborn, Claudia. 2007a. Das Zustandspassiv: Grammatische Einordnung- Bildungsbeschränkung - Interpretationsspielraum. Zeitschrift für ger-manistische Linguistik 35. 83–144.

Maienborn, Claudia. 2007b. On Davidsonian and Kimian states. In IleanaComorovski & Klaus von Heusinger (eds.), Existence: Semantics andSyntax, 107–130. Dordrecht: Springer.

Maienborn, Claudia. 2009. Building Ad Hoc properties: On the interpretation

Sinn und Bedeutung 15 - [PDF Document] (258)

Stative Passives and Event Kinds 257

of adjectival passives. In Arndt Riester & Torgrim Solstad (eds.), Pro-ceedings of SuB 13, 35–49. University of Stuttgart.

Moens, Marc & Mark Steedman. 1988. Temporal ontology and temporal ref-erence. Computational Linguistics 14.2. 15–28.

Parsons, Terence. 1990. Events in the Semantics of English: A Study in Sub-atomic Semantics Current Studies in Linguistics Series 19. Cambridge,MA: MIT Press.

Rapp, Irene. 1996. Zustand? Passiv? Überlegungen zum sogenannten “Zus-tandspassiv". Zeitschrift für Sprachwissenschaft 15.2. 231–265.

Rapp, Irene. 1997. Partizipien und semantische Strukur: Zu passivischen Kon-struktionen mit dem 3. Status. Tübingen: Stauffenburg.

Rothstein, Susan. 2004. Structuring Events: A Study in the Semantics of LexicalAspect. Oxford: Blackwell.

Schlücker, Barbara. 2005. Event-related modifiers in German adjectival pas-sives. In Emar Maier, Corien Bary & Janneke Huitink (eds.), Proceed-ings of SuB 9, 417–430. Radboud University Nijmegen.

von Stechow, Arnim. 1996. The different readings of wieder ‘again’: A struc-tural account. Journal of Semantics 13. 87–138.

von Stechow, Arnim. 1998. German participles II in Distributed Morphology.Ms. Unversity of Tübingen.

Wasow, Thomas. 1977. Transformations and the lexicon. In Peter Culicover,Thomas Wasow & Adrian Akmajian (eds.), Formal Syntax, New York:Academic Press.

Welke, Klaus. 2007. Das Zustandspassiv: Pragmatische Beschränkungen undRegelkonflikte. Zeitschrift für germanistische Linguistik 35. 115–145.

Sinn und Bedeutung 15 - [PDF Document] (259)

Sinn und Bedeutung 15 - [PDF Document] (260)

Events in Adjectival Passives*

Helga Gese University of Tübingen


Abstract. The paper presents a new analysis of the semantics of adjectival passives mainly concentrating on the status of the implicit event, its agent participant and the dependency relation between eventive base and derived property. On the basis of data from two questionnaire studies it proposes an adjectival ∅-affix, modifying Maienborn’s (2009) analysis in two respects: First, it does not involve existential quantification over an event particular but reference to an event kind. Second, the dependency relation between the event kind and the property denoted by the participle of an adjectival passive sentence is not a causal or temporal one. It is a relation of lexical supervenience in the sense of Kim (1990) which leaves room for the pragmatic character of adjectival passives, for their specific ‘surplus in meaning’.

1 Introduction Adjectival passives such as (1) are combinations of a form of the copula to be plus an adjectivized past participle. In German, copula and auxiliary differ in form: Adjectival passives use the copula sein (‘to be’) whereas verbal passives are built with a form of werden (‘to become’).

(1) Die Tür ist geschlossen. The door is[COP] closed ‘The door is closed.’

(2) Die Tür wird geschlossen. The door becomes[AUX] closed ‘The door is closed.’

* Work on this paper was supported by the German Science Foundation (DFG) as part of the project A1 “The semantics and pragmatics of combinatory meaning variation” within the collaborative research center SFB 833, Universität Tübingen, and by the Ministry of Science, Research and the Arts Baden-Württemberg and the European Social Fund (ESF) within the Schlieben-Lange-Programm. I would like to thank Claudia Maienborn, Sebastian Bücking and Berit Gehrke for helpful discussion and comments. I would especially like to thank Britta Stolterfoht for having shared with me her psycholinguistic expertise.

In Reich, Ingo et al. (eds.), Proceedings of Sinn & Bedeutung 15, pp. 259–273. Universaar – Saarland University Press: Saarbrücken, Germany, 2011.

Sinn und Bedeutung 15 - [PDF Document] (261)

260 Gese

Semantically, copula adjective sentences are property ascriptions to their subject referent (Maienborn 2005). What is special in adjectival passives, in contrast to copula sentences with genuine adjectives, is that the ascribed property is a complex property which has to be derived from the verbal base of the participle. Most semantic accounts assume that the property causally results from the occurrence of the event denoted by the adjectival passive’s verbal base (cf., e.g., Maienborn 2009, Kratzer 2000; for an exception see Gehrke 2010). There are three problems with this assumption. First, the possibility of schon immer (‘always’) modification, as in (3) and (4), and other non-event-based uses of adjectival passives, such as (5),1 cast some doubt on whether we really have to deal with a causal relation between an event particular and its resulting state.

(3) Die linke Bronchie war schon immer verengt. ‘The left bronchial tube had always been[COP] constricted.’

(4) Der Bildhauer meißelt aus dem Stein Figuren hervor, die nach seinem Verständnis dort schon immer verborgen waren. ‘The sculptor carves shapes out of blocks of granit that he believes have always been[COP] hidden inside of them.’

(5) Bei den Glattnasen sind die Lidspalten bei der Geburt noch geschlossen. ‘The eye-lid slits of the vesper bat are[COP] still closed at birth.’

Second, adjectival passives, contrary to verbal passives, do not regularly combine with agent modifiers (cf. (6) vs. (7)). The acceptability of agentive modification depends on whether or not it is relevant for the ascribed property (cf. Rapp 1997, Maienborn 2010, for more details on this). These restrictions cannot be easily accounted for if we assume that adjectival passives contain an event particular with agent participant as their base.

(6) Der Brief wurde von Gabi / von einem Experten geschrieben. The letter became[AUX] by Gabi / by an expert written ‘The letter was written by Gabi / by an expert.’

(7) Der Brief ist ???von Gabi / von einem Experten geschrieben. The letter is[COP] ???by Gabi / by an expert written ‘The letter was written by Gabi / by an expert.’

Third, there is some evidence that the relation between the eventive base and the derived property cannot be a classical causal one. Typical causal relations

1 The sentences (3) – (5) are adapted from real occurrences found in the world wide web.

Sinn und Bedeutung 15 - [PDF Document] (262)

Events in Adjectival Passives 261

(cf. Engelberg 2005, for more details on this) hold between events but not between states. Adjectival passives with stative base verbs such as (8) do not fit into this pattern. Moreover, causal relations normally correlate with tem-poral succession. This doesn’t hold for adjectival passives like (9) in which the relation is one of simultaneity.

(8) Das Haus ist von Studenten bewohnt. the house is[COP] by students occupied ‘The house is occupied by students.’

(9) Die Gefangenen sind streng bewacht. the prisoners are[COP] strongly guarded ‘The prisoners are closely guarded.’

Examples like (3) to (9), schon immer (‘always’) modification, restrictions on agent modification and properties inconsistent with classical causal relations pose three questions that will be our starting point in this paper: (i) whether the occurrence of the event is in fact a necessary part of the truth conditions of adjectival passives, (ii) what role the agent plays and (iii) how event and ascribed property can be related, if not by a CAUSE or RESULT predicate. After a short presentation of the formal account which constitutes the background of the investigations and the analysis carried out in this paper, section three presents two questionnaire studies with a truth value judgment task (TVJT) which investigate the status of the event and its agent participant in adjectival passives. The results of these studies point towards an analysis relying not on event particulars but on event kinds. The remaining part of the paper concentrates on formalizing the relation between this event kind and the derived property. Borrowing a dependency relation widely used in the philosophy of mind, section four argues for analyzing the link between eventive base and derived property as a case of lexical supervenience. Super-venience allows connecting the derived property to an event kind and captures the intuition that the derived property is ‘more’ than the result state of the eventive base. It leaves room for the specific role pragmatics plays in the formation of adjectival passive by capturing the rather subjective, pragmatic nature of the derived property.

2 Background Maienborn’s (2009) account of the formal semantics of adjectival passives sets the frame for the investigations and modifications presented in this paper. Modeling the semantics of an adjectival passive sentence as the ascription of

Sinn und Bedeutung 15 - [PDF Document] (263)

262 Gese

an underspecified property to its subject referent, Maienborn (2009: 44) proposes the following adjectival affix:

(10) Adjectival ∅-affix: λP λx λs ∃e [s: Q(x) & result (e,s) & P(e)]

In (10) the underspecification of the ascribed property is rendered by using the free variable Q for which pragmatics has to provide a suitable value. The only restriction on Q is that it results from the event denoted by the verbal base of an adjectival passive. There are mainly three reasons which speak in favour of an underspecification account and against an account which identifies the ascribed property with the result state contained in the event structure of an adjectival passive’s eventive base. First, the formation of adjectival passives is not restricted to verbs with lexically given result states (which should be the case if their semantics relied on result states), cf. (11).2

(11) Er war geschmeichelt, als der Personenkult um ihn schließlich groteske Formen annahm. ‚He was flattered when the cult of personality surrounding him finally veered into the grotesque.’ (Der Spiegel 40/1994, 10/3/1994)

Second, adjectival passives and genuine adjectives are not distributed in a complementary way, cf. (12) and (13). As „the output of a lexical rule may not be synonymous with an existing lexical item“ (Kiparski 1983: 15), blocking should occur if adjectival passives referred not to an underspecified property but to the lexically given result state. Third, some sentences show a clear meaning difference between derived property and lexically given result state. In (14) geöffnet (‘opened’) cannot be identified with its result state offen (‘open’). The value assigned to Q by the context is not ‘open’, it is the property of ‘not being in the original packaging state’.

(12) Die Tür ist geöffnet. ‘The door is opened.’

(13) Die Tür ist offen. ‘The door is open.’

(14) Das Spiel ist geöffnet, aber unbespielt und absolut neuwertig. ‘The game is opened but unplayed and in pristine condition.’ (Maienborn 2010: 9, my translation)

2 In general, these constructions with non-resultative base verbs need contextual support, cf. Gese, Stolterfoht & Maienborn (2009: 136), Kratzer (2000: 4), Rapp (1998: 243).

Sinn und Bedeutung 15 - [PDF Document] (264)

Events in Adjectival Passives 263

In Maienborn’s formulation of the ∅-affix involved in adjectival passive formation the only constraint on the ascribed property Q is that its value result from the occurrence of the base verb’s event referent. In (10), this is rendered by existential quantification over an event particular. Yet, as already suggested by the non-event-based uses of adjectival passives referred to in the introduction, there are some reasons for doubting that this formalization is on the right track.

3 Experiments on the Status of the Event

Experiment 1: “Status of the Event“

To test whether the occurrence of the event denoted by the verbal base of an adjectival passive sentence is indeed a necessary ingredient of its truth conditions – as it is in the case of a verbal passive – a questionnaire study with a truth value judgment task (TVJT) was conducted (see, e.g. Crain and Thornton 1998, for more details on this method). The study tested how the judgments of verbal passives and adjectival passives can be affected by a context which excluded the occurrence of the events denoted by their base verbs. It compared adjectival passives, verbal passives and, in a control condition, copula sentences with genuine adjectives as utterances in two types of context: an eventive one and a purely stative one, in which any eventive component is excluded: Eventive context:

(15) Pünktlich um 17 Uhr leert der Postbote den Briefkasten. Anna, die das vom Fenster aus beobachtet, freut sich, denn sie ist pleite und hat morgens einen Brief an ihren reichen Onkel eingeworfen. Zufrieden sagt sie zu Erwin: „Der Briefkasten ist geleert / wurde geleert / ist leer.“ ‘At 5 p.m. right on schedule the mailman empties the mailbox. Anna, watching from her window, is glad to see this because she is broke and she had put in a letter addressed to her rich uncle that morning. She tells Erwin: “The mail box is[COP] emptied / is[AUX] emptied / is empty.”’

Stative context:

(16) Soeben ist der fabrikneue Briefkasten an der Eugenstr. aufgestellt wor-den. Der Bürgermeister darf feierlich den allerersten Brief einwerfen. Die Frage, ob er denn wirklich der erste sein wird, der einen Brief in diesen Kasten einwirft, bejaht der anwesende Postfilialleiter und sagt: „Der Briefkasten ist geleert / wurde geleert / ist leer.“

Sinn und Bedeutung 15 - [PDF Document] (265)

264 Gese

‘A brand new mailbox has just been installed in Eugenstr. The mayor is expected to post the very first letter as part of a small ceremony. The manager of the post office who is in attendance confirms when asked whether the mayor will in fact be the first one to put a letter in this mailbox: “The mail box is[COP] emptied / is[AUX] emptied / is empty.”’

Given the existence of adjectival passives with schon immer (‘always’) modification noted at the beginning of this paper the following hypotheses were formulated for Experiment 1:

(H1) Adjectival passives should be less dependent on the occurrence of the events denoted by their base verbs than verbal passives. Compared to verbal passives, they should receive more TRUE ratings in purely stative contexts.

(H2) (control condition) As the semantics of genuine adjectives does not con-tain any eventive component there should be no difference between the two sorts of contexts in the ADJ condition.

Method Fourty-two undergraduate students of Tübingen University parti-cipated for course credits or monetary reimbursem*nt. All participants were native speakers of German. Materials consisted of thirty-six experimental sentences in six versions and thirty-six filler sentences. All experimental items began with a context in which the occurrence of an event is described in condition EVENT or denied in condition NoEVENT and ended with the utterance of an adjectival or verbal passive or with a copula sentence with a genuine adjective. The base verbs of the adjectival and verbal passive utterances matched the event used in the EVENT-context. The copula adjective sentences contained genuine adjectives which corresponded to the result state of this event (e.g. EVENT-context: X leert Y (‘X empties Y’), target-utterance: Y ist geleert / wurde geleert / ist leer (‘Y is[COP] emptied / is[AUX] emptied / is empty’)). The filler items presen-ted different sorts of sentences in contexts: filler sentence plus context were either tautologous or contradictory, or the sentence was true but pragma-tically odd in the context. Six presentation lists were constructed in which the 36 experimental items were randomly mixed with the 36 filler items. The six lists were counterbalanced across items and conditions: Each participant saw only one version (AP / VP / ADJ) of each of the target utterance embedded in one type of context (EVENT or NoEVENT). The questionnaires were distributed in an introductory linguistics class. Participants had one week to complete the questionnaire. They were told to read the narratives carefully and to judge the

Sinn und Bedeutung 15 - [PDF Document] (266)

Events in Adjectival Passives 265

truth value of the sentence in the described situation (“Is the utterance in the described situation true?”; possible answers: yes / no). Results and Discussion The results of the questionnaire study are presented in Table 1.

context type sentence type

Event NoEvent

AP 84,1% 20,6%

VP 86,1%


Adj 83,3% 80,6%

Table 1. Percentage of TRUE responses to the TVJT

A repeated measures ANOVA revealed a significant main effect of context type and sentence type and, more importantly, a significant interaction of the two factors (F1(2,82) = 161.89, p1 ≤ .001; F2(2,70) = 155.18, p2 ≤ .001). As predicted by (H2), for copula sentences with genuine adjectives (adj) there was no significant difference between the event and the NoEVENT contexts (all Fs < 1). For the two other conditions, the judgments differed significantly depending on sentence type in the NoEVENT contexts (F1(1,41) = 17.25, p1 ≤ .001; F2(1,35) = 21.83, p2 ≤ .001) but not in the event contexts (all Fs < 1). As predicted by (H1) adjectival passives were judged true more often than verbal passives in NoEVENT contexts (F1 (1,41) = 31.74, p1 ≤ .001; F2 (1,35) = 20.90, p2 ≤ .001). Even though, descriptively, the percentage of true judg-ments for adjectival passives in purely stative contexts is not very high, the difference between adjectival and verbal passives is highly significant. Moreover, participants’ judgments of the filler items in contradictory context (which were judged true only in 8,7% of the cases) were clearly different from the judgments for adjectival passives in the NoEVENT condition but not from verbal passives. Even if event-occurrence is strongly preferred in adjectival passives, this difference from contradictory sentences calls the truth-conditional relevance of event-occurrence into question.

Experiment 2: “Status of the Agent“

The experiment reported in this section focuses on the agent participant. A first hint to the status of the agent in adjectival passives comes from the restrictions on agent modification alluded to in the introduction. As agent modification serves to make explicit the implicit agent argument, the question is whether the restrictions on agent modification point to the absence of implicit agents in the semantics of adjectival passives. In order to test this assumption another TVJT experiment was conducted. Participants judged conditional sentences such as (17) to (19) which contained in their consequent either an adjectival passive or a copula sentence

Sinn und Bedeutung 15 - [PDF Document] (267)

266 Gese

with a genuine adjective. The antecedent of the experimental sentences referred either to the whole event denoted by the base verb of the adjectival passive (17) or the participation of an agent was excluded (18). In a third condition which parallelled the NoEVENT-condition of Experiment 1 the antecedent referred to a non-event-based state (19).

(17) Wenn Karla gerade alle Rollläden in ihrem Zimmer heruntergelassen hat, dann ist das Zimmer abgedunkelt / dunkel. ‘If Karla just lowered all the blinds of her room, the room is[COP] darkened / dark.’

(18) Wenn es in einem Zimmer keine Beleuchtung gibt und nach Sonnen-untergang kein Licht mehr von draußen reinscheint, dann ist das Zimmer abgedunkelt / dunkel. ‘If there is no light source in the room and, after sunset, no light falls through the window from outside, the room is[COP] darkened / dark.’

(19) Wenn ein Kellerzimmer schon immer weder Fenster noch Beleuchtung hat, dann ist das Zimmer abgedunkelt / dunkel. ‘If a basem*nt room never had a window or any kind of indoor lightning, the room is[COP] darkened / dark.’

The hypotheses tested in Experiment 2 were based on the restriction on agen-tive modification in adjectival passives and on the results of Experiment 1 which call into question the truth-conditional relevance of event-occurrence and thus the necessity of identifying event participants:

(H1) Adjectival passives should receive more TRUE judgments in non-agentive contexts than in non-eventive contexts.

(H2) (control condition) There should be no differences between the three sorts of contexts in the genuine adjective condition.

Method Thirty-six undergraduate students of Tübingen University partici-pated and received a monetary reimbursem*nt. All participants were native speakers of German and none of them participated in Experiment 1. Materials consisted of 36 experimental sentences in six versions and 36 filler sentences. Filler as well as experimental sentences were of the type Wenn X, dann Y (‘If X than Y’) where Y was in condition AP an adjectival passive sentence and in condition ADJ a copula sentence with a genuine adjective. In condition EVENT X referred to the whole event denoted by the base verb of the adjectival or verbal passive, in condition NoAG the participation of an agent was excluded and in context NoEVENT the whole event was missing.

Sinn und Bedeutung 15 - [PDF Document] (268)

Events in Adjectival Passives 267

The base verbs of all adjectival passives were causative accomplishments such as abdunkeln (‘to darken’), zähmen (‘to tame’), kürzen (‘to shorten’), verschönern (‘to embellish’) or räumen (‘to evacuate’). In the control condition (ADJ), the adjective denoted the result state of the corresponding adjectival passive’s base verb (e.g. abdunkeln – dunkel (‘to darken’ – ‘dark’). Filler sentences were either tautologous or contradictory, or they were true but required, in their consequent part, the cancellation of an implicature which contradicted the antecedent of the sentence (e.g. If all students passed the exam some students passed it.). Design and Procedure Six presentation lists were constructed in which the 36 test items were randomly mixed with the 36 fillers. The six lists were counterbalanced across items and conditions: Each list included only one version of each experimental sentence. Sentences were presented in a self-paced fashion on a PC using E-Prime software (Psychology Software Tools, Inc.). After each sentence the participants were asked to judge its truth by answering the question “Stimmt das?” (‘Is it right?’) by yes or no. Data Analysis and Results Reading times and truth-value judgments were analyzed. Reading times for the adjectival passive sentences were significantly higher in the two non-standard context conditions NoAG and NoEVENT (EVENT 1125 ms., NoAG 1648 ms., NoEVENT 1595) whereas there were no significant differences in the ADJ condition. Due to space limitations and lack of theoretical relevance, I will not report the reading times in detail here. The results of the TVJT are presented in Table 2.

context type sentence type

Event NoAG


AP 92,1% 59,7% 31,0%

Adj 88,9% 87,0% 86,6%

Table 2. Percentage of TRUE responses to the TVJT

For the TVJT, a repeated measures ANOVA revealed a significant main effect of context type and of sentence type. More importantly, the interaction of context type and sentence type was highly significant (F1(2,70) = 47.68, p1 ≤ .001; F2(2,70) = 49.900, p2 ≤ .001). Whereas, as predicted by (H2), there were no significant differences in the genuine adjective condition (all Fs < 1), in the adjectival passive condition the judgments differed significantly depending on context (F1(2,70) = 75.47, p1 ≤ .001; F2(2,70) = 81.25, p2 ≤ .001): Adjectival passive sentences received more TRUE judgments in the EVENT condition than in the two other conditions and, as predicted by (H1),

Sinn und Bedeutung 15 - [PDF Document] (269)

268 Gese

they received more TRUE judgments in condition NoAG than in condition NoEVENT (F1(1,35) = 31.35, p1 ≤ .001; F2(1,35) = 35.51, p2 ≤ .001). Discussion In Experiment 1, participants gave more TRUE judgments for adjectival passives in NoEVENT contexts than for verbal passives in the same contexts (20,6% vs. 5,2%). In both experiments, there were clearly more TRUE judgments for adjectival passives in the NoEVENT condition than for contradictory filler sentences (Experiment 1: 20,6% vs. 8,7%; Experiment 2: 31% vs. 5,2%). In Experiment 2, the percentage of TRUE judgments increased if not the whole event but only the agent component of the base verb’s event structure was excluded by context. Interestingly, adjectival passive sentences in such NoAG condition received even more TRUE judgments than the filler sentences which were true but contained a generalized implicature which contradicted the antecedent of the sentence (59,7% vs. 51,5%). There are two major conclusions which can be drawn from the two experiments presented above. First, adjectival passives in event-occurrence excluding contexts are no clear cases of contradiction. In the light of existing formal accounts of the semantics of adjectival passives, this result is some-what surprising: The adjectival affixes proposed in the literature (e.g., Maienborn 2009, Kratzer 2000) all involve existential quantification over an event particular, i.e. concrete instantiation of the event. If these analyses were correct, the occurrence of the base verb’s event would be a prerequisite for an adjectival passive sentence to be true. The results of Experiment 1 and 2 show that this is not the case. Second, the agent is less important for the interpretation of an adjectival passive sentence than expected for a regular event participant. The agent contained in an event particular is accessible via the event argument. It should thus be equally important for the interpretation of a sentence as the event argument itself. The results of Experiment 2 showed that the agent is less important for the interpretation of an adjectival passive sentence than the event itself. In view of these results, it seems plausible to assume that the semantics of an adjectival passive sentence does not contain an event particular and that concrete instantiation of the event and identification of its participants might just be pragmatic issues. This conclusion receives further support by the results of Experiment 2 where NoAG adjectival passives received descriptively even more TRUE judgments than true filler sentences which required the cancellation of a generalized implicature.

Sinn und Bedeutung 15 - [PDF Document] (270)

Events in Adjectival Passives 269

4 Formal Analysis and Conclusion We now turn to the consequences of these results for the formal semantics of adjectival passives. As already noted, event particulars necessarily have event participants and they are instantiated. On the other hand, reference to an event kind does not need identification of its participants as these are generically bound and it does not require the actual occurrence of the event. It is thus plausible to assume that the semantics of adjectival passives involves event kinds rather than event particulars. Let us briefly return to the problem of schon immer (‘always’) modification mentioned at the beginning of this paper. Under an event kind analysis, the property denoted by the participle in an adjectival passive sentence depends on a generic evidentiality which does not contain information either about a possible instantiation or about specific agent participants.3 This provides a simple explanation for the acceptability of schon immer: the modification specifies that the event kind was not instantiated.4 In the remainder of the paper I will propose a formal analysis of adjectival passives that follows Maienborn’s (2009) underspecification account but modifies it by adding reference to event kinds and by choosing the appropriate link between event kind and derived property. In Maienborn’s adjectival affix, repeated here in (20), the underspecified property Q an adjectival passive sentence ascribes to its subject referent is causally linked to an existentially bound event particular.

(20) Adjectival ∅-affix: λP λx λs ∃e [s: Q(x) & result (e,s) & P(e)] (Maienborn 2009: 44)

To account for the reference to event kinds, a first idea might be to simply replace the event particular P(e) in (20) by the respective event kind ↑P5. There are two reasons why such a solution would be too simplistic. First, it is technically impossible for a property particular to directly result from an 3 In sentences such as Kratzer’s (2000) The blood vessel was obstructed the fact that the event kind does not identify event participants leads to the possibility of referring either to an agentive or to a stative obstructing kind. 4 Under this view, the question why adjectival passives are in fact often interpreted as referring to a concrete event instance has to be answered pragmatically. The explanation amounts to saying that event kinds pragmatically implicate their concrete instantiation under certain conditions (e.g. in post state contexts). The exact spellout of this pragmatically implicated instantiation will be the matter of another paper. 5 This is Link’s (1995: 376) notation for an “up-arrow” operation converting predicates into kind-denoting terms; “for instance if TIGER is a one-place predicate denoting the set of tigers ↑TIGER is a singular term that denotes the kind Tiger.” For a similar operator see, e.g., Chierchia 1998.

Sinn und Bedeutung 15 - [PDF Document] (271)

270 Gese

event kind.6 Second, as already noted in the introduction, typical properties of causal relations such as temporal precedence of cause and effect are not shared by all adjectival passives. In fact, one of the most important properties of causal relations is their denseness, i.e. their tendency to form causal chains. Typical causal sentences, such as (21a) can be, potentially infinitively, expanded to more fine grained causal chains like (21b).

(21) a. Smoking causes an increase of blood pressure. b. Smoking causes an increase of adrenalin level, this causes an

increase of blood pressure.

Adjectival passives do not function this way. In (22) the relation between the event kind Mowing the grass in (22a) and the derived property in (22b) cannot be paraphrased by a causal sentence such as (22c) as this would imply the possibility of expansion to a more fine grained causal chain. The deviance of (22d) shows that such an expansion is impossible: There simply is no intermediate cause X which could be inserted.

(22) a. Mowing the grass b. The grass is mown. c. Mowing the grass causes the grass to be mown. d. ???Mowing the grass causes X, this causes the grass to be mown.

This lack of denseness in adjectival passive sentences leads to an explanatory gap which should not be present in causal relations: Causes fully determine their effects and effects are fully predictable on the basis of their causes. In adjectival passives, however, the property denoted by the participle is not fully determined by its eventive base. This was already demonstrated by (14), in which the property denoted by the participle geöffnet (‘opened’) cannot be identified with the lexically given result state ‘open’. The same holds for (4), repeated here as (23), in which the property is not only the result state of hiding, which would be ‘to be out of sight’. Rather, it is something like ‘to be inherently present‘. Similarly, in (24a) eingereicht (‘submitted’) has an additional meaning component, namely ‘to be of high value’ (compared to a contextually salient alternative, e.g. an article which is still in preparation). In (24b) the same participle denotes the property of being of low value.

6 Gehrke (to appear) chooses to solve this problem by an analysis not only referring to event kinds but also to state kinds which are instantiated via a realization operation. The problem with this solution is that there are no independent, non-technical reasons for assuming state kinds in adjectival passives.

Sinn und Bedeutung 15 - [PDF Document] (272)

Events in Adjectival Passives 271

(23) Der Bildhauer meißelt aus dem Stein Figuren hervor, die nach seinem Verständnis dort schon immer verborgen waren. ‘The sculptor carves shapes out of blocks of granit that he believes have always been[COP] hidden inside of them.’

(24) a. Der Artikel ist nicht in Vorbereitung, er ist eingereicht. ‘The article is not in preparation, it is submitted.’ b. Der Artikel ist nicht akzeptiert, er ist eingereicht. ‘The article is not accepted, it is submitted.’

As (23) to (24) show, there is a surplus in the meaning of the participle of an adjectival passive which is not fully determined by its eventive base. In Maienborn’s adjectival affix (20) above, this underdetermination is rendered by using a free variable Q for the property denoted by the participle in an adjectival passive sentence. There is an explanatory gap between Q and its underlying base, the event kind ↑P. Q is ‘more’ than (parts of the event structure of) its eventive base, therefore it is unpredictable from ↑P alone. A similar explanatory gap is known in philosophy as the mind-body problem. There are several formulations of this problem which all rely on the intuition that the mental is somehow determined by the physical but that it cannot be reduced to it. On the one hand, physical events such as firing of C-fibers are more basic than mental states such as pain feeling in the sense that mental states depend on physical events but not vice versa. On the other hand mental properties are unpredictable or unknowable from information concerning their physical base-level phenomena (cf. Jackson’s (1982) ‘knowledge argu-ment’). This explanatory gap led philosophers such as Jaegwon Kim to reject a reductive explanation of mental properties by neurophysiological processes and to opt for a non-temporal, non-dense dependency relation which they called “supervenience” (cf., e.g., Kim 1990). The common core of all defini-tions of supervenience can be captured by the slogan ‘A supervenes on B if there is no A-difference without a B-difference (everything else being equal)’. Applied to the mind-body problem this means that a mental property supervenes on a set of neurophysiological processes in the sense that one cannot imagine differences in mental properties without neurophysiological differences (everything else being equal). The advantage of supervenience over other dependency relations (e.g. classical CAUSE) is that it only partially determines the supervenient property. It thereby leaves room for the subjective, non-reducible character of supervenient properties which cannot be deduced from their underlying base-level phenomena. This dependency leaving room for underdetemination, for an irreducible ‘surplus in meaning’, is precisely what we need for the semantics of

Sinn und Bedeutung 15 - [PDF Document] (273)

272 Gese

adjectival passives. Borrowing a term from Engelberg (2005), we will call supervenience in the domain of the lexicon Lexical Supervenience. An informal definition is given under (25):

(25) Lexical Supervenience For any state s and set of events e7, [[ LSV (s, e) ]] = 1 iff there can be no change in s without a change in e (everything else being equal).

In the case of adjectival passives the property denoted by the participle lexically supervenes on the event kind of the verbal base in the sense that if two adjectival passive sentences which are maximally similar, i.e. which share the same subject and the same context, refer to two different properties they also have to differ in their eventive base. On the other hand, lexical supervenience makes it possible to derive different properties from the same eventive base if the context differs too, cf. (24a) and (b). It thereby accounts for the subjective, pragmatic character of adjectival passives. Applying the LSV-relation to Maienborn’s adjectival affix (20) and replacing event particulars by event kinds finally yields our new version of the ∅-affix:

(26) Adjectival ∅-affix: λP λx λs [s: Q(x) & LSV (s, ↑P)]

According to (26) the adjectival affix introduces an underspecified property Q which is ascribed to the subject referent x of the sentence. The assignment of a value to Q must be done by pragmatics with the only restriction that it lexically supervenes on the event kind derived from the verbal base. This means that Q depends on its eventive base ↑P, but that there is an explanatory gap between the two.8 Lexical supervenience accounts for this specific gap in the meaning derivation of adjectival passives. It leaves room for the role pragmatics plays in the interpretation of adjectival passives and for their characteristic ‘surplus in meaning’.

7 Supervenience is usually defined as holding between a property and a set of properties. As kinds can be roughly characterized as the set of all their instances (cf. Chierchia (1998)), this makes the above definition particularly suitable for the dependency relation between event kind and ascribed property in adjectival passives. 8 Depending on the sort of base verb and the context of the sentence this gap may be smaller or bigger. If the meaning component which is supplied by the eventive base is informative enough in the given context pragmatics may choose to identify Q with it, but in other cases Q is a pragmatically derived complex property which contains a characteristic ‘surplus in meaning’ compared to its verbal base.

Sinn und Bedeutung 15 - [PDF Document] (274)

Events in Adjectival Passives 273

References Chierchia, Gennaro. 1998. Reference to kinds across languages. Natural

Language Semantics 6. 339–405. Crain, Stephen and Rosalind Thornton. 1998. Investigations in universal

grammar: a guide to experiments in the acquisition of syntax and semantics. Cambridge, Mass: MIT Press.

Engelberg, Stefan. 2005. Stativity, supervenience, and sentential subjects. In Claudia Maienborn & Angelika Wöllstein (eds.), Event Arguments: Foundations and Applications. 45–68. Tübingen: Niemeyer.

Gehrke, Berit. to appear. Passive states. To appear in Violeta Demonte & Louise McNally (eds.), Telicity, Change, and State: A Cross-Categorial View of Event Structure. Oxford: Oxford University Press.

Gese, Helga, Britta Stolterfoht & Claudia Maienborn. 2009. Context effects in the formation of adjectival resultatives. In Susanne Winkler & Sam Featherston (eds.), The Fruits of Empirical Linguistics, vol. 2, Product, 125–155. Berlin, New York: Mouton de Gruyter.

Jackson, Frank. 1982. Epiphenomenal Qualia. Philosophical Quarterly 32. 127–136.

Kim, Jaegwon. 1990. Supervenience as a philosophical concept. Metaphilosophy 21. 1–27.

Kiparsky, Paul. 1983. Word formation and the lexicon. In Proceedings of the 1982 Mid-America Linguistics Conference, 3–29. Lawrence, Kansas: University of Kansas.

Kratzer, Angelika. 2000. Building statives. In Berkeley Linguistic Society 26. 385–399.

Link, Godehard. 1995. Generic information and dependent generics. In Gregory Carlson & Francis Pelletier (eds.), The Generic Book. 358–382.

Chicago: University of Chicago Press. Maienborn, Claudia. 2005. On the limits of the Davidsonian approach: the

case of copula sentences. Theoretical Linguistics 31(3). 275–316. Maienborn, Claudia. 2009. Building event-based ad hoc properties: on the

interpretation of adjectival passives. In Arndt Riester & Torgrim Solstad (eds.), Proceedings of Sinn und Bedeutung 13, 35–49. Stuttgart: University of Stuttgart.

Rapp, Irene. 1997. Partizipien und semantische Struktur. Zu passivischen Konstruktionen mit dem 3. Status. Tübingen: Stauffenburg.

Sinn und Bedeutung 15 - [PDF Document] (275)

Sinn und Bedeutung 15 - [PDF Document] (276)

Facts and Ideals: On the Role of doch in Conditionals and Optatives*

Patrick Grosz

MIT [emailprotected]

Abstract. This paper addresses optative constructions, constructions that ex-press a wish or desire without a modal that means ‘wish’ or ‘want’. Focusing on German, I argue that “expressing a wish” is a possible use of a conditional antecedent that is freely available. The question is how certain elements, such as the contrastive particle doch (or the focus particle nur ‘only’), which are typical for optative constructions, interact with this wish. I argue that they interact with the wish indirectly; they have a meaning that is independent from optativity, but which can be used to bring out an already available wish reading. This is achieved by eliminating alternative readings. Discussing German doch as a case study, I show how this interaction can be made precise.

1 The Puzzle Optative constructions (Scholz 1991, Rosengren 1993, Rifkin 2000, Asarina & Shklovsky 2008, Biezma 2010, Gärtner 2010) express a wish or desire without containing a modal that means wish or want. This is illustrated in (1); (1a) conveys a wish that appears equivalent to the wish described in (1b).

(1) a. If only John had come to the party! b. I wish John had come to the party.

In many languages, optatives seem to have the shape of conditional antecedents that contain the particle only (Rifkin 2000). In some languages, other particles are prototypical markers of optative constructions; in German, (2a), the unstressed contrast particle doch (cf. Thurmair 1989) seems to support an optative reading; as shown in (2b), Dutch toch can do so as well.

* For helpful discussions, I thank Norbert Corver, Kai von Fintel, Hans-Martin Gärtner, Noah Goodman, Martin Hackl, Irene Heim, Sabine Iatridou, Angelika Kratzer, Pritty Patel-Grosz, Paul Portner, Josh Tenenbaum and Coppe van Urk; I would also like to thank the audience at MIT’s ling-lunch.

In Reich, Ingo et al. (eds.), Proceedings of Sinn & Bedeutung 15, pp. 275–289. Universaar – Saarland University Press: Saarbrücken, Germany, 2011.

Sinn und Bedeutung 15 - [PDF Document] (277)

276 Grosz

(2) a. Wenn Hans doch einmal auf Maria gehört hätte! German if Hans doch once to Maria listened had ‘If only Hans had listened to Maria once!’ b. Als Jan toch eens naar Marie had geluistered! Dutch if Jan toch once to Marie had listened ‘If only Jan had listened to Marie once!’

A question at the core of research such as Rifkin (2000) is what such particles contribute to an optative clause, specifically whether they compositionally contribute optativity. The present paper analyzes German doch in optatives as a case study, based on the semantics of doch in declaratives (Abraham 1991, Doherty 1987, Grosz in press, Karagjosova 2001, 2004, Lindner 1991, Ormelius-Sandblom 1997). In declaratives, doch roughly marks the modified proposition p as an established fact (in the sense of Kratzer & Matthewson 2009), which in the evaluation context is presupposed to contradict a salient alternative r, as shown in (3).

(3) Hans kocht oder putzt immer, aber [¬[p∧r] nie beides]. Also wissen Hans cooks or cleans always but never both thus know wir, dass [¬r Hans nicht gekocht hat], weil [p er doch geputzt hat]. we that Hans not cooked has because he doch cleaned has ‘Hans always cooks or cleans, but never both. Therefore, we know that Hans didn’t cook, because he [doch] cleaned.’

presuppositions triggered by “doch”: It is an established fact that [p = Hans cleaned], and there is a salient alternative proposition [r = Hans cooked], such that ¬[p∧r].

This paper addresses two questions. First, can we devise a uniform semantics of doch that covers both its optative use and its use in non-optative constructions? Second, how is doch linked to optativity? I answer the first question in the affirmative and present a generalized analysis of doch that covers its optative and its non-optative uses. The second question is answered as follows: The contribution of doch to optative constructions is indirect in the sense that it can block non-optative readings in out-of-the-blue contexts.

2 The Proposal 2.1 A Uniform Analysis of doch Before discussing optatives with the form of conditional antecedents, this section provides the general background for my analysis of doch in optative clauses. Consider first the difference in German between the root clause with

Sinn und Bedeutung 15 - [PDF Document] (278)

Facts and Ideals 277

verb-second movement in (4a) and the unembedded dass-clause in (4b), Truckenbrodt (2006). By virtue of convention, a verb-second clause typically has the force of an assertion, (4a). In contrast, an unembedded dass-clause in German can be exclamative or have the optative force of a command or wish, (4b). I treat such conventional forces as reflecting different uses of the expressed proposition (i.e. as different speech acts, cf. Levinson 1983).

(4) a. Ich hätte Rom noch einmal gesehen. I had Rome still once seen ‘I would have seen Rome once more.’ (assertion) b. Dass ich Rom noch einmal gesehen hätte! that I Rome still once seen had ‘I wish I had seen Rome once more!’ (wish / #assertion)

I will henceforth call utterances like (4a) declarative statements and cases like (4b) dass-optatives. Both utterance types allow for the presence of doch, (5). Again, the verb-second clause in (5a) is used as a (reinforcing) statement, whereas the dass-clause in (5b) expresses a wish. By virtue of doch, (5a) conveys that in a given set of circ*mstances, the speaker takes it to be granted that she would have seen Rome once more. The less transparent contribution of doch in (5b) is discussed further down. The core question is how to account for the presence of doch in both utterances in a uniform way.

(5) a. Ich hätte Rom doch noch einmal gesehen. I had Rome doch still once seen ‘(As we know,) I would have seen Rome once more.’ (statement) b. Dass ich Rom doch noch einmal gesehen hätte! that I Rome doch still once seen had ‘I wish I had [doch] seen Rome once more!’ (wish)

In order to posit a uniform analysis of doch, we need to relativize its meaning to the type of the utterance that it occurs in. To do so, I pursue the following strategy. First, I assume that there are at least two contextually given sets of propositions that are used to manage the discourse (which I will henceforth call context sets): The common ground is the set that contains propositions that are treated as mutual knowledge by the discourse participants (Stalnaker 1974, 1978). The ideal list of a discourse participant i is the set that contains propositions that reflect i’s ideals (subsuming i’s wishes, i’s goals, and laws that i abides to). The ideal list replaces Han’s (1998) Plan Set and Portner’s (2005) To-Do List (which are reminiscent of Lewis’ 1979 sphere of

Sinn und Bedeutung 15 - [PDF Document] (279)

278 Grosz

permissibility), which it is based on1. It serves to order possible worlds into better worlds and less optimal worlds and thus behaves like an ordering source (Kratzer 1981). Having assumed that discourse contexts involve both a common ground and different participants’ ideal lists, it is natural to assume that assertions (and statements in general), like (4a) and (5a), are speech acts that operate on the common ground, whereas wishes, like (4b) and (5b), are speech acts that operate on ideal lists. In other words, utterance types come with conventionalized instructions on where to assign the modified proposition.

We also need to assume that apart from speech acts that add to the common ground or to an ideal list (like assertions or commands), there must be speech acts that reactivate propositions from a context set, to make them salient in the discourse (e.g. as a premise for something else), cf. (6). An assertion can be rejected as inappropriate if the expressed proposition is shared knowledge, (6a); in contrast, a reactivating statement, marked by the particle ja cannot be rejected in this way, (6b). (Cf. Repp 2009, in the spirit of Krifka’s 2007 common ground management.)

(6) a. adding p to the common ground A: Im März 1968 war Thatcher noch nicht an der Macht. in March 1968 was Thatcher yet not in the power ‘In 1968, Thatcher wasn’t in power yet.’ B: Jaja, das weiß ich eh! ‘Duh, I know that!’ b. reactivating p from the common ground A: Im März 1968 war Thatcher ja noch nicht an der Macht. in March 1968 was Thatcher ja yet not in the power ‘As we all know, in 1968, Thatcher wasn’t in power yet.’ B: # Jaja, das weiß ich eh! ‘Duh, I know that!’

By virtue of the (unstressed variants of the) German particles doch and ja, an utterance can be marked as reactivating old information, rather than adding new information. This is illustrated in (7a) versus (7b). If the modified proposition is shared knowledge of the speaker and hearer, as in (7a), doch and ja are possible, and a declarative without such particles (the lack of which is symbolized by ‘Ø’) is pragmatically odd. In contrast, if the modified information is new information, as in (7b), unstressed doch and ja are odd.

1 The label (i’s) ideal list is chosen (as opposed to Plan Set or To-Do List) to reflect the fact that it can contain propositions that i has no control over, such as that it rains tomorrow.

Sinn und Bedeutung 15 - [PDF Document] (280)

Facts and Ideals 279

(7) a. Context: H is well aware that she’s been to Paris and S wants to make this shared fact salient in order to follow up on it.

Du warst ja / doch / #Ø schon in Paris. you were ja doch #Ø already in Paris ‘You’ve (ja / doch / #Ø) already been to Paris.’

b. Context: H is an amnesiac and believes that she has never been to Paris; S discovers an old flight ticket to Paris with H’s name on it.

Du warst #ja / #doch / Ø schon in Paris. you were #ja #doch Ø already in Paris ‘You’ve (#ja / #doch / Ø) already been to Paris.’

We can now give an analysis of doch, relativized to speech acts, cf. (8). The ‘familiarity’ component discussed above is captured by (8a). Furthermore, (8b) captures the fact that doch differs from ja in that it presupposes that there is a salient alternative proposition r, which contradicts the modified proposition p in the utterance context (see Grosz in press for a recent discussion). I use the term indicates instead of ‘presupposes’ or ‘implicates’, as it is not clear how these terms apply at the speech act level.

(8) Semantics of “doch” (simplified and generalized to speech acts) For any proposition p used in a speech act φ, a. doch p indicates that the speaker considers p to be established as part

of the context set targeted by φ. b. doch p indicates that there is a contextually salient proposition r, such that the common ground entails ¬[p∧r].

Having established a distinction between ideal list and common ground and a uniform analysis of doch, we can now provide an analysis for (5a+b), in (9), omitting the meaning component in (8b) for ease of exposition.

(9) a. Ich hätte Rom doch noch einmal gesehen. I had Rome doch still once seen i. speech act: retrieve [(in certain salient circ*mstances) the speaker would have seen Rome once more] from the common ground.

ii. doch ⇒ [(in such circ*mstances) the speaker would have seen Rome once more] is an established part of the common ground.

b. Dass ich Rom doch noch einmal gesehen hätte! that I Rome doch still once seen had i. speech act: retrieve [the speaker has seen Rome once more] from the speaker’s ideal list. ii. doch ⇒ [the speaker has seen Rome once more] is an established part of the speaker’s ideal list.

Sinn und Bedeutung 15 - [PDF Document] (281)

280 Grosz

Having shown the analysis at work, it is worth pointing out a further parallel between doch in optatives and doch in non-optative clauses. Grosz (in press) argues that doch interacts with focus, positing that the salient proposition r that conflicts with the modified proposition p (in (8b)) must be a focus alternative, illustrated in (10).

(10) Context: Georg and Peter see a blue Peugeot parked in front of the pub. Peter: Schau, der Hans ist da! ‘Look, Hans is here!’ Georg: Nein, der Hans hat doch einen [GRÜNEN]F Peugeot. no the Hans has doch a green Peugeot ‘No, (as we both know) Hans has [doch] a GREEN Peugeot.’

i. doch indicates that Georg considers [p Hans has a green Peugeot] to be an established part of the common ground (i.e. not under de-bate).

ii. doch indicates that there is a salient focus alternative r = [Hans has a blue Peugeot], such that ¬[pΛr] (given that Hans only has one car).

While I have omitted this feature from (8b) as it is not at the core of the present discussion, such interaction with focus can also be observed in the case of optatives, illustrated in (11). Here, focus indicates which aspect of reality the speaker would like to change.2 This can be taken as further evidence for a uniform contribution of doch.

(11) a. Dass doch [OTTO]F die Nachtschicht mit Anna geteilt hätte! that doch OTTO the night.shift with Anna shared had ‘If only it had been OTTO who shared the night shift with Anna!’ b. Dass Otto doch [die NACHTschicht]F mit Anna geteilt hätte! ‘If only it had been the NIGHT shift that Otto shared with Anna!’ c. Dass Otto die Nachtschicht doch [mit ANNA]F geteilt hätte! ‘If only it had been ANNA that Otto shared the night shift with!’

Having posited a uniform analysis of doch, section 2.2 considers conditional antecedents with doch. The remainder of this paper argues that a uniform approach to doch extends to optative and non-optative if-clauses and sheds light on why doch in an if-clause prefers an optative reading.

2.2 Doch in Conditional Antecedents This section discusses what doch adds to conditional antecedents, covering both non-optative and optative cases. (12a) is a baseline example of a non-counterfactual conditional clause, (12b) is a counterfactual conditional clause.

2 Replacing dass ‘that’ by wenn ‘if’ in (11) does not change the judgments, as counterfactual dass-optatives and counterfactual wenn-optatives are roughly equivalent (cf. Scholz 1991).

Sinn und Bedeutung 15 - [PDF Document] (282)

Facts and Ideals 281

(12) a. Wenn Karl gewinnt, (dann) wird gefeiert! if Karl wins then celebrated ‘If Karl wins, we celebrate.’ b. Wenn Karl gewonnen hätte, (dann) hätten wir gefeiert! if Karl won had then had we celebrated ‘If Karl had won, we would have celebrated.’

Adding doch to a conditional antecedent can have two effects. The first effect is illustrated in (13). Here, by means of using unstressed doch (typically in indicative conditional antecedents) the speaker conveys that the truth of the antecedent proposition is established (cf. also Iatridou 1991 on factual conditionals and since-clauses). For now, I assume that the antecedent is used in a secondary speech act (as an embedded root clause, cf. Hooper & Thompson 1973, Haegeman 2003), by means of which the antecedent proposition is reactivated from the common ground.

(13) a. Wenn Karl doch gewinnt, dann wird gefeiert. if Karl doch wins then celebrated ‘Since Karl is obviously going to win, we will celebrate.’ b. primary speech act: assert [if Karl wins, we celebrate] c. secondary s.a.: reactivate [Karl will win] from the common ground

The second effect of doch in conditional antecedents is shown in (14), (glossing over possible prosodic differences between (13) and (14)). Here, by virtue of unstressed doch (typically in counterfactual conditional antecedents) the speaker conveys a wish for the truth of the antecedent proposition3.

(14) a. Wenn Karl doch gewonnen hätte, dann hätten wir gefeiert! if Karl doch won had then had we celebrated ‘If only Karl had won, then we would have celebrated!’ b. primary s.a.: assert [if Karl had won, we would have celebrated] c. secondary s.a.: reactivate [Karl won] from the speaker’s ideal list

The analysis presented in (8) derives the following. Assume that the sec-ondary speech act is retrieval in both (13) and (14); in (13) it operates on the common ground, in (14) it operates on the speaker’s ideal list. It follows that in (13) doch conveys that the antecedent proposition is an established part of the common ground. By analogy, in (14) doch indicates that the antecedent proposition is an established part of the speaker’s ideal list. In both cases,

3 It is beyond the scope of this paper why optative if-clauses can occur without a consequent, as in (2a) above, whereas non-optative if-clauses require a consequent.

Sinn und Bedeutung 15 - [PDF Document] (283)

282 Grosz

doch supports (or even triggers) retrieval of this proposition from some context set. I now provide evidence for such a contribution of doch.

(15) and (16) show that doch in a non-optative conditional antecedent indicates that the truth of the antecedent has been established (i.e. made part of the common ground). Therefore, doch can be placed into the antecedent in (15), where it has already been established that the antecedent is true.

(15) A: Es regnet! – B: Und? ‘It is raining!’ ‘So?’ A: (Na,) wenn es doch regnet, müssen wir die Party absagen. well if it doch rains must we the party cancel ‘(Well,) since it’s [doch] raining, we have to cancel the party.’

In contrast, doch cannot be used in the conditional antecedent in (16), where it is still under debate whether the antecedent is true.

(16) A: Regnet es? – B: Ich weiß nicht. Warum? ‘Is it raining?’ ‘I don’t know. Why?’ A: (Na,) wenn es (# doch) regnet, müssen wir die Party absagen. well if it doch rains must we the party cancel ‘(Well,) if / #Since it’s (#doch) raining, we have to cancel the party.’

This shows that doch can only be used in conditional antecedents if the truth of the antecedent is established. For doch in optative antecedents, an analogous point can be made. In example (17b), the doch-marked optative antecedent is ill-formed (as opposed to the straight imperative in (17a)). This follows if doch requires the hearer, Stefan, to accommodate that the modified proposition is already on the speaker’s ideal list. While hearers will often accommodate for such information (explaining that optatives are usually good in out-of-the-blue contexts, cf. Scholz 1991), Stefan has good reasons (e.g. social norms) to refuse to accommodate in (17b). This example thus feels inappropriate, as it conveys that Stefan should have known all along.

(17) Context: Stefan is at Thomas’s place and Thomas has made no suggestion whatsoever that he doesn’t want Stefan to stay for longer. Stefan: Stört es dich, wenn ich mir noch ein Bier nehme? ‘Does it bother you if I have another beer?’ a. Thomas: Ach, Stefan, geh jetzt bitte. Mir wird es zu spät. oh Stefan leave now please me becomes it too late ‘Oh Stefan, please leave now. It’s getting too late for me.’

Sinn und Bedeutung 15 - [PDF Document] (284)

Facts and Ideals 283

b. Thomas: # Ach, Stefan, wenn du doch jetzt gehen würdest. oh Stefan if you doch now leave would # ‘Oh Stefan, if only you’d leave now.’

Contrast (17) with (18); in (18), Thomas’s initial suggestion plausibly adds the proposition that Stefan leaves to Thomas’s ideal list. Therefore, it is felic-itous in (18) to reactivate this proposition in the subsequent optative.

(18) Context: Thomas is sick. Stefan is looking after him even though there is a great party on for tonight. Thomas: Stefan, geh ruhig auf die Party. Das stört mich nicht. ‘Stefan, please do go to the party. That doesn’t bother me.’ Stefan: Nein, nein, ich bleibe bei dir. ‘No, no, I’ll stay here with you.’ Thomas: Ach, Stefan, wenn du doch jetzt gehen würdest. oh Stefan if you doch now leave would ‘Oh Stefan, if only you’d leave now.’ Du hättest so viel Spaß! ‘You would have so much fun!’

2.3 Why Does doch Seem to Cause Optativity? At this point, we can return to the question of what doch contributes to an optative conditional. Specifically, why does doch trigger optativity in (19a), in the sense that (19a) is typically understood as an optative and a non-optative reading is not even considered, even though it is possible, cf. (19b) .

(19) a. Wenn Hans doch geblieben wäre … if Hans doch stayed were ‘If only Hans had stayed!’ b. Wenn Hans doch geblieben wäre, wäre Fürchterliches passiert. if Hans doch stayed were were horrible.things happened ‘Since (under certain circ*mstances) Hans would have stayed, horrible things would have happened.’

To account for this pattern, I propose that the contribution of doch in con-ditional antecedents makes an optative reading more accessible whenever the context does not explicitly favor a non-optative reading. To see this approach at work, we need to consider minimally contrasting pairs of utterances. So far, we have only considered indicative cases of doch in non-optative con-ditional antecedents, as in (20a). Crucially, such constructions are possible in the subjunctive, as shown in (20b) (and also in (19b) above).

Sinn und Bedeutung 15 - [PDF Document] (285)

284 Grosz

(20) a. Wenn Karl doch gewinnt, dann wird gefeiert. if Karl doch wins then celebrated ‘Since Karl is obviously going to win, we will celebrate.’ b. Wenn Karl doch gewonnen hätte, hätten wir gefeiert. if Karl doch won had had we celebrated ‘Since Karl would have won, we would have celebrated.’

Evidently, (20b) cannot be counterfactual, as doch requires the truth of the antecedent to be established whereas counterfactual antecedents are implied to be false. A context for (20b) is given in (21); as indicated, the antecedent in (20b) and (21) must be implicitly conditionalized. The implicit conditional is made overt (in parentheses) in (21).

(21) Berti: I’m so annoyed that the race was canceled. – Susi: But why? – Berti: Because Karl would have won. – Susi: So? Why do you care? You don’t even like Karl. – Berti: But I like to celebrate and … … wenn Karl (, wäre das Rennen nicht abgesagt worden,) doch if Karl were the race not canceled been doch gewonnen hätte, hätten wir gefeiert. won had had we celebrated ‘Since(, had the race not been canceled,) Karl would have won, we would have celebrated.’

We can now construct an example that allows for both an optative reading and a non-optative reading, given in (22), (23) and (24).

(22) Wenn Karl doch gewonnen hätte … dann hätten wir gefeiert! if Karl doch won had then had we celebrated ‘If Karl doch had won … then we would have celebrated.’

What (22) conveys on its non-optative reading is given in (23).

(23) Non-optative reading of (22): If Karl had won, we would have celebrated. Karl didn’t win (or lose) because the race was canceled. We didn’t celebrate. doch ⇒ I reactivate from the common ground that [Karl would have won if the race had not been canceled].

Contrast this with the optative reading of (22), given in (24).

(24) Optative reading of (22): If Karl had won, we would have celebrated. Karl didn’t win.

Sinn und Bedeutung 15 - [PDF Document] (286)

Facts and Ideals 285

We didn’t celebrate. doch ⇒ I reactivate from my ideal list that [Karl did win].

The fact that doch in such ambiguous conditional antecedents typically gives rise to an optative reading can now be derived as follows. Under an optative use, (24), doch triggers a presupposition/implicature with respect to the speaker’s ideal list. Given that the speaker is the highest authority with respect to her own ideal list, this will by default be self-fulfilling, i.e. the hearer will accommodate unless there are good reasons to refuse to accommodate (cf. Stefan in (17b)). In contrast, in non-optative cases, e.g. (23), doch triggers a presupposition/implicature with respect to the common ground, i.e. with respect to shared knowledge between speaker and hearer. This will fail in all contexts in which no such shared knowledge persists and cannot be easily accommodated. Furthermore, given that optatives with doch are typically in the subjunctive, a further asymmetry arises. Non-optative subjunctive antecedents that contain doch must be implicitly conditionalized, as in (21) and (23), whereas optative antecedents do not have such a requirement. It follows that non-optative readings, like (23), are further restricted to contexts in which the implicit conditionalization of the con-ditional antecedent can be successfully resolved. Therefore, by virtue of placing the particle doch in a conditional antecedent, as in (22), non-optative readings are restricted to very specific contexts and blocked in all other contexts. In contrast, doch in an optative conditional antecedent imposes restrictions that are typically self-fulfilling (in the sense that a hearer will accommodate a presupposition with respect to what the speaker wishes for). This makes doch acceptable in an optative conditional antecedent even when uttered out of the blue, deriving the fact that doch biases an optative reading.

2.4 Against a Strictly Compositional Approach I have argued that doch has a uniform semantics that is sensitive to the type of utterance it occurs in. The meaning of doch is thus in some sense independent from optativity, which predicts that typical optative features such as the particle doch are neither sufficient nor necessary conditions of optativity. We have already seen that particles like doch do not automatically give rise to an optative reading when placed into a subjunctive if-clause, (25).

Sinn und Bedeutung 15 - [PDF Document] (287)

286 Grosz

(25) Wenn du doch so einfach aufhören könntest zu rauchen, if you doch so easily stop could to smoke warum machst du’s dann nicht? why make you’it then not ‘If, as we’ve established, you could stop smoking that easily, then why don’t you do it?’ (slightly sarcastic)

Similarly, my analysis predicts that optativity should be possible in the absence of any particle. This prediction also seems to be correct, as the bare conditional antecedent in (26)4 can be understood as expressing a wish.

(26) Rico schaute die Blumen an und dachte: ‘Rico looked at the flowers and thought:’ Wenn Stineli diese sehen könnte! if Stineli these see could ‘If Stineli could see these!’ und stand lange unbeweglich am Zaun. ‘and stood at the fence for a long time without moving.’

2.5 Why Optatives without any Cues Fail In sum, I have argued for a particular view of doch in conditional antecedents that can be summarized as in (27), where doch is viewed as an optativity cue.

(27) Summary – Cue for a wish An optativity cue is an element that cues a wish reading for a conditional antecedent as follows: i. Its semantic contribution is independent from optativity. ii. Its meaning is compatible with a conditional antecedent that expresses a wish in a non-specific (or even out-of-the-blue) context. iii. Its semantic contribution to a non-optative conditional antecedent requires a very specific context (which cannot be out-of-the-blue). iv. Therefore, if the context does not determine whether a wish speech act is intended or not, the optativity cue conveys that a wish speech act is intended by blocking alternative readings, due to (iii).

An interesting aspect of optative constructions is that optatives without any cue are typically somewhat marked, (28), making the presence of particles seem obligatory (but see Rosengren 1993, cf. also (26) above).

4 From Johanna Spyri (1878): Heimatlos. Geschichten für Kinder und auch für solche, welche die Kinder lieb haben. Acceptability in modern-day German verified with native speakers.

Sinn und Bedeutung 15 - [PDF Document] (288)

Facts and Ideals 287

(28) in an out-of-the-blue context: Wenn ich ??(doch) reich wäre! if I doch rich were ‘If ??(doch) I were rich!’

While (26) suggests that optatives without particles are not always ill-formed, the question arises why optatives without any particles are dispreferred. I propose that this follows from treating such particles as optativity cues, if we make standard assumptions on rational discourse participants (cf. Lewis’s 1969 signaling games). If a speaker has to decide whether to use optativity cues and the hearer has to decide how to interpret conditional antecedents without such cues, the most successful strategies are typically those where speakers always use cues and hearers always interpret antecedents without cues as true conditionals. It follows that hearers will typically understand (28) without doch as a (fragmentary) non-optative conditional, unless the context overrides this preference. In cases like (26), an optative intention can be inferred from other information (such as the inferred friendship between Rico and Stineli, the description of the context, the verb that is used, etc.).

3 Conclusion I addressed the meaning and role of particles such as German doch in optative constructions. I argued that conditional antecedents can express a wish by virtue of a secondary speech act; particles do not encode this wish, but act as cues that bring out a possible wish reading (i.e. optative reading) by eliminating competing non-optative readings. I showed that this analysis correctly predicts that such particles are neither necessary nor sufficient conditions of optativity. Finally, I showed that this analysis can explain that unmarked conditional antecedents are typically understood as non-optative fragments. This follows, as rational discourse participants will usually pursue strategies where optative cues are used when optativity is intended and conditional antecedents without such cues are understood as non-optative.

References Abraham, Werner. 1991. Discourse particles in German: How does their

illocutive force come about? In: Werner Abraham (ed.): Discourse Particles, 203–252. Amsterdam: John Benjamins.

Asarina, Alya & Kirill Shklovsky. 2008. Optativity in English and Other Languages. Handout for MIT Ling-Lunch.

Sinn und Bedeutung 15 - [PDF Document] (289)

288 Grosz

Biezma, Maria. 2010. Optatives: conditionals with reversed topicality. Handout for Sinn und Bedeutung 15.

Doherty, Monika. 1987. Epistemic Meaning. Berlin: Springer. Gärtner, Hans-Martin 2010. SAI and NPI-Licensing: The Case of Optatives.

Slides for Oberseminar Neuere Arbeiten zu Syntax und Semantik, Universität Tübingen.

Grosz, Patrick. in press. German doch. An Element that Triggers a Contrast Presupposition. In Proceedings of CLS 46.

Han, Chung-Hye 1998. The Structure and Interpretation of Imperatives. Mood and Force in Universal Grammar: UPenn PhD. Thesis.

Haegeman, Liliane. 2003. Conditional Clauses: External and Internal Syntax. Mind and Language 18. 317–339.

Hooper, Joan & Sandra Thompson. 1973. On the applicability of root transformations. Linguistic Inquiry 4. 465–497.

Iatridou, Sabine. 1991. Topics in Conditionals: MIT PhD. Thesis. Karagjosova, Elena. 2001. Towards a comprehensive meaning of German

doch. In Proceedings of the Sixth ESSLLI 2001 Session. 131–142. Karagjosova, Elena. 2004. German doch as a marker of given information.

Sprache & Datenverarbeitung 28. 71–78. Kratzer, Angelika & Lisa Matthewson. 2009. Anatomy of two discourse

particles. Handout for SULA 5. Kratzer, Angelika. 1981. The Notional Category of Modality. In Hans-Jürgen

Eikmeyer & Hannes Rieser (eds.): Words, Worlds, and Contexts, 38–74. Berlin / New York: de Gruyter.

Krifka, Manfred. 2007. Basic Notions of Information Structure. In Caroline Féry, Gisbert Fanselow & Manfred Krifka (eds.): The Notions of Information Structure. 13–55. Potsdam: Universitätsverlag Potsdam.

Levinson, Stephen C. 1983. Pragmatics. Cambridge: CUP. Lewis, David. 1969. Convention. Cambridge, MA: Harvard University Press. Lewis, David. 1979. A problem about permission. In Esa Saarinen, Risto

Hilpinen, Ilkka Niiniluoto & Merril Provence Hintikka (eds.): Essays in honour of Jaakko Hintikka. 163–175. Dordrecht: Reidel.

Lindner, Katrin. 1991. “Wir sind ja doch alte Bekannte.” The use of German ja and doch as modal particles. In Werner Abraham (ed.): Discourse Particles. 163–201. Amsterdam: John Benjamins.

Ormelius-Sandblom, Elisabet. 1997. Die Modalpartikeln ja, doch und schon. Zu ihrer Syntax, Semantik und Pragmatik. Stockholm: Almqvist & Wiksell International.

Portner, Paul. 2005. The Semantics of Imperatives within a Theory of Clause Types. In Proceedings of SALT 14. New York: CLC Publications.

Sinn und Bedeutung 15 - [PDF Document] (290)

Facts and Ideals 289

Repp, Sophie. 2009. On the interplay of discourse particles and illocutionary negation. Handout for the workshop on expressives and other kinds of non-truth-conditional meaning, DGfS 31. University of Osnabrück.

Rifkin, Jay. 2000. If only if only were if plus only. In CLS 36–1. 369–384. Rosengren, Inger. 1993. Imperativsatz und “Wunschsatz” – zu ihrer Gram-

matik und Pragmatik. In Inger Rosengren (ed.): Satz und Illokution Bd. 2. 1–47. Tübingen: Niemeyer.

Scholz, Ulrike. 1991. Wunschsätze im Deutschen – Formale und funktionale Beschreibung. Satztypen mit Verberst- und Verbletztstellung. Tübingen: Niemeyer.

Stalnaker, Robert. 1974. Pragmatic presupposition. In Milton Munitz & Peter Unger (eds.): Semantics and philosophy. 197–213. New York: New York University Press.

Stalnaker, Robert. 1978. Assertion. In Peter Cole (ed.): Syntax and semantics 9: Pragmatics. 315–332. New York: Academic Press.

Thurmair, Maria. 1989. Modalpartikeln und ihre Kombinationen. Tübingen: Niemeyer.

Truckenbrodt, Hubert. 2006. On the semantic motivation of syntactic verb movement to C in German. Theoretical Linguistics 32. 257–306.

Sinn und Bedeutung 15 - [PDF Document] (291)

Sinn und Bedeutung 15 - [PDF Document] (292)

In Reich, Ingo et al. (eds.), Proceedings of Sinn & Bedeutung 15,pp. 291–305. Universaar – Saarland University Press: Saarbrücken, Germany, 2011.

Conventional and Free Association with Focus inNgamo (West Chadic)∗

Mira GrubicUniversity of [emailprotected]

Malte ZimmermannUniversity of Potsdam


Abstract. The paper discusses association with focus in Ngamo (West Chadic,Afro-Asiatic). We present evidence from this non-Indoeuropean language infavour of Beaver & Clark (2008)’s claim that different kinds of focus-sensitive el-ements interact with the meaning of focus in different ways, namely conventionalassociation with focus vs free association. We show that exclusive particles (only)in Ngamo, as in English, conventionally associate with focus. (Scalar-) Additiveparticles (also, even), by contrast, do not pattern like their English counterparts:Same as Q-adverbials, they are more free in their association behaviour, and canalso associate with non-focused elements under certain conditions.

1 Association with FocusFocus-sensitive elements depend on the grammatical placement of focus fortheir interpretation. This semantic dependency is often referred to as associ-ation with focus and can be seen clearly in sentences containing the focus-sensitive particle only (cf. (1)): only is an exclusive particle, it leads to an ex-clusion of the alternatives induced by focus. In (1a), because focus is on Bill,it is excluded that John likes other people, whereas in (1b), because focus is onlikes, it is excluded that John loves or admires Bill. In the case of only, associ-ation with focus actually makes a truth-conditional difference: (1a) is false inthe given context, whereas (1b) is true.

(1) (Context: John likes Mary and Bill, but he loves Sue.)a. # John only likes [BILL]F .

(excluded alternatives: {John likes Mary, John likes Sue})b. John only [LIkes]F Bill.

(excluded alternative: {John loves Bill})

∗ This research was conducted as part of project A5 “Focus realization, focus interpretation and fo-cus use from a cross-linguistic perspective” of the Collaborative Research Center 632 “InformationStructure”, funded by the German Science Association (DFG). We would like to thank the GermanScience Association, as well as our Ngamo informants Jibir Audu Janga Dole, Zakar Yusuf Babash*tte and Hadiza Abdullahi.

Sinn und Bedeutung 15 - [PDF Document] (293)

292 Grubic & Zimmermann

Analogous, but non-truth-conditional effects can be seen with the focus-sensi-tive particles also and even. Also is an additive particle, which includes alter-natives to the focused element. (2a) says that John sent a letter to somebodybesides Bill, whereas (2b) expresses that John sent something besides a letterto Bill.

(2) a. John also sent a letter to [BILL]F .(included alternative e.g. {John sent a letter to Sue})

b. John also sent a [LETTer]F to Bill.(included alternative e.g. {John sent a package to Bill})

Even also has this additive meaning component, e.g. that John sent a letter tosomeone else in (3a). In addition, it has a scalar meaning component (Karttunen& Peters 1979): the presence of even expresses that on a scale of expectability,Bill is less likely to receive a letter by John than any of the implied (focus)alternatives to Bill.

(3) a. John even sent a letter to [BILL]F .(included alternative e.g. {John sent a letter to his mother})

b. John even sent a [LETTer]F to Bill.(included alternative e.g. {John sent an email to Bill})

Again, the alternatives that are added and the corresponding scale shift whenthe focus is shifted: in (3b), what is added is that John sent something elseto Bill – e.g. an email, and that sending this other object is more expectedthan sending a letter. Each of these particles thus associates with focus in anintuitive sense. The main question of this paper is whether the particles only,also and even associate with focus in the same way, intra- as well as cross-linguistically. Chapter 2 describes two unified approaches to this associationprocess, conventional and free association, and presents the mixed approach ofBeaver & Clark (2008), whose general framework and terminology we adoptin this paper. Their main argument is that in English, different focus-sensitiveelements associate with focus in different ways. The focus particles only, alsoand even all conventionally associate with focus, whereas other focus-sensitiveelements like the q-adverbial always freely associate with focus. In section3, we present some data from the West Chadic language Ngamo that showthat, as proposed by Beaver & Clark for English, there are also different kindsof association with focus in Ngamo. The data also suggests, however, that,unlike in English, additive(-scalar) particles associate only freely with focusin Ngamo. Section 4 presents the formal semantic analysis of focus-sensitiveparticles in Ngamo. Section 5 concludes.

Sinn und Bedeutung 15 - [PDF Document] (294)

Association with Focus in Ngamo 293

2 Conventional and Free Association with FocusThere are two kinds of unified approaches to the association process: Conven-tional association accounts and free association accounts. Conventional andfree association are the terms used by Beaver & Clark (2008), and correspondto Rooth’s (1992) weak and strong association, respectively.In conventional association accounts (Rooth 1985; Jacobs 1983), focus-sensi-tive elements relate directly to the alternatives introduced by focus. This rela-tion is semantic, i.e. the dependency on focus is coded directly into their lexicalmeaning, as shown for only in (4a). In free association (Rooth 1992; von Fintel1994), on the other hand, focus-sensitive elements quantify over an implicitfree variable C, the reference of which is fixed by the context (4b).

(4) a. JonlyKw = λq ∀p ∈ JqK f : p(w)→ p = JqK0

b. JonlyCKg,w = λq ∀p ∈ g(JCK) : p(w)→ p = q

C usually resolves to the focus alternatives, since these are contextually salient,so the difference is mainly one of empirical elegance: According to Rooth(1992), a free account is stronger because it does not tie the semantic effectsof focus to the meaning of specific lexical items. Moreover, the free associa-tion account makes a prediction that the conventional account does not make:(apparent) association with non-focused elements should be possible if the con-text provides a value for C that differs from the focus alternatives. As Beaver& Clark (2008) point out, this prediction is borne out for some focus-sensitiveelements, but not for others. While only can never associate with grammati-cally non-focused constituents, always can occur in contexts in which it seemsto associate with non-focused material. This can be seen in example (5) fromBeaver & Clark (2008: 193), which tests for association with weak, unstress-able, and hence inherently unfocused pronouns. The test sentence enforces areading in which the focus-sensitive element associates with the weak DO-pronoun, because other possible readings (association with the verb or VP) areexcluded by the context. This reading is not accepted for a sentence with only,but it is fine for the parallel sentence with always.

(5) a. ??People who grow rice only eat it. # ‘People who grow rice eatnothing but rice’

b. People who grow rice always eat it. ‘Whenever people who growrice eat, they eat rice’

Based on the different behaviour of always and only, Beaver & Clark (2008)propose that different focus-sensitive elements associate with focus in differentways, with some elements conventionally associating with focus (e.g.only), and

Sinn und Bedeutung 15 - [PDF Document] (295)

294 Grubic & Zimmermann

others freely associating with focus (e.g. always). According to them, the classof conventionally associating elements in English contains the focus-sensitiveparticles only, also and even discussed above, whereas the class of freely asso-ciating expressions contains, for example, quantificational adverbs like always,generics, counterfactuals, and modals.Another candidate for a freely associating element is the stressed additive par-ticle AUCH (“also”) in German, which appears to associate with the precedingcontrastive topic (Krifka 1999).

(6) (I know that Pia visited the exhibition. But what did Peter do?)Peter hat die Ausstellung AUCH besucht. ‘Peter visited the exhibition,too.’

In these cases, the contrastive topic introduces the alternatives that are rele-vant for the resolution of the free variable C. This is another instance in whicha “focus-sensitive” element can associate with a prosodically weak or evenempty element. In example (7a), the associate of the stressed additive particlecan be elided, because it is given in the preceding context question. This isnot possible for prefocal unstressed auch, which must associate with an overtfocus-accented element (7b).

(7) (You did the dishes. And the garbage?)a. Hab




b. Auchauch





‘I took care (of it) too.’

The next section presents data from Ngamo (West Chadic) that support Beaver& Clark’s claim that there are different kinds of focus-sensitive elements. More-over, the Ngamo data provide further evidence that additive particles, at leastin some languages, do not belong in the same class as exclusives. This is notfully unexpected given the behaviour of German additives mentioned above.

3 Focus & Focus-Sensitive Particles in NgamoThis section gives an overview of grammatical focus marking and focus-sen-sitive particles in Ngamo. Ngamo is a West Chadic language of the ‘A’ sub-branch spoken in NE Nigeria by about 60’000 speakers (Gordon 2005). It hastwo major dialects, Yaya Ngamo and Gudi Ngamo (Schuh 2005). The data inthis paper come from the Gudi dialect.

Sinn und Bedeutung 15 - [PDF Document] (296)

Association with Focus in Ngamo 295

3.1 Focus in NgamoAs in many other African languages (Fiedler et al. 2010), overt focus mar-king in Ngamo is asymmetric: Focused subjects must be syntactically marked,whereas focused non-subjects need not be explicitly marked for focus. Thecanonical word order is SVO, but when a subject is questioned or focused, thesubject is inverted to the right edge of vP. The subject cannot occur betweenthe verb and the object (8a), but its word order is free with respect to other con-stituents following the direct object (cf. (8b)). Non-subject focus is preferablyrealized in-situ (9a), but ex-situ test sentences are also accepted (9b).

(8) a. *Salkobuild-PFV







b. Sàlkobuild-PFV












‘Who built a house in Nigeria last year?’

(9) a. ShuwaShuwa




‘Who did Shuwa call loudly?’b. Èsha




‘(she) called JAJEI loudly.’

We suggest that the reason behind the subject inversion to the right edge of vPis an interface requirement that forces the focused element to be right-alignedwith a phonological phrase boundary projected by the right edge of vP (Samek-Lodovici 2005; Truckenbrodt 1999; Zimmermann 2006). Since objects and ad-juncts are canonically realized at the right edge of vP anyway, it follows thatthey can remain in-situ when focused.

Inverted subjects are obligatorily preceded by a morphological markeri/ye (10a), which is again optional with focused non-subjects (10b). Schuh(2005: 27) suggests that i/ye is not a focus marker, but a background marker,which is historically derived from the definite determiner ye’e. This suggestionis supported by the fact that it can occur twice in an utterance, thus markingbackgrounded material following the focused constituent (10c).

(10) a. [Context: Hasha called Yura]O’ò,No,





‘No, KULE called Yura.’

Sinn und Bedeutung 15 - [PDF Document] (297)

296 Grubic & Zimmermann

b. [Context: Who did Shuwa call?]ShùwaShuwa




‘Shuwa called JAJEI.’c. Sàlko








‘Who built a house in Nigeria last year?’

A standard test for exhaustivity in (11) shows that answers with i/ye are inter-preted as exhaustive, or maximal, in contrast to answers without i/ye: In thequestion context in (11), the complete answer A entails the partial answer A1without i/ye, but it does not entail the partial answer A2 with i/ye, which signalsmaximality, making A2 infelicitous in the given context.

(11) [Context: Who did Kule call?] A: Kule called Shuwa and Dimza.→ A1:Kule



‘There was an event of Kule calling Dimza.’9 A2:#Kule




‘The maximal calling event by Kule involves Dimza as a callee.’

We suggest that i/ye is a definiteness marker on events (Larson 2003; Holesubmitted) that introduces a presupposition that there is a maximal contextuallysalient event exemplifying the vP-denotation (excluding the focus denotation)as in (Kratzer 2007), cf. answer A2 in (11):

(12) J-i/yeK = λ f<v,t>: there is a maximal salient event e, s.t. f(e) = 1. f

This analysis of the maximality effect is supported by the fact that it is can-cellable in cases in which we talk about separate events (13).

(13) Sàlkobuild-PFV








‘(lit.) DIMZA did the house-building, and Umar built a(nother) house.’

To sum up, subject focus in Ngamo is obligatorily marked by inversion to theright edge of vP, with background marking of the backgrounded part precedingthe focused subject. Non-subject focus is only optionally marked. The back-ground marker is a definiteness marker on events which introduces a maximal-ity presupposition on the backgrounded vP-denotation.

Sinn und Bedeutung 15 - [PDF Document] (298)

Association with Focus in Ngamo 297

3.2 Focus-Sensitive Particles in NgamoThis section presents the distribution and the association behaviour of threefocus-sensitive particles in Ngamo: the additive ke (‘also’), the additive-scalarhar (‘even’), and the exclusive particle yak (‘only’).When associating with non-subjects, the three particles behave alike. All threefocus-sensitive particles can occur sentence-initially, in immediately preverbalposition and after the VP, but not between the verb and the direct object.

(14) (Har)even






‘Baba even SOLD the house.’

(15) (Ke)also









‘Dimza also gave a watch TO ABU.’

(16) (Yak)only








‘She only called him YESTERDAY.’

These examples show that the particles can precede or follow their associate,without a change in meaning. In addition, they have a pre-focal (e.g. har) and apost-focal form (e.g. har’i), the distribution of which appears to be conditionedby prosodic factors, such as the presence of a subsequent prosodic boundary.The examples also show that there is no adjacency requirement: all three parti-cles can associate from an adjacent or distant position. Non-adjacent preverbaland post-VP particles are illustrated in (17).

(17) [Context: Kule built a house.]a. si




‘He also built a SCHOOL.’b. Kule




‘Kule also SOLD the house’

We propose that focus particles in Ngamo typically denote adverbial operatorsthat are adjoined to the extended vP-projection, but there appear to be adnom-inal counterparts as well (e.g. the third yak in (16)).

Crucially, the three particles however behave differently when it comesto association with subjects. Recall that in cases of conventional associationwith focus, the associate of the focus-sensitive particle is obligatorily focus-

Sinn und Bedeutung 15 - [PDF Document] (299)

298 Grubic & Zimmermann

marked, whereas in free association with focus this is not necessarily the case.Since focused subjects are marked by inversion, we therefore predict that con-ventionally associating focus-sensitive elements will only be able to associatewith inverted – and, thus, focus-marked – subjects! This expectation is borneout for exclusive yak, which can only associate with inverted subjects.

(18) a. Sàlkobuild-PFV





‘Only KULE built a house.’b. #Yak






(intended:) ‘Only SHUWA built a house.’

In contrast, ke/har cannot associate with inverted subjects. We suggest that thisis due to a clash of the maximality presupposition of the background markerwith the additive presupposition of the particles. In (19), background markingintroduces the presupposition that there is a salient maximal event of build-ing a house involving Kule (and nobody else), whereas the additive particleke (‘also’) presupposes that someone else took part in the contextually salientevent of house-building, in violation of maximality. According to one con-sultant, this structure is only permitted in a context in which the (maximal)house-building event is juxtaposed to a (maximal) event of a different type.

(19) [Context: Hawwa built a house]a. #Salko





(intended) ‘KULE also built a house.’(Consultant comment: ‘Where there is ‘salko bano-i’, this meansthat the other person did something else.’)

b. *Salkobuild-PFV





(intended): ‘Even KULE built a house.’

Instead, the only way of expressing what looks like association with subjectswith these particles is to leave the subject in its canonical preverbal position.

(20) a. Kè/Haralso/even




b. KulèKule




Sinn und Bedeutung 15 - [PDF Document] (300)

Association with Focus in Ngamo 299

c. KulèKule




‘KULE built a house, too.’

Since focused subjects are banned from this position (see above), we are led toconclude that the particles ke and har do not operate directly on the focus valueof the subject in such cases, but interact with the subject denotation in a moreindirect way. In short, we capture the difference between the additive particlesand the exclusive particle by suggesting that the former freely associate withfocus, whereas the latter conventionally associates with focus.

Further evidence for this proposal comes from the association of the dif-ferent types of particles with weak and strong pronouns. In Ngamo, indirectobject pronouns are usually incorporated into the verb. When the pronoun isfocused, it must occur in its strong form, which is headed by the prepositionki (‘to’). Crucially, additive ke can associate with the weak, incorporated, andthus non-focused form of the pronoun (21), whereas yak can only associatewith strong, and hence focused pronouns (22).

(21) [Context: Whom did Kule give a watch?]‘Onkogive-PFV








‘He gave a watch to Dimza, and also gave a watch to me.’

(22) [Context: Did Kule give a watch to all of them?]a. #O’o,





(intended:) ‘No, Kule only gave a watch to HER.’b. O’o,








‘No, Kule only gave a watch to HER.’

Moreover, ke can freely associate with zero subjects, an option excluded forthe exclusive particle, as is shown for the contrastive topic context in (23),which is modeled after an example from (Krifka 1999). Here, an answer witha contrastive topic subject is enforced by explicitly giving a partial answer to asuperquestion of the form “Who did what?”, which is followed by a request forinformation concerning a second individual, functioning as a contrastive topicin this context. As the subject is given, it can be dropped from the answer,although the additive particle ke seemingly associates with it. This is anotherinstance in which the additive particle associates with unfocused material, andthus evidence in favour of free, and not conventional, association with focus.

Sinn und Bedeutung 15 - [PDF Document] (301)

300 Grubic & Zimmermann

(23) a. NèI








‘I know that Kule built a house, but what about Shuwa?’b. Kè



‘(He) also built a house.’

Summing up, additive(-scalar) particles show a different association behaviourwith subject focus when compared to the exclusive particle: While yak (‘only’)must associate with focus-marked (inverted) subjects, ke (‘also’) and har (‘e-ven’) cannot associate with such inverted subjects. Instead, they appear to as-sociate with preverbal subjects, which are never focused. This leads us to con-clude that yak conventionally associates with focus, while ke and har freelyassociate with focus.

4 Analysis & DiscussionThis section presents the formal analysis of the focus-sensitive particles in-troduced in the previous section. First, the framework used for the analysisis presented in section (4.1), then the exclusive particle yak will be discussed(4.2), then the additive(-scalar) particles ke and har (4.3).

4.1 A QUD Approach to Information StructureIn a QUD-approach (e.g. Roberts 1996, Büring 2003), the idea that the goalof discourse is to share previously unshared information is captured by mod-eling discourse as driven by implicit (hearer-) questions: The goal of eachnew conversation is to cooperatively answer the super-question “What is theway things are?”. This question is tackled by splitting it up into subquestions,which each ask for a partial answer to the superquestion. Each new declarativeutterance answers the lowest question in the tree – the Current Question. Inthis model, information structural categories like focus and topic are used fordiscourse-management. They indicate what the implicit questions under dis-cussion are. This is done through question-answer congruence: According toRoberts (1996), the focus alternative set (cf. Rooth 1985; 1992) of the utteranceis congruent to the set of possible answers indicated by the Current Question,e.g. (24) for answer A1 in (25a).1

(24) JCQK0 = JA1KF = {John likes Bill, John likes Mary, John likes Sue}

1 For weaker constraints on question-answer congruence, see Büring (2003: 517) and Beaver &Clark (2008: 47).

Sinn und Bedeutung 15 - [PDF Document] (302)

Association with Focus in Ngamo 301

(25) (a) What is the way things are?


CQ. Whom does John like?

A1. [John likes [Bill]F

(b) What is the way things are?


Who likes whom?

CQ. Whom does John like?

A2. [John]CT likes [Bill]F

Contrastive topics indicate the presence of alternatives raised by questionsabove the Current Question (Büring 2003). So in (25b), the focus on Bill inA2 indicates that Bill must be replaced by a wh-element in the Current Ques-tion CQ, but the contrastive topic accent on John indicates that there is anadditional relevant question immediately above the Current Question, in whichthe subject John is also replaced by a wh-element. Büring (2003) proposes thatutterances like this do not only have a normal and a focus value, but also a CTvalue, which marks them as partial answers to this higher question.

4.2 The Exclusive Particle yakIn order to account for the conventional association behaviour of the exclusiveparticle yak ‘only’ in Ngamo, we follow suggestions by Beaver & Clark (2008)on the semantic function of exclusives. According to these authors, the mainfunction of exclusive particles is not to exclude alternatives, but to indicate thatthe proposition modified by the exclusive particle in an answer to an explicit orimplicit CQ indicated by the focus structure is less strong (on a salient scale)than expected by the hearer. By uttering (26A) in response to the explicit CQin (26Q), for instance, the speaker signals that he takes the hearer to expecta stronger alternative out of the question denotation to be true, e.g. that Johninvited Mary, Sue, Bill and John, among others.

(26) Q. Whom did he invite?A. He invited only MaryF .

Technically, this effect can be modelled by assigning yak the lexical entry in(27), which can be conceived of as a variant of the lexical entry for only inBeaver & Clark (2008):

(27) JyakKw(p) = 1 iff ¬ ∃ q ∈ CQyak: p ≤ q ∧ q(w);defined iff ∀q ∈ CQyak p ≤ q; where ‘≤’ stands for ‘weaker on acontextually salient scale’.

Sinn und Bedeutung 15 - [PDF Document] (303)

302 Grubic & Zimmermann

According to (27), the semantic effects of yak are twofold. First, the presenceof yak imposes a restriction - in form of a presupposition - of the original CQindicated by grammatical focus marking. According to the presupposition in(27), the new CQyak contains only propositional alternatives that are at leastas strong (on a contextually salient scale) as the proposition expressed by theutterance containing yak. Restricting the original CQ to CQyak in this way cap-tures the intuition that the hearer expects stronger alternatives to p to be true. Atthe same time, the presupposition excludes alternatives of equal strength (e.g.,Bill and John in the case of (26)) as it requires all the alternatives in CQyak to beeither identical to p, or stronger than p on a relevant scale. The second effect ofyak takes place at the truth-conditional level by specifying that p is indeed thestrongest true alternative in the yak-modified CQyak. Importantly, this analysiscontinues to treat yak as conventionally associating with focus, as yak makesdirect reference to an – albeit modified – CQ as indicated by grammatical focusmarking in its lexical entry.

4.3 The Additive(-Scalar) Particles ke and harThe additive(-scalar) particles ke and har do not refer to the focus alternativesdirectly. Their central semantic constribution consists in presupposing the ex-istence of another contextually salient situation, in which an alternative propo-sition out of a contextually bound variable C holds.

(28) J keCKg,w(p)(s) = p(s)(w);defined iff ∃ s’ in w, s’ 6= s : ∃ q ∈ g(JCK), q 6= p: [q(s’)(w)]

Har additionally presupposes that its complement is relatively unlikely.

(29) J harCKg,w(p)(s) = p(s)(w), defined iffi. ∃ s’ in w, s’ 6= s : ∃ q ∈ g(JCK), q 6= p: [q(s’)(w)]ii. p is (relatively) unexpected compared to other elements in g(JCK).

Since the Current Question is typically salient in a given context, the contextvariable C is usually resolved to it; e.g. in (30), C is resolved to the CurrentQuestion “What did Kule build?”, giving rise to the meaning in (31).

(30) [CQ: What did Kule build? Kule built a school, and...]KuleKule



‘Kule also built a HOUSE.’

(31) J keCKg,w(J Kule built a school K)(s) = 1 iff Kule built a school in s inw, defined iff ∃ s’ in w, s’ 6= s : ∃ q ∈ {Kule built a school, Kule built

Sinn und Bedeutung 15 - [PDF Document] (304)

Association with Focus in Ngamo 303

a house, Kule built a shed, ...}, q 6= p: [q(s’)(w)]

As seen in connection with example (19), in the case of focus-marked (in-verted) subjects (32), we see that if C is contextually resolved to the CurrentQuestion: “Who built a house?”, this will normally lead to a clash between thepresuppositions of ke and the background marker (33) (but see below).

(32) [CQ: Who built a house? Hawwa built a house, and...]#salkobuild-PFV





(intended:) ‘KULE also built a house.’

(33) J keCKg,w(J Kule built a houseK)(s) = 1 iff Kule built a house in s in w,defined iff ∃ s’ in w, s’ 6= s : ∃ q ∈ {Hawwa built a house, Kule built ahouse, Shuwa built a house, ...}, q 6= p: [q(s’)(w)]Presupposition of i-marking: There is a maximal salient house-buildingevent whose agent is Kule.

Instead, as argued above, any association of a (scalar-) additive particle witha subject is indirect in nature, constituting an instance of free association withfocus. There are at least two ways for such free associations with non-focusedsubjects to arise. First, the Current Question can be resolved to wide-scopequestions of the form “What happened?” (34) in contexts in which the VPis given, as illustrated in (35). Notice that the nuclear accent in the Englishparaphrase falls on the subject because the VP is given.

(34) [What happened? Hawwa built a house, and...]KuleKule



‘KULE also built a house.’

(35) J keCKg,w(J Kule built a houseK)(s) = 1 iff Kule built a house in s in w,defined iff ∃ s’ in w, s’ 6= s : ∃ q ∈ {Hawwa built a house, Hawwabought a car, Kule built a house, ...}, q 6= p: [q(s’)(w)]

The second strategy involves apparent association with a (contrastive) topic,which is possible since the canonical preverbal position of subjects is the de-fault topic position. In this case, the ke-sentence with topical subject relates tothe super-question “Who did what?”, which splits up into VP-subquestions asin (36), leading to (37).

(36) [(What did Hawwa do?) Hawwa built a house, and... (CQ: What didKule do?)]

Sinn und Bedeutung 15 - [PDF Document] (305)

304 Grubic & Zimmermann




‘KULE built a house, too.’

(37) J keCKg,w(J Kule built a houseK)(s) = 1 iff Kule built a house in s in w,defined iff ∃ s’ in w, s’ 6= s : ∃ q ∈ {Hawwa built a house, Hawwabought a car, Kule built a house, ...}, q 6= p: [q(s’)(w)]

Additional evidence for this second option comes from marked discourse-contexts in which ke actually does occur together with a focus-marked subject(38). In such cases, the focus/background marking suggests that the meaning ofthe i/ye-marked VP ‘building a house’ forms the (contrastive) topic event. Theantecedent super-question is again “Who did what?”, but this time it is split upinto subject questions ranging over contextually given events (38). The resultis shown in (39).

(38) [(Who bought a car?) Hawwa bought a car, and... (CQ: Who built ahouse?)]Salkobuild-PFV





‘Kule (also) built a house.’

(39) J keCKg,w(J Kule built a houseK)(s) = 1 iff Kule built a house in s in w,defined iff ∃ s’ in w, s’ 6= s : ∃ q ∈ {Hawwa built a house, Hawwabought a car, Kule built a house, ...}, q 6= p: [q(s’)(w)]Presupposition of i-marking: There is a maximal salient house-buildingevent whose agent is Kule.

In sum, we have shown that the proposed analysis of ke and har with the de-notations in (28) and (29) can account for the behaviour of these particles ob-served in section 3.2.

5 ConclusionThe Ngamo data presented suggest that there are different kinds of associationwith focus, similar to what was found for English by Beaver & Clark (2008).However, in contrast to English also and even, Ngamo additive particles do notconventionally associate with focus. This corresponds to findings from otherlanguages, e.g. Bura (Hartmann & Zimmermann 2008) and Thompson Salish(Koch & Zimmermann 2010). What remains to be seen is whether the analysisof additive particles in Ngamo can be extended to provide a unified account ofstressed and unstressed additive particles in German.

Sinn und Bedeutung 15 - [PDF Document] (306)

Association with Focus in Ngamo 305

ReferencesBeaver, D. & B. Clark. 2008. Sense and Sensitivity. Oxford: Wiley-Blackwell.Büring, D. 2003. On D-Trees, Beans, and B-Accents. Linguistics & Philosophy

26. 511–545.Fiedler, I. et al. 2010. Subject focus in West African languages. In M. Zimmer-

mann & C. Féry (eds.), Information structure, 234–257. Oxford: OxfordUniversity Press.

von Fintel, K. 1994. Restrictions on Quantifier Domains: UMass Amherstdissertation.

Gordon, R.G. 2005. Ethnologue: Languages of the World. Dallas, Texas: SIL.Hartmann, K. & M. Zimmermann. 2008. Not only ’only’, but ’too’, too.

Alternative-sensitive particles in Bura. In A. Grønn (ed.), Proceedingsof Sinn und Bedeutung (SuB) 12, 196–211.

Hole, D. submitted. The deconstruction of Chinese ’’ clefts revisited.Jacobs, J. 1983. Fokus und Skalen. Tübingen: Niemeyer.Karttunen, L. & S. Peters. 1979. Conventional implicature. In C.K. Oh &

D. Dineen (eds.), Presupposition, New York: Academic Press.Koch, K. & M. Zimmermann. 2010. Focus-sensitive operators in

Nlhe7kepmxcin (Thompson River Salish). In M. Prinzhorn et al. (eds.),Proceedings of Sinn und Bedeutung 14, Vienna, .

Kratzer, A. 2007. Situations in natural language semantics. In E.N. Zalta (ed.),Stanford encyclopedia of philosophy, Stanford: CSLI.

Krifka, M. 1999. Additive Particles under Stress. In Proceedings of SALT 8,111–128. Cornell: CLC Publications.

Larson, R.K. 2003. Event descriptions in Fon and Haitian Creole. In D. Adone(ed.), Development in Creole Studies, Tuebingen: Niemeyer.

Roberts, C. 1996. Information Structure in Discourse: Towards an IntegratedFormal Theory of Pragmatics. OSU Working Papers in Linguistics 49.

Rooth, M. 1985. Association with focus. UMass. Amherst: Graduate Linguis-tics Student Association.

Rooth, M. 1992. A Theory of Focus Interpretation. Natural Language Seman-tics 1. 75–116.

Samek-Lodovici, V. 2005. Prosody syntax interaction in the expression offocus. Natural Language and Linguistic Theory 23. 687–755.

Schuh, R. 2005. Yobe State, Nigeria as a Linguistic Area. Manuscript, UCLA.Truckenbrodt, H. 1999. On the Relation between Syntactic Phrases and Phono-

logical Phrases. Linguistic Inquiry 30. 219–255.Zimmermann, M. 2006. Focus in Western Chadic: A Unified OT-Account. In

C Davis et al. (eds.), Proceedings of NELS 36, UMASS, Amherst.

Sinn und Bedeutung 15 - [PDF Document] (307)

Sinn und Bedeutung 15 - [PDF Document] (308)

In Reich, Ingo et al. (eds.), Proceedings of Sinn & Bedeutung 15,pp. 307–320. Universaar – Saarland University Press: Saarbrücken, Germany, 2011.

The Concept of Semantic Phase and the Different Readings ofa*gain

Sabine GründerUniversity of Bielefeld


Abstract. The paper offers a new kind of approach to the semantic contrastbetween repetitive and restitutive again. The heart of the theory is the new con-cept of Semantic Phase. It parallels the syntactic concept and is motivated asan instance of the Principle of Hierarchical Abstraction. The concept refers toa switch from imperfective to perfective view of a situation at the level of vP.Applying the modifier before or after phase transition derives the two readingswithout stipulation of lexical ambiguity. The framework used is Finite-state Tem-poral Semantics of Fernando. The syntactic background is an Orphan analysis ofright-peripheral adverbials. Syntactic underspecification is resolved by the useof pragmatic information reflected locally by the prosody of the utterance.

1 Introduction1.1 The Basic DataThe paper is concerned with the ambiguity that can arise in connection withthe interpretation of the adverb again (or German ‘wieder’). While in connec-tion with atelic states and activities again expresses plain repetition, appliedto a telic accomplishment or achievement, the adverb either expresses repeti-tion of the entire event (repetitive reading) or repetition of its result state only(restitutive reading). Consider the following telic standard example.

(1) John opened the door again. (rep./rest.)

The restitutive interpretation presupposes that the door has been open some-time before, but it does not require that is has been opened by John or anybodyelse. For the repetitive interpretation to be true, the door must have been openedby John at some point of time in the past.

The surface position of the adverb seems to have influence on the avail-ability of the possible interpretations. On the one hand there is the ambiguoussentences in (1) where we find the adverb in sentence final position, on theother hand the second sentence in (2) with sentence initial adverb for whichonly the repetitive interpretation is available.

Sinn und Bedeutung 15 - [PDF Document] (309)

308 Gründer

(2) Again, John opened the door. (rep.)

For the ambiguous first construction intonation can be used as a disambigua-tion device. While unmarked intonation goes with the restitutive reading, mainaccent on the adverb forces the repetitive interpretation.

(3) a. John opened the DOOR again. (rest.)b. John opened the door AGAIN. (rep.)

1.2 Some Existing ApproachesThe repetitive/restitutive duality of again/‘wieder’ is probably the most thor-oughly discussed example of the syntactic-semantic flexibility that is charac-teristic of adverbial adjuncts in general. Against this background, the outcomeof the several approaches was not purely descriptive, but was at the same timeaiming at giving us some deeper and more general insights into the syntax andsemantics of verb phrases and into the constitution of interfaces. Therefore, theanalysis of again/‘wieder’ has always been like a measure of what has beenachieved in the linguistic theory of adjuncts. Up to now the issues are far frombeing settled. The controversy primarily concerns the question of where to lo-cate the source of the ambiguity: in semantics, syntax or pragmatics.

The classical treatment of Dowty (1979) presupposes decomposition ina conceptual semantic language. In the representations of the two readings ofthe ambiguous sentence, the adverb occupies the same structural position. Theinterpretational contrast arises from two different semantic representations thatbelong to different syntactic categories: a sentence modifier and verb phrasemodifier. A meaning postulate accounts for the semantic relationship betweenthe two. Although the semantic contrast derives correctly in this framework,no explanation can be given for the influence of syntax or prosody, and thesolution is based on stipulation of lexical ambiguity .

Later approaches that include the concept of lexical ambiguity (with orwithout meaning postulates) are, for instance, Fabricius-Hansen (2001), Reyle,Rossdeutscher & Kamp (2008), Jaeger & Blutner (2003).

The most principled alternative, that tries to do without theoretically costlystipulations on the lexical semantic side, is the theory of von Stechow (1996).It is based on the following kind of data for German ‘wieder’.

(4) a. (weil) Fritz wieder das Fenster öffnete (rep.)b. (weil) Fritz das Fenster wieder öffnete (rest./rep.)

Assuming a single lexical semantic entry, von Stechow claims that the ambigu-ity can be resolved entirely in terms of syntactic scope. Decomposition in the

Sinn und Bedeutung 15 - [PDF Document] (310)

Semantic Phase and again 309

style of Generative Semantics is located in the syntax. The theory is based on arather complex and abstract syntactic theory and uses movement of argumentsto Case positions. The leading idea is, that a structural accusative position haswide scope with respect to the agent relation expressed by the head of the VoicePhrase. If ‘wieder’ precedes an accusative object, a repetitive reading is oblig-atory, if it follows the accusative object, two readings are available due to twopossible positions of the adverbial. Since the arguments have moved to theirCase positions outside the Voice Phrase, the D-position of the adverb is nolonger uniquely identifiable from the surface, hence the ambiguity.

Another scope-based account was given by Pittner (2003). On the base ofa single lexical entry for the adverb, the different readings are determined bythe syntactic entity the modifier is related to. Although this assumption seemsvery natural and promising, it is not a trivial matter from a theoretical point ofview. If one accepts that an adjunct that is assigned one and only one semanticrepresentation can modify different types of entities, one would have to accountfor the interaction between the semantic contribution of the adjunct and thesemantic properties of the modified entity in a principled way.

Pointing to German word-order effects in connection with indefinite ob-jects as well as to the disambiguating effects of intonation, Jaeger & Blutner(2003) offer an alternative approach to the syntactically based theories, thatuses the framework of Optimality Theory. Disambiguation is the result of aprocess of pragmatic strengthening, which selects optimal candidates from ahighly underspecified relation between form and meaning. The word-order ef-fects involve scrambling of definite noun phrases. Optional from the point ofview of syntax, scrambling of nominal arguments plays an important role inthe information structural partition of an utterance into background and focus.Semantic material that is known or in some sense anaphoric relative to the con-text is moved out of the focus domain. On this basis it can be explained whythe semantics of example (5) below, with the adverbial preceding an indefiniteobject, is similar to the one given before where the adverbial follows a definiteobject (4b).

(5) (weil) Fritz wieder ein Fenster öffnete (rest./rep.)

Furthermore, considering the connection between pragmatic and prosody, itcan be stated that de-accented constituents are given. Accordingly, de-accentinga verb phrase in a syntactically ambiguous ‘wieder’-construction triggers arepetitive interpretation; in this case the sentence accent ends up on the ad-verbial. Unmarked intonation, on the other hand, places the main accent on theobject if it is verb adjacent and on the verb otherwise, and causes a restitutive

Sinn und Bedeutung 15 - [PDF Document] (311)

310 Gründer

interpretation. As was shown at the beginning (3), for the English examplesthere are similar disambiguation effects arising from the presence or absenceof an accent on the adverbial.

(6) a. (weil) Fritz das Fenster WIEDER öffnete (rep.)b. (weil) Fritz das Fenster wieder ÖFFNETE (rest.)

Although the approach does justice to the influence of context and intonation,the different interpretations need to be based on different lexical entries again,and there is no compositional semantics available in the theory.

1.3 The Aims of the PaperIn this paper we are going to present a new type of approach that tries to com-bine the virtues of the existing theories while avoiding their drawbacks.

It uses a concept of semantic decomposition that is less abstract than theone introduced by Dowty. There is no assumption of lexical ambiguity. It is ascopal approach but it can do with a much simpler structure and without theconcept of syntactic decomposition. Furthermore, scope is not syntacticallybut pragmatically determined, and mediated by the assignment of constituentsto the information structure domains of focus and background. This kind ofcontext-sensitive scope resolution gets formally implemented using prosodicinformation that is locally available on the constituents of the utterance. Thedifferent readings of the adverbial are determined with respect to the samesituation seen from different aspectual viewpoints. The idea of systematicallychanging the view of a situation follows from an independently motivated andgeneral cognitive principle: the Principle of Hierarchical Abstraction.

The proposal is part of a more general approach to left/right contrastsin the interpretation of English temporal adverbials that was offered in Grün-der (2009). Accordingly, the focus of the investigation is on finding a theo-retically well-motivated and general analysis for the standard cases of repet-itive/restitutive ambiguity for again. Technical solutions for special cases orexceptions, that are known in the literature, will have to be part of a more de-tailed future work. We take the fact that the general strategy motivated in thepaper can also be used to derive several other contrasts in adverbial modifica-tion as giving further weight to the proposal made here.

2 Semantic Phase Theory2.1 Hierarchical AbstractionThe Principle of Hierarchical Abstraction is seen as one of the most fundamen-tal and general cognitive principles to reduce complexity of problem-solving

Sinn und Bedeutung 15 - [PDF Document] (312)

Semantic Phase and again 311

tasks. It is a means to reduce details and condense information through step-wise merge of several elements in the problem space into one.

In modern linguistic theory, language too is described as a ‘system of dis-crete infinity, consisting of hierarchically organized objects’ (Chomsky 2008:137). Additionally, the Principle of Hierarchical Abstraction is included inform of Syntactic Phases (Chomsky 2001, 2008).

Phases mark points in the derivation where syntactic material is trans-ferred to the phonological component of the language system. At the level ofCP and vP (or VP if one does without assumption of the v-head), only materialin the head and the specifier is kept available for further syntactic processing;the information contained in the complement is spelled out respectively. In con-sequence, complexity of syntactic processing is reduced by minimizing searchspace and unloading working memory.

Interestingly, pragmatics assumes a similar transition point too. In infor-mation structure theory, the verbal domain is often considered the new infor-mation focus domain of the utterance. This view rests on the assumption thatthe syntactic tree undergoes partition into areas which are treated differently insemantics (for instance, Diesing (1992)).

Now the idea is to integrate this pragmatic differentiation into semanticsand make vP a relevant transition point for semantic composition too. The con-cept of Semantic Phase, that is proposed in this paper is considered an instanceof the general Principle of Hierarchical abstraction. Taking into account theparallel to the syntactic as well as the pragmatic concept, phase abstractionwould become a candidate for a general interface principle.

2.2 Perfective vs. Imperfective ViewpointIn case of the semantic phase concept, phase transition is supposed to consistin a change of the temporal granularity of the model when leaving vP. Moreprecisely, while at a point of semantic processing inside vP, the situation ap-pears to be internally structured into different temporal phases, outside vP it isseen as an unstructured single whole.

These two different views of a situation can be considered a structuralrealization of the concepts of imperfective and perfective aspect. According toComrie (1976), aspectual categories are different ways of viewing the internalconstituency of a situation. ‘Perfectivity indicates the view of the situation asa single whole, without distinction of the various separate phases that makeup that situation; while the imperfective pays essential attention to the internalstructure of the situation’ (ibid.).

Obviously, changing from imperfective to perfective view at the level ofvP is a way of hierarchically abstracting from details and reducing complexity

Sinn und Bedeutung 15 - [PDF Document] (313)

312 Gründer

of the model and of the representation.

2.3 The Proposal in a NutshellIn the context of the paper, the capacity to describe the world at different levelsof granularity is considered a symptom of the context-sensitivity of natural lan-guage. It is explained as a means to reduce complexity of semantic processingby applying the Principle of Hierarchical Abstraction.

Granularity shifts proceed via underspecifying situations and situationaldescriptions with respect to certain temporal aspects of their interpretation.Conceptual details concerning the internal temporal constituency of a situationare only available during local processing inside the new information focus ofan utterance. They are abstracted away as soon as processing reaches the back-ground domain of the utterance, where the information needs to be globallyhandled and brought into relation to the overall semantic context.

To have available two different conceptual views of a situation makesnatural language a very flexible descriptive means that can be very precise andvery effective at the same time.

In connection with the problem of ambiguities in temporal adverbial mod-ification the idea is the following. The puzzling flexibility in the semantics oftemporal adverbials is due to the granularity of temporal meaning. The im-perfective or perfective view of a situation is chosen as an attaching point forthe adverbial depending on the context of its use inside or outside the new in-formation focus of an utterance. Applying the identical adverbial to the samesituation represented at different levels of granularity causes the entire groupof characteristic interpretational contrasts.

3 The Semantic Framework3.1 Situations as Regular LanguagesWhat is needed to formally analyze the semantic contrast for the adverb againin the way sketched above is a semantic framework that is decompositional. Asituation has to be represented not just as an indivisible atom, but its differenttemporal parts need to be taken into account and made accessible by the for-malism. Additionally, there should be the possibility to implement the idea ofdifferent levels of granularity. Thereby, internal structure of a situation can beincluded or abstracted away by decision.

A modern approach to event semantics that could serve well as a basis forformal implementation is Finite-state Temporal Semantics of Fernando (2003,2004, 2006) Fernando (2003, 2004, 2006). In Fernando‘s theory, a situationalconcept is formalized as a Regular Language.

Sinn und Bedeutung 15 - [PDF Document] (314)

Semantic Phase and again 313

Given a finite set Φ of formulas, a symbol σ of such language consists ofa non-contradictory subset of Φ , which non-exhaustively describes what holdstrue at some single point in time. The symbols are combined via the basicregular operations concatenation σ1σ2, alternation σ1+σ2 and iteration σ∗1 (orσ+1 for non-empty iteration) to form regular expressions which define a regular

language as a set of strings. Negation of symbols is defined in the style of DeMorgan as: ¬ = Φ ; ¬ φ1,...,φn = ¬φ1 + ... + ¬φn .

A simple example (by Fernando) is given below. Take the symbols to besnapshots of a camera; then each string can be viewed as a temporal sequenceof such snapshots. With respect to the given example ‘rain from dawn to dusk’this means, that the formalization of the situational concept starts with a pictureon the left, on which there can be seen rain and dawn, followed by a finitenumber of pictures in the middle on which there is rain, and ended by one onthe right that shows rain and dusk.

(7) Λ (rain from dawn to dusk) = rain, dawn rain ∗ rain, dusk

Therefore, a situation is represented not just as an atom, but its internal statesare taken into account as well. No abstract BECOME-operator needs to beused, since the concept just directly mirrors the temporal path of the event.

The model of such a language is given by a Kripke Frame with partialvaluations. More precisely, the interpretational basis consists of a set of statesthat are partial valuations over a set of variables A, the carrier of a first-orderstructure.

Additionally, Fernando includes time variables in language and groundsthem in the model by the help of δ -points. That means, instead of the contin-uum of the real numbers, moments in time get modeled by non-open intervals(r- δ

2 , r+ δ

2 ). This strategy is motivated by the intuition that the precision of ac-tual observations always is finite. The choice of the extension of the δ -pointsdetermines a certain temporal granularity of the model.

3.2 Situational ClassesFernando‘s central idea for a definition of aspectual features is to formallybase it on the symbols α(L) and ω(L) that start and finish a given language,respectively. They serve to encode the property of a situational type of beinginitially or finally bounded or unbounded. If the condition α(L) is immediatelyswitched after the first stage an initial boundary is marked; if α(L) is preservedthe concept is initially unbounded. In the same way ω(L) can be used to marka final boundary, reading the string from right to left in that case. Aspectualfeatures, according to Fernando, then just enumerate all the possibilities for acorresponding concept to be bounded or unbounded in that sense.

Sinn und Bedeutung 15 - [PDF Document] (315)

314 Gründer

Definition 1. Aspectual Features:

telic (L) = ¬ω(L)+

iter (L) = ω(L)+

prog (L) = ¬α(L)+

reten (L) = α(L)+

Let us assume a situational concept has a minimal length of three sym-bols, in other words, every situation consists of a beginning, a middle part, andan end. On this perspective, the four classical aspectual classes are derivable asthe set of logically possible cross-combinations of the four aspectual featuresas defined above.

Below, the corresponding properties of being initially or finally bounded,are marked by using a short binary code, with the first digit referring to thebeginning, the second to the ending, and 1 and 0 indicating the presence orabsence of a boundary, respectively. If we let a and o refer to the two boundarymarking propositions inside the symbols α(L) and ω(L) we get the abstractcharacterizations on the very right.

Definition 2. Aspectual Classes:

state: reten, iter (0 0) a a, o + o

activity: prog, iter (1 0) a ¬a, o + ¬a, o

achievement: reten, telic (0 1) a, ¬o a, ¬o + o

accomplishment: prog, telic (1 1) a ¬a, ¬o + ¬a, o

The following translations, which give formalizations within the frame-work of some concrete examples, may serve as an illustration. (For the sake ofabbreviation, ¬a is suppressed in presence of o on the basis of obvious entail-ment relations.)

(8) a. Λ (be silly) = be silly(x) be silly(x) + be silly(x)

b. Λ (swim) = ¬∃y 6= /0 (swim(y)) ∃y 6= /0 (swim(y)) + ∃y 6= /0 (swim(y))

c. Λ (reach the summit) = ¬(be at summ.(x)) ¬(be at summ.(x)) + be at summ.(x)

d. Λ (build a tower) = ¬∃y ≤ t (build(y)) ∃y ≤ t (build(y)), ¬build(t) + build(t)

Sinn und Bedeutung 15 - [PDF Document] (316)

Semantic Phase and again 315

In example (8a) the relevant proposition - being a and o at the same time- refers to a state of mind of the subject. The valuation of the proposition re-mains unchanged for finitely many states, and so no boundaries are markedfor the state-concept. In the representation of the activity concept in (8b) thevariable is referring to parts of a spatial path the subject is taking. Here, an ini-tial boundary exists due to the change of the truth value of the a/o-propositionfrom the first to the second state. Since after this immediate switch the valu-ation of the proposition then remains unchanged for finitely many states, nofinal boundary is marked, and the situational concept therefore is an iterativeor atelic one. In (8c) the spatial position of the subject is what matter for acharacterization of the phases of the achievement. The change in truth valueof the o-proposition from the second last to the last state marks a final bound-ary, and therefore makes the situational concept telic. But there exists no initialboundary. Finally, in (8d) the constant t is referring to the tower, the values ofthe variable y are the parts of the tower that were already constructed. Condi-tion α (¬∃y ≤ t (build(y))) changes its truth value right after start; conditiono (build(t)) just before the end. Accordingly, the accomplishment-concept isinitially as well as finally bounded.1

4 The Semantic Analysis4.1 The Basic ConceptsThe possibility within the framework to change the granularity of the modelallows a direct formal implementation of the concept of semantic phase, thatwas proposed in Section 2. For a representation of the imperfective view ofa situation, that takes into account its internal temporal structure, the repre-sentation mainly looks like the formula presented in the previous section, justwith the subject included. Below there is the imperfective version of the con-cept ‘John open the door’ (infinitive) from the initial examples (o refers to theopening-angle of the door, j the subject).

(9) ¬∃x ≤ o (open (j, x)) ∃x ≤ o (open (j, x)), ¬open (o) + open (o)

1 In order to differentiate result states that are reversible (‘open the door’) from those that arenot, Fernando marks a set of inertial formula, that hold until a force is applied to stop them holding(‘build a tower’) or cannot be stopped at all (‘write an article’). For Fernando this concept of inertiais relevant in connection with the definition of the perfect, but it also plays a role for temporaladverbial modification. For instance, application of again should be blocked in cases where aresult is strictly inertial. In what follows this problem will be of minor interest though. The mainfocus of the formalization of the initial examples will be on the contrast between telic and atelicconcepts with respect to the interpretational effects they show in connection with the adverb again.

Sinn und Bedeutung 15 - [PDF Document] (317)

316 Gründer

In order to get a perfective view of the same situation, that abstracts awayfrom its internal structure, one has to increase granularity of the model untilthe entire situation just falls into one state. Metaphorically, the snapshot takenby our camera has an increased duration of exposure such that the entire eventcan be taken by a single picture. Accordingly, after of phase transition at thelevel of vP, the above representation in (9) gets changed and condensed in thefollowing way.

(10) * j open the door

For an analysis of the meaning of the adverb again I assume the followingsingle representation. The formula refers to the last symbol ω(L) of the regularexpression that represents the situational concept the adverb is supposed tomodify.

(11) Λ (again) = ω(L) *

The adverbial concept will get combined with the situational concept by simpleconcatenation. By just writing the two respective string one after the other inthat way, the temporal presuppositional character of the meaning of again iscaptured directly.

4.2 Again and Telic Situational ConceptsNow let us see how the contrast between the repetitive and restitutive reading,which again shows when applied to a telic situational concept, derives on thatbasis.

In case of the restitutive reading, the adverb gets interpreted before phasetransition and with respect to the imperfective view of the situation. In this con-stellation, the symbol ω(L), that is used by the modifier, is just the result stateof the event. Consequently, after modification the concept mirrors a temporalcourse where the result of the event held already at an earlier point in time. Be-low this is spelled out for the initial example ‘John open the door again’. Here,an accomplishment is chosen, but for an achievement the mechanism wouldobviously work quite parallel.2

(12) open (o) * ¬∃x ≤ o (open (j, x)) ∃x ≤ o (open (j, x)), ¬open (o) + open (o)

In contrast, the repetitive reading follows from applying the adverb after phasetransition. Now, the situation is represented from a perfective point of view. Inresult, the relevant symbol ω(L), that it taken by the adverbial, comprises the

2 Tempus is not included in this article, but can easily be following the formalization available inFinite-state Temporal Semantics.

Sinn und Bedeutung 15 - [PDF Document] (318)

Semantic Phase and again 317

description of the full situation, that was condensed into a single state of themodel. Accordingly, the regular expression that comes up after modificationdescribes a temporal course where the entire event with the identical subjectalready took place at some previous point in time.

(13) j open the door * j open the door

This means, the identical semantic representation for the adverb again can beused to derive both possible readings. Note, that since the imperfective and theperfective view of the a situation are both just regular expressions, no problemsof type-shift arise in the formalism used here.

4.3 Again and Atelic Situational ConceptsAs was said at the beginning, in connection with atelic situational conceptsno similar semantic contrast appears, but again just always expresses plainrepetition. Let us check whether this empirical fact can be accounted for by theproposed theory.

In the case of atelic states and activities, the last symbol of the imperfec-tive concept does not mark a final boundary. This means that no result state isdescribed but just a continuation of the state or activity phase that characterizesthe situation, respectively. Consequently, it does not make a real difference ininterpretation whether the adverbial gets applied to the imperfective or perfec-tive view of an atelic situation. Below both readings are spelled out for a stativeconcept ‘John be silly again’ and an activity concept ‘John swim again’.


a. be silly(x) * be silly(x) be silly(x) + be silly(x)

b. j silly * j silly


a. ∃y 6= /0 (swim(j, y)) * ¬∃y 6= /0 (swim(j, y)) ∃y 6= /0 (swim(j, y)) + ∃y 6= /0 (swim(j, y))

b. j swim * j swim

5 Syntax (Informal Sketch)5.1 General BackgroundFor reasons of space, the syntactic part of the theory can not be presented informal detail, but we will have to restrict ourselves to giving some generalideas and informal explanations. For a full formal analysis we refer the readerto Gründer (2009) or other material to appear. The framework used there is

Sinn und Bedeutung 15 - [PDF Document] (319)

318 Gründer

an adapted and extended version of Dynamic Syntax of Kempson, Meyer-Violand Gabbay (2001).

To supplement this semantic approach by a fitting syntactic theory, theright-peripheral adverbials in question are assumed to be orphans. Orphansare constituents that are independent of their host sentence in syntax. Prosodicinformation, reflecting the contextual status of the constituents, is mediatingbetween underspecified syntactic and specified semantic structure. Since therelevant information is read off locally on the constituents of the utterance,contextual principles can be put at work without having to implement heavymechanisms on discourse level.

This idea is methodologically challenging, since in the standard gram-matical systems, following the example of Chomsky, semantic interpretation isfully determined syntactically. Pragmatic considerations take place outside thereal grammar formalism and after semantic processing only. And there is, forprinciple reason, no interaction between the different components of the gram-mar in a way that phonetics would mirror pragmatics and would interact withsyntax in order to determine semantics. But following the considerations fromthe beginning of the paper, this kind of interaction is just what is needed for ananalysis of the meaning of adverbial modifiers, for instance again.

5.2 Dynamic Scope Resolution by ProsodyIn the formalism of Dynamic Syntax, that is used in Gründer (2009) as a formalbasis, syntactic processing is seen as progressive and goal-driven enrichmentof some partial, underspecified structure through stepwise parse of a string ofwords. Information is built up on a left-to-right basis relative to some con-text against which choices may be made as the construction process proceeds.Words are the processing units of the parser, and they include their syntacticinformation in form of a simple program that effects changes in a tree-structurethat is growing top-down. For the placement of the elements in the tree, tree-addresses are used. A number of processing rules govern the integration andfurther processing of information.

Among the several adaptions and extensions we made to the original sys-tem of Dynamic Syntax, the most relevant one is the inclusion of prosodicinformation into the parsing process. This means that the input of words comesmarked with respect to accent, and accent marks have influence on the process-ing and placement of the information inside the tree structure.

In case of right-peripheral again the disambiguation process intuitivelyworks like this. At the point where scanning of the modifier is triggered, itdoes not actually get integrated by the syntactic rule, but its structural posi-tion remains underspecified. Now, depending on the prosodic marking of the

Sinn und Bedeutung 15 - [PDF Document] (320)

Semantic Phase and again 319

adverbial as plus or minus accented, the semantic information carried by theword is placed at different position inside the tree. Lower, at the level of vP, ifthere is no accent, or higher, at sentence level, if there is an accent on again.Consequently, the adverbial gets involved into the semantic form at the rightplace relative to the phase transition point, and therefore applies to either theimperfective or perfective view of the situation.

6 ConclusionThe paper offered an explanation of the semantic flexibility of the adverb againon the basis of the concept of granularity. Temporal granularity of meaningrefers to changes in the degree of conceptual detail for time-related aspectsof interpretation. Application of the non-ambiguous modifier to either the im-perfective or perfective view of the same situation can cause the two differentreadings. The choice between both interpretations is not syntactically deter-mined, but it is made relative to the semantic context that is reflected locally bythe prosody of the utterance.

If one considers the results of the paper from a more general theoreticalperspective, then the investigation of adverbial modifiers was shown to havetheoretical depth as well as the capacity to illuminate systematic processes atthe interfaces between syntax, semantics and pragmatics. Additionally, it couldgive a clear and concrete example for the relationship between linguistic theoryand general principles of cognition.

ReferencesChomsky, N. 2001. Deriation by phase. In Kenstowicz (ed.), Ken hale: A life

in language, 1–52. Cambridge: Mit Press.Chomsky, N. 2008. On phases. Ms., MIT.Comrie, B. 1976. Aspect. Cambridge: Cambridge University Press.Diesing, M. 1992. Indefinites. Cambridge, MA: MIT Press.Dowty, D. 1979. Word meaning and montague grammar. Dordrecht: Kluwer.Fabricius-Hansen, C. 2001. Wi(e)der and again(st). In C. Fery & W. Sternefeld

(eds.), Audiatur vox sapientia. a festschrift for arnim von stechow, 101–130. Berlin: Akademie Verlag.

Fernando, T. 2003. Finite-state descriptions for temporal semantics. In Pro-ceedings of the fifth international workshop on computational semantics(iwcs-5), 122–136.

Fernando, T. 2004. A finite-state approach to events in natural language se-mantics. Journal of Logic and Computation 14(1). 79–92.

Fernando, T. 2006. Situations as strings. Electronic Notes in Theoretical Com-

Sinn und Bedeutung 15 - [PDF Document] (321)

320 Gründer

puter Science 165. 23–36.Gründer, S. 2009. Die Granularität zeitlicher Bedeutung - Ein dynamischer

Ansatz zur Syntax und Semantik Temporaler Adverbiale: University ofLeipzig, Faculty of Mathematics and Computer Science dissertation.

Jaeger, G. & R. Blutner. 2003. Competition and interpretation: The german ad-verb wieder („again“). In E. Lang, C. Maienborn & C. Fabricius-Hansen(eds.), Modifying adjuncts, 393–416. Berlin/New York: deGruyter.

Kempson, R., W. Meyer-Viol & D. Gabbay. 2001. Dynamic syntax. Oxford:Blackwell.

Pittner, K. 2003. Process, eventuality, and wieder/again. In E. Lang,C. Maienborn & C. Fabricius-Hansen (eds.), Modifying adjuncts, 365–392. Berlin/New York: deGruyter.

Reyle, U., A. Rossdeutscher & U. Kamp. 2008. Ups and downs in the theoryof temporal reference. Linguistics and Philosophy 5. 565–635.

von Stechow, A. 1996. The different readings of wieder „again“: A structuralaccount. Journal of Semantics 13. 87–138.

Sinn und Bedeutung 15 - [PDF Document] (322)

Degree Modification in Russian Morphology: The Case of the Suffix -­‐ovat

Olga Kagan

Ben Gurion University of the Negev [emailprotected]

Sascha Alexeyenko University of Osnabrück


Abstract. In this paper, we investigate the semantics of the adjectival suffix -ovat in Russian. We argue that this suffix constitutes a morphological degree modifier and propose for it a formal analysis formulated within the framework of degree semantics. The suffix specifies that the degree to which a property holds of an object is slightly higher than the standard of comparison. A detailed consideration of different types of adjectives and standards of comparison available for these adjectives, in combination with the proposed analysis, allows us to account for the distribution of the suffix and for the range of arising interpretations.

1 Introduction: Data In this paper, we investigate the semantics of the adjectival suffix -ovat in Russian as in the following examples where it is applied to the adjectives dorogoj ‘expensive’ and vysokij ‘high’:

(1) a. Etot restoran okazalsja dlja nas dorog-ovat-ym. this restaurant turned_out for us expensive-ovat-M.INSTR ‘This restaurant turned out to be somewhat expensive for us.’ b. Takije kabluki dlja menja vysok-ovat-y. such heels for me high-ovat-PL.NOM ‘Such heels are somewhat too high for me.’

Intuitively, the interpretation associated with -ovat comes close to “a little bit too”: The prices in the restaurant in (1a) slightly exceeded the speaker’s expectations or average prices for restaurants of that type, but were still not simply too expensive such that the speaker was not able to pay them. Similarly, the heels in (1b) are somewhat too high for the speaker as to be absolutely comfortable or to look completely appropriate, however, they are only somewhat too high rather than just too high.

However, it seems that -ovat can make different contributions with different adjectives. While with dorogoj ‘expensive’ and vysokij ‘high’ in (1)

In Reich, Ingo et al. (eds.), Proceedings of Sinn & Bedeutung 15, pp. 321–335. Universaar – Saarland University Press: Saarbrücken, Germany, 2011.

Sinn und Bedeutung 15 - [PDF Document] (323)

322 Kagan & Alexeyenko

its meaning is comparable to that of the English slightly too, this is not the case with adjectives like sladkij ‘sweet’ or vlažnyj ‘wet’ as below:

(2) a. Po utram on pjot proxladnyj sladk-ovat-yj čaj. at mornings he drinks cool sweet-ovat-M.ACC tea ‘In the mornings, he drinks cool sweetish tea.’ b. Lena protjorla mebel’ vlažn-ovat-oj trjapkoj. Lena wiped furniture wet-ovat-F.INSTR duster ‘Lena wiped the furniture with a wettish duster.’

In examples in (2), -ovat implies that the property lexicalized by the stem holds of the argument to an intuitively low degree, e.g., the tea in (2a) is not really sweet, rather it is only somewhat sweetish. In other words, ‘sweet + -ovat’ does not entail ‘sweet’. The same holds for vlažnovatyj in (2b), which implies that objects, of which it is true, are not properly wet, but are not really dry either. Again, this means the lack of entailment to the meaning of the unmodified positive form vlažnyj ‘wet’.

Another interesting fact concerning the distribution of -ovat is that it can be attached to some adjectives but not to others: (a) it is incompatible with non-gradable adjectives, such as žyvoj ‘alive’,

mjortvyj ‘dead’, or čjotnyj ‘even’; thus, *žyvovatyj, *mertvovatyj, and *čjotnovatyj are not acceptable forms;

(b) in many pairs of positive and negative adjectives that lexicalize scales with the same dimension, the suffix can be attached to one member of the pair only, namely, to the one that conventionally has a negative connotation, e.g. grjaznovatyj (dirty+ -ovat) / *čistovatyj (clean+ -ovat); ploxovatyj (bad+ -ovat) / *xoroševatyj (good+ -ovat); dorogovatyj (expensive+ -ovat) / *deše(vo)vatyj (cheap+ -ovat); slabovatyj (weak + -ovat) / *sil’novatyj (strong + -ovat);

(c) in some other pairs, by contrast, both the positive and the negative member can combine with the suffix, e.g. dlinnovatyj (long+ -ovat) / korotkovatyj (short+ -ovat); šyrokovatyj (broad + -ovat) / uzkovatyj (narrow + -ovat); tjaželovatyj (heavy + -ovat) / legkovatyj (light + -ovat).

Finally, we assume that in some cases the suffix cannot attach to a root due to purely morpho-phonological factors, such as, e.g., the length of the word or euphony, cf. *interesnovatyj ‘interesting + -ovat’, *agressivnovatyj ‘aggressive + -ovat’, *prostodušnovatyj ‘simple-minded + -ovat’. For in-stance, the suffix is unlikely to combine with a stem that consists of more than two syllables. However, in what follows we will ignore such cases and concentrate on the semantic-pragmatic nature of the suffix.

Sinn und Bedeutung 15 - [PDF Document] (324)

The Russian Suffix -ovat as Degree Modifier 323

The core idea of our analysis is that the suffix -ovat functions as a degree modifier, similarly to comparative morphemes. We argue that it imposes a relation between two degrees on the scale lexicalized by the adjectival root. One of them is the maximal degree to which the property holds of the individual argument of the adjective. It is entailed to slightly exceed the other one, namely, the standard of comparison.

The paper is structured as follows. In Section 2, we briefly discuss the necessary theoretic considerations about scales, degrees, and standards of comparison, mainly based on work by Kennedy & McNally (2005) and Heim (2000). In Section 3, we set forth our analysis that accounts for the data from Section 1. We systematically discuss different types of adjectives, both non-gradable adjectives and various sub-classes of gradable adjectives, and different types of standards of comparison, which -ovat can apply to. Finally, Section 4 concludes the discussion.

2 Scales and Standards 2.1 Types of Scales Following a number of studies on the semantics of gradable adjectives (Cruse 1980, Winter & Rotstein 2004, Kennedy & McNally 2005, Kennedy & Levin 2007, among many others), we assume that the meanings of gradable adjectives can be characterized in terms of scales and degrees, defining a scale as a set of degrees totally ordered along some dimension. Depending on the structure of the scale, the following subtypes of scales have usually been distinguished: (a) totally open scales: such scales do not have minimal or maximal

points, and, therefore, adjectives that map their arguments along such scales are not compatible with degree modifiers that pick out end points, e.g., absolutely and completely for the maximal degree, slightly and partially for the minimal degree: - tall, expensive, deep, glad, heavy, etc.

(b) upper-bound closed scales: the property has a maximal possible degree, which constitutes the upper bound of the scale; the corresponding adjectives can be modified by absolutely and completely: - clean, dry, flat, straight, etc.

Sinn und Bedeutung 15 - [PDF Document] (325)

324 Kagan & Alexeyenko

(c) lower-bound closed scales1: the property is instantiated to at least a smallest value, which follows the zero degree at the lower bound of the scale; adjectives with underlying scales of this type can be modified by slightly and partially: - dirty2, wet, bumpy, dangerous, etc.

Applying this distinction to Russian adjectives, we can see that adjectives like dorogoj ‘expensive’ / dešovyj ‘cheap’ and vysokij ‘high’ / nizkij ‘low’ lexicalize totally open scales lacking both a minimal and a maximal degree, since neither soveršenno ‘absolutely’ nor slegka ‘slightly’ is compatible with either of them:

(3) a. #soveršenno vysokij / #slegka vysokij #absolutely high / #slightly high b. #soveršenno nizkij / #slegka nizkij #absolutely low / #slightly low c. à highness nizkij vysokij d. à lowness vysokij nizkij

Adjective pairs like grjaznyj ‘dirty’ / čistyj ‘clean’ and sladkij ‘sweet’ / nesladkij ‘not sweet’ lexicalize partially closed scales. More precisely, ‘clean’ and ‘not sweet’ map their arguments along upper-bound closed scales (with the maximal degree at the upper bound), while ‘dirty’ and ‘sweet’ map their arguments along lower-bound closed scales (with the minimal degree at the lower bound):

(4) a. soveršenno čistyj / #slegka čistyj absolutely clean / #slightly clean b. #soveršenno grjaznyj / slegka grjaznyj #absolutely dirty / slightly dirty c. à cleanness (max) grjaznyj čistyj d. à dirtiness (min) čistyj grjaznyj 1 Yoon (1996) and Rotstein & Winter (2004) alternatively use the terms “total” and “partial” for antonymous adjectives lexicalizing upper- and lower-bound closed scales respectively. 2 Antonymous members in pairs of gradable adjectives (such as clean and dirty) map their arguments onto scales with the same dimension and the same degrees. However, their scales are different, since the respective orderings are inverse (but see Kennedy 2001 and Kennedy & McNally 2005 for an alternative view, on which such antonyms lexicalize the same scale but involve positive versus negative degrees).

Sinn und Bedeutung 15 - [PDF Document] (326)

The Russian Suffix -ovat as Degree Modifier 325

An important implication that Kennedy & McNally (2005) draw from the fact of differences in the scale structure concerns the nature of the standard of comparison.

Gradable adjectives, which map their arguments along totally open scales, are claimed to have a context-dependent standard of comparison, therefore they are called relative gradable adjectives. This type of standard of comparison, called distributional standard throughout the paper, is determined with respect to the domain of the adjective, i.e., based on the distribution in the class of objects, which constitute the comparison set in the respective context. The objects, which the positive form is true of, “stand out” with respect to the property that the adjective encodes.

By contrast, gradable adjectives, which map their arguments onto degrees on scales closed from at least one end, are supposed to have a context-independent absolute standard of comparison that corresponds to the minimal or maximal degree on the scale. These adjectives have been dubbed absolute gradable adjectives. The standard of adjectives with upper-bound closed scales corresponds to the maximal degree, i.e., they require their arguments to possess a maximal amount of property they describe (e.g. čistyj ‘clean’ and nesladkij ‘not sweet’). Adjectives with lower-bound closed scales have their standard at the minimal degree on the scale, i.e., they require their arguments to possess some minimal degree of the relevant property (e.g. grjaznyj ‘dirty’ and sladkij ‘sweet’).

In addition, we assume the existence of a functional standard for all types of gradable adjectives, which is determined relative to some purpose relevant in the context of utterance.

2.2 Functional Standard The functional standard is the maximal degree on the interval consisting of degrees that are compatible with the requirements of the situation. The most typical case when this standard is invoked is the modification by the degree modifier too (Heim 2000, Meier 2003). A sentence of the form x is too P means, roughly, that x is characterized by the property P to a degree that is higher “than is compatible with certain (contextually given) goals or desires” (Heim 2000: 19). These goals or desires can be provided explicitly, as in (5) and (6), or need be inferred from the context, as in (7):

(5) Our truck is too tall to go through this tunnel.

(6) This concert is too long to burn to a single CD.

(7) These heels are too high.

Sinn und Bedeutung 15 - [PDF Document] (327)

326 Kagan & Alexeyenko

The contextually relevant goals for the sentences in (5)-(6) are packed in a to-phrase: to go through a certain tunnel in (5) and to burn a concert to a CD in (6). The corresponding functional standards in these contexts, i.e., the maximal degrees of height and length that are compatible with the respective goals, although not provided explicitly, are recoverable from the situation. For the sentence in (5), this degree corresponds to the height of the tunnel; for the sentence in (6), it corresponds to the volume of the CD.

By contrast, the contextually relevant goal/desire for the sentence in (7) remains implicit and can vary from context to context. It may be the desire that shoes should be comfortable or the goal to look appropriate in a society, or, essentially, any other contextual requirement.

Heim (2000) captures the meaning component contributed by too by assigning this item a modal semantics (see also Meier 2003 for a similar modal analysis). The analysis she proposes is provided in (8):

(8) [[ too ]] w = λP<s,dt> . max(P(w)) > max{d: Ǝw' ∈ Acc(w): P(w')(d) = 1}

The construction x is too P implies that the maximal degree to which P holds of x in the reference world w is higher than the maximal degree to which P holds of x in any possible world that stands in a particular accessibility relation to w. The accessibility relation Acc maps a world w to a set of worlds in which the contextually specified purposes or desires are achieved or satisfied, and which are similar to w in other relevant respects. As de-monstrated above, the nature of the accessibility relation varies from context to context.

What has been called the functional standard throughout this paper is represented in Heim’s analysis in (8) as max{d: Ǝw' ∈ Acc(w): P(w')(d) = 1}, i.e., it is the maximal degree that is compatible with the situation re-quirements. For the sake of simplicity, below we will abbreviate this formula simply as C to refer to the functional standard, following Nakanishi (2004).

3 A Unified Analysis of -­‐ovat We propose that the suffix -ovat is a morphological degree modifier. It provides information regarding the degree to which the argument possesses the property lexicalized by the stem. The suffix imposes a relation between this degree and the standard of comparison. The semantics of -ovat is provided in (9):

(9) λP<d,et>λd’dλxe . max{d: P(d)(x)} > d’ ˄ (max{d: P(d)(x)} – d’ < dc)

Sinn und Bedeutung 15 - [PDF Document] (328)

The Russian Suffix -ovat as Degree Modifier 327

In prose, the suffix specifies that the maximal degree d to which a property P holds of an individual x is higher than another degree d’, i.e., the standard of comparison, whose source will be discussed below. It further specifies that the difference between the two degrees is relatively low, i.e., lower than dc, which represents a contextually provided expectation value. Thus, the suffix fulfills the double function of (i) imposing a relation between two degrees on a scale and (ii) vaguely measuring the difference between these degrees.

Below, we argue that -ovat consistently contributes the semantics in (9). The different sub-meanings of the suffix, discussed in Section 1, arise by virtue of the fact that the suffix can apply to different types of standards of comparison. In what follows, we systematically discuss the application of -ovat to adjectives with underlying scales of different types and different standards of comparison.

3.1 Non-­‐Gradable Adjectives The analysis predicts correctly that -ovat cannot attach to non-gradable adjectives. Degree modifiers require their adjectival argument to be gradable (Kennedy & McNally 2005). If it is not gradable, a type mismatch occurs. The adjectives *žyvoj ‘alive’, mjortvyj ‘dead’, and čjotnyj ‘even’ are not gradable and, therefore, they are of type <e,t> (the property type). But the suffix requires an argument of type <d,<e,t>>. Hence the unacceptability of such forms as *žyvovatyj, *mjortvovatyj, and *čjotnovatyj.3 On a more intuitive level, the adjectival stems do not provide a degree which could then be compared to the standard of comparison.

3.2 Gradable Adjectives: Absolute Standard

3.2.1 Lower-­‐Bound Closed Scales If the scale lexicalized by the stem is lower closed and, thus, has a minimal value, it is to this value that the suffix applies. Thus, the lowest degree on the scale functions as the standard of comparison.

Let’s illustrate the application of -ovat to an adjective with an underlying lower-bound closed scale, such as, for instance, sladkij ‘sweet’. The compatibility with slegka ‘slightly’ but not with soveršenno ‘absolutely’ indicates that this adjective lexicalizes a lower-bound closed scale, which has

3 Note that the unacceptability of *žyvovatyj and *mjortvovatyj cannot be explained phonologically by the fact that the stem ends in the consonant -v-. This is shown by the acceptability of such adjectives as krivovatyj and čerstvovatyj, whose stems end in -v- as well. Further, the same kind of phonological explanation could not apply to the non-existence of such adjectives as *čjotnovatyj.

Sinn und Bedeutung 15 - [PDF Document] (329)

328 Kagan & Alexeyenko

a minimal value and no maximal value (an entity can be absolutely not sweet, but not absolutely sweet):

(10) Čaj slegka / #soveršenno sladkij. tea slightly / absolutely sweet ‘The tea is slightly / #absolutely sweet.’

The adjective sladkovatyj denotes the property of being slightly sweet, e.g. sladkovatyj čaj is tea that contains a small amount of sugar. The argument of sladkovatyj is entailed to possess sweetness to a degree that is slightly higher than the minimum. This meaning is derived in the following way:

(a) The semantics of sweet is provided in (11a). (b) The result of application of -ovat to the stem sladk- reveals the

representation in (11b). The maximal degree to which the argument of the resulting adjective is sweet slightly exceeds the standard of comparison, i.e., the minimal degree on the scale.

(c) The resulting function applies to the standard of comparison associated with the stem, and we get the meaning in (11c).

(11) a. λdλx . sweet(d)(x) b. λd’λx . max{d: sweet(d)(x)} > d’ ˄ (max{d: sweet(d)(x)} – d’ < dc) c. λx . max{d: sweet(d)(x)} > min(Ssweet) ˄ (max{d: sweet(d)(x)} –

min(Ssweet) < dc) d. à sweetness

nesladkij sladkij

sladkovatyj The figure in (11d) graphically represents the relations between the denotations of nesladkij ‘not sweet’, sladkij ‘sweet’, and sladkovatyj. We assume that the lower boundary on the scale of sweetness represents zero sweetness, i.e., corresponds to the absence of the property. In order for an object to fall under the denotation of sladkij ‘sweet’, it has to reach a particular degree of sweetness. Finally, an object counts as sladkovatyj if the degree of its sweetness is higher than the minimal point on the scale, but not considerably higher than this point.4

4 That is, we assume that sladkovatyj is outside of the denotation of sladkij. For further discussion of this issue, see Kagan & Alexeyenko (2010), Section 4.2.

Sinn und Bedeutung 15 - [PDF Document] (330)

The Russian Suffix -ovat as Degree Modifier 329

A further example that illustrates the application of -ovat to an adjective with a lower-bound closed scale is grjaznyj ‘dirty’. The scale of dirtiness has a minimal value (corresponding to zero dirtiness, or absolute cleanliness) but no maximal value (there is no limit to how dirty one can get). The suffix -ovat applies to the minimal value on the underlying scale of this adjective, so that the resulting adjective, grijaznovatyj, denotes the property of being slightly dirty, i.e., slightly dirtier than an absolutely clean entity.

(12) [[ grijaznovatyj ]] = λx . max{d: dirty(d)(x)} > min(Sdirty) ˄ (max{d: dirty(d)(x)} – min(Sdirty) < dc)

Several additional examples of adjectives that lexicalize a scale with a lower boundary and can be modified by the suffix -ovat are provided below:

(13) vlažnovatyj (wet-ovat) ‘slightly wet’, gor’kovatyj (bitter-ovat) ‘slightly bitter’, ostrovatyj (spicy-ovat) ‘slightly spicy’, krivovatyj (crooked-ovat) ‘slightly crooked’, strannovatyj (strange-ovat) ‘somewhat strange’, grustnovatyj (sad-ovat) ‘a little bit sad’, etc.

3.2.2 Upper-­‐Bound Closed Scales When a scale has a maximal value, its upper boundary constitutes another potential standard of comparison for the application of -ovat. However, it turns out that -ovat fails to apply to this standard. Recall that the suffix en-sures that the property holds of an argument to a degree that is higher than the standard of comparison. Trivially, no degree can be higher than the maximal element on the scale.

An example of an adjective that lexicalizes an upper-bound closed scale is čistyj ‘clean’, which lacks a minimal value and whose maximal value corresponds to absolute cleanliness. This scale is almost identical to the one lexicalized by the antonymous adjective grjaznyj ‘dirty’ discussed in Section 3.2.1 above, except for the fact that the two scales are characterized by inverse ordering relations. Roughly, the higher an object is on the scale of cleanliness (i.e., the cleaner it is), the lower it is on the scale of dirtiness. We noted above that the scale of dirtiness has a minimal value but no maximal one. Correspondingly, the scale of cleanliness has a maximal but not a minimal value.

The adjective *čistovatyj does not exist. Formally, the unacceptability of this form can be explained as follows. The standard of the adjective čistyj ‘clean’ is the upper scale boundary. Thus, the application of the suffix -ovat to this adjective would render the semantics in (14a):

Sinn und Bedeutung 15 - [PDF Document] (331)

330 Kagan & Alexeyenko

(14) a. λx . max{d: clean(d)(x)} > max(Sclean) ˄ (max{d: clean(d)(x)} – max(Sclean) < dc) b. à cleanness

grjaznyj čistyj čistovatyj

Since no degree on the scale of cleanliness can be higher than max(Sclean), the requirement max{d: clean(d)(x)} > max(Sclean) cannot be satisfied. Therefore, -ovat cannot be felicitously applied.

We now have an explanation of the contrast between the existing grijaznovatyj and the non-existing *čistovatyj. The adjectives grjaznyj and čistyj are antonyms that lexicalize scales with the same dimension. The scales come with one and the same standard (absolute cleanliness), which cor-responds to the minimal value on Sdirty and the maximal value on Sclean. For both adjectives, this standard is a potential candidate for -ovat to apply to. Given the ordering that characterizes each scale, we get the following result. With grjaznyj, the application of the suffix produces the meaning ‘slightly dirtier than the minimum’, or ‘slightly dirtier than an absolutely clean entity’. This is an acceptable interpretation, and the adjective grjaznovatyj exists. With čistyj, the resulting meaning would be ‘slightly cleaner than the maxi-mum’, or ‘slightly cleaner than an absolutely clean entity’. This interpretation is ruled out, and so the adjective *čistovatyj does not exist.

Additional upper-bound closed adjectives that cannot combine with -ovat are provided below:

(15) *rovnovatyj (straight-ovat)5, *ploskovatyj (flat-ovat)6, *sveževatyj (fresh-ovat)7, etc.

It should be pointed out, however, that some upper-bound closed adjectives (such as e.g. suxoj ‘dry’) can combine with -ovat. We argue that this is possible because in such cases, the suffix applies to a different type of standard, namely, to the functional standard. This issue will be addressed in Section 3.3.2 below.

5 Compare to the acceptable krivovatyj (crooked-ovat). 6 The adjective exists under a different, idiomatic meaning. 7 The intended meaning here is one of being a fresh product; presumably, under this meaning, the adjective lexicalizes an upper closed scale, as it is possible to say Jeda absolutno svežaja ‘The food is absolutely fresh’.

Sinn und Bedeutung 15 - [PDF Document] (332)

The Russian Suffix -ovat as Degree Modifier 331

3.3 Gradable Adjectives: Functional Standard In Sections 2.1 and 2.2, we demonstrated that both relative gradable adjectives and absolute gradable adjectives have a standard of comparison other than the absolute and the distributional one respectively. This standard is not computed relative to the distribution in the class, nor does it constitute a (minimal or maximal) boundary on the underlying scale. Rather, it is determined relative to some contextually relevant goal or desire. In this section, we consider cases in which -ovat applies to the functional standard.

3.3.1 Open Scales Relative adjectives lexicalize scales that lack both a minimal and a maximal value. Thus, no absolute standard is available. Still the suffix -ovat is com-patible with adjectives of this kind. With such adjectives, -ovat applies to the functional standard. In particular, it specifies that the degree to which the property holds of an argument is slightly too high to be compatible with the requirements of the situation.8

The adjective velikovatyj ‘big/great-ovat’ illustrates our point. This adjective lexicalizes an open scale and denotes a property of being slightly too big for the present purpose.

(16) a. [[ velikovatyj ]] = λx . max{d: big(d)(x)} > C ˄ (max{d: big(d)(x)} – C< dc) b. à bigness

f velikovatyj

For instance, this adjective can be used to describe shoes that are too big for a given individual, or a piece of furniture that is too big to fit in a particular room. At the same time, the argument of velikovatyj is only en-tailed to be slightly too big for the current purposes. Therefore, if this prop-erty characterizes the shoes that one is trying for size, this may not yet be a reason not to buy them: insoles or socks could solve the problem.

Along with velikovatyj, -ovat applies to the functional standard with numerous relative adjectives, including the following:

(17) vysokovatyj (tall-ovat) ‘slightly too tall’, nizkovatyj (short/low-ovat) ‘slightly too short/low’, šyrokovatyj (wide-ovat) ‘slightly too wide’,

8 Interestingly, -ovat cannot apply to a distributional standard. In Kagan and Alexeyenko (2010) (cf. Section 4.1), we suggest that this is a result of the inherently vague nature of this standard.

Sinn und Bedeutung 15 - [PDF Document] (333)

332 Kagan & Alexeyenko

dlinnovatyj (long-ovat) ‘slightly too long’, dorogovatyj (expensive-ovat) ‘a little bit too expensive’, starovatyj (old-ovat) ‘somewhat too old’...

The negative connotation sometimes associated with -ovat comes from the cases when it applies to the functional standard. An excess and the resulting incompatibility with the requirements of the situation create the negative flavour.

3.3.2 (Partially) Closed Scales The suffix -ovat can also apply to the functional standard with some adjectives that lexicalize scales with a boundary. For instance, the adjective suxoj ‘dry’ lexicalizes an upper-bound close scale (an entity can be absolutely dry, but not absolutely wet). -ovat cannot take the maximal value as the standard, for reasons discussed in Section 3.2.2. An object cannot be drier than absolutely dry. However, -ovat can attach to this adjective applying to the functional standard. The resulting adjective denotes a property of being slightly too dry for the present purposes (for instance, a duster may be too dry for an efficient cleaning). The adjective suxovatyj thus receives the semantics in (18)9:

(18) a. [[ suxovatyj ]] = λx . max{d: dry(d)(x)} > C ˄ (max{d: dry(d)(x)} – C <dc) b. à dryness

f suxoj (abs. stnd.) suxovatyj

Turning to adjectives that lexicalize a lower-bound scale, they, too, appear to allow the application of -ovat to the functional standard. For instance, it has been mentioned above that the adjective ostrovatyj (spicy-ovat) can mean ‘slightly spicy’. However, it may also mean ‘somewhat spicier than desirable in the given context’, as illustrated in (19):

(19) Etot sup dlja menja neskol’ko ostrovat. this soup for me somewhat spicy-ovat ‘This soup is somewhat too spicy for me.’

9 Of course, this raises the question of why such adjectives as *čistovatyj do not exist. We have seen why the suffix cannot take the maximal value on the scale as the standard, but why can it not apply to a functional standard, triggering an entailment that the argument is too clean for some purpose? This issue is addressed in Section 3.3.3.

Sinn und Bedeutung 15 - [PDF Document] (334)

The Russian Suffix -ovat as Degree Modifier 333

We therefore propose that ostrovatyj is ambiguous between the following two readings, which differ in terms of the standard selected by the suffix:

(20) a. [[ ostrovatyj1 ]] = λx . max{d: spicy(d)(x)} > min(Sspicy) ˄ (max{d: spicy(d)(x)} – min(Sspicy) < dc) b. [[ ostrovatyj2 ]] = λx . max{d: spicy(d)(x)} > C ˄ (max{d: spicy(d)(x)}

– C < dc)

Under (20a), the adjective denotes the property of being just a little bit spicy. Objects that are included in its denotation do not lack the property of spiciness but have it to a low degree. In turn, (20b) represents the property of being slightly exceeding the functional standard for spiciness, i.e., being slightly more spicy than desirable in the given context.

3.3.3 Conventionalized Gaps: Adjectives with a Positive Connotation If -ovat can apply to the functional standard with absolute adjectives, as demonstrated in Section 3.3.2, why do the words *čistovatyj (clean-ovat) or *rovnovatyj (straight-ovat) not exist? Further, why do we get the asymmetry with such relative antonyms as the following: ploxovatyj (bad-ovat) - *xoroševatyj (good-ovat), slabovatyj (weak-ovat) - *sil’novatyj (strong-ovat), glupovatyj (stupid-ovat) - *umnovatyj (clever-ovat)?

Note that in these pairs the stems consistently denote properties one of which is conventionally viewed as positive and the other one, as negative. That is, by default, it is good to be clever but not to be stupid, and being strong is judged to be preferable over being weak. Analogously, clean is better than dirty. Once such a conventional opposition is present, -ovat is typically compatible only with the member of the pair that carries a negative connotation. Apparently, with these pairs of adjectives, the attachment of the suffix and the resulting interpretation is governed not only by contextual but also by conventional considerations. It is conventionally determined for certain dimensions an excess in what direction is likely to be undesirable. Roughly, ‘worse than desirable’ is much more likely than ‘better than desirable’, ‘weaker than desirable’ is more likely than ‘stronger than desirable’, etc. Conventionally, by default, a high degree of cleanliness, cleverness, goodness, etc. is judged as a good thing, which makes these adjectives less easily compatible with the negative flavour of “a higher degree than desirable”, which is contributed by -ovat.

Of course, in an appropriate context, it is possible to conceptualize of an individual being “too good”, “too strong”, and even “too clever”. Therefore, the degree modifier too is perfectly compatible with such adjectives. However, due to the fact that -ovat is a derivational morpheme, which com-

Sinn und Bedeutung 15 - [PDF Document] (335)

334 Kagan & Alexeyenko

bines with the stem in the course of word formation, it is more sensitive to lexical and conventional restrictions. Therefore, it does not easily apply to properties whose degree is unlikely to be higher than desirable. In contrast, too, which is an independent lexical item that combines with an adjective at a much higher level of the derivation, can override the conventionalized preferences of the stem in an appropriate context.

Interestingly, if a polysemous adjective is inherently likely to receive a negative connotation under only one of its sub-meanings, this sub-meaning will be compatible with -ovat. For instance, the adjective prostovatyj (simple-ovat) sounds strange when modifying a problem or a question. Here, we have the positive/negative contrast *prostovatyj/složnovatyj (simple-ovat /strong-ovat) of the kind discussed above. But the adjective prostoj may also be used to modify one’s personality, in which case it receives the meaning ‘simple-minded’. This sub-meaning inherently receives a negative conno-tation, and the word prostovatyj is perfectly acceptable if used in this sense.10

4 Conclusion To sum up, in this paper we have investigated the semantics of the adjectival suffix -ovat in Russian. We argued that this suffix constitutes a morpho-logical degree modifier and proposed for it a formal analysis formulated within the framework of degree semantics. The suffix specifies that the degree to which a property holds of an object is slightly higher than the stand-ard of comparison. A detailed consideration of different types of adjectives and standards of comparison available for these adjectives, in combination with the proposed analysis, allows us to account for the distribution of the suffix and for the range of arising interpretations.

References Cruse, David. 1986. Lexical semantics. Cambridge, UK: Cambridge

University Press. Heim, Irene. 2000. Degree Operators and Scope. In Proceedings of SALT 10,

40–64. Ithaca: CLC Publications. Kagan, Olga & Sascha Alexeyenko. 2010. Degree Modification in Russian

Morphology: The Case of the Suffix -ovat. In Proceedings of IATL 26. Kennedy, Chris. 2001. Polar Opposition and the Ontology of ‘Degrees’.

Linguistics and Philosophy 24. 33–70.

10 Similarly, mjagkij ‘gentle’ used with respect to people generally has a positive connotation. By contrast, mjagkovatyj (gentle-ovat) refers to someone who is too gentle, getting spineless.

Sinn und Bedeutung 15 - [PDF Document] (336)

The Russian Suffix -ovat as Degree Modifier 335

Kennedy, Chris & Beth Levin. 2007. Measure of change: The adjectival core of degree achievements. In L. McNally and C. Kennedy (Eds.), Adjectives and adverbs: Syntax, semantics and discourse. Oxford: Oxford University Press.

Kennedy, Chris & Louise McNally. 2005. Scale Structure, Degree Mod-ification, and the Semantics of Gradable Predicates. Language 81(2). 345–381.

Meier, Cécile. 2003. The Meaning of too, enough and so… that. Natural Language Semantics 11. 69–107.

Nakanishi, Kimiko. 2004. On Comparative Quantification in the Verbal Domain. In Proceedings of SALT 14, 179–196. Ithaca: CLC Publications.

Rappaport Hovav, Malka. 2009. Scalar Roots and Their Results. Handout of a talk given at Roots Workshop at Stuttgart, June 2009.

Rotstein, Carmen & Yoad Winter. 2004. Total adjectives vs. partial adjectives: Scale structure and higher-order modifiers. Natural Language Semantics 12. 259–288.

Yoon, Youngeun. 1996. Total and partial predicates and the weak and strong interpretations. Natural Language Semantics 4. 217–236.

Sinn und Bedeutung 15 - [PDF Document] (337)

Sinn und Bedeutung 15 - [PDF Document] (338)

On the Relation between Coherence Relations and Anaphoric Demonstratives in German*

Elsi Kaiser

University of Southern California [emailprotected]

Abstract. Recent research suggests that pronoun interpretation is guided by the semantic coherence relations between clauses. However, it is not yet well-understood whether coherence effects extend to other anaphoric expressions beyond pronouns. We report an experiment on German, a language in which human antecedents can be referred to both with personal and demonstrative pronouns. How do anaphoric demonstratives – whose referential properties have been argued to be complementary to pronouns – fit into coherence-based views? Our results suggest that although coherence does not modulate the antecedent choice of anaphoric demonstratives to the same extent that it in-fluences pronoun interpretation, demonstratives interact with coherence-related processing by guiding comprehenders’ expectations of coherence relations.

1 Introduction One of the most-researched challenges of language comprehension has to do with the interpretation of pronouns and other ‘underspecified’ referring expressions. An expression such she, it or this is semantically under-informative: on its own, it does not provide sufficient information to identify the intended referent. However, we encounter these kinds of forms very frequently in both written and spoken language and are able to interpret them without difficulties. In this paper, I report a psycholinguistic experiment that aims to shed light on the processes involved in reference resolution by investigating the referential properties of two kinds of underspecified forms in German, namely personal pronouns (er, sie ‘s/he’) and demonstrative pronouns (der, die). The results show that to understand the referential properties of these forms, we need to take into account the semantic coherence relations between sentences, but that pronouns and demonstratives interact with coherence-related processing in different ways. * Thanks to Petra Horvath, Erik Bresch and Samuli Heilala for assistance with coding the data. Thanks are also due to Caroline Féry, Thomas Weskott and Robin Hörnig who made it possible for me to collect data at the University of Potsdam. I would also like to thank the SuB audience for valuable comments. This research was partly supported by NIH grant 1R01HD061457.

In Reich, Ingo et al. (eds.), Proceedings of Sinn & Bedeutung 15, pp. 337–351. Universaar – Saarland University Press: Saarbrücken, Germany, 2011.

Sinn und Bedeutung 15 - [PDF Document] (339)

338 Kaiser

Various approaches to anaphor resolution have been proposed. In this section, I review some key approaches and summarize recent evidence regarding the importance of inter-sentential semantic relations. According to attention-based approaches, the interpretation of pronouns and other forms is guided by a correlation between referring expressions and the salience/accessibility of the antecedent (e.g., Givón 1983, Ariel 1990, Gundel et al. 1993): The most reduced referring expressions (e.g. unstressed pronouns) refer to the most salient/accessible entities, and demonstrative pronouns and other fuller forms refer to less salient entities. Salience is often regarded as being influenced by grammatical role, with subjects more salient than objects, or topicality, with topics more salient than non-topics. However, there are empirical complications for the subjecthood=salience view. In particular, Smyth (1994) and Chambers & Smyth (1998) found structural parallelism effects: pronouns prefer referents in parallel syntactic positions.

Recent evidence suggests that neither attention-based nor parallelism-oriented approaches are sufficient, and argues for coherence-based ap-proaches. According to coherence accounts, the use and interpretation of pronouns depends on the semantic relation between the pronoun-containing clause and the antecedent-containing clause. These approaches view anaphor resolution as a by-product of general inferencing/reasoning about relations between clauses (Hobbs 1979, Kehler 2002, Kehler, Kertz, Rohde & Elman 2008). To see how coherence relations influence pronoun interpretation, con-sider ex.(1). In principle, ‘him’ could refer to Phil or to Stanley. However, if the relation between the sentences is semantically parallel ex.(1a), people tend to interpret ‘him’ as referring to the parallel argument, Stanley (Kertz, Kehler & Elman 2006): Comprehenders construe the two events as similar, i.e., Stanley was tickled and was poked. In contrast, if the relation between the two clauses is a result relation ex.(1b), people are more likely to interpret ‘him’ as referring to the subject Phil (Kertz et al. 2006).

(1) a. Phil tickled Stanley, and (similarly) Liz poked him. Parallel relation: him => bias to object (Stanley) b. Phil tickled Stanley, and (as a result) Liz poked him. Result relation: him => bias to subject (Phil)

As shown in ex.(2), subject pronouns are also sensitive to coherence. When the relation between the two clauses is result/cause-effect, as in (2a), subject-position pronouns prefer the preceding object (Kertz et al. 2006). In contrast, when the relation between the two clauses is a temporal narrative relation (one event preceded the other but did not cause it), Kehler (2002) notes that we may observe a subject bias ex.(2b), see also Kertz et al. (2006).

Sinn und Bedeutung 15 - [PDF Document] (340)

Coherence Relations and Anaphor Resolution 339

(2) a. Phil tickled Stanley, and as a result he laughed uncontrollably. Result relation: he => bias to object (Stanley) b. Phil tickled Stanley, and then he laughed at Mark’s joke. Narrative relation: he => bias to subject (Phil)

It is important to note that particular coherence relations do not always push pronouns to antecedents with certain grammatical roles. What matters is the semantics of the clauses and their relation. E.g., a subject pronoun in a result relation does not have to refer to the preceding object: Both (3a) and (3b) involve a result relation but he can refer to the preceding subject or object:

(3) a. Peter snapped at Ethan, and he sulked the rest of the afternoon. Result relation: he => bias to object (Kertz et al. 2006) b. Peter snapped at Ethan, and he felt guilty the rest of the afternoon. Result relation: he => bias to subject (Kertz et al. 2006)

As a whole, a number of studies indicate that a successful account of pronoun interpretation needs to take into account the semantic coherence relations that hold between clauses (e.g. Wolf, Gibson & Desmet 2004, Kertz et al. 2006, Kehler et al. 2008, Rohde & Kehler 2008, Kaiser 2009).

2 What about Other Anaphoric Forms? Existing work on coherence effects has focused primarily on the behavior of overt pronouns. However, other referential forms are also used to refer to previously-mentioned entities, including null pronouns, demonstratives and definite NPs. This brings up the question of whether coherence sensitivity also extends to other referring expressions. Are coherence effects a core property of all kinds of reference tracking, regardless of form, or are they a specific phenomenon that only occurs with certain anaphoric forms? In particular, could it be the case that only the default anaphoric form in a particular language exhibits sensitivity to coherence relations, and that other forms are governed by factors such as grammatical role?

On a general level, existing work suggests that referring expressions can indeed differ in how sensitive they are to different kinds of information. For example, although Kaiser & Trueswell (2008) did not look specifically at coherence, they found that Finnish personal pronouns and demonstratives differ in how much they ‘care’ about a potential antecedent’s grammatical role vs. its linear position/discourse-status (see also Kaiser (in press) on Dutch). Recently, Ueno & Kehler (2010) found that Japanese null pronouns are primarily sensitive to grammatical role whereas overt pronouns are more sensitive to verb aspect (see also Rohde & Kehler 2008).

Sinn und Bedeutung 15 - [PDF Document] (341)

340 Kaiser

Thus, as a whole, existing research indicates that referring expressions can be asymmetrical in terms of what kind of information they are sensitive to. In the present paper, to test whether referring expressions differ in how sensitive they are to coherence information, I compared the behavior of personal pronouns and demonstrative pronouns in German. Although this research is on German, it is potentially relevant to other languages as well, since demonstrative pronouns are used to refer anaphorically to human antecedents in many languages (e.g. Kibrik 1996 on Russian, Comrie 1997 on Dutch, Kaiser & Vihman 2010 on Estonian, Himmelmann 1996).

2.1 Existing Work on Pronouns and Demonstratives In German, both personal pronouns and demonstrative pronouns can be used to refer back to human antecedents1, but their referential biases are different: Pronouns are described as preferring an antecedent in subject position, while demonstratives prefer non-subject antedecents, as shown in (4a,b).

(4) a. Paul1 wollte mit Peter2 Tennis spielen. Paul1 wanted to play tennis with Peter2. b. Doch {er1/der2} war krank. But {he1/DEM2} was sick. (Bosch & Umbach 2007)

Personal pronouns can be regarded as more default/less marked than demonstratives, based on relative frequency (Bosch, Rozario & Zhao 2003).2 Before investigating the referential properties of these forms, let us consider some background facts. Demonstratives often look like definite determiners (e.g. der Mann ‘the man’), but differ in certain cases/numbers (see Bosch, Katz & Umbach 2007). Although demonstratives are sometimes felt to have a pejorative tone and are more common in informal registers, they also occur in written text (Bosch et al. 2007) and are not consistently felt to be pejorative.

Let us now take a look at the referential properties of pronouns and (short) demonstratives. Given that both forms can refer to human antecedents, what guides the division of labor between them? Based on the NEGRA corpus of written German, Bosch, Rozario and Zhao (2003) found that when the antecedent is in the immediately preceding sentence, pronouns refer to a nominative element in 86.7% of the cases, whereas demonstratives refer to a nominative element in only 23.6% of the cases. (Nominative is the default subject case). Based on these findings, Bosch et al. put forth the

1 Following Bosch et al (2003, 2007), I refer to anaphoric der and die as demonstrative pronouns, 2 German also has longer demonstrative pronouns (e.g., diese(r), jene(r)), which Bosch et al. (2007) describe as less frequent than the short forms (see Abraham 2006 for further discussion of diese(r)). I do not discuss these forms here, but regard them as an important area for future work.

Sinn und Bedeutung 15 - [PDF Document] (342)

Coherence Relations and Anaphor Resolution 341

Complementarity Hypothesis: “Anaphoric personal pronouns prefer referents that are established as discourse topics, while demonstratives prefer non-topical referents.” They regard nominative case (subjecthood) as signaling topicality, so the Complementarity Hypothesis treats pronouns and demonstratives as having complementarity grammatical-role and infor-mation-structural preferences. However, psycholinguistic experiments by Bosch, Katz & Umbach (2007) found that although demonstratives prefer object antecedents, pronouns do not exhibit a clear subject preference. Consequently, Bosch & Umbach (2007) argue that pronouns prefer discourse topics and demonstratives avoid topics. Topicality is also mentioned by Zifoun et al. (1997, vol.1: 558), who suggest pronouns are used for referents already established as topics, and demonstratives for referents that are new information or contrastive. Abraham (2006) also analyzes pronouns as involving topic continuation, and demonstratives as markers of topic shift.

3 How Do Anaphoric Demonstratives Fit into Coherence-­‐Based Views?

Let us now consider how demonstrative pronouns could fit into coherence-based views of reference resolution. As we saw in Section 1, a growing body of work on English points to a close relation between the interpretation of pronouns and the coherence relations that hold between the pronoun-containing clause and the antecedent-containing clause.

One of the key aims of the experiment reported in this paper is to test whether German personal pronouns show the coherence sensitivity observed in English. In addition to providing evidence of crosslinguistic replicability, this question is of interest given that German has a particular anaphoric form specialized for object reference, i.e., the demonstrative pronoun. English has no comparable expression.3 Thus, perhaps English subject-position pronouns can be pushed towards object interpretations by result relations (see ex.2) because there exists no dedicated object-referring anaphor? Continuing with this reasoning, one might expect that the existence of a special object-referring form in German, the demonstrative, means that personal pronouns cannot be pushed to refer to an object antecedent.

3 A possible candidate for English, former/latter, is rare and highly marked. The distinction between stressed/unstressed pronouns is sometimes mentioned as being similar to the pronoun/ demonstrative distinction (see Bosch et al. 2003). However, existing work on English stressed pronouns led to conflicting claims: Some (e.g. Kameyama 1999) argue for a salience-based approach, but others claim use of stressed pronouns is driven by contrast (e.g. de Hoop 2003).

Sinn und Bedeutung 15 - [PDF Document] (343)

342 Kaiser

Furthermore, we wanted to investigate how referential dependencies influence comprehenders’ assumptions about coherence. More specifically, even if the interpretation of anaphoric demonstratives is strongly object-biased and perhaps not influenced by coherence, can demonstratives nevertheless influence comprehenders’ expectations about coherence? If a connective is ambiguous between a result relation and a narrative relation, can com-prehenders’ assumptions about which relation to activate be influenced by the referential biases of the anaphor? I discuss this below.

3.1 Inferring Coherence Relations from Anaphoric Dependencies The question of whether particular referential dependencies can shape comprehenders’ expectations about coherence relations relates to work by Rohde (2008) and Rohde & Kehler (2008). They noted that, if different coherence relations are associated with different referential dependencies, we might expect that encountering a particular referential pattern will lead people to expect a particular coherence relation. In other words, we might find that not only do coherence relations influence the interpretation of pronouns, as argued by Kehler (2002) and Kehler et al. (2008), but that the interpretation of pronouns also influences the construal of coherence relations. Indeed, in a series of sentence continuation studies, Rohde and Kehler showed this to be the case, and thus argued for a bidirectional relation between pronoun interpretation and coherence establishment.

Because the logic of their experiments is relevant for my work, let us take a closer look at one of their studies. In Rohde (2008)’s sentence-completion study, participants read short fragments consisting of a sentence and the first word of the next sentence (ex.5a,b), and wrote continuations. The verbs in the first clause were NP1 implicit causality verbs (Garvey & Caramazza 1974). Prior work has shown that when a sentence with an NP1 implicit causality verb is followed by an ‘explanation’ continuation (ex.5a), the continuation is likely to start with reference to the first noun in the initial clause (the subject). Given this well-known pattern, Rohde hypothesized:

If comprehenders use cues about who has been mentioned next to determine which coherence relation is likely to be operative, then an NP1-referring pronoun is predicted to shift comprehenders’ expectations in favor of NP1-biased coherence relations, whereas an NP2-referring pronoun is predicted to shift expectations in favor of NP2-biased coherence relations. (Rohde 2008:87)

(5) a. John infuriated Mary. He… cheated at Scrabble. b. John infuriated Mary. She… told him to take a hike.

Sinn und Bedeutung 15 - [PDF Document] (344)

Coherence Relations and Anaphor Resolution 343

The results showed that (i) when the gender of the pronoun signalled a subject antecedent, participants were more likely to provide a continuation that constituted an explanation relation (ex.5a), and (ii) when the gender of the pronoun signalled an object antecedent, participants wrote more result continuations (ex.5b). Rohde (2008) concludes that “comprehenders use information about which referent has been mentioned next to update their expectations about the operative coherence relation” (p.97).

As a class of referential forms, pronouns are known to be rather flexible; a pronoun can be used to refer to a preceding subject or preceding object. Thus, one could argue that a pronoun that refers clearly to the preceding subject (or preceding object) provides information about the coherence relation, because the form could also have referred to the other potential antecedent. In contrast, demonstrative pronouns are ‘pickier’ and more rigid in that they have a strong preference for the object antecedent. Thus, I wanted to find out how rigid demonstratives actually are and whether a rigidly object-referring form could also influence participants’ inferences about what coherence relation is operative.

4 Experiment To look at the scope of coherence effects in German, I used a sentence completion task where participants read a sentence followed by a prompt word (e.g. X tickled Y and then he…) and provided a continuation sentence. In critical items, the prompt word was a pronoun or a demonstrative. This task is a combination of comprehension and production: Participants need to interpret the prompt anaphor before they can provide a continuation.

4.1 Methods, Design Twenty native German speakers (mostly students at the University of Potsdam, Germany) participated in a sentence-completion task with 16 targets and 32 fillers. Targets consisted of an initial transitive clause followed by a connective and either a personal pronoun or a demonstrative pronoun:

(6) Die Schauspielerin hat die Schneiderin gekitzelt und dann hat The actress has the seamstress tickled and then has

{sie/die} {pronoun/demonstrative}… ‘The actress tickled the seamstress and then {PRO/DEM}…’

Participants were asked to provide natural-sounding continuations. All target sentences mentioned two same-gender characters in the first clause (e.g. der Bauer ‘the farmer’, der Feuerwehrmann ‘the fireman’, die Kellnerin ‘the

Sinn und Bedeutung 15 - [PDF Document] (345)

344 Kaiser

waitress’, die Friseurin ‘the hairdresser (f.)’). The verbs were action/ agent-patient verbs (as defined by Stevenson, Crawley & Kleinman, 1994).4 As shown in (6), targets contained the connective dann ‘then’ which – like English ‘then’ – is ambiguous between a narrative interpretation and a result interpretation. Use of a connective that is ambiguous between these two readings is crucial, as it allows us to see whether participants’ interpretation of the connective is influenced by the nature of the anaphoric form.

In addition to using the ambiguous connective dann ‘then’, I tested a clearly causal connective, demzufolge ‘therefore, as a result’. However, some native speakers find this connective to be unnatural/odd-sounding with certain types of causal sequences. In some respects, this connective is perhaps akin to English ‘thus’ or ‘hence.’ Thus, while I will briefly mention the results with demzufolge, I focus mostly on dann. More generally, the ways in which German resultative connectives (demzufolge, folglich, deswegen, infolgedessen, etc.) map to different kinds of causal relations is an interesting question (see also Pander Maat & Sanders 2001 on Dutch).

4.2 Research Questions First, to test whether German pronouns show the coherence sensitivity exhibited by pronouns in English, I wanted to see whether result relations would be associated with an increased proportion of object interpretations. When faced with a pronoun prompt, if a comprehender chooses to treat two clauses as being connected by a result relation (recall that the connective dann ‘then’ is ambiguous), does this push the pronoun away from the preceding subject – presumbaly prominent both due to its syntactic position and due to structural parallelism– and boost the rate of object interpretations? To investigate this, I analyzed the antecedents of pronouns depending on whether the relation between the clauses was a result or non-result relation.

Second, I wanted to find out whether the strong object bias that had been previously observed with German demonstrative pronouns would persist regardless of coherence relation and whether it would influence participants’ inferences about coherence relations. More specifically, in light of the behavior of English subject-position pronouns – namely that result relations tend to be associated with object reference (see Rohde 2008) and narrative relations tend to be associated with subject reference (suggested by Kehler 2002) – I wanted to see whether in German, demonstratives push comprehenders to expect a result relation. This is shown schematically in (7):

4 The perfect tense (aux + past participle) allowed us to include a verb + anaphor sequence in the second clause (German is verb-second) without constraining participants’ continuation options.

Sinn und Bedeutung 15 - [PDF Document] (346)

Coherence Relations and Anaphor Resolution 345

(7) a. X verbed Y and then pronoun… => then/next (narrative) b. X verbed Y and then demonstrative… => result

Thus, the prediction is that subject-preferring pronouns may trigger the ex-pectation that we are dealing with a narrative ‘next’ relation, and object-referring demonstratives may trigger an expectation of a result relation.

In addition to shedding light on the referential properties of German pronouns and demonstratives, the issues investigated here can contribute to our understanding of whether and how coherence effects relate to grammatical roles. Recall that work by Rohde (2008) and Rohde & Kehler (2008) suggests that encountering a particular grammatical-role-based ref-erential pattern (e.g. mention of preceding object) leads people to expect a particular coherence relation (e.g. result). However, as discussed with respect to ex.(3), a particular coherence relation does not force a pronoun to ‘point to’ a certain grammatical role: Following a cause-effect relation, a subject-position pronoun can refer to a preceding subject or object. This flexibility raises questions regarding the nature and robustness of the associations between certain kinds of referential dependencies and certain coherence relations. My experiment on German allows us to contribute to these issues by investigating how robustly a particular referential pattern leads people to expect a particular coherence relation – especially when the cue is in the form of a rather rigidly object-referring demonstrative pronoun.

4.3 Data Analysis Participants’ continuations were analyzed independently by two native German speakers blind to the aims of the experiment. A third blind coder’s analyses were used to resolve any disagreements. The continuations were analyzed for (i) whether the anaphoric expression (the prompt word) referred to the preceding subject, preceding object, or whether the antecedent was unclear. Coders also noted (ii) whether the demonstrative was used anaphor-ically or as a definite article, since the demonstrative prompts are ambiguous between these two construals. Furthermore, since the connective dann ‘then’ is ambiguous (ex.8a, b), coders analyzed each dann token individually to see (iii) whether it involved a result or non-result relation.

(8) a. The actress tickled the seamstress and then she sat down and learned her lines. [non-result, narrative relation]

b. The actress tickled the seamstress and then she laughed really hard for 10 minutes. [result relation]

Sinn und Bedeutung 15 - [PDF Document] (347)

346 Kaiser

5 Results and Discussion In this section, we first consider the results for the pronoun conditions (Section 5.1) and then the results for the demonstrative conditions (Section 5.2). At the end of this section, we consider the occasions on which participants used the demonstrative prompt as a definite article (Section 5.3), which occurred frequently, on 75.6% of all demonstrative trials.

5.1 Pronouns Overall, when dann is followed by a pronoun, there are more subject continuations (73.4%) than object continuations (26.6%). The proportion of subject continuations is significantly higher than chance (one-sample t-test, hypothesized mean 0.5 (50%), t1(18)=4.9, p<.001, t2(15)=2.699, p<.02).5 Now, taking a closer look at the data, Figure 1 (next page) shows the percentage of trials on which participants used the pronoun to refer to the preceding subject or object, grouped by whether the relation between the clauses was result or non-result. (20% of dann+pronoun trials were coded as ‘unclear antecedent’; they are excluded from analysis.)

Figure 1 reveals a clear relationship between coherence and choice of antecedent: When participants use the pronoun to refer to the preceding subject, we find mostly non-result relations (4.7% result relations, 68.8% non-result relations). However, when participants use the pronoun to refer to the preceding object, result relations are more frequent (23.4% result re-lations, 3.1% non-result relations). Looking separately at subject and object continuations, we find that the distribution of result vs. non-result relations differs significantly from chance for both kinds of continuations (p’s<.02).

5.2 Demonstratives Used Anaphorically Figure 2 (next page) shows the behavior of demonstrative pronouns when they are preceded by dann ‘then’ and used anaphorically. (Five percent of dann+anaphoric demonstrative trials had an unclear antecedent; they are excluded from these analyses.) Now, contrary to what we saw with pronouns, there are more object continuations than subject continuations: the gray bar is taller than the black bar (88.88% object continuations vs 11.11% subject

5 Some degrees of freedom vary due to empty cells. Also, recall that I also tested the more marked, specifically causal connective demzufolge ‘therefore’; these results are not shown in Figure 1. When pronouns are preceded by this connective, there is no subject advantage: There are 49.33% subject continuations and 50.66% object continuations. This asymmetry between the ambiguous dann ‘then’ and the causal demzufolge ‘therefore’ already suggests that result relations are associated with a boost in object interpretations.