Interfaces and Interface Conditions 9783110926002, 9783110195477

Citation preview

Interfaces and Interface Conditions

Language, Context, and Cognition Edited by Anita Steube

Volume 6

≥ Walter de Gruyter · Berlin · New York

Interfaces and Interface Conditions Edited by Andreas Späth

≥ Walter de Gruyter · Berlin · New York

앝 Printed on acid-free paper which falls within the guidelines 앪 of the ANSI to ensure permanence and durability.

Library of Congress Cataloging-in-Publication Data Interfaces and interface conditions / edited by Andreas Späth. p. cm. ⫺ (Language, context, and cognition, v. 6) Includes index. ISBN 978-3-11-019547-7 (alk. paper) 1. Grammar, Comparative and general ⫺ Syntax. 2. Semantics. 3. Prosodic analysis (Linguistics) 4. Psycholinguistics. I. Späth, Andreas, 1964⫺ P291.I43 2007 415⫺dc22 2007019552

ISBN 978-3-11-019547-7 Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available in the Internet at http://dnb.d-nb.de. ” Copyright 2007 by Walter de Gruyter GmbH & Co. KG, D-10785 Berlin All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording or any information storage and retrieval system, without permission in writing from the publisher. Printed in Germany Cover design: Christopher Schneider, Berlin Printing and binding: Hubert & Co. GmbH & Co. KG, Göttingen

Contents Introduction........................................................................................VII Semantic Form as Interface...................................................................1 Manfred Bierwisch Verb Meaning: How much Semantics is in the Lexicon?...................33

Johannes Dölling & Tatjana Heyde-Zybatow

'Was noch?' Navigating in Question Answer Discourse.....................77 Regine Eckardt Old and new propositions....................................................................97 Kerstin Schwabe Deriving selectional properties of 'exclamative' predicates..............115 Klaus Abels German w-clauses at the left and right periphery of copular sentences..........................................................................141 Ilse Zimmermann On the Syntax of Prepositional Phrases............................................157 Josef Bayer & Markus Bader The Structure of Events in Word Formation.....................................181 Rosemarie Lühr The lexical content of connectors and its interplay with intonation. An interim balance on sentential connection in discourse................199 Ewald Lang & Marcela Adamíková Contextual Boundness and Contrast in the Prague Dependency Treebank.......................................................................231 Eva Hajiþová, Petr Sgall & KateĜina Veselá On the Semantic Foundations of the Contrastive Focus Within a Lexicalist Approach...........................................................245 Andreas Späth

VI

Contents

Focus on focus: The brain's electrophysiological response to focus particles and accents in German..........................................277 Stefan Heim & Kai Alter Corpus- and psycholinguistic investigations of linguistic constraints on German object order..................................................299 Sandra Pappert, Johannes Schließer, Dirk P. Janssen & Thomas Pechmann Some contextual effects of aboutness topics in German...................329 Werner Frey Interfaces, (Non-)Compositionality, Information Structure and Presupposition............................................................................349 Ingolf Max Subject Index.....................................................................................379

Introduction In June 2004, the research group 'Linguistic Foundations of Cognitive Science: Linguistic and Conceptual Knowledge', funded by the Deutsche Forschungsgemeinschaft (FOR 349), organized the workshop 'Interface and Interface Conditions'. The workshop discussed questions that arise when a theory draws a sharp line between linguistic and extralinguistic knowledge. The meeting coincided with the 65th birthday of the leader of the research group, Prof. Dr. Anita Steube. The workshop was focussed on questions of the interface between the linguistic and the conceptual system on the one hand, and on questions of the borders between the various levels of the linguistic system, on the other hand. Since this complex of problems has always been of special interest to Anita Steube, both in her research work and in her teaching practice, the participants took this occasion to pay respect and gratitude to the meritorious works of Anita Steube. Within a modular conception of the linguistic system, interfaces are central to the theory of grammar. In accordance with the Minimalist Grammar of Noam Chomsky, the linguistic system is embedded in systems of performance, i.e. language is a part of the cognitive competence of human beings. With respect to the human language faculty, it is necessary to distinguish two systems of performance: the system of articulation and perception (A/P) and the conceptual-intentional system (C/I). The formation of linguistic structures provides well-formed representations as the input into the performative systems. Linguistic representations of each level of the grammatical system are determined by interface conditions. The syntactic derivation gets to a point, i.e. Spell-out, at which the formation of linguistic structure branches into two distinct representational levels. The two levels provide instructions for the systems of performance A/P and C/I. Firstly, the formation of a linear structure takes place at the surface structure of a sentence, i.e. linearization becomes manifest for syntax at Spell-out and at the same time in the Phonological Form (PF). Secondly, the interpretation of linguistic utterances results from hierarchical structures. At the level of Logical Form (LF), operations apply that do not influence the linearization in overt syntax any longer. These operations contribute to the formation of hierarchical relations and result in a rep-

VIII

Introduction

resentational form that provides the input information into the Semantic Form (SF). The articles of the present volume refer to the interface conditions of A/P and C/I, respectively. In particular, the authors focus on the conditions that provide the linguistic system for the interface conditions. The introductory article of Manfred Bierwisch 'Semantic Form as Interface' starts from the assumption that an interface is a formal means which enables the interaction of two separate systems. With respect to linguistic performance, we can reckon with an interface between articulation and perception in the sense that A/P is a complex system controlling articulation and perception. This interface regulates the production and recognition of phonetic signals. C/I is a more complex system of conceptual and intentional representations. By means of C/I the human organism interacts with its external and internal environment. The difference between A/P and C/I is, inter alia, that elements of A/P essentially occur as linear structures. However, C/I containts structures of different types. The article exemplifies the problem by means of data from English and German. For instance, the author discusses a German example, where it is possible to differentiate two types of activity on the base of different objects. In English, these activities are expressed by the same verbal predicate: Peter took off his hat/his pullover vs. Peter nahm seinen Hut ab/zog den Pullover aus. There are obviously general conditions that associate one and the same activity with systematically different semantic representations in different languages. Distinctions on the level of Semantic Form (SF) don't necessarily correspond with different interpretations on C/I. The article makes clear that an automatic correpondence between SF and the representations in C/I is indispensable and that it is in accordance with the essence of the interface-function. In the next contribution, Johannes Dölling and Tatjana HeydeZybatow focus on those conditions of C/I that units of the lexicon require. The considerations start from the fact that many verbs have more than one conceptual meaning representing the conventionalized meaning of the verbal lexical entry. The authors ask how much of the conceptual meaning contained in the utterance is supplied by the lexical meaning of a verb. The article follows different strategies. The goal is to derive as much information as possible from the lexicon. Further, the assumption that verbal semantics is at most confined to a few abstract fixings in the lexicon has to be confirmed. Finally, the number of lexical semantic representations for one and the same verb has to be reduced. The following example illustrates the aim of the article: Luisa

Introduction

IX

kochte die Suppe. 'Luisa cooked the soup.' vs. Die Suppe kochte. 'The soup cooked.' vs. Luisa kochte [irgendetwas]. 'Luisa cooked [something].' The sentences suggest that a verb like kochen 'cook' allows for a transitive and for two intransitive uses at the same time. So, what are the consequences for the definition of the lexical entry of the verb kochen in German? The authors suppose that the content of an utterance is encoded in a fragmentary, schematic structure called semantic form and can only be accessed by means of inferential operations. These operations provide the necessary enrichment of the conceptually underspecified semantic representation. Semantics and pragmatics are treated as two independent systems of information processing. Semantics as a part of the grammar computes the context-invariant meaning of an utterance. The subsequent pragmatic process of interpretation generates more specific meaning structures that depend on the context. This process is based on inferential operations and supports the combinations of various pieces of information originating from different cognitive sources. The article by Regine Eckardt 'Was noch' – Navigation in Question Answer Discours' follows. It discusses the German temporal particle noch and its semantic differentiation depending on the grammatical context of a sentence. Elements of the category of temporal particles refer to subsequent phases in a time interval, where a given proposition switches its truth value from true to false and vice versa. All uses of the particle noch rest on a scale that contains a referential interval that the speaker takes in focus. The article makes clear that not only LF seems to provide the conditions for C/I: Depending on question-answer contexts sentences have a certain prosodic structure, which becomes manifest at PF: ELSE hat noch ein Taschenmesser dabei 'ELSE has 'noch' a pocket knife'. In the present example, the focus placement triggers a switch of the meaning of the particle distinct from its temporal meaning. Thus, variation of the prosodic structure of one and the same sentence can contribute to the conditions for C/I and influence the conceptual interpretation of an utterance. In her contribution 'Old and new propositions', Kerstin Schwabe argues that the illocutionary operator ASSERT is associated with a declarative root clause. ASSERT always provides a new proposition. A new proposition is an argument of the matrix predicate which contributes to the denotation or the presuppositions of an old proposition, which corresponds to the given information. Therefore, a new proposition updates the Common Ground of the interlocutors. The article shows why certain German propositional predicates can embed verb-second clauses, while a negated proposition, for instance, can not: Anna glaubt,

X

Introduction

Hans kommt gerade 'Anna believes Hans is just coming' vs. *Anna glaubt nicht, Hans kommt gerade 'Anna doesn't believe Hans is just coming'. The article makes clear that declarative root clauses and verbfinal clauses do not have the same semantic meaning, and that their ordinary meaning does not denote a proposition. The author introduces the concept of a propositional situation sp, which is considered to be a discourse referent. The distinction between propositional situations sp and non-propositional situations si enables one to differentiate sppredicates such as believes, which refer to propositional situations, from si-predicates, which refer to non-propositional situations. The article by Klaus Abels 'Deriving selectional properties of 'exclamative' predicates' discusses restrictions that hold for the embedding of predicates in questions. The author explains why exclamative predicates do not embed polar interogatives and why the embedding of intensified questions in interrogatives is blocked: Peter is surprised at/*is wondering that Erna was at the party. The starting point of the analysis is the claim that polar questions are singleton sets of propositions. In this point, the author distinguishes himself from the analysis of Hamblin (1973) where polar questions are treated as two membered sets of propositions, i.e. the prosition p and its complement ™p. In the second part of the article, the author discusses presuppositions of exclamative predicates and questions with intensifiers. The article by Ilse Zimmermann 'German w-clauses at the left and right periphery of copular sentences' discusses cleft constructions. The author follows the general assumption that clefts contain a relative clause that is extraposed to the right periphery. Pseudoclefts, however, contain an interrogative clause at the left periphery. The syntactic analysis is based on Percus (1997) who considers an extraposed relative clause as belonging to a DP representing the expletive pronoun es at the level of Phonetic Form, e.g.: Was Philip schreibt, (das) ist ein Roman 'What Philip is writing (that) is a novel'. The syntactic structure of the example is the following: [CP [DP +def [CP Was Philip schreibt ]]+Top [CP[DP das ] [C' [C ist C ][ModP Mod [ ... [VP [CP daß Philip [DP einen Roman ]+Foc schreibt ]+Foc [V' tDP tV ]]]]]]]. If the resumptive pronoun is left out, the topical DP with the embedded interrogative w-clause will figure in SpecCP. The topical DP is base generated as an adjunct of CP, and tDP in V' is the trace of the pronoun in SpecCP. The following contribution from Josef Bayer und Markus Bader 'On the Syntax of Prepositional Phrases' analyses the grammatical status of a preposition (P) in comparison with other lexical heads. In addition, the distinction between P and PP is thematised. For instance, P obligatorily pied pipes, other categories don't. P can form a complex

Introduction

XI

head with a simple wh-item in copy movement: [Mit wem] glaubst du, [mit wem] Anita meint, [mit wem] wir uns treffen sollten? 'Who do you believe Anita thinks we should meet with?' A further characteristic, for example, is the phenomenon that a dative-assigning P combines with uninflected indefinites as their complements. The authors argue that P shares some features with lexical categories and some features with functional categories. P agrees with its complement with respect to certain features. When both P, as the probe, and XP, as the goal, have the same feature, then this feature is deleted on P. P and XP need not agree in all their features. If one of them lacks a feature that the other has, the former will be specified with this feature. In this respect, P differs from truly lexical categories like V. The article by Rosemarie Lühr 'The Structure of Events in Word Formation' analyses the internal structure of abstract verbal nouns in ancient Greek. In the historical development, gradual intrusion of nominal expressions at the cost of their verbal pendants took place. The author pursues the following question: Are there contextual elements in Thucydides' use of abstract verbal nouns that indicate how an event is anchored with respect to its internal or external temporal dimension as well as to its spatial dimension? Further, the contribution shows which complements of the basic verb are passed on to the abstract verbal noun. In Greek, the infinitive construction is a linguistic competitor for abstract verbal nouns amongst the non-finite verb. An infinitive construction can take complements. At the same time, both constructions have an article. The author asks the question whether both constructions are different linguistic means. If so, the question should arise, under what contextual conditions did Thucydides prefer the abstract verbal noun construction over the infinitive phrase? Ewald Lang and Marcela Adamíková discuss the relation between the grammatical properties of certain connectors and the prosodic make-up of the respective sentences. The authors restrict their considerations to a selection of adversative connectors. They analyse data from German and Slavic. The article shows again how the phonological structure of sentences influences the conceptual interpretation of an utterance in its discourse context. The main claim of the article is the following: what the connectors lack in specified lexical content might be compensated by syntactical and/or prosodical means. The authors show, for instance, that the following Slovak sentences have different discourse pragmatic functions only because of their different prosodic information: (i) Môj manžel [NE-vlastní]F auto, [ale má MOTORKU]F 'My husband [does not own]F a car, but [he has a motor-cycle]F'. (ii) Môj manžel [NE-vlastní]F auto, ale má [MOTORKU]F 'My husband [does

XII

Introduction

not own]F a car, but [a motor-cycle]F'. The authors put both sentences into restricted contexts and show that utterance (i) indicates Contrast, while (ii) is interpreted as a Correction. The different wideness of the focus in both sentences is supported by empirical data. The authors show that conjuncts of a co-ordinate construction form a special sort of structural context for each other. This context is available within a complex sentential category and is relevant for the morphosyntactic and the semantic interpretation of a co-ordinate construction. It holds at the same time that the connectors that link the conjuncts of a co-ordinate construction specify the syntactic and the semantic relations between the conjuncts as parts of a complex sentential category. But they are also sensitive to the discourse context outside the complex sentential category. The article by Eva Hajiþová, Petr Sgall and KateĜina Veselá 'Contextual Boundness and Contrast in the Prague Dependency Treebank' discusses the relationship between information structural notions on the conceptual and the grammatical levels. It explains and illustrates the Prague Functional Generative Description in connection with the Prague Dependency Treebank. The authors take a closer look on notions like 'contextually bound' and 'contextually free'. They distinguish between the layer of cognition and the language system. The opposition of 'given' vs. 'new' information and 'contrastiveness' are part of the layer of cognition. The notions of 'contextually bound' and 'contextually nonbound' items as well as 'topic' and 'focus' are part of the language system. The word order on the surface structure of a sentence is assumed to be a part of the morphemic structure, in which the representation of a sentence has the form of a string rather then of a tree. In his contribution 'On the Semantic Foundations of the Contrastive Focus Within a Lexicalist Approach' Andreas Späth scrutinizes the question, whether the mere assumption of an alternative set corresponds to the truth conditional meaning that the contrastive focus contributes to the sentence meaning. On the assumption that elements of a set of alternatives are extensional objects of the extralinguistic world, the author develops a functional lexical entry for contrastive focus. The entry as a part of the meaning of a contrasted constituent allows an inferential operation whose result is the assumption of an alternative set that has to contain at least two elements. The functional lexical entry for contrast enriches the meaning of the contrastively focused constituent. On the basis of selected phenomena, the article exemplifies how contrast operates on the level of lexical structure, e.g.: Otto trägt DIEses Hemd 'Otto is wearing THIS shirt'. The truth of the sentence depends on the existence of a proximal object, but not on the existence of a given distal

Introduction

XIII

object in the given discourse model. Thus, the partial meaning of the contrastive focus has to state the existence of a proximal object and of a distal object as well. What holds at the same time is that only the referent of the contrasted constituent fulfils the proposition. The starting point of the discussion in the article by Stefan Heim and Kai Alter 'Focus on focus: The brain's electrophysiological response to focus particles and accents in German' is the fact that prosodic pitch accents can be used as clues for the comprehension of syntactically ambiguous sentences. In a series of ERP-studies, the authors investigated whether the expectancy for pitch accents can also be established in sentences which are isolated from their context. The authors use the focus particle sogar 'even' that guides sentence interpretation by assigning both focus and pitch accent to the right-adjacent constituent. In German, this particle can also assign focus to an accented word to the left: Peter verspricht sogar ANNA zu arbeiten und das Büro zu putzen 'Peter promises even Anna to work and to clean the office vs. PETER verspricht sogar Anna zu arbeiten und das Büro zu putzen 'Even Peter promises Anna to work and the office to clean'. A second experiment shows effects of inappropriate accents in sentences containing the focus particle sogar 'even'. The aim of the article by Sandra Pappert, Johannes Schließer, Dirk P. Janssen and Thomas Pechmann 'Corpus- and psycholinguistic investigations of linguistic constraints on German object order' is to evaluate certain linguistic constraints as predictors of the frequency with which word order variants of double object sentences occur. The authors refer to a series of empirical investigations with respect to the relative order of accusative and dative objects in German. On the base of corpus queries, completion questionnaires, and self-paced reading experiments the authors empirically scrutinized to what extent linguistic constraints on German word order variation are valid. The experiments examine the validity of the following constraints: The Dative constraint [DAT < ACC], i.e. datives tend to precede accusatives; the Animacy contraint [+ANIM < –ANIM], i.e. constituents referring to animate objects tend to precede constituents referring to inanimate objects; the Definiteness contraint [+DEF < –DEF], i.e. definite constituents tend to precede indefinite ones. The first position of an argument-DP corresponds to the German prefield (Vorfeld). The subsequent DP is a constituent of the middle field. Werner Frey continues the discussion about the complex of word order. His article 'Some contextual effects of aboutness topics in German' investigates discourse related properties of topics in the middle field of a sentence. The author focuses, for instance, on the resolution

XIV

Introduction

of anaphoric elements and the compatibility of DPs in the middle field with the functions of a discourse topic. One example is constructed by means of cataphoric pronouns. They show that a sentence topic may have effects outside of its sentence: Eri war zufrieden. Heute wurde Kochi erwartungsgemäß nominiert 'He was satisfied today was Koch as expected nominated' but *Er1 war zufrieden. Heute wurde erwartungsgemäß Koch1 nominiert. The author gives some German examples which support the claim that the designation of an aboutness topic may have easily detectable effects on a preceding or a following sentence. For this reason, the author argues for a designated topic position in German. The article gives further evidence for the assumption that there are textual configurations which are only coherent if certain phrases are designated as topics. The final contribution 'Interfaces, (Non-)Compositionality, Information Structure and Presupposition' of Ingolf Max considers issues of information structure from the point of view of the focus-background structure of sentences. The starting point is the congruency of declarative sentences in the context of questions-answers pairs: WHO slept? – The CHILD slept but: *The child SLEPT. The author questions how we can formally explain this mismatch between an answer with respect to its question and how we can explain the congruency of questions and answers. Here the question arises whether information structure is an external pragmatic phenomenon or a purely linguistic one and whether information structure effects the meaning of simple sentences. The article shows how information structure can be explicated within a two-dimensional framework. Within this methodical frame, the author attempts to explain the mismatch in question-answer pairs. He differentiates between potential background presuppositions induced by information structure and real background presuppositions. This differentiation is based on a negation test using a negation operation, which is sensitive to the background of a sentence. Sentence negation is treated in the sense that the whole sentence is in its scope. Finally, the author demonstrates that potential background presuppositions are not necessarily potential lexical presuppositions. The present volume represents a survey of approaches to the complex problems of the differentiation in linguistic and extralinguistic, i.e. in conceptual knowledge and the interface conditions that make it possible to transform the former into the latter by means of the human performative systems. As mentioned above, the volume is the result of the workshop 'Interface and Interface Condition', held in honour of Anita Steube. The participants of the workshop and contributors of the pre-

Introduction

XV

sent volume of 'Language, Context & Cognition' thankfully dedicate their contributions to Anita Steube. We all hope for further prosperous cooperation and new discussions in order to solve similar puzzles on the interaction of the linguistic and extralinguistic systems in a broad theoretical perspective. Finally, we would like to thank all those who organized the workshop and the publication of this volume. In particular, we thank Andreas Bulk, Michael Rießler, David Dichelle and Ryan Young and for their technical assistence and for the proofreading of the English texts. We also thank the authors and reviewers of the articles for their friendly and reliable cooperation. Leipzig, May 2007 Andreas Späth

Manfred Bierwisch (Berlin)

Semantic Form as Interface 1 Preliminaries The term interface had a remarkable career over the past several decades, motivated largely by its use in computer science. Although the concept of a "surface common to two areas" (Oxford Advanced Learner's Dictionary, 1980) is intuitively clear enough, the range of its application is not very sharp and well defined, a "common surface" is open to a wide range of interpretations. As an initial guideline for the following considerations, the definition in (1) might be useful: (1)

Interface: a point at which independent systems or diverse groups interact; computer science: the point of interaction or communication between a computer and any other entity, such as a printer or human operator. (The American Heritage Dictionary 1992)

It should be clear enough that the notion of interface does not only have a strong metaphorical flavor, but that its application is therefore in most cases just a heuristic move. With this proviso in mind, it might be useful to explore some possibilities of its application with respect to a specific domain. To this effect, I will consider an interface as a formal means, by which two separate systems interact. Hence I will not use it for the link between two domains or parts of the same system. I will not talk e.g. of the interface between words and sentences or between sentence and text, just as we would not talk about the interface between a book and its pages. But it might be appropriate to assume an interface between e.g. articulation and perception or between language and space. More explicitly, I will assume that two systems A and B have an interface with the structure C, if the following holds: (2)

a. b.

A and B are structurally different in relevant respects. A and B share a system C, which represents a restricted isomorphism between partial structures and processes of A and B.

2

Manfred Bierwisch

c.

A interacts with B in terms of C, i.e. there are systematic effects of A on B (and vice versa), which are mediated through C.

This is still a rather provisional characterization, but it should be sufficient for the present purpose. For the sake of illustration, think of a pocket calculator and the obvious, but by no means trivial prerequisites on the basis of which it is to be operated. Evidently, the internal organization of the pocket calculator differs essentially from that of its putative user. But they share a subsystem by means of which they systematically interact. This system consists of the calculator's keys and display, both using the digits 0 to 9 with decimal coding. Further keys indicate arithmetical operations, the result of which is represented by the display. It is to be noted, by the way, that in this case the interface, by which A influences B is not identical to that by which system B returns the effects to A. In other words, the interface between two systems is not necessarily symmetrical. Notice furthermore that, in line with (2a), the interface structure will usually be related to rather different operations and structures within each of the two interacting systems. This is not at variance with a necessary partial equivalence, on which the interface is based. Without partial equivalence, systematic interaction between two systems would in fact be impossible. Thus the pocket calculator clearly realizes processes that are quite different from those of the human user of the machine, even though both are equivalent to the extent to which they are bound to and controlled by principles of arithmetic and decimal coding of numbers. In what follows, I will suggest that the heuristic use of the notion interface may shed some light on problems that are traditionally dealt with in terms of different levels of representation, such as Conceptual Structure and Logical or Semantic Form.

2 The Conceptual Framework According to uncontroversial assumptions, a natural language relates the form of its expressions to their meaning, and vice versa. Hence a natural language is a system of knowledge that creates a systematic correspondence between the elements of two rather different domains, viz. the form of linguistic expressions and their meanings. The structure of these two domains is based on corresponding systems of mental organization, that I will abbreviate, following Chomsky (1995), as A/P und C/I, respectively, where A/P is the complex system controlling articulation and perception, i.e. the production and recognition of signals, and C/I abbre-

Semantic Form as Interface

3

viates the even more complex systems of conceptual and intentional representations, in terms of which the human organism deals with its external and internal environment. Schematically, this can be represented as follows: (3) Signal A/P C/I Environment

In view of this schema, one might be tempted to apply the interfacemetaphor to the system of language as a whole, observing that its task is to connect the systems of A/P and C/I. This move would be heavily misleading, though, for at least two reasons. First, the basic principles of A/P must be essentially different from those of C/I, because elements of A/P are basically linear patterns of linguistic signals, while the structures of C/I are of a radically different type, capable to integrate conditions of all aspects of conceptually organized experience, as will be discussed in somewhat more detail below. Hence no reasonable subset of common elements provides a system of shared representations. Second, this heteromorphy of A/P and C/I is not an incidental deficit, but rather the fundamental condition for the unprecedented nature of the human language capacity, namely the emergence of a recursive mapping between two completely heterogeneous domains. The crucial point is that natural languages are combinatorial systems of symbolic signs, i.e. signs with a necessarily conventional relation between form and meaning and inherent recursive compositionality. In other words, the fundamental difference between A/P and C/I does not only prevent an interface between form and meaning, but it is for the same reason the source of the unique nature of the language faculty and its special status among the mental capacities. Hence a natural language must be construed as a mental system, that accounts for the relationship indicated in (3) in two respects, that must be clearly distinguished. First, it must accommodate the incommensurable structures of two different systems A/P and C/I, and second, it must compute a bridge between the accommodated structures. These different requirements are represented in (4), following the Minimalist Program of Chomsky (1995), which construes the internal or I-Language as a system that generates pairs of representations belonging to the Phonetic Form PF and the Logical Form LF, respectively, assuming that the correspondence between A/P and C/I is mediated through these linguistic levels of representation: (4)

A/P PF mo LF C/I  

I-Language

4

Manfred Bierwisch

The present paper will focus on the first of these requirements, i.e. on linguistic representations that are connected to extra-linguistic structures. Even though this connection is clearly determined by the respective extra-linguistic systems, it still depends crucially on intra-linguistic conditions, notably those on the sub-systems of Phonetic and Logical Form, to which the interface-metaphor most naturally applies. The conditions imposed by the system of I-Language on the Interface Representations include in particular the Principle of Representational Economy, which precludes the appearance of vacuous elements. With respect to the interface levels, Representational Economy amounts to the requirement (5): (5) Full Interpretation: LF and PF must not contain elements that are not interpretable in C/I and A/P, respectively.

The rationale of this principle is the assumption that there are elements or features of I-Language that have no extra-linguistic realization, but merely govern the computational processes that relate LF and PF. The details of these operations will not concern us here. As a consequence of the principle of Full Interpretation, neither PF nor LF allow for elements that are vacuous with respect to the external systems they interact with. Thus PF cannot contain e.g. degrees of stress on obstruents, as only syllables can participate in stress patterns. Conversely, though, interpretations of PF in A/P may exhibit systematic as well as accidental distinctions that cannot be reflected in PF. A simple case in point are the properties of voices of male as opposed to female speakers, or the different mood accompanying an utterance of Good morning! that the listener might well identity. This suggests that the relation between A/P and PF is a homomorphism in so far as PF abstracts away from A/P in systematic respects. Although this seems to be correct in crucial respects, it is not the whole story, though, as we will see shortly. Before turning to the main topic of this paper, viz. the interaction of I-Language with C/I, a terminological adjustment is to be made. Within the Minimalist Program, the interface of I-Language with C/I is taken to be LF, which is a syntactic level, made up from elements and categories to be chosen from the store of lexical items. Now, while at PF lexical items are considered as complex structures, consisting of features, segments, and syllables, at LF they are treated as basic elements – verbs, nouns, prepositions, etc. –, essentially without internal structure, although from a different perspective they are nothing but collections of features, including semantic ones. As the internal structure of lexical items and its systematic consequences will turn out to be essential for

Semantic Form as Interface

5

the conceptual interpretation of I-Language, I will assume it to be represented by the Semantic Form SF, a representational level that actually provides the interaction of I-Language with the conceptual and intentional conditions of language use. More specifically, I will consider lexical items like long, short, in, under, open, etc. not just as sets of features with no semantically relevant internal structure, but as systematic configurations, which are relevant for I-Language as well as its interpretation in C/I. Hence the schema (4) should be changed to (6): (6)

A/P PF mo SF C/I

This modification raises the question of how SF relates to syntax in general and LF in particular. Without going into the details, I will assume, however, that SF reduces to LF, if the distinction between lexical items and the SF-primes of which they are made up is ignored.1 To put it the other way round, SF emerges from LF, if the internal organization of lexical items with respect to their semantic interpretation is made explicit – just as PF "spells out" the internal structure of lexical items with respect to their phonetic interpretation. These remarks leave untouched a large number of questions, which are essential for the computational mapping between PF and SF. But the basic condition concerning the SFC/I-Interface should be clear enough: The conceptual interpretation of linguistic expressions has to take into account the internal structure of words like tall as opposed to short – just as the articulatory interpretation has to deal with the structure of tall as opposed to call.

3 Some Empirical Observations Besides obvious differences, there are important similarities between the two interfaces of I-Language. It might therefore be useful to have a look at the interpretation of PF, before the more controversial issues concerning interpretation of SF are taken up. It is a well known observation, that PF is generally more restrictive than A/P: It abstracts away from various types of information that are present in the signal and play a systematic role in speech processing. Properties distinguishing male and female voices, characteristics of personal identity, age or mood of the speaker are among the obvious factors 1

Taking the structure of SF, to which we turn below, as input for the mapping from SF to PF, we can conversely construe LF as the result of adding the morpho-syntactic categorization on the basis of the lexical items from which the SF-representation is made up.

6

Manfred Bierwisch

that appear in A/P and are recognized by speaker and hearer alike, but do not enter PF. This suggests that PF is a homomorphism with respect to the patterns belonging to A/P. The interaction of I-Language and A/P would thus reduce the information under perception, while in speech production A/P adds amendments to the information of PF. This is not the whole story, though. It is easy to imagine conditions of impoverished or defective acoustic conditions that nevertheless lead to complete, normal representations at PF. More interestingly, PF may systematically reflect distinctions that are not always realized in A/P. A case in point are stress patterns, whose systematic distinctions in PF often go beyond usual phonetic realizability, nevertheless becoming relevant, e.g. under metrical conditions of poetry. In short, PF reconciles conditions of I-Language and articulation/ perception. Its representations are made up from basic elements, which have a systematic status in both domains and are combined by principles that both domains allow for. Each of these domains may involve specifications, which the other ignores, but the shared conditions nevertheless guarantee automatic and reliable matching.2 Turning now to the semantic interface, we notice first a well known and extensively documented phenomenon, viz. the richness of conceptual distinctions compared to their rather sparse reflection in linguistic representations. Like PF, although in rather different respects, SF abstracts away from specifications provided by the conceptual interpretation. The point might be illustrated by two simple examples: (7)

I finished the book a while ago.

Depending on whether the speaker is the author or a reader of the book, whether the book is a novel or a research report, whether it comes from a public library or is private property, quite different situations can be referred to. Similarly for an appropriate utterance of (8), Paul may be disappointed by the new building he was shown around, or he may dislike the atmosphere among the colleagues he was working with or the job he had as an employee of the institute. Several other interpretations might come to mind, which are conceptually, but not linguistically distinct. (8) 2

Paul didn't like the institute. This observation is only apparently at variance with the principle of Full Interpretation indicated in (6). The point is not that PF may contain elements that are not interpretable in A/P, but rather that not all relations and distinctions among interpretable elements of PF are reflected in A/P.

Semantic Form as Interface

7

Notice that the different interpretations of cases like (7) or (8) do not turn on ambiguity of any sort, neither structural like (9a) nor lexical like (9b). (9)

a. b.

He had secret plans to leave. The post was unreliable.

Thus while the conceptually different interpretation of cases like (9) is systematically reflected in SF, either by different choice or by different combination of elements, no such account is assumed for (7) or (8), which instantiate one type of what is usually called underspecification. It is easy to see that this sort of lacking specificity is an ubiquitous phenomenon of natural language.3 What is less well established is the fact that SF and its foundation in I-Language may support distinctions and relations that are not reflected in C/I, in other words, SF can also be more specific than its interpretation in C/I. Cases of this sort become visible in case of differences projected on the same situation by different languages. An obvious case in point is (10), an inscription presented at the Berlin subway entrance: (10) Please validate your ticket! Bitte das Ticket entwerten!

Every dictionary correctly characterizes entwerten (devaluate) and validate as antonyms, or rather as designating inverse processes, but the two expressions in (10) are nevertheless understood as requesting the same activity. Thus different aspects of the same state of affairs are (conventionally) fore-grounded in English and German. A different, and more complicated pattern of distinctions can be seen in (11): (11) a. a'. b. b'. c. c'. 3

Peter took off his hat. Peter nahm/setzte den Hut ab. Peter took off his pullover. Peter zog den Pullover aus. Peter undressed (completely). Peter zog sich (ganz) aus.

Not only everyday utterances with obvious conceptual context rely on this sort of unspecificity. An important part of poetry is based on just this phenomenon, as is obvious from e.g. Hölderlin's verses: Die Linien des Lebens sind verschieden (The lines of life are different Wie Berge sind und wie der Länder As mountains are and as the countries' Grenzen. borders)

8

Manfred Bierwisch

In German, two types of activity, which English lumps together in (a) and (b), are distinguished on the basis of different objects in (a') and (b'). On the other hand, getting rid of specified objects and of the clothes altogether, as in (b') and (c'), respectively, are covered by the same verb in German, while they are clearly distinguished by (b) and (c) in English. Actually, things are far more complex than the examples in (11) indicate – not only because ausziehen is multi-ways ambiguous with respect to readings not relevant here and in ways that differ essentially from comparable ambiguities of take off, as every dictionary shows, but also because the contrast between etwas ausziehen in (b') and sich ausziehen in (c') is not just an instance of underspecification, but rather a special type of ambiguity.4 Another type of semantic distinctions that does not represent conceptual differences turns on locative relations of the following sort: (12) a. a'. b. b'.

the cup on the table die Tasse auf dem Tisch the handle on the door der Griff an der Tür

While direct contact between an object and a supporting surface is realized by the same relation in English, in German requires a distinction to be made between the relation to a horizontal surface and the relation to vertical surface. Again, further complications show up, other relations and additional conditions are taken into account. Bowerman (1996) discusses a wide range of distinctions made in different languages with respect to spatial relations. Semantic distinctions like those in (10), (11), or (12) turn on the content of individual lexical items, not on general principles. There might be more general conditions, however, which associate the same state of affairs with systematically different semantic representations in different languages. For instance, Levinson (1996) describes three types of spatial 4

The two readings that need to be lexically distinguished can be made visible by the ambiguity of (i), where ihn can be replaced by den Mantel 'the coat', as in (ii), or seinen Sohn 'his son', as in (iii), with clearly different relations in SF. (i) Peter zog ihn aus (ii) Peter zog den Mantel aus (iii) Peter zog seinen Sohn aus Both relations are realized by grammatically different arguments in (iv): (iv) Peter zog seinem Sohn den Mantel aus. Peter took his son the coat off It might be noted that these intricacies are not merely a lexical anomaly of the verb ausziehen – they are mirrored by parallel phenomena that appear with the verb anziehen.

Semantic Form as Interface

9

orientation by which different languages identify the front/back- and left/right-direction of an object. In systems with intrinsic orientation the direction depends on the object, in systems of relative orientation, it depends on the observer, and in systems of absolute orientation, it depends on the overall environment. Levinson's observations are, once again, more intricate than this rough summary indicates,5 but the main point should be clear enough: There are alternative ways, according to which different languages assign spatial structure to the same situation. Notice, however, that more or less systematic distinctions in SF, which don’t have different interpretations in C/I, are not at all in conflict with principle (5) of Full Interpretation. (5) prevents conceptually vacuous elements or relations in SF, but it doesn't require distinct interpretations for different elements.6 And the fact that, conversely, SF ignores a great number of differences present in C/I, as indicated in (7) and (8), is even less in conflict with Full Interpretation. As a matter of fact, if a conceptual distinction does appear in C/I, then it can in principle be represented also in SF.7 In sum, the interaction between conceptually represented experience and the linguistic representations that most directly correspond to it is open to interpolation in various ways. But in spite of this flexibility, there must be a reliable common core C in the sense noted in (2) above, by means of which the interaction succeeds. The following sections will be concerned with minimal assumptions to be made in this sense with respect to the interface between C/I and linguistic knowledge.

5

6

7

It is, in particular, not obvious, to what extent different linguistic representations of the same spatial scenario induce conceptual differences that go beyond their verbalization, including inferences about spatial relations. But this is a separate issue that need not concern us here. There are, of course, numerous ways to map different elements onto identical realizations. To give an analogy from music, enharmonic keys like Cis and Des major have a well defined, different position in the tonal system, although their acoustic realization is indistinguishable. See also fn 2 with respect to PF. In other words, different thoughts can in principle be expressed by different words. More precisely: for every conceptual distinction C/I provides for, there is in any natural language L, a (possibly complex) expression whose SF explicitly represents that distinction. This is, by the way, the principle defended in Searle (1969) under the term expressibility as a defining property of natural language. It must be added, though, that this principle is confined to conceptual distinctions – whatever that means. It does not apply to representations that are inherently restricted to analogical representation. Thus, one can distinguish two persons by name, but one cannot verbally express the difference of their faces. These are intricate problems, which cannot appropriately be pursuit here.

10

Manfred Bierwisch

4 Minimal Conditions on SF Interfaces like SF and PF are subject to two types of requirements: First, they must share to some extent the format of the system they interact with. Second, they must be composed of basic elements that are motivated by and can be interpreted in the interfacing systems. In case of PF, the representational format derives from the time course determining the articulation and perception of the signal. More specifically, the skeleton of PF is a sequence of time slots (13), where each X is a slot to which the PF-elements are linked: (13)

XXXXXXXXX….

The linear ordering inherent in this sequence provides also the combinatorial principle underlying the PF-representations. The basic elements in terms of which PF is organized are features (or feature values), which are systematically linked to the skeletal positions, such that formally threedimensional representations emerge, organized along the skeletal core. The feature values, functioning as primes within PF, represent properties, which in A/P impose conditions on signal production and perception, i.e. on place, manner, and agent of articulation. Although this sketch is oversimplified in various respects,8 the way in which PF is related to A/P should be clear at least in principle: The sequential organization of PF as well as the "content" of its primes is due to A/P, the relations within PF are determined by the rules and principles of I-Language. The situation is comparable for SF, although remarkably more complex in at least two respects. First, there is no evident base line which determines a straight and obvious representational format. There is nothing comparable to the time course of the signal (which, by the way, controls spoken as well as signed languages, and holds even for the derivative format of written language). No direct and obvious principle of this type organizes the totality of situations or states of affairs which linguistic expressions are about.9 Hence no natural condition on the dimensionality of 8

9

For a more detailed exposition, see e.g. Halle (1983, 1995). – It should be added, that although A/P is to be construed as normally dealing with spoken language, it must alternatively include the basis for sign language, i.e. visual perception of gestural articulation. See Bierwisch (2001) for some discussion of pertinent consequences. It must be stressed, that the temporal nature of mental operations, including the production or comprehension of utterances is an undeniable fact, which must not be confused, however, with the completely independent (a)temporal nature of their meaning. An utterance like (i) is, of course, formulated and interpreted in real time, but that does not imply that John preceded the woods or the period of three hours in any sense: (i) John walked through the woods for three hours. See Bierwisch & Schreuder (1992) for some discussion of these matters.

Semantic Form as Interface

11

SF seems to be given in advance. Second, the variety of domains in which primitive elements of SF are to be interpreted exceeds by far those of A/P, as is easily to be recognized: A/P has to recruit motor control of signalproduction and perceptual analysis of the signals thus produced, while C/I must integrate all aspects that are involved in the organization and control of behavior. I will return to the second problem in section 5 and 6, turning first to the representational format of SF. The general nature of C/I can, unlike that of A/P, neither emerge from the temporal structure of expressions, nor from spatial or any other basic dimensionality, it must rather encompass conceptual configurations of any kind, including a-temporal and non-spatial structures. Hence a sufficiently abstract combinatorial framework must be assumed to support and control SF. A general principle that meets this condition is the functor-argument-combination, according to which a functor X applies to an argument Y, with no linear ordering between X and Y. This principle implies that elements of SF be assigned to types which specify their combinatorial properties. The relevant type system and the pertinent assignment can be defined as follows: (14) a. b.

If D and E are types, then ¢D,E² is a type. If X is of type ¢D,E² and Y is of type D, then [X Y] is of type E.

This seems to be the minimal framework which the organization of SF could rely on in view of the indefinite or open dimensionality of C/I.10 The next point to be noted is the fact that a functor X may have properties according to which it must apply to two (or more) arguments. The characterization of those types is made possible by the recursive nature of definition (14a), which allows both D and E to be complex types of the sort defined in (14a): If D and ¢D,E² are types, then ¢D,¢D,E²² is again a type, viz. the type of a functor that combines with two elements of type D to get a complex element of type E. As a matter of fact, (14) allows for combinations of unlimited complexity.11 It is an empiri10

11

One might object that an even more elementary framework is the system of Bare Phrase Structure proposed in Chomsky (1995), according to which two elements X and Y are combined into [X Y] by the operation Merge, which assigns [X Y] simply the categorization (i.e. the "label") of its head, say X. Hence no functor types of the sort defined in (14a) must be assumed. Notice, however, that the condition by which the head X selects its complement Y must now be given in some other way. Therefore the framework as a whole cannot be simpler or more general than the one proposed in (14). The proposal to characterize the combinatorial structure of natural languages by recursive type systems of the sort adopted here has been criticized e.g. in Schnelle (1999) as making the claim that knowledge of language is based on explicit mastery of fully fledged modern logic and type structure. However, by the same token, one would have to reject the claim that knowledge of natural numbers is appropriately characterized by the

12

Manfred Bierwisch

cal question, to what extent these possibilities are made use of, and what boundaries might actually constrain their exploitation. The most important point in this respect is the constraint already embodied in (14), viz. the condition that functor types can only be binary, combining a functor with just one argument. A functor with more than one argument, actually a ubiquitous reality, must therefore belong to a type of the sort (15), indicating that a many-placed functor combines with its arguments in hierarchical steps. (15) ¢D1,¢D2,¢…¢Dn,E²²²²

This condition corresponds directly to the fact that no sequential ordering is available to represent structural relations in SF. Thus in order to keep apart the different relations a functor has to its different arguments, hierarchical distinctions must be made. If for instance (16a) and (16b) are assumed to be different according to standard notational conventions, according to which a functor R is placed between its first and second argument, then (17a) and (17b) would be the corresponding hierarchical representation, where hierarchy is indicated by bracketing, while linear order is irrelevant.12 (16) a. (17) a.

aRb [a [R b]]

b. b.

bRa [[a R] b]

Strictly speaking, in SF linear order does simply not exist. As already noted, the time dependence of actual processes of the formation of thoughts and inferences must not be misconstrued as a structural property of SF – or C/I, for that matter.13

12

13

recursive successor operation, arguing that elementary calculation must not imply explicit recourse to e.g. Peano's axioms. In other words, the power of a theoretical system is not to be confused with its use to under restrictive boundary conditions. Hence (17a) could just as well be written as [ [ R b ] a ]. – To avoid notational confusion, I would emphasize, that according to (14), the type ¢D,E² is strictly different from ¢E,D². What is important is the fact, that the type system imposes no ordering on the elements combined, the claim is not, that the notation of types has no ordering. This consequence could be avoided, if we would write e.g. ¢D¢E²² instead of ¢D,E², which could then be construed as equivalent to ¢¢E²D². Notations like these would deviate, however, from standard notational conventions for type systems, which I have adopted here. Consider e.g. a case like Who did he talk about, where the time course of production and comprehension is subject to different conditions, as the hearer normally assigns the Theme-role to who only when about is processed, while the speaker does it in advance, although the SF of the sentence is the same under both perspectives. See also fn.9. – Notice, incidentally, that no such asymmetry holds for PF, where linear ordering is after all a structural property.

Semantic Form as Interface

13

Less clear constraints delimit the set of types accessible to SF. This concerns first the inventory of basic types, from which functor types are derived by (14a), and second the finite subset of functor types that are actually needed for proper semantic representations. As to the first issue, it is a fundamental (and ultimately empirical) question, which basic types must be assumed for natural languages. I will adopt here the standard view that at least two basic types e and t must be distinguished, with e as the type of entities, i.e. objects, substances, and events, and t as the type of states of affairs or situations.14 Further basic types might be imagined, but I will leave it at that. It should be clear, however, that basic types must not be confused with primitive elements. Semantic configurations assigned to e or t can in fact be of arbitrary complexity, while on the other hand functors of complex types may be primitive elements of SF. Functor types based on e and t include n-place predicates (18a), n-place functions from individuals to individuals (18b), functions from propositions to propositions (18c), functions from predicates to predicates (18d), etc.: (18) a. ¢e,t ² ; ¢e,¢e,t²² ; ¢e,¢e,¢e,t²²² ; . . . b. ¢e,e² ; ¢e,¢e,e²² ; . . . c. ¢t,t² ; ¢t,¢t,t²² ; . . . d. ¢¢e.t²,¢e,t²² ; . . .

This open list directly leads to the question whether and how the set of functor types occurring in actual representations might be restricted, irrespective of the fact that (14) defines an infinite set of potential functors. While there is no point in setting up arbitrary limits for virtual structures, restricting e.g. the length of words or sentences or the complexity of concepts, there is an interesting limit deriving from the system's architecture. Remember, that the functor-argument-combination defined in (14b) necessarily reduces the complexity of the functor-type by applying the functor to its argument. Hence even in arbitrary complex configurations, types cannot become more complex than those of the basic functors the combination starts with. The range of types is thus automatically restricted by the set of primitive elements and their types.15 14

15

The considerations underlying this choice can be traced at least to Frege (1879) and have been pursued within the tradition of formal semantics, developed especially in Montague (1974), where the notational variant t for truth-value bearer and e for entity has been proposed. To adopt this convention is not an arbitrary choice, as the distinction seems to be fundamental. The general tenets of formal semantics, however, which sharply distinguish logical (objective) from mental (subjective) aspects of meaning, differ essentially from the present perspective, for which semantics is a matter of mental phenomena. Notice that this does not restricts the complexity of possible configurations in SF, since

14

Manfred Bierwisch

If the set of primes is fixed and finite, an issue to which we will turn shortly, the range of types is obviously finite.16 In any case, the type system determines a restricted system that supports a potentially infinite set of representations with completely abstract dimensionality. Notice that the type-system does not consist of elements of SF, just as the ordering of segments does not consist of elements of PF. The type system determines the format, not the content of representations of SF in much the same way in which formation and ordering of segments determines the format, but not the content of PF. In view of the interface-nature of SF, we are nevertheless faced with the question, how this representational framework corresponds to the structure of C/I which SF interacts with, as the format of the interfacing systems must be compatible, as noted earlier. For obvious reasons, a simple and direct answer, comparable to the time course as basis for the ordering of segments in PF, is largely impossible. As a simple illustration, consider situations like (19), which could easily be extended by further variations: (19) a. c. e. g.

he crossed it he knew it he prevented it he denied it

b. d. f. h.

he followed him he forgot it he destroyed it he hated him

Whatever he, him, and it might represent, it is obvious, that the different dimensionality and conceptual nature of the situations, their spatial, temporal, causal, cognitive, emotional relations cannot be distinguished and reflected by the representational format, but only by the content of the elements from which the SF of the different verbs is made up. To put

16

due to the recursive nature of the combination defined by (14b), functors as well as arguments can be arbitrarily complex, irrespective of the restricted complexity of their type. Two further constraints may be worth noticing: First, functors of type ¢D,E² with basic D and E define the indispensable minimum, since otherwise there would be no combination. Second, if definition (14b) of functional application – abbreviated in (i) – is supplemented by functional composition (ii), in order to provide a more general mode of combination, then every n+2-place functor can in principle be built up from n 2-place functors, as sketched in (iii) for a 3-place functor with arguments of type G, J, and H: (i) D x ¢D,E² œ E (ii) ¢J,D² x ¢D,E² œ ¢J,E² (iii) ¢G,¢J,D²² x ¢D,¢H,E²² œ ¢G,¢J,¢H,E²²² (Combination is indicated by 'x', functors follow their argument for perspicuity.) As a consequence, 2-place functors would be the upper bound for the complexity of types to which all configurations can be reduced. – For more systematic exposition of the technicalities, see e.g. Partee et al. (1990). Whether these formal matters are empirically relevant or not, must be left open here.

Semantic Form as Interface

15

it the other way round: Combining elements in SF means nothing but the bare connection. Whatever shows up as a distinction in SF must either be a distinction in terms of combinatorial hierarchy, or must derive from the content of semantic elements entering this combination. These considerations were in fact the reason for adopting the abstract functor-argument-structure as the underlying format of SF. This leaves us with the final and in many respects central problem of the interface character of SF, viz. the nature of its basic elements and their interpretation. It first seems necessary to distinguish between constants and variables. The nature of variables is an important and intricate problem within the tradition of logic. Without entering these discussions, I will just make the following assumptions, which should be uncontroversial: (20) a. b. c. d.

SF provides an open set of variables x, x1, x2, … for every type D, in SF. Any two variables xi and xk are different – within and across types. Every variable x allows for an unlimited number or tokens, i.e. x can have more than one occurrence within the same SF-representation. A variable x of type D can be substituted according to systematic principles by basic or complex elements M of type D, including variables of type D.

Substitution and binding of variables belong to the operations by which the SF of complex expressions is constructed from the SF of their constituent parts. As the present paper focuses on interpreting, rather than constructing SF-representations, the principles of variable binding and substitution will be left aside, noting only that occurrences of the same variable must be treated alike. This holds also for the way in which variables function as interface-elements with respect to C/I, a problem to be taken up below. Turning now to the proper semantic primes, i.e. the basic constants of SF, a number of crucial questions is to be faced. Is there a fixed and finite, possibly universal repertoire of basic semantic elements? What would be the size of this repertoire? What would be the origin and the interpretation of its elements? While there is more or less agreement about these questions with respect to PF, this is by no means the case for SF. Thus the assumption of a fixed, universal repertoire as part of the language faculty is widely accepted with respect to PF,17 but essentially doubted or practically denied with respect to SF, e.g. in Fodor (1981). A 17

In view of the revealing discoveries about sign language, I have argued in Bierwisch (2001) that the basis for primes of PF in the Language Faculty should presumably be assumed to determine only the organizational principles, instead of the specific content of the repertoire, such that the structural conditions can be realized by alternative modalities, relating e.g. to motor control of either the vocal tract or the hand.

16

Manfred Bierwisch

closer look at the problems involved urges one to distinguish primes of different kinds. In fact, Katz and Fodor (1963) distinguished already between socalled semantic markers like (Human), (Male), (Adult) with an essentially systematic status, and distinguishers like [has never married] or [lowest academic degree].18 This rather provisional distinction reappeared in various forms and with different claims associated with basic elements of different sorts. Thus to a large extent, Generative Semantics, as reported e.g. in Dowty (1979) concerned itself with the systematic properties of primitives like CAUSE, BECOME, HUMAN, some of which reappeared as functional heads and light verbs in various versions of LF, such as Hale and Keyser (1993) and later work. I will return to these matters in section 6. To conclude this outline, two examples illustrate SF-representations as discussed so far. (21a) indicates the lexical SF of the transitive verb open, (21b) is that of the adjective red. The variables x, y, z mark positions to be filled in by constituents the lexical items combine with, while the variable s in (21a) eventually refers to the situation that instantiates the condition specified by the verb:19 (21) a. s e

inst ¢t, ¢e, t²²

x e

act

cause

¢e, t² ¢t, ¢t, t²²

become open ¢t, t²

¢e, t²

t

y e

t t ¢t, t² t ¢e, t² t

18

19

These are, of course, highly provisional examples for the purpose of illustration only. In later work, e.g. Katz (1972), the theoretical relevance of distinguishers was (inappropriately, I believe) very reduced, while semantic markers were considered as potentially complex, systematic configurations. The initial distinction was nevertheless a helpful proposal. It inspired necessary disputes that are not really settled yet. Technically, this role of variables is to be expressed by operators, which make the variables available for substitution by the SF of complements. Thus (i) sketches a more complete version of the entry for transitive open : (i) / open / [ +V ] Oy Ox Os [t s [¢e,t² inst [t x act ] [¢t,t² cause [t become [t open y ] ] ] ] ] For details about the operators specifying argument positions see e.g. Bierwisch (1997).

Semantic Form as Interface

b.

color

red

¢e, ¢e, t²² e

17

z e

¢e, t² t

These examples also illustrate the general assumption that the SF of major lexical categories like Noun, Verb, Adjective are configurations of type t, specifying the more or less complex propositional condition that a lexical entry imposes on a situation it applies to. To combine these conditions – primarily by substitution and binding of variables – according to the syntactic structure of complex expressions is a matter of the computational system of I-Language. To relate the complex conditions thus specified, and their basic ingredients to the external and internal environment of the speaker/hearer is a matter of the C/I, to which we turn now.

5 Basic Assumptions about C/I The systems of conceptual, intentional representation which SF interacts with are not a single, self-contained module of mental organization, but a highly complex aggregate of partially autonomous components with special biological foundation and different phylogenetic history. This aggregate includes the various perceptual capacities, the organization of motor control, of spatial and temporal orientation, the "Theory of Other Minds" supporting inter-individual coordination, and the effects of emotional evaluation, to give an incomplete and unsystematic list. It is clear that there must be essential modes of interaction among the various subsystems, such as visual, auditory and tactile perception controlling movement and action, or the various systems jointly supporting spatial orientation. The question, which systems are interrelated in which way, or which specific types of representation different sub-systems rely on, cannot be pursued here in even remotely adequate way. We can only identify two types of problems that directly bear on the interface aspect of SF, indicating the differences and the resulting limits of the representational systems. It must first be noted that most semantic theories simply do not acknowledge a representational difference between SF and C/I (or whatever would correspond to this distinction). Therefore the problem of characterizing representations of C/I simply does not arise e.g. in stan-

18

Manfred Bierwisch

dard versions of formal semantics.20 Assumptions about conceptual structure, if they are made explicit at all, are to a large extent plausible extrapolations from analyses dealing with SF (or Logical Form, for that matter). Exceptions to this claim can be found, however, in the domain of vision and spatial orientation. The vast literature on these matters warrants the claim that an essential distinction is to be made between spatial and semantic representations.21 A crucial point to be noted is the different format on which propositional and spatial systems (and may be others) are based, supporting different types of inferences and problem solving normally classified as thinking. A concrete example which highlights the latter point is the notion of mental models discussed in Johnson-Laird (1996). A mental model is essentially an internal representation or image the structure of which corresponds in crucial respects to that of the object or situation it represents. Thus a mental model is similar to a physical model of the represented situation, somewhat like a chemist's model of a molecule, or an architect's model of a house, and therefore different from a linguistic expression that describes a situation. As a consequence, a mental model can represent only those properties that can be preserved under the condition of similarity. Remember that to avoid this kind of restriction was a central point motivating the principles of SF. Although SF does not preclude similarity altogether, it does not require it (beyond analogy of hierarchical organization). This freedom allows for SF-representations that in some way correspond to mental models. For the sake of illustration, suppose that (22) is a mental model that represents a situation that could variably be described by (23)(a), (b), or (c), among others: (22)

q a

(23) a. b. c. 20

21

Œ

A circle is on the left of a square and a triangle is on the right of the square A square is on the right of a circle and a triangle on the right of the square A circle is on the left of a square, which is on the left of a triangle

Notice that the distinction between extension and intension or Frege's Sinn and Bedeutung, familiar from formal semantics, must not be confused with the present distinction, even though their point is to deal with different aspects of meaning, albeit in a different respect. It is not surprising that terminological matters are not very clear in this respect. Besides the distinction between Logical Form and Semantic Form discussed above (both of which must be construed as versions of propositional structure), Jackendoff (1996) uses the term Conceptual Structure CS to capture essentially the same aspect of meaning that I have called SF. Reviewing relevant parts of the literature on visual perception, Jackendoff makes a sharp distinction between CS and Spatial Representation SR. This distinction is very similar to that between SF and spatial representations advocated below, or between propositional representations and mental models.

Semantic Form as Interface

19

(24) [ s INST [ [ H x [ CIRCLE x ] ] [ LOC [ LEFT [ H y [ SQUARE y ] ] ] ] ] ] AND [ s' INST [ [ H z [ TRIANGLE z ] ] [ LOC [ RIGHT [ L y [ SQUARE y ] ] ] ] ] ]

In (24), a simplified SF of (23a) is given, where the three objects of (23), represented by a circle, a square, and a triangle in (23a), are identified by indefinite descriptions, viz. [ H x [ CIRCLE x ] ], [ H y [ SQUARE y ] ], and [ H z [ TRIANGLE z ] ], respectively. The operator for indefinite reference H, which appears in these descriptions, is an element of type ¢e,¢t,e²², that combines a variable of type e and a proposition of type t into a configuration of type e. Its interpretation introduces an individual, which corresponds to the variable bound by the operator and meets the conditions specified in the proposition. Likewise, the definite description build up by the L-operator takes up an individual already identified. According to these provisional comments, definite and indefinite descriptions may correspond directly to objects of mental models, though both have absolutely different internal structures, corresponding to different aspects of mental organization. This holds equally for the conditions the objects are subject to, in our example the spatial relation represented as is on the left/right of in (23). The mental model (22) exhibits these relations not by elements of representation, but directly by the mode of combination, similar to the way in which e.g. /tik/ and /kit/ are different in PF. In SF, however, [ x LOC [ LEFT y ] ] and [ z LOC [ RIGHT y ] ] do not only turn these spatial relations into explicit representational elements, but do, moreover, split them up into the common (and completely abstract) condition of localization, represented by the element LOC of type ¢e,¢e,t²², and the specification LEFT vs. RIGHT of type ¢e.e², which assigns the corresponding region to its argument. As one easily realizes, the correspondence between SF and mental models is anything but simple and obvious, even in artificially simple cases like (23). It should therefore be emphasized, that the format of these representations is determined by independently motivated crucial conditions: SF has to accommodate, as already noted, the full range of modules participating in C/I, and must adapt them to the computational system of recursive symbol combination. The principles of mental models on the other hand depend on central aspects of the sensory input, and they directly support certain types of reasoning. Thus, Johnson-Laird (1996) has shown, that for instance decisions about spatial relations between objects are made on the basis of "direct inspection" applied to mental models like (22), rather than by means of logical rules operating on propositional representations deriving from (23).22 A different aspect of 22

The type of reasoning explored by Johnson-Laird (ibid.) is illustrated by (i) – (v), where the answer to (v) is based on the mental model in (vi):

20

Manfred Bierwisch

mental models and a more intricate kind of operation than direct inspection is involved in the phenomena called "mental rotation", reported e.g. by Shepard and Cooper (1982). According to these studies, the time to decide about the identity of two geometrical objects, presented by twodimensional diagrams, depends directly on the size of the angle by which one of these objects is rotated against the other. The full range of insights and problems related to visual images and mental models goes far beyond the present limits, as is clear e.g. from the concise survey given by Osherson et al. (1990). But whatever aspects might turn out to be relevant for mental representations of visual and related experiences, their interaction with the language dependent Semantic Form is subject to at least two un-escapable limitations. First, there is a fundamental, inherent boundary for mental models of any kind relative to SF (or other sorts of propositional representation), due to the condition that the structure of mental models must correspond to the structure of what they represent. The most elementary consequence is that mental models cannot represent propositional negation. Thus although (22) and numerous other mental models would be compatible with e.g. (25), there is, by definition, no mental model that actually represents this situation. (25) There is no circle to the right of the triangle.

The crucial point is the operator NOT, provided by SF and realized in PF in various, grammatically determined ways. This possibility depends on the fact that natural languages are combinatorial systems of symbols with arbitrary, non-iconic correspondence between SF and PF. It is exactly the iconic, non-arbitrary nature of mental models, which excludes an explicit counterpart of the operator NOT in the representational system of mental models. It is worth noting, moreover, that NOT differs substantially even from SF-elements like LOC or LEFT and RIGHT. Although these elements do – like NOT – lack explicit counterparts in mental models, they still correspond to relations (or regions) within the spatial structure from which the representational format of mental models de(i) The knife is on the right of the plate. (vi) S P K (ii) The spoon is on the left of the plate. (iii) The fork is in front of the spoon. F C (iv) The cup is in front of the knife. (v) What's the relation between the fork and the cup? That in fact models like (vi) are used to answer (v) is shown by descriptions that do not yield an unequivocal model representation, forcing the use of rules for propositional inferences, which causes remarkable delay and errors.

Semantic Form as Interface

21

rives,23 while NOT cannot correspond to any particular aspect of a mental image. It merely determines its "validity" as a whole, relative to some other representation. This limitation of mental models infects a large range of conditions and distinctions that can be represented in SF, but lack direct counterparts in mental models. This includes central aspects of modality, time, conditionality, and abstract properties like familiarity, moral value, etc. Consider simple cases like (26): (26) a. b. c. d.

The car does not move. The car can move. The car will move shortly. The car would move, if necessary

These sentences would all be compatible with, but not represented by the mental image of a car that doesn't move. And the difference between The car is moving and The car was moving is hardly represented by two systematically differing mental models. How far these limitations extend, is difficult to decide. A mental model that explicitly represents the content of The next war will be awful is hard to imagine.24 In any case, even if mental models could use space as a kind of analogical structure, representing e.g. social hierarchies as configurations in a "social space", there are systematic limits with respect to the representational capacity of mental models. Second, there is an equally principled limitation of SF with respect to characteristic distinctions and conditions appearing in mental models or visual images. The most striking, but certainly not the only case in point is the domain of faces: Every normal person can distinguish, remember, and identify the faces of an indefinite number of individuals under extremely different circumstances. Many of these individuals may be named and described, but their faces can never be verbalized. Faces can be characterized by diagnostic features and properties, but the actual 23

24

It must be added, that even elements like square or triangle, which incidentally are not true primes, but configurations of SF, do not correspond directly to objects of mental models, but rather to their charateristic properties. The actual correspondence between SF and mental models in the case of objects involves, as already mentioned, the (in)definite descriptions and the referential variables on which they are based. These are important aspects of the different representational principles on which visual images and propositional structures are based. The problem is, of course, not the possibility to imagine any sort of horrible scenarios from former or fictitious wars, but the abstract qualification of a potential future war. Without stretching the notion of structural correspondence far beyond the reliable conditions assumed so far, no mental model that properly corresponds to what it represents is available for a statement about abstract states of affairs.

22

Manfred Bierwisch

identification of a face succeeds only by visual means. Similar observations apply, incidentally, to voices, identifying people by auditory means. Cartoons and caricatures furthermore indicate possibilities for genuine, non-propositional abstractions and cognitive operations on mental images representing faces. The point here is not the interesting question whether and to what extent faces are special, as opposed to other domains of visual patterns, but that C/I represents sensory distinctions that do not correspond to elements or distinctions represented in SF. An important consequence of these observations concerns the principle of expressibility mentioned earlier. What can be expressed in a given natural language in the sense that the Phonetic Form of an expression corresponds to a distinct representation in SF includes only the propositional aspect of all mental representations. In other words, the "mental world" may well contain thoughts, which can only indirectly be represented by a linguistic expression. The problems indicated so far would easily multiply, if one thinks of further specialized modules participating in C/I, such as music, emotions, or interpersonal relations, to mention just three fairly well studied and rather different domains. To enter the details of the respective representations and mental operations would not only go far beyond the present limits; it is actually unfeasible, since explicit proposals, as far, as they are available, are too disparate to allow different principles of representation to be compared and interrelated. For the domain of music, Jackendoff and Lehrdal (2005) have made revealing proposals about the mental representation of the formal structure of musical utterances. It seems to be parallel and in fact related to the prosodic aspect of PF in relevant respects, i.e. to meter and stress of natural language. Whether and how the structures of music represent somehow those of motion and emotion, is an intriguing, but unsettled question.25 I must refrain here from even more inconclusive speculations about representational conditions of emotions, Other Minds, and further domains of mental representation. Two things should be clear, however, even from these highly provisional remarks. First, the different modules of C/I must have specific, representational systems with different elements and modes of combination, complying, among others, with their different types of input. They cannot, under this perspective, constitute a single, homogeneous mode 25

In Bierwisch (1979), I have suggested that the formal structure of music is iconically related to what might be called "Gestural Form", comparable to the fact that Phonetic is symbolically related to Semantic (or Logical) Form. Gestural Form would thus be in a way the meaning assigned to musical form. In any case, music has a specific relation to emotion. It is not about emotion, but represents it.

Semantic Form as Interface

23

of mental organization.26 And second, in spite of the alleged heterogeneity, the separate sub-systems must interact in specific, systematic, and normally spontaneous ways, independent of linguistic expressions. In other words, there must be appropriate representations in terms of which different modules can cooperate. Visual control of motoric patterns, or the activity of different systems jointly constituting the system of spatial orientation are obvious examples.27 The intriguing question emerging from these considerations is: How do different systems of C/I interact with SF? Does e.g. the visual system, which must integrate, among others, colors, shapes, positions and motion of objects, interact with SF directly, i.e. in terms visual patterns and images, or does it contribute to a more comprehensive, not domain-specific sort of representation, which eventually provides the interpretation of SF? This leads to at least two alternative possibilities: (27) Assuming, that human behavior is based on a complex array of autonomous systems of mental organization with domain specific types of representation requires that either a. SF serves as the interface in terms of which the different, otherwise separate systems of C/I interact, or b. C/I provides different types of interaction for its systems, hence SF draws on (at least partially) integrated C/I-representations.

Although these options seem to define a rational alternative – SF either creates or is imposed on the interaction of different modules –, it is difficult to see, how it could be decided on empirical grounds.28 However 26

27

28

There is, of course, after all we know a single and uniform manner of representation in terms of neurons, their synaptic connection, and the patterns of activation they allow for. But that is not the relevant level of analysis in the present context. A parallel case in point is the integration of articulation and auditory perception constituting the interface-partner for PF, as noted earlier. A decided position is advocated by Fodor (1983) on theoretical grounds. He conceives of the human mind as consisting of autonomous, domain-specific input/output-systems, which support a central system, which is global, fully interactive and mediates between the otherwise encapsulated input-modules. On Fodor's view, language is one of the autonomous input-output-modules, which do not interact with other modules, but only with the central system. In spite of its clear and definite position, it is unclear, however, how SF (or LF, for that matter) is to be conceived under this view: Either SF is assumed to participate in mediating among the various subsystems of the mind, analogous to (27a). This would mean, however, that it must belong to the central system and could by definition not be part of language, which is one of the autonomous input-output modules. Or SF belongs to language, as independent reasons clearly suggest, but then it cannot subserve the interaction with other modules, similar to (27b). In order avoid these and other problems, Jackendoff (1997) proposed, what he calls "Representational Modularity". On this view, modules are not encapsulated input- or output-systems, like language comprehension or visual perception. Modules are rather construed as autonomous

24

Manfred Bierwisch

these problems will be solved or perhaps reformulated, an automatic correspondence between SF and the representations in C/I is indispensable, it is in fact the essence of the interface-function. The correspondence must obviously satisfy different requirements for the mapping from SF into C/I and from C/I into SF. In what follows, I will nevertheless keep to one perspective as far as possible. As already noted, the combinatorial structure of SF is open to any interpretation that meets the conditions imposed by the type-based hierarchy. For the sake of illustration, look at the configuration [ LOC [ LEFT [ H y [ SQUARE y ] ] ] ] in (24), where the interpretation of the operator H y introduces an instance of y, i.e. an individual of type e, which combines with the condition SQUARE identifying a property of this individual. Next the functor LEFT combines with this individual, assigning to it the (spatial) region on its left, which again functions as an entity of type e. Now the element LOC of type ¢e,¢e,t²² combines with this individual and creates the property of being located within this region. In (24), this property is then assigned to the individual x which has the property CIRCLE. As noted above, one way to think of the interpretation corresponding to (24) is the mental model indicated in (22), which shows that elements of SF get interpretations with rather different status in C/I. Thus, while the combinatorial structure of SF is transparent in the sense that it does not fix anything beyond connectedness, the "content" of this connection as well as any other condition entering the conceptual interpretation is determined by the basic elements of SF. Hence any condition an SF-representation imposes on its interpretation within the framework of type structure is ultimately due to the basic elements of SF.

6 Three Kinds of Basic Elements of SF The nature of these basic semantic elements is among the most controversial issues in semantic theory. There is little reason to assume that it can be settled just by inspecting the interface properties of semantic representations. It might be useful, though, to sort out some distinctions

representational systems. Thus PF or the Syntactic Structure SS are modules, alongside with e.g. visual or musical representation. How SF (or the Conceptual Structure CS, which corresponds to it in Jackendoff's system) as a module relates to other representations in C/I is not obvious, though. The view on Spatial Representation and its relation to CS presented in Jackendoff (1996) seems to favor the position of (27b). The issue requires further discussion, though.

25

Semantic Form as Interface

with respect to the way basic elements participate in the organization of SF-representations and their interpretation in C/I. Although lexical items are the building blocks from which complex linguistic expressions are made up, their Semantic Form, as noted above, is not in general a single basic element, but a more or less complex configuration of more elementary units. One kind of these elementary units are elements that have repeatedly been identified because of their fairly obvious syntactic or semantic effects. Thus characteristic syntactic and semantic properties of causative verbs such as kill, give, show, change, close and many others are captured by elements like ACT and CAUSE, as indicated in (21a), and extensively discussed e.g. in Wunderlich (1997). Likewise, characteristic semantic (and syntactic) properties of color terms like red, blue, etc. are determined by the element COLOR, as shown in (21b).29 Kinship terms like father, sister, son, parent are another domain the syntactic and semantic properties of which are determined by systematic combinations of primes like MALE, PARENT, ANIMATE. Further areas of lexical structure, like terms for spatial dimensions, (change of) possession, would easily provide additional evidence for an array of systematic primes. The morpho-syntactic effects of elements of this sort include, besides Argument Structure with categorial and semantic selection, conditions on Person, Number, Gender, Tense, Mood, and others. A provisional cut-out of the system of these elements is given in (28), where elements are grouped according to their typemembership. (28) Systematic Primes HUMAN, MALE, ALIVE, OPEN, ACT, …. a. b. LOC, COLOR, PARENT, SEE, HAVE, …. CAUSE, AND, …. c. d. NOT, BECOME, POSSIBLE, …. LEFT, RIGHT, MAXIMAL, VERTICAL …. e.

¢e.t² ¢e,¢e,t²² ¢t,¢t,t²² ¢t,t² ¢e,e²

The combinatorial options within SF follow from the type these elements belong to. Their interpretation in C/I however depends on varying conditions determined by the modules of C/I. Some of these conditions 29

Notice that color terms differ from other adjectives also with respect to their syntactic behavior. Thus while for (ia) there is the parallel (ib), no parallel of this sort exists for (iia): (i) a. the blue book, the red shirt b. the blue of the book, the red of the shirt (ii) a. the interesting book, the long shirt b. *the interesting of the book, the long of the shirt There is, of course, the length of the shirt, but that is a nominalization, which clearly differs from (ib).

26

Manfred Bierwisch

were mentioned with respect to mental models as one of the candidates for C/I-interpretation. There are, of course, lots of intriguing questions to be answered with respect of the conditions fixed by HUMAN, MALE, HAVE, OPEN, etc. What is the core of interpretation, what is the range of variation, how is it determined in different combinations or contexts? Besides these questions, which need to be answered for each of the elements in particular, there are problems, which concern the system as a whole. What is the basis of the elements, could there be arbitrary amendments, or is the system closed and fixed, how is it determined ontogenetically, i.e. what input is needed to activate its elements? To assume a systematic, perhaps universal status for the overall system, of which (28) is an incidental illustration, does neither imply nor exclude, however, that these elements are part of the biologically fixed language faculty. It might well be the case, that the language faculty determines the dimensions or coordinates along which – within the combinatorial type system – basic elements of SF can be differentiated, without providing a repertoire that is fixed a priori. The items illustrated in (28) are systematic basic elements, because of their characteristic role within SF and in I-Language in general. Several of these elements have been analyzed under various perspectives in the literature. The precise delimitation of this system is, of course, not obvious at all. The core of it is fairly clear, however. Whatever the correct choice of the systematic elements for a given language might be, it is equally clear that the full range of lexical items can hardly be characterized in terms of these elements alone, if the system (28) belongs to is not open for just arbitrary expansion. As mentioned earlier, idiosyncratic distinctions are indispensable, in order to keep apart lexical items like dog vs. cat, horse, tiger, or lion, all of which are based on the systematic element ANIMAL. Likewise red must be distinguished from pink and brown within the set of terms specified by the systematic element COLOR. Since this kind of specificity holds for large parts of the lexicon, recourse to analyses in terms of systematic elements has been rejected as circular e.g. by Fodor (1981), because the system of basic elements would approach the size of the set of lexical items. Hence decomposition of lexical items in terms of basic elements would be circular and should hence be discarded in favour of lexical items without lexical decomposition. However, the fact that idiosyncratic elements might be necessary for many, or possibly most lexical items does by no means preclude them from exhibiting compositional structures based on idiosyncratic as well as systematic elements, where the letter do in fact provide the frame for idiosyncratic distinctions. The

Semantic Form as Interface

27

case of color terms illustrated by (21b) is one of the more obvious examples.30 Idiosyncratic elements, introduced for the first time as "distinguishers" in Katz and Fodor (1963), reappeared in a somewhat different guise in Pustejovsky (1995) under the slightly misleading title "Qualia structure". As a matter of fact, two fairly different kinds of idiosyncratic elements must presumably be distinguished according to a number of criteria, among them the different background or origin in C/I. As a rough orientation, idiosyncratic elements might either be irreducible, i.e. routed in primary sensory qualities, or inducible, i.e. constructed on the basis of specific and cumulative information. I will call them Qualifying elements and Dossiers, respectively, and briefly illustrate them in turn. To begin with, Qualifying elements are meant to specify for example values within perceptual domains corresponding to lexical items like sweet, sour, bitter, cold, hot, hard, soft, dark, bright, red, blue, etc. The perceptual domains as such might be identified by Systematic elements, which set the stage for idiosyncratic distinctions marked by be Qualifying elements, as hinted at for color terms. There is a wide range of intriguing problems concerning the origin, the status, and the effect of the perceptual domains and distinctions, most of which, though clearly part of C/I, are outside the range of language and SF. On the other hand, problems of metaphorical use of lexical items relying on synesthetic conditions, if e.g. dark and bright shift from sensory to intellectual, ethical or mood qualities, may well be a particular issue of interface effects. A further intricate issue is the status of complex perceptual patterns such as the voice or the face of a person or the shape of particular objects like leaves, flowers, or animals. As noted earlier, sensory clusters of this sort are available for cognitive operations – they can be remembered, recognized, modified, labeled –, but their content can generally not be verbalized, i.e. it cannot be represented by a compositional verbal expression. This raises the question whether clusters of this sort are to be considered as Qualifying elements, in case they are connected to a representation in I-Language. If so, the representation of a familiar person in SF includes a Qualifying element, whose interpretation in C/I is the visual pattern representing the face (and the voice and perhaps characteristic gestures etc.) of that person. These considerations lead directly to the intriguing question, how visual (and other sensory) patterns serving as 30

Semantic elements representing different color shades are not only classified by means of the systematic, relational component color, they are furthermore subject to systematic constraints governing their possible verbalization. As shown by Berlin & Kay (1969), conceptualization and verbalization are strongly dependent on cardinal colors structuring the perceptual field of color.

28

Manfred Bierwisch

prototypes for conceptual classification are to be dealt with. This is, of course, a far-reaching problem that cannot seriously be taken up here. Turning next to what I have called Dossiers, i.e. basic elements constructed and enriched according to experience. An obvious way to illustrate the nature of these elements is to look at proper names, which have an SF, that must characteristically include an element of the kind in question. Starting with the assumption that Berlin means something like the city named "Berlin",31 one notes that the SF of a name must comprise four characteristic components, as indicated in example (29), where (29d) consists of the conditions that make up the Dossier about Berlin, its geographical position, its function, etc., which is not only necessary to distinguish the various places with the same name, but also to fix the properties the speaker does rely on in identifying the place. (29) The SF of a normal Proper Name specifies a. Referential definiteness, i.e. representation of a fixed individual x b. Classification of x as an instance of a certain class (like PERSON, PLACE, etc.) c. Assignment of the name as a property of x d. Assignment of whatever is (assumed to be) known about x

To illustrate this schema, the lexical entry for Berlin would look like (30), where Lx identifies a specific individual within the representational space of C/I, which is classified as a place, for which the Phonetic Form is made accessible by PF to the naming component, i.e. the SF-condition, by which a proper name is assigned to x,32 and BERLIN is a set of conditions which jointly characterize x: (30) /berli:n/ [+N, -Masc, -Plur…] [ Lx [ [ PLACE x ] AND [ PF NAME-OF x ]

AND [ BERLIN

x ]]]

Three points are to be noted with respect to BERLIN and Dossiers in general, including in particular those about persons. First, a Dossier might contain all types of information that are available in C/I and permanently associated with the individual x. The Dossier might in particular include visual or other sensory representations bound to x, but just as well abstract or propositional conditions, like age, or time and manner of acquaintance. Second, a Dossier in SF is a prime, which means that it 31

32

There is a vast literature about the diverse problems of proper names, which need not be taken up here. For some discussion of the relevant aspects see e.g. Geurts (1997). This is a provisional convention, which must eventually account for potential PF-variants of a name, and also for diagnostic conditions, to correlate and distinguish graphemic variants or alternatives. But these matters can be ignored here.

Semantic Form as Interface

29

functions as a single, unanalyzed, encapsulated element, just like Systematic or Qualifying elements. But it can, so to speak, become transparent to the extent to which the content of the Dossier can be matched by the representational format of SF. That is possible for propositional conditions, but barred for the internal structure of perceptual patterns. Hence a Dossier may be a kind of sluicing device which allows information from C/I to appear in SF. In other words, Dossiers are primes like other basic elements, their internal make up is closed off, but it is available for representation in SF as far as it meets the representational format of SF. And third, it follows from the internal structure of Dossiers, that the system of these elements can be expanded not only with respect to each individual Dossier as a specific element – by new information to be incorporated – but also with respect to the whole repertoire, by introducing new elements, which characterize additional individuals. Now, the next step is to observe that proper names are not special in using dossiers. Idiosyncratic conditions CANINE, FELINE, or EQUINE, distinguishing dog, cat, and horse on the basis of the systematic element ANIMAL, are in all respects just noted Dossiers like those of BERLIN, PARIS, NAPOLEON, or HANS CASTORP. They would include, in particular, the visual pattern of a prototypical cat, dog, or horse, and they would indicate conditions on typical behavior and species specific sounds. In a similar vein, the Dossier of book or knife collects conditions on shape and functional aspects. A great deal of the revealing analyses in terms of qualia structure, co-composition, and coercion in Pustejovsky (1995) are based on conditions of this sort. Thus, in order to interpret Mary finished the book, the qualia structure must provide the condition that a book can be written and read, and that both activities can begin and have an end, so Pustejovsky argues, which supports the (at least) twofold conceptual interpretation of the sentence. Let me finally mention two non-trivial ways in which Dossiers and Qualifying elements are interrelated. First, as just noted, visual and other sensory representations – faces, prototypical shapes, voices, characteristic calls – and hence potential Qualifying elements may naturally be included as proper parts into Dossiers. Second, Qualifying elements, initially construed as distinctions within basic domains of sensory representations and hence as simple perceptual values directly determined by receptors or effectors, can obviously acquire fairly complex structure, resulting from higher order integration of sensory input or motoric control. Thus faces and complex shapes are anything but simple values of sensory domains, they rather result from exposition to complex information under varying conditions. Hence the borderline between Qualifying elements and (certain aspects of) Dossiers might not be as sharp as

30

Manfred Bierwisch

their essentially different nature would suggest. And Pustejovsky might not be too far off the mark in lumping them together under a single heading.

7 Instead of Conclusions It is easy to see that there are unanswered questions, even with respect to the central issues of the present considerations. It nevertheless seems to me fairly safe to assume that semantic representations are based on an abstract, transparent, combinatorial structure. As a matter of fact, all reasonable approaches to semantics are based on this assumption in one way or the other. It is also obvious that the representations making use of a framework of this kind must interact with different mental subsystems in terms of which the human mind organizes its experiences and actions. The remaining question concerns the origin and nature of the packaging, so to speak, of the units that mediate between SF (or its equivalent in other approaches) and its external interpretation. The conjecture defended here is the assumption that the elements realizing this interface business are of roughly three sorts, rooted either in the computational system of language (systematic elements), or in the external, primarily sensory systems (Qualifying elements), or participating in both, as a kind of sluicing mechanism (Dossiers). The assumption of Dossiers, living on some sort of double currency, is less illegal as it might appear at first glance. If one recognizes the fact, that the language faculty and the representations it supports, is phylogenetically as well as ontogenetically imposed on other, independently operating mental systems, recruiting them as domains of interpretation, then a compromise is natural under certain conditions, and it is likely to take a form like the one envisaged here. The need to spell out the details and consequences leaves us with a lot of open questions.

References Berlin, B. & P. Kay (1969): Basic Color Terms. Berkeley and Los Angeles: University of California Press. Bierwisch, M. (1979): Musik und Sprache. In: Jahrbuch Peters 1978, Leipzig: Edition Peters, 9–102. Bierwisch, M. (1997): Lexical Information from a Minimalist Point of View. In: Chr. Wilder, H.-M. Gärtner & M. Bierwisch (eds): The Role of Economy Principles in Linguistic Theory. Berlin: Akademie-Verlag, 227–266.

Semantic Form as Interface

31

Bierwisch, M. (2001): Repertoires of Primitive Elements – Prerequisite or Result of Acquisition? In: J. Weissenborn & B. Höhle (eds): Approaches to Bootstrapping. Vol 2, Amsterdam: John Benjamins, 281–307. Bierwisch, M. & R. Schreuder (1992): From concepts to lexical items. In: Cognition 42, 23–60. Bloom, P., M. Peterson, L. Nadel & M. Garrett (eds) (1996): Language and Space. Cambridge, Mass.: MIT Press. Bowerman, M. (1996): Learning How to Structure Space for Language: A Crosslinguistic Perspective. In: P. Bloom et al. (eds), 385–436. Chomsky, N. (1995): The Minimalist Program. Cambridge, Mass.: MIT Press. Dowty, D. R. (1979): Word Meaning and Montague Grammar. Dordrecht, Holland: D. Reidel. Fodor, J. A. (1981): Representations. Cambridge, Mass.: MIT Press. Fodor, J. A. (1983): The Modularity of Mind, Cambridge, Mass.: MIT Press. Frege, G. (1879): Begriffsschrift, eine der arithmetischen nachgebildete Formelsprache des reinen Denkens. Halle a. S.: L. Nebert. Geurts, B. (1997): Good News about the Description Theory of Names. Journal of Semantics 14, 319–348. Hale, K. & S. J. Keyser (1993): On Argument Structure and the Lexical Expression of Syntactic Relations. In: K. Hale & J. Keyser (eds): The View from Building 20, Cambridge, Mass.: MIT-Press, 53–109. Halle, M. (1983): On Distinctive Features and their Articulatory Implementation. Natural Language and Linguistic Theory 1, 91–105. Halle, M. (1995): Feature Geometry and Feature Spreading. Linguistic Inquiry, 26, 1–46. Jackendoff, R. S. (1996): The Architecture of the Linguistic-Spatial Interface. In: P. Bloom et al. (eds): Language and Space, 1–30. Jackendoff, R. S. (1997): The Architecture of the Language Faculty, Cambridge, Mass.: MIT Press. Jackendoff, R. S. & F. Lehrdal (2005): The Capacity for Music: What Is It and What's Special About It? To appear in: Cognition. Johnson-Laird, P. N. (1996): Space to Think. In: P. Bloom et al. (eds), 437–462. Katz, J. J. (1972): Semantic Theory. New York: Harper & Row. Katz, J. J. & J. A. Fodor (1963): The Structure of a Semantic Theory. Language 39, 170–210. Levinson, St. C. (1996): Frames of Reference and Molyneux's Question: Crosslinguistic Evidence. In: P. Bloom et al. (eds): Language and Space 109–169. Montague, R. (1974): Formal Philosophy, Selected Papers. Edited by Richmond H. Thomason. New Haven: Yale University Press. Osherson, D. N., St. M. Kosslyn & J. M. Hollerbach (eds) (1990): Visual Cognition and Action. Cambridge, Mass.: MIT Press.

32

Manfred Bierwisch

Partee, B. H, A. ter Meulen & R. Wall (1990): Mathematical Methods in Linguistics. Dordrecht: Kluwer Academic Publishers. Pustejovsky, J. (1995): The Generative Lexicon. Cambridge, Mass.: MIT Press. Schnelle, H. (1999): Mental Computation – A Critical Analysis of some Proposals by M. Bierwisch. Theoretical Linguistics 25, 257–281. Shepard, R. N. & L. A. Cooper (1982): Mental Images and Their Transformations. Cambridge, Mass.: MIT Press. Searle, J. R. (1969): Speech Acts. Cambridge: Cambridge University Press. Wunderlich, D. (1997): Cause and the structure of verbs. Linguistic Inquiry 28, 27–68. The American Heritage Dictionary of the English Language (1992). Boston: Houghton Mifflin Co.

Johannes Dölling & Tatjana Heyde-Zybatow (Leipzig)

Verb Meaning: How much Semantics is in the Lexicon? 1 Introduction Obviously, verbs contribute fundamentally to the meaning of utterances which they are part of. But there is a continuous discussion about the concrete nature of their contribution. Particularly, the fact that many verbs exhibit more than one literal, i.e. conventionalized meaning gives rise to a number of questions concerning their semantics. Thus, in view of the variability of use it is reasonable to ask how much of the conceptual information – conveyed by utterances – is actually supplied by the lexical meaning of the verb. More specifically, one could ask to what extent the interpretation of verbs is triggered by semantic information from the lexicon. Or, to say it in another way, what do we know about the meaning of a verb when we know its lexical semantics? Three strategies of analyzing the problem can be roughly distinguished: the maximalist, the intermediate and, the minimalist strategy. Maximalist approaches try to derive as much information as possible from the lexicon and, therefore, assume rather rich semantic characterizations of verbs. In particular, they suppose that each meaning variant of a verb should be represented with a separate lexical semantic entry. By way of contrast, minimalist approaches give prominence to the assumption that verbal semantics is at most confined to a few abstract fixings in the lexicon. As a consequence of attributing more importance to contextual modulation in the interpretation, they negate the existence of lexical polysemy. Finally, intermediate approaches also aim at reducing the number of lexical semantic representations for one and the same verb. However, as intermediate approaches are projectionist in nature this reduction only works at the cost of expanding the part of general semantic assumptions in the lexicon. In this paper, we review those strategies and argue for an analysis that adopts a minimalist attitude towards the lexical semantics of verbs. Our considerations are founded on the assumption that the relation between the grammatically determined meaning of a linguistic expression and its interpretation in a particular context is a complex and many-lay-

34

Johannes Dölling & Tatjana Heyde-Zybatow

ered one. Particularly, we suppose that the content of an utterance is not conveyed directly by this utterance. Instead, it is encoded in a rather fragmentary, schematic structure called semantic form and can only be accessed via inferences that provide the necessary enrichment of the conceptually underspecified semantic representation on the basis of world knowledge. Thus, the understanding of utterances always involves a pragmatic process of inferential completion of the linguistically given information. We want to investigate this general phenomenon with respect to the lexical semantic form of verbs and their contextual specification in utterance interpretation. The paper is organized as follows. Chapter 2 gives a survey of relevant data of meaning variation of verbs. We show that there have to be distinguished two main types of variation of primary meaning within the verbal domain. Chapter 3 discusses the maximalist attitude towards the lexicon and examines in more detail a particular proposal that is prototypical for the strategy of lexical meaning specification and, hence, of lexical polysemy. Chapter 4 serves the comparison of some intermediate analyses. Thereby we turn our attention to the framework of lexical decomposition, because it is the most important and most influential device of this strategy of the semantics of verbs. The discussion of the three strategies is concluded in chapter 5 by examining several minimalist approaches. Particularly, we evaluate a more recent, syntactic line of meaning decomposition that ultimately results in a disposal of the lexical verb semantics as such. Chapter 6 offers our proposal of a radically underspecified semantic representation of verbs, which is viewed as the indispensable lexical base for contextually varying the verbal meaning. In chapter 7 we sketch how the differentiation between several meaning variants can be accomplished by pragmatic enrichment of semantic forms.

2 Meaning Variation of Verbs One of the most controversial topics of the research on verb meaning is the phenomenon traditionally called polysemy (cf. e.g. Cruse 1986, Pinkal 1995, Ravin & Leacock 2000 and Pethö 2001). The main characteristic of a verb being subject to a variability of this kind is that it has a set of different meanings which are related to each other in nontrivial ways and are equally conventionalized. Such variations of pri-

Verb Meaning: How much Semantics is in the Lexicon?

35

mary (or literal) verb meaning1 differ essentially from the ones that involve a transfer to a secondary (or non-literal) meaning i.e. shifts of verb meaning.2 Moreover, there have to be distinguished two main types of variations of primary verb meaning – the systematic (or regular) and the non-systematic (or non-regular) one. Instances of both types come along with alternations in the argument structure of the verbs, i.e. in number or order of their argument positions or in sortal restrictions on their arguments. Systematic variations of primary verb meaning are characterized by general relations between the several readings and, therefore, a more or less large number of verbs exhibit this type of variation in similar fashion. They are of major concern for the semantic and syntactic research for quite some time (cf. e.g. Grimshaw 1990, Jackendoff 1990, Levin 1993, Goldberg 1995 and Levin & Rappaport Hovav 1995). A typical kind of systematic meaning variation of verbs is the alternation between a causative and an inchoative reading exemplified by sentences as in (1a) and (1b), respectively. (1)

a. b. c.

Anna schmolz das Eis. Anna melted the ice. Das Eis schmolz. The ice melted. *Anna schmolz [irgendetwas]. *Anna melted [something].

Whereas (1b) describes a change of state of the ice (1a) describes Anna's bringing about of such a change. It appears that the causative reading of the verb schmelzen 'melt' is connected to its transitive use and the inchoative reading is connected to its intransitive one. In the 1

2

Generally, we will use the more neutral term variation of primary meaning instead of polysemy for reasons that will become clear later in the text. Examples of meaning shift are given by sentences like (i) and (ii). (i) #Die Flasche gefror. #The bottle froze. (ii) #Das Kind nieste zehn Minuten lang. #The child sneezed for ten minutes. Whereas the arguments of the verb gefrieren 'freeze' in its primary meaning are restricted to NPs of liquids in (i) it has a reading in which its argument refers to a container of liquid. Analogously, whereas originally niesen 'sneeze' applies only to occurrences of a single sneezing in (ii) the verb refers to a process of repeated sneezing. As is indicated by #, both verbs are subject to a particular coercion by means of which they are getting a derived, secondary reading. In this paper we will not deal with such variations of verb meaning. For a general approach see e.g. Dölling (2003, 2005) and Egg (2005).

36

Johannes Dölling & Tatjana Heyde-Zybatow

causative the nominative NP refers to the agent of the eventuality3 described and the accusative NP to the theme, but in the inchoative the nominative NP refers to the theme of the eventuality. The most essential observation is that a sentence like (1a) entails one like (1b). In this manner, the entailment relation makes it clear that the two meaning variants of the respective verb are systematically related to one another. Moreover, the case of (1c) indicates that a verb like schmelzen has no intransitive use in which the theme is left implicit. Other verbs of this type are e.g. verbrennen 'burn' and zerbrechen 'break'. A second kind of systematic meaning variation connected to the transitive/intransitive alternation of a verb is instantiated by fegen 'sweep' in (2) (and, analogously, e.g. by wischen 'wipe', essen 'eat' or schreiben 'write'). (2)

a. b. c.

Hans fegte den Flur. Hans swept the floor. *Der Flur fegte. *The floor swept. Hans fegte [irgendetwas]. Hans swept [something].

Here the difference between the two meaning variants of the verb follows from the fact that in (2a) the theme of sweeping is referred to, but in (2c) it is not. As the last sentence is entailed by the first one both uses of fegen stand in a general meaning relation to each other. Furthermore, (2b) shows that unlike to (1b) no reading of the verb exists where the subject NP refers to the theme. Moreover, we have to distinguish between systematic meaning variations of verbs like fegen and wischen on the one hand and such of verbs like essen and schreiben on the other hand. In (3a) the verb fegen is applicable either to processes or to events and, thus, has two possible readings, in (3b) it can only apply to processes. By way of contrast, in (4a) the verb essen may be predicated of events only whereas in (4b) it shows an behaviour analogously to fegen in (3b). Essentially, in each case the (b)-sentence follows from the respective (a)-sentence. (3)

a. b.

3

Maria fegte zwei Flure in zehn Minuten/zehn Minuten lang. Maria swept two floors in ten minutes/for ten minutes. Maria fegte Flure *in zehn Minuten/zehn Minuten lang. Maria swept two *floors in ten minutes/for ten minutes.

The term eventuality is used in a general sense (cf. Bach, 1986) comprising events, changes, actions, processes, activities, states etc.

Verb Meaning: How much Semantics is in the Lexicon?

(4)

a. b.

37

Maria aß zwei Äpfel in zehn Minuten/*zehn Minuten lang. Maria ate two apples in ten minutes/*for ten minutes. Maria aß Äpfel *in zehn Minuten/zehn Minuten lang. Maria ate apples *in ten minutes/for ten minutes.

The data in (4) indicate that the meaning of essen alternates depending on whether the accusative NP has a quantized reference  (4a)  or a cumulative reference  (4b). Such a meaning alternation is based on the fact that the objects referred to by accusative NPs in sentences like (4) figure as an incremental theme in the eventualities described.4 Another kind of systematic variation is demonstrated by (5). (5)

a. b. c.

Luisa kochte die Suppe. Luisa cooked the soup. Die Suppe kochte. The soup cooked. Luisa kochte [irgendetwas]. Luisa cooked [something].

The sentences in (5b) and (5c) suggest that a verb like kochen 'cook' (as well as e.g. rauchen 'smoke' and backen 'bake') allows for two intransitive uses. It appears that the relation between (5a) and (5c) is the same as that between (2a) und (2c). More importantly, in (5a)  as well as in (5c)  the verb has an agentive (or causative) reading, whereas in (5b) it has a non-agentive one. Similarly to the relation between (1a) and (1b), the sentence in (5a) entails that in (5b). As illustration of a last kind of systematic meaning variation, which can be connected to alternation between the transitive and intransitive use of the respective verb, consider the sentences in (6). (6)

a. b.

Fritz trank den Alkohol. Fritz drank the alcohol. Fritz trank [, d.h. er war Alkoholiker]. Fritz drank [, i.e. he was an alcoholic].

Unlike the verbs in (2c) and (5c) trinken 'drink' in sentence (6b) does not exhibit an episodic but a habitual reading. This means that there are at least two different meaning variants of the verb, which are related: in (6a) trinken denotes a property of events, in (6b) it denotes a property

4

See Krifka (1989, 1992) for a seminal explanation of this kind of systematic meaning variation. Cf. also e.g. Filip (1999) and Rothstein (2004).

38

Johannes Dölling & Tatjana Heyde-Zybatow

of states that are realised by such events (cf. Parsons 1990). An analogous variation can be observed, e.g., with the verb rauchen 'smoke'. Contrary to the meaning variations discussed so far non-systematic variations of primary verb meaning are in each case specific for a particular verb and, therefore, idiosyncratic in a way. So, the sentences in (7) illustrate various readings that the verb öffnen 'open' can have.5 (7)

a. b. c. d. e. f.

Der Pförtner öffnete die Tür.6 The porter opened the door. Die Mutter öffnete das Paket. The mother opened the parcel. Der Angestellte öffnete die Datei. The clerk opened the file. Die Dame öffnete die Halskette. The lady opened the necklace. Der Junge öffnete das Taschenmesser. The boy opened the pocket knife. Der Verkäufer öffnete das Geschäft. The shop assistant opened the shop.

All the situations described above have in common that the object referred to by the accusative NP is open in a special way after the respective event of opening is accomplished. At the same time there are important differences between the various actions depending on the sort and the properties of the respective object. The several readings of öffnen can be clearly separated from each other so that it would be not appropriate just to assume that the lexical meaning of öffnen is simply vague. Instead there are different meaning variants which cover a rather large range of more or less similar domains of eventualities. Therefore, the respective meaning variants are connected to each other by relations of similarity. Meaning variations based on similarity between the domains covered can be found with many verbs. The examples in (8) and (9) show that the several meaning variants of a verb do not only depend on the

5 6

Cf. Searle (1983) for an analogous discussion of the English verb open. Notice that the verb can also be used intransitively, but only if it has a meaning that is systematically related to such one given in (7a): (i) Der Pförtner öffnete [die Tür]. The porter opened [the door]. (ii) *Der Pförtner öffnete [das Paket/die Datei/...]. *The porter opened [the parcel/the file/...].

Verb Meaning: How much Semantics is in the Lexicon?

39

different accusative NPs but also on the kind of the nominative NPs (in the case of nehmen 'take')7 or the PPs (in the case of verlieren 'lose').8 (8)

a. b. c.

(9)

a. b.

Der Verkäufer nahm das Geld. The shop assistant took the money. Der Dieb nahm die Juwelen. The thief took the jewels. Die Frau nahm den Mann. The woman took the man. Hans verlor sein Geld beim Wandern. Hans lost his money at hiking. Hans verlor sein Geld an der Börse. Hans lost his money at the stock exchange.

Besides the general world knowledge associated with the constituents of a sentence we often need to know the special context of an utterance to get the right meaning of a particular occurrence of a verb. (10) Hans verlor sein Geld. Hans lost his money. (11) Maria verließ die Schule. Maria left (the) school.

In the case of (10) the context has to determine whether Geld 'money' is used in a more concrete sense like in (9a) or in a more abstract sense like in (9b). Similarly, the circumstances of uttering a sentence like (11) have to clarify whether Maria has left the building or the institution. Accordingly, the verb verlassen 'leave' denotes a property of local or of social changes. To conclude, we point out that like systematic meaning variations non-systematic ones can be correlated with alternations in syntactic structure as the following sentences show. (12) a. b.

7

8

Maria schlug dem Jungen auf die Schulter. Maria hit the boy on the shoulder. Maria schlug ihren Vater im Schach. Maria beat her father in chess.

Obviously, this is true of öffnen in a similar way. Cf. e.g. sentences (i) and (ii) where the nominative NP does not refer to an agent. (i) Der Schlüssel öffnete die Tür. The key opened the door. (ii) Der Wind öffnete die Tür. The wind opened the door. Cf. e.g. Bierwisch (1983, 1996) and Ruhl (1989) for a discussion of similar examples.

40

Johannes Dölling & Tatjana Heyde-Zybatow

c. d.

Maria schlug ein Loch in die Wand. Maria hit a hole into the wall. Maria schlug einen Bogen um das Hindernis. Maria compassed the fence.

The different meaning variants of the verb schlagen 'hit' ('beat', 'compass') are associated with specific distributions of thematic roles and respectively with particular syntactic realisations. After having outlined the two main types of variation of primary verb meaning, we will examine how proposals of the three different strategies of lexical semantics of verbs – maximalist, intermediate and minimalist – account for them.

3 Maximalist Analyses The basic assumption of maximalist analyses is that variations of primary verb meaning are instances of lexical polysemy, i.e. a phenomenon of lexical semantics. Consequently, the characterization of a single verb in the lexicon is very rich in order to cover all its different readings. In particular, the several meaning variants are listed as separate semantic representations. On the one hand, this strategy of lexical specification of verb meaning seems to reflect most adequately our common experience and is viewed as being in accordance with the lexicographical practice. On the other hand, however, maximalist approaches have to face the objection that they lack generalizations and, therefore, systematicity and cognitive economy.9 A prime example for an analysis of this kind is Engelberg (2000). The author assumes that the lexical entry of a verb consists of at least two specifications for each of its variants: SYN contains the syntactic subcategorization and hence also the complement structure of the verb variant. SEM contains its semantic representation in form of a O-abstract as well as a set of associated meaning postulates. These meaning postulates determine the semantic entailments between the different meaning variants of the respective verb and furthermore provide the sortal restrictions of the verbal arguments. How many variants of a verb have to be assumed depends on the number of the different syntactic constructions the verb can be part of as well as on the number of the various sortal restrictions on the explicit and implicit arguments of the verb. For instance, Engelberg differentiates between five variants of the verb 9

Cf. e.g. Pustejovsky (1995) for a critical examination.

Verb Meaning: How much Semantics is in the Lexicon?

41

spülen 'wash' (or 'rinse'), which are realised with the help of spülen1 – spülen5 in (13). (13) a. b. c. d. e.

Karl spülte1 die Wäsche/den Mund/das Geschirr/die Haare/die Wunde/... . Karl washed the clothes/the mouth/the dishes/the hair/the wound/... . Karl spülte2 seiner Mutter die Wäsche/den Mund/das Geschirr/... . Karl washed the clothes/the mouth/the dishes/...for his mother. [Mit Hinweis auf Geschirr:] Karl spülte3 gerade. [In connection to dishes:] Karl just washed. [Beim Zahnarzt:] Karl spülte4 gerade. [At the dentist:] Karl just rinsed. Die Waschmaschine spülte5 gerade. The washing machine just washed.

As the sentences suggest, the determination of the lexical entry for spülen has to take into account three different subcategorization frames (spülen1 vs. spülen2 vs. spülen3  spülen5) and, additionally, in view of (13c–d) three different sortal restrictions on implicit arguments of the verb (spülen3 vs. spülen4 vs. spülen5). In (14) we give a slightly modified version of Engelberg's (2000: 154) entry, where x, y, z are variables for objects, e is a variable for eventualities, WASH1  WASH5 are predicate constants representing the meaning of spülen1  spülen5, respectively, and … is the operator of conceptual necessity. (14) spülen1: SYN: V, /NPacc/NPnom SEM: OyOxOe. WASH1(x, y, e) MPspülen1: …xye[WASH1(x, y, e) o CLOTHES(y) › MOUTH(y) ›…] spülen2: SYN: V, /NPacc/NPdat/NPnom SEM: OyOzOxOe. WASH2(x, y, z, e) MPspülen2 : …xyze [WASH2(x,y, z, e) o ANIMATE(x) & ANIMATE(z)] MPspülen3: … xyze [WASH 2(x, y, z, e) o WASH 1(x, y, e)] spülen3: SYN: V, /NPnom SEM: OxOe. WASH 3(x, y, e)] MPspülen4: … xye [WASH 3(x, y, e) o DISHES(y)] MPspülen5: … xye [WASH 3(x, y, e) o WASH 1(x, y, e)] spülen4: SYN: V, /NPnom SEM: OxOv. WASH 4(x, y, e)] MPspülen6: … xye [WASH 4(x, y, e) o HUMAN(x)] MPspülen7: … xye [WASH 4(x, y, e) o MOUTH(y)] MPspülen8: … xye [WASH 4(x, y, e) o WASH 1(x, y, e)]

42

Johannes Dölling & Tatjana Heyde-Zybatow

spülen5: SYN: V, /NPnom SEM: OxOe. WASH 5(x, y, e)] MPspülen9: … xye [WASH 5(x, y, e) o WASH_MACHINE(x)] MPspülen10: … xye [WASH 5(x, y, e) o CLOTHES(y)] MPspülen11: … xye [WASH 5(x, y, e) o WASH 1(x, y, e)]

Engelberg uses the Davidsonian format for the semantic representation of verbs.10 Thus, the predicate constants WASH1  WASH5 denote different relations between eventualities of washing and the objects participating in them. More specifically, there are four three-place predicate constants WASH1, WASH3  WASH5 and one four-place predicate constant WASH2. As it is indicated by the semantic representations of spülen3 – spülen5, each of the verb variants has an implicit argument position, i.e. the arity of the O-abstract does not correspond to that of the predicate constant being part of it. Further properties of the predicate constants are determined by the meaning postulates. Especially, postulates MPspülen3, MPspülen5, MPspülen8 and MPspülen11 define the relations between the various predicate constants by determining that each of WASH2  WASH5 entails WASH1. It is the existence of these relations, which ensures that the meaning variations are connected to each other. The remaining postulates characterize the sortal restrictions that have to be met by the explicit and implicit thematic arguments. It is essential that Engelberg's analysis consists in a detailed and precise formal characterization of the lexical semantics of a great number of verbs. Nevertheless, there have to be made at least two major critical points with respect to it. The first problem is that a lexical entry like (14) does not represent all the possible meaning variants of spülen. On the one hand, both spülen1 and spülen2 can be further differentiated in their meaning. It is evident that the verb has different readings – similar to the meaning variants of öffnen in (7) – if it is combined with alternating accusative NPs like in (13a) or (13b). On the other hand, for instance, the reading of spülen in a sentence like (15) – the German counterpart of flush – is not covered by the entry in (14). This reading is highly context dependent and can be not expressed by spülen3 or spülen4 as (13c) and (13d) show. (15) [Mit Hinweis auf die Toilette:] Karl spülte gerade. [In connection to toilet:] Karl just flushed. 10

The Davidsonian representational format differs from the traditional one in that the verbs dispose of an eventuality argument in addition to their thematic arguments.

Verb Meaning: How much Semantics is in the Lexicon?

43

Furthermore, because the respective predicate constant would have only one thematic argument position it cannot be related to the other ones. The second and more serious problem concerns the sortal restrictions as part of a verb's entry. It is a consequence of the proposal that, for instance, the decision between the different variants of spülen can only be drawn on the basis of world knowledge. As an example, in sentences like (16a) and (16b) the sortal restrictions of the verb can only be met if additional information about the NPs er 'he'/sie 'she'/es 'it' and Nummer 5 'number 5', respectively, is available. (16) a. b.

Karl spülte ihn/sie/es gerade. Karl just washed him/her/it. Nummer 5 spülte gerade. Number 5 just washed.

Similarly, this is true of the differentiation between the readings of spülen like in (13c), (13d) and (15), where information about the specific situation of utterance is necessary to select the adequate variant. There are two possible solutions to the latter problem: either one has access to non-linguistic context knowledge during the process of the semantic combination in order to chose the appropriate variant of the verb or one has to compute all the possible variants in parallel and decides at the end of the computation which variant is the adequate one in the given context. The first solution contradicts the principle of semantic compositionality and is not valid in the Montagovian tradition. The second solution implies a rather extensive (and costly) computation, which is undesirable for economical reasons. Other maximalist approaches can be found within the field of Cognitive Grammar. For instance, Langacker (1987, 1988) and Taylor (1995, 2000) argue for a lexical network model to represent the meaning variants of verbs. This means that every polysemous verb is represented in the lexicon by the entirety of its readings which are linked with each other in manifold ways. Admittedly, researchers adopting this position have particularly delivered extensive data on non-systematic meaning variation of many verbs. Contrary to Engelberg's approach, however, the network theory negates the fundamental difference between grammatical and encyclopedic knowledge. Rather, the assumption is that lexical meanings of verbs always constitute themselves on the basis of the whole general knowledge about the respective eventualities. But such a holistic approach to linguistic meaning restricts the validity of the principle of semantic compositionality fundamentally. Thus, it seems to be impossible to explicate the contribution that a par-

44

Johannes Dölling & Tatjana Heyde-Zybatow

ticular verb makes to the grammatically determined meaning of expressions.

4 Intermediate Analyses Proposals that can be characterized as intermediate make up by far the largest group of approaches to verb semantics. Whereas they are lexicalistic by nature  like maximalist analyses  as verbs are assumed to have lexical semantic representations from which syntactic structures are projected11, they also exhibit similarities with the minimalist strategy in that generalizations on meaning variants help to reduce the number of lexical semantic representations of a verb. Roughly speaking, there are two types of intermediate analysis: The first type supposes that lexical representations of verbs are underspecified and, with it, somehow abstract. The particular meaning variant of a verb results then from specification in the respective context. In contrast to it, approaches of the second type assume one meaning variant of a verb as the basic one. This concrete meaning has to be lexically represented and all other readings are contextually derived from it. The most important strategy in the realm of intermediate analyses is that of lexical decomposition, which can be traced back to the Generative Semantics tradition and has been developed within a more formal framework by Dowty (1979). Its fundamental assumption is that in the lexicon verb meanings can be represented in terms of a relatively small set of general primitives, typically drawn from ontological categories. The major advantage of the strategy is seen in that the lexical semantic representation of a verb contains some information about the structure of the eventualities referred to immediately. So this information is accessible without possibly long-winded inferential processes and, more importantly, can determine the syntactic behaviour of the verb. The Conceptual Semantics of Jackendoff (1983, 1990, 2002) is not only an example of the first type of intermediate approaches but also a protoypical instance of the decompositional strategy. Jackendoff regards the conceptual structure CS of a verb as a configuration that is constituted of conceptual primitives and represents the lexical meaning

11

For different lexicalist approaches that view stored semantic information about verbs as central for the syntactic derivation see e.g. Jackendoff (1983), Grimshaw (1990), Williams (1994), Levin & Rappaport Hovav (1995), Heim & Kratzer (1998) and Reinhart (2003).

Verb Meaning: How much Semantics is in the Lexicon?

45

of the verb. The verbal argument structure as well as the sortal requirements on the arguments are defined with respect to these primitives. We illustrate the approach by means of the verb climb, which appears in sentences like (17a)  (17c) in three syntactic contexts: with a null complement, with a NP, and with a PP. (17) a. b. c.

John climbed. John climbed the mountain. John climbed to the top of a mountain.

Jackendoff (1990) unifies the different uses into the lexical entry in (18), where the meaning of the verb is assumed as being decomposed into the conceptual primitives GO, TO and TOP-OF. (18)

PF: /climb/ SYN: V, ___ ¢XPj² CS: [Event GO ([Thing ]i, [Path {TO ([Place TOP-OF ([Thing ]j )])} ]{j} )]

In SYN the angled brackets indicate that the postverbal complement is optional; in CS the two possible specifications of the path-constituent as [Path TO ([Place TOP-OF ([Thing ]j)])] or as [Path ]j  depending on whether the postverbal complement is a NP or a PP  are marked as mutually exclusive by using braces. In the syntactic derivation the single meaning variants of climb emerge from CS by either specifying the path-constituent in the case of (17b) and (17c) or leaving the path-constituent undetermined in the case of (17a). A deficit of the proposal is that a lexical entry like (18) does not cover non-systematic variations of primary meaning of the verb like those in (19), i.e. variants of the kind of e.g. öffnen in (7) or nehmen in (8). (19) a. b. c.

Bill climbed [up] the mountain. Bill climbed down the mountain/the rope/along the ridge/ through the tunnel/…. The snake climbed [up] the tree.

Jackendoff (1990, 2002) suggests for these cases to identify the lexical meaning of verbs with a system of preference rules being part of the encyclopedic knowledge. These yield depending on the particular use the adequate reading. Unfortunately, it is completely left open how these two perspectives on verbal meaning can be unified. Another deficit of Jackendoff's approach is that he acknowledges the necessity of

46

Johannes Dölling & Tatjana Heyde-Zybatow

generalized and, therefore, underspecified lexical CSs but does not draw any further consequences from that for his theoretical frame. By contrast, the Two-Level Model of Meaning – established by Bierwisch (e.g. 1983, 1996, 1997, 2002), Bierwisch & Lang (1989) and related work (cf. e.g. Lang 1994, Kaufmann 1995, Ehrich 1996, Stiebels 1996, Steube 1997, Wunderlich 1997, 2000 and Zimmermann 1999) – is based on the assumption of a separate level of semantic form SF alongside that of the level of conceptual structure CS. Whereas in terms of CS the context dependent, conceptual interpretation of expressions is organized, SF has the task of representing the purely grammatically calculated meaning. Consequently, representations of this level are determined compositionally by syntactic and lexical semantic information and have to be conceptually underspecified with respect to the concrete content of utterances. Thus, SF plays the role of an interface between syntax proper and what Chomsky (2004) calls the Conceptual-Intentional System. As is suggested, the lexically stored semantic information reflects only those parts of verbal meaning that are grammatically relevant. This means that the lexical SF of a verb has to be understood as a schema imposing grammatical restrictions on meaning variation of the particular verb and, thus, giving rise to restrictions on syntactic structure projected. Concerning the concrete representational format of SF, the TwoLevel Model relies on a framework of lexical decomposition, which is essentially an eventuality-based adaptation of the classical proposal of Dowty (1979). According to Bierwisch (1997, 2002), the lexical entry for öffnen can be given as in (20). (20)

PF: /œffn-/ SYN: V SF: OyOxOe. e : (ACT(x) (CAUSE (BECOME (OPEN(y)))))

The SF in (20) is a decompositional structure that includes the oneplace predicate ACT(ivity), the propositional operators CAUSE and BECOME as well as the one-place predicate OPEN as its primitive components. The colon ':' is assumed as a two-place predicate expressing that an eventuality is an instance of the proposition represented by the structure following after the colon. A lexical entry as in (21) is supposed to yield the basis for handling systematic meaning variations like such ones exemplified by schmelzen in (1).

Verb Meaning: How much Semantics is in the Lexicon?

(21) PF: SYN: SF:

47

/³melts-/ V Oy[Dœ Ox]Oe. e : [D (ACT(x) (CAUSE] (BECOME (LIQUID(y))))[D )]

Here the index D ranges over + and ; the notation [DX] indicates that X is present if D is +, and absent otherwise. Accordingly, the causative and the inchoative reading of schmelzen can be derived depending on whether an argument for the position labeled by Ox is syntactically available or not. The two readings are represented by the structures in (22a) and (22b), respectively. (22) a. b.

OyOxOe. e : (ACT(x) (CAUSE (BECOME (LIQUID (y))))) OyOe. e : (BECOME (LIQUID (y)))

Moreover, a solution for cases of non-systematic variation of primary meaning has been offered by the approach too. Bierwisch (1996) argues that the decompositional strategy allows a unified SF for verbs like verlieren or verlassen. The verb verlieren, for instance, could have the following lexical entry: (23) PF: SYN: SF:

/verli:r-/ V OyOxOe. e : (BECOME (™HAVE(x, y)))

The assumption is that depending on the contextual specification of the two-place predicate HAVE either as denoting a relation of concrete, physical possession or as denoting a relation of abstract possession on the level of CS the different readings of the verb in (9) and (10) emerge. Our attitude towards the Two-Level Model is a twofold one. On the one hand, we agree that levels of meaning of different specificity should be distinguished. Particularly, our proposal outlined in chapter 6 and 7 adopts this idea and wants to advance it in several respects. On the other hand, we observe some difficulties that come with the implementation of decomposed meaning structures:12 Firstly, it appears that in decompositional analyses like that one proposed the eventuality structure of verbs is not really represented in meaning structure. CAUSE and BECOME are used as propositional operators, although one would expect that they are directly related to eventualities. Furthermore ACT, representing the activity of the agent, is not used as a predicate of activities but of objects. Thus, the connec-

12

See also the discussion in Dölling (2005).

48

Johannes Dölling & Tatjana Heyde-Zybatow

tion to eventualities is only indirectly established via instantiation of 'eventuality-free' propositions.13 Secondly, it is a well-known problem of the decomposition strategy that structures containing CAUSE are logically weaker than the meaning of the verbs those structures should be the representations of (cf. e.g. Chierchia & McConnell-Ginet 2000 and Engelberg 2000). So, in conflict with the decompositional structure in (20), it is not the case that in every situation in which an action of x causes that y opens one can say that x opened y.14 In addition, it is not clear how under this condition, for instance, the meaning of verbs denoting eventualities of more specific openings, i.e. verbs like aufreißen 'rip open', aufbrechen 'break open', aufriegeln 'unbolt' or auftrennen 'undo' can be represented. Thirdly, it is an open question whether decompositional structures of the given kind can adequately cover all variations of primary meaning. Generally, we doubt that such representations yield sufficient conditions for the contextual identification of the several non-systematic meaning variants of a verb. For instance, it is not clear how the actual differentiation of meaning like that in (7) works if the SF of öffnen in (20) is taken as a starting point. Obviously, to specify only the oneplace predicate OPEN is not enough for getting the right reading of the verb. But if so, which of the other primitives occurring in the decompositional structure should be specified too and how should it be done in coordination with each other? Moreover, as according to (20) the eventualities of opening have to involve an activity the SF is not qualified for uses of the verb in sentences that describe situations lacking an agent.15 Our general comments on lexical decomposition hold analogously for the approach of Rappaport Hovav & Levin (1998, 2003), which is at present very influential. Its basic idea is that the lexical meaning of verbs has to be divided into an idiosyncratic and a structural part. The latter one is a composition of lexical semantic templates that are based on the decompositional format of Dowty (1979). (24) illustrates such schematic eventuality structures:

13

14

15

For a Dowty-style procedure of semantic decomposition that in contrast reflects the eventuality structure immediately see e.g. Parsons (1990). For instance, it is possible that a person unintentionally causes the opening of a door (e.g. by means of an air draft), but one would not say that the person opened the door. Cf. examples given in footnote 7. Egg (1994) demonstrates that an exclusively decompositional approach to lexical SF has to fail, because often such structures set the boundaries for possible meaning variations too narrow. He proposes to combine it with a lexical network approach.

Verb Meaning: How much Semantics is in the Lexicon?

(24) a. b. c. d. e.

49

activity: [x ACT] states: [x ] achievements: [BECOME [x ]] accomplishments1: [[x ACT]CAUSE [BECOME [y ]]] accomplishments2: [x CAUSE [BECOME [y ]]]

The idiosyncratic information of singular verbs, i.e. their constant 'core' meaning enters the templates at the position designated by the angled brackets. Levin and Rappaport Hovav use these kinds of semantic representations to explain, for instance, the different types of transitive/intransitive alternations. Thus, a verb like sweep has the representation in (25a) whereas a verb like melt the one in (25b). (25) a. b.

[x ACT y] [x CAUSE [BECOME [y ]]]

According to the authors (25a) shows that activities of sweeping have a prototypical theme (indicated by y) which is the reason for the grammaticality of the intransitive use of sweep (cf. (2c)). Furthermore, because the verb denotes activities it always realises an agent. Thus the theme can never appear as subject NP (cf. (2b)). Contrary to the behaviour of sweep, the representation of melt is determined by the fact that this verb denotes changes of state or activities bringing about of such changes. The theme of the respective eventualities can never be omitted (cf. (1c)), but it can be realised as subject (cf. (1b)). The approach also takes the extension of argument structures into account in that the semantic templates in (24) can be integrated in more complex ones by a process called template augmentation. This mechanism is supposed to determine the possibility of alternations like that in (26a) and (26b) and the impossibility of those in (26c) and (26d). (26) a. b. c. d.

Otto swept the floor clean. Otto swept the crumbs in the corner. *Otto swept the crumbs. *Otto swept in the corner.

Essentially, such template augmentations and, with it, differentiations of verb meaning are performed in the lexicon.16 We concede that the investigations by Levin and Rappaport Hovav have delivered extensive data and, moreover, interesting generalizations 16

Another lexical procedure of argument extension is suggessted in Wunderlich (2000).

50

Johannes Dölling & Tatjana Heyde-Zybatow

on systematic meaning variation of verbs. But we have to say that the proposed analysis is in several respects too informal and vague for considering it a definitive explanation of the phenomena.17 In addition, the approach does not address the determination of the idiosyncratic part of verbal meaning and, with it, the specific potential of non-systematic variation of verbs. So far we have discussed theories that all belong to the first type of intermediate approaches to verb semantics. A representative of the second type – taking a meaning variant as the basic meaning of the verb – is Pustejovsky's Theory of the Generative Lexicon (1991, 1995).18 The author starts out from the assumption that in the lexicon a verb can have a single semantic representation that reflects only one of its primary meaning variants. For example, he supposes that the verb bake denotes in its lexical meaning a property of changes of state as in (27a). (27) a. b.

Paul baked a potato. Paul baked a cake.

By way of contrast, the reading in (27b), where bake denotes a property of creative actions, is the result of a coercion of meaning, i.e. a shift to another meaning variant of the verb. Although the idea of a generative lexicon seems rather appealing, two major problems of Pustejovsky's approach have to be pointed out: Firstly, it is absolutely unclear why one primary meaning variant of a verb should be the most basic one and the other variants have to be derived from it. There is, for instance, no explanation for the differing significance of the two occurrences of bake in (27). Secondly, the postulated operation of coercion impairs the grammatically determined process of combination, which results in a violation of the principle of semantic compositionality. This violation is due to Pustejovsky's assumption that the lexicon contains elements of genuine encyclopaedic knowledge, which can be inserted into semantic structure by means of coercion.19 What this short survey about intermediate analyses should demonstrate is that in spite of attempts to reduce the number of lexical semantic representations for one and the same verb, there are still quite a few non-trivial problems with the formalisms used in the several proposals. 17 18

19

See for a more detailed analysis e.g. Engelberg (2000) and Piñón (2001). It should be noted that in other respects Pustejovsky is also an adherent of the strategy of lexical decomposition. See for a general criticism Fodor & Lepore (1998).

Verb Meaning: How much Semantics is in the Lexicon?

51

5 Minimalist Analyses Minimalist approaches give up on the lexicalist position and, hence, try to shift the major burden of meaning determination away from the lexicon. The lexical semantics of verbs is reduced as much as possible to an abstract minimum  prototypically presented in Ruhl (1989)20  or even it is regarded as generally inaccessible during the grammatical computation  as recently argued by Borer (2005). The determining factor for the respective meaning variant of a verb is the syntactic and conceptual context in which the verb occurs. This means that syntactic and pragmatic aspects of verbal meaning gain considerably in importance compared to lexical semantic specifications. For instance, Marten (2002) explicitly argues for an approach in which the utterance meaning is the mere result of an interaction of syntactic and pragmatic processes.21 In the realm of minimalist analyses the constructionist strategy is an important one. All theories of Construction Grammar share the assumption that the meaning of a verb in a particular syntactic context results from an interrelation between the lexical meaning of the verb and the meaning of the respective construction. Especially, Goldberg (1995) argues for an analysis of argument structure as an abstract meaning of syntactic constructions and, therefore, as independent of the lexical semantic representation of verbs (cf. also Fillmore, Kay & O'Connor 1988 and Fillmore, Kay, Michaelis & Sag 2003). The verbal predicate itself possesses a concrete participant structure that has to agree with the particular argument structure. Otherwise the verb can not be embedded in the respective syntactic construction. Essentially, the constructions themselves can also contribute argument positions that have no corresponding participant role in the lexical entry of the verb. As an example, the representation of the 'caused motion' construction analyzed by Goldberg is given in (28). (28) SEM: CAUSE-MOVE PREDICATE SYN: V

> OBLdirectional

A verb like throw can occur in a sentence like (29), which instantiates the construction. 20 21

As is noted by the author, this work is done subsequent to Bierwisch (1983). Marten's framework is based on the Dynamic Syntax of Kempson, Meyer-Viol & Gabbay (2001) and the Relevance Theory of Sperber & Wilson (1986/1995).

52 (29)

Johannes Dölling & Tatjana Heyde-Zybatow

Sam threw the ball against the wall.

The two participant roles of the verb – 'thrower' and 'throwee' – are compatible with the two argument positions of the construction  agent and theme, respectively. Furthermore, the construction adds the path argument position and, with it, specifies the meaning of the verb. Although till now it is controversial what the source of these abstract meanings of syntactic constructions should be the idea to separate argument positions from the verb seems to be a promising one. Therefore we want to discuss this strategy of separation in a little more detail. Traditionally, argument structure is analyzed as part of the lexical entry of a verb. Consequently, the projection of syntactic structure is viewed as depending on the kind of argument structure the verb brings in (e.g. transitive vs. intransitive). All the known linking theories are developed in this realm of thinking. Over the past years, however, such lexicalist approaches have been come under attack by reason of owning the disadvantage to be fundamentally redundant in that the projected syntactic structure has to agree with the lexical properties of the listed items (cf. e.g. Borer 1994, 2005, Ritter & Rosen 1998 and Ramchand 2006). The redundancy originates from the double notation of the argument positions, first in the lexical argument structure and second in the syntactic subcategorization. Thus, linking mechanisms have to be a necessary part of those theories because otherwise it could not be assured that an (semantic) argument position is connected to the adequate (syntactic) complement. In view of the massive variability that verbs show with respect to their arguments a further difficulty has been pointed out. If argument positions are represented as part of the lexical semantic representation it is necessary to include all the different possibilities of argument structures and, hence, of meaning variants for one verb in the lexicon. Kratzer's (1996, 2003) proposal to sever the external argument from its verb can be regarded as a response to this challenge. Following Marantz (1984), she argues that there are many sentences like those in (7) where a particular sort of internal argument triggers a particular meaning variant of the verb, and claims that there are only a few instances where an external argument does the same. According to her, this observation can be explained if external arguments are not any longer analyzed as true arguments of verbs. Instead, Kratzer proposes that a (non-overt) functional head called voice carries the thematic information about the subject argument. The subject NP itself is located in the specifier position of voiceP.

Verb Meaning: How much Semantics is in the Lexicon?

53

As the external argument positions are introduced in the syntactic derivation they have not to be included in the lexical meaning structure of the verbs. More specifically, Kratzer uses a mixture of the (classical) Davidsonian and the neo-Davidsonian22 format of semantic representation. External arguments are represented as arguments of separate neoDavidsonian predicate constants like AG(ent), whereas all other thematic arguments (e.g. object NPs) are represented as true arguments of Davidsonian verbal predicates. The lexical semantic representation of a verb like open in its transitive variant is now a two-place predicate without an agent argument position. We illustrate a simplified derivation of sentence (30a) in (30b). (30) a. b.

John opened the door. voiceP Oe. AG(john, e) & OPEN(def_door, e) NP john

voice' OxOe. AG(x, e) & OPEN(def_door, e) voice OxOe. AG(x, e) NP def_door

VP Oe. OPEN(def_door, e) V OxOe. OPEN(x, e)

The introduction of the external argument is only successful if there is a mechanism that ensures the identity of the two eventualities in the above structure (the eventuality that comes with the verb and the one the AG-predicate is applied to). To this end, a principle of event(uality) identification is assumed. While numerous researchers agree on the sketched treatment of external arguments,23 there is an ongoing debate about the status of the internal arguments. For instance, Williams (2007) counters Kratzer's claim that they should be generally treated as true arguments of the 22

23

The term neo-Davidsonian is used for a representational format which separates a verb meaning into a predicate constant of eventualities and one or more thematic predicate constants denoting relations between eventualities and their participants. Cf. Parsons (1990, 1995) and Schein (1993). See e.g. von Stechow (1996), Pylkkänen (1999, 2006) and Beck & Johnson (2004). Von Stechow argues extensively for a separation of the external arguments on the basis of an analysis of the adverb again. He assumes that again in its repetitive reading is higher located than voiceP, whereas in its restitutive reading it must be below voiceP. If the agent would be a true argument of the verb the sensitivity of the adverb for its syntactic position could not be explained.

54

Johannes Dölling & Tatjana Heyde-Zybatow

verb. William's analysis of data from Igbo and Mandarin advocates that patient predicates should be introduced by the VPs, but not by the verbs. Based on this, a conclusion of broader relevance is abstracted: We need to distinguish between what sort of eventuality a verb is related to, and what combinatory requirements are associated with the verb lexically. A verb need not have as many lexical argument positions as the eventualities referred to have participants. Borer (2005) extends the idea of separation to any internal arguments. According to her, there is no need for a lexically represented argument structure because all thematic predicates are imposed by the syntactic construction in which the verbs occur. Generally, the argument structures of verbs are computed syntactically by means of various functional heads and independently of lexical information. In fact, Borer goes further and develops a rather radical minimalist theory. She is interested in the question why lexical items are so much more flexible with respect to the context they can be used in and, thus, in their concrete meaning than the syntactic structures themselves. To explain this notorious difference in flexibility, Borer reasons that the role syntactic configurations play in the determination of meaning has to be considered to a greater extent. The analysis of the constructionally determined parts of meaning is one of the main purposes of Borer's (2005) neo-constructionist approach to grammar, called the Exo-Skeletal Model.24 Within this framework the assumption of a lexicon as a store for word meanings is questioned altogether. Borer understands verbs  as well as nouns and adjectives  as concealed 'conceptual packages' which are embedded in the syntactic structure without affecting the structure or being affected themselves. Only after the syntactic structure has gained a grammatically determined meaning the 'conceptual packages' open and their content has to be verified with the structural meaning. This verification process takes place at a 'making sense' component, which is different from both the grammatical and the conceptual system. It is at this point that the meaning variant of a verb is determined with the help of the information of its thematic arguments. As a consequence, for instance, the choice between the causative and the inchoative variant of verbs like melt or open is licensed through syntactic structure. Actually, the proposal by Borer is closely related to a second, more recent line of the strategy of meaning decomposition, which follows the 24

Notice that Borer follows the generative tradition whereas most adherents of the classical constructionist strategy adopt basic positions that are closely connected to Cognitive Grammar.

55

Verb Meaning: How much Semantics is in the Lexicon?

idea that eventuality structure expressed by a verb is explicitly encoded in syntax. Generally, lexically stored decompositional representations of verb meaning are not anymore supposed. Instead, for instance, Lin (2004) argues that meaning variants of verbs should be viewed as being compositionally constructed from a universal inventory of primitives linked immediately to functional heads and an open class of 'verbal roots'.25 Following the author, a sentence like (31a) which contains the causative variant of melt has the syntactic 'eventuality' structure (31b) and can be interpreted as (31c). (31) a. b.

Mary opened the window. voiceP NP Mary

voice' vDOP

voice

vGP

vDO + DYN  INCHO

NP

vG' vBEP

the window vG + DYN + INCHO

—

vBE  DYN

open [state open]

c. Oe. ARGext(mary, e) & DO(e, [activity undef]) & e's [BE(s, [state open]) & BECOME(e', s) & ARGG(def_window, e') & CAUSE(e, e')]

In (31b) the verbalizing heads vDO, vG and vBE can be viewed as corresponding to the traditional primitives DO (or ACT), BECOME and BE, respectively. Moreover, it appears that the semantic representation in (31c) is similar to decompositional structures proposed in Parsons (1990). The labels ARGext and ARGG denote thematic relations which involve the external argument (introduced by voice) and the internal argument associated with vG, respectively. Finally, [state open] is used to represent the idiosyncratic part of meaning of the verb open, i.e. its 'verbal root'. 25

For further approaches being partly very distinct from Lin's one as well as from each others cf. e.g. Hale & Keyser (1993), von Stechow (1995, 1996), Ritter & Rosen (1998), Travis (2000) and Ramchand (2006).

56

Johannes Dölling & Tatjana Heyde-Zybatow

What can be concluded from the consideration of the various approaches is the tendency to reduce the lexical information to an absolute minimum in order to keep the variability in the use of verbs. On the one hand, we think that in pursuing this intention minimalist analyses are on the right track. Obviously, beside conceptual contexts of utterance the syntactic constructions a verb occurs in have more influence on meaning variation than it is commonly assumed. For this reason, lexical semantic representations of verbs should contain as few as possible structural restrictions. In particular, we sympathize with the idea that thematic information is not simply projected from the lexicon. On the other hand, however, there have to be made some critical points with respect to actual implementations of the minimalist strategy. Generally, we disbelieve the adequacy of structures that essentially result from a transfer of the device of lexical decomposition into syntax. Such structures are not only subject to most of our comments made above in regard to versions of Dowty's framework. They have to be also questioned for the fact that now verbs own no independent lexical semantic representation at all. Thus, it seems that the syntactic line of meaning decomposition overshoots the mark. Especially, Borer's (2005) theory ultimately aims at the complete elimination of verb semantics as part of the lexicon and, with it, of the grammatical system. By way of contrast, we claim that there has to be a minimal residual of lexical semantic representation of the verbs to gain a conceptually adequate use of them. In the next chapter we outline a minimalist approach which takes into account this requirement.

6 Lexical Semantic Form of Verbs and Grammatical Computation Our point of departure is the observation that the linguistic form drastically underdetermines the conceptual information that is conveyed with an utterance. Therefore, we distinguish between two major phases of the process of understanding: Firstly, in the phase of semantic composition, which is part of the grammatical computation, the context-independent and, hence, merely formal meaning of the utterance is calculated. Secondly, in the phase of pragmatic interpretation its contextually specified meaning is obtained by means of inferences integrating the formal meaning base with accessible elements of conceptual knowledge. As a consequence, each understanding process contains some procedures of

Verb Meaning: How much Semantics is in the Lexicon?

57

pragmatic enrichment, which decisively contribute to determining the kind of meaning called traditionally the semantic content of utterance.26 More specifically, we suppose a Multi-Level Model of Meaning (see Dölling 1995, 1997, 2003, 2005), which can be understood as an exploration of the Two-Level Model. Its basic idea is that the understanding process comprises a number of different levels over which the content of an utterance has to be spelled out step by step. Two of the various meaning levels are of particular interest to the topic dealt with here  the level of semantic form SF and that one called parameter-fixed structure PFS. Structures at the SF level, which represent the formal meaning of expressions and, therefore, form the starting point for interpretation, have two crucial properties: Firstly, they are built up strictly compositionally, i.e. they are determined exclusively by the lexical items occurring in that utterance and the manner of their syntactic combination. Thus, any interference with the autonomously organized level of SF by making reference to conceptual knowledge is excluded. Secondly, SF structures are radically underspecified with regard to the conceptual information conveyed.27 They contain a lot of parameters, i.e. free variables that can be substituted by constants from a particular domain assigned to the singular parameter. It is just such a contextually licensed fixing of SF-parameters with values that allows for considerable variation in the meaning of utterances. The level of PFS succeeds immediately that of SF. Structures at this level are conceived as the first intermediate results in the course of contextually specifying the meaning of expressions. They differ from SF structures insofar as the parameters occurring in the latter are now replaced by suitable constants. It appears that several mechanisms of defeasible reasoning have an important part in this process of pragmatic enrichment.28 By means of them information being not present in SF but necessary for grasping the concrete meaning of the respective ex26

27

28

For a similar attitude towards the understanding process and, with it, the roles semantics and pragmatics have to play in cf. e.g. Hobbs et al. (1993), Carston (1999, 2002), Levinson (2000) and Recanati (2004). Underspecified meaning structures are used in semantic research already for some time to explain various phenomena like scope ambiguity, anaphora, lexical ambiguity and 'reinterpretation' of expressions. See e.g. Reyle (1992), Pinkal (1996), van Deemter & Peters (1996) and Egg (2005). As inference procedures of this kind one can think of, for instance, such of abduction (e.g. Hobbs et al. 1993, Dölling 1997, Blutner 1998, Maienborn 2001, 2003), explicature (e.g. Carston 1999, 2002) and generalized conversational implicature (e.g. Levinson 2000, Blutner 2002).

58

Johannes Dölling & Tatjana Heyde-Zybatow

pressions is inferred from the conceptual knowledge base. In this manner, PFS represents the very level on which meaning variations of the kind we are concerned with take place. In contrast to other minimalist analyses and, particularly, to Borer (2005), we take it for granted that along the lines of the phonological and syntactic information of a verb its lexical semantic representation participates in the grammatical computation as an autonomous unit. Essentially, our assumption is that the SF of the verb should be analysed as a basic building block of the SF of utterances in which it occurs. But as the verbs are open to a plurality of variations of their primary meaning, one has to take care of that that their lexical SFs allow for each of these differentiations by fixing SF-parameters respectively. For illustrating, in (32) we propose a lexical entry for öffnen, which is provisional in several respects. (32)

PF: SYN: SF:

/œffn-/ V, [D/NPacc][E/NPdat]/NPnom Oe. OPEN(e)

PRAG:

(i) (ii) (iii) (iv)

GRAMMAR

D =  and E =  iff = {OPENk1, OPENk2,…} D = + and E =  iff = {OPENl1, OPENl2,…} D =  and E = + iff = {OPENm1, OPENm2,…} D = + and E = + iff = {OPENn1, OPENn2,…}

Unlike the analyses considered so far, the entry consists of four components. Whereas the lexical components PF, SYN and SF constitute its grammatical part, i.e. that one which is the basis for the grammatical computation including the composition of SF, the component PRAG is an additional, extra-grammatical part which is relevant for the process of pragmatic enrichment. Following Krifka (1989, 1992), in order to represent the lexical meaning of verbs we use the non-decompositional neo-Davidsonian format. Therefore, in (32) the SF of öffnen is identified with the oneplace predicate Oe. OPEN(e) over eventualities, i.e. no thematic argument positions are included in it. Thus, the lexical semantic representation is independent of the varying number and sort of arguments that are available in the syntactic structures in which the verb is embedded. However, for restricting the set of such syntactic constructions in a way, we assume that the component SYN clarifies the primary subcategorization frames of the verb. The notation [D/NPacc][E/NPdat]/NPnom indicates that öffnen can be used in a ditransitive, a transitive as well as an intransi-

Verb Meaning: How much Semantics is in the Lexicon?

59

tive variant.29 Consequently, by default, the verb is allowed to occur in syntactic structures exemplified by sentences like in (33). (33) a. b. c. d.

Hans öffnete der Mutter die Tür. Hans opened the door for the mother. Hans öffnete der Mutter. Hans opened for the mother. Hans öffnete die Tür. Hans opened the door. Hans öffnete. Hans opened.

Supposing such constraints over possible subcatgorization frames does not mean that it is impossible that the verb appears with other argument structures. But then the respective structures have to be licensed by additional conditions on its syntactic context. Another essential property of the SF in (32) is that OPEN 30 is a parameter which has yet to be contextually fixed with one-place predicate constants denoting various types of opening. Thus, in the lexicon the meaning of öffnen is not represented by means of either a specific constant OPEN or a decompositional structure like such ones considered above. For this reason, along with the separation of thematic argument structure, the lexical SF of a verb can be regarded as being not more than a 'pointer' that helps to identify its appropriate meaning variant under the respective contextual conditions. It is in this sense that SFs are radically underspecified. PRAG, finally, is the lexical component that contributes significantly to the fixation of OPEN. It comprises the conditions for determining the domain of values of the SF-parameter in dependence on the syntactic context in which the verb occurs. As an example, condition (i) specifies that the domain of values of OPEN has to be identified with the set containing predicate constants like OPENk1 or OPENk2 as members if the verb is used intransitively. Analogously, according to condition (ii) a member of the set containing OPENl1, OPENl2 etc. has to be substituted for OPEN if the verb is subcategorized for a nominative and an accusative NP. Which of the values out of the particular domain then is actually selected in the process of pragmatic enrichment is dependent on the kind of entities which the argument NPs refer to and, possibly, on further contextual conditions. 29

30

In respect of [DX] and [EX] we suppose an analogous convention made for the entry in (21). In the following SF-Parameters are always marked by italics.

60

Johannes Dölling & Tatjana Heyde-Zybatow

Evidently, a lexical entry like in (32) faces the problem how the several meaning variants of the verb can combine with their argument structure. Our proposal is that the information about the respective argument positions is brought in by means of independent configurations that are inserted in the process of SF composition and contain further SF-parameters. The insertion of these SF structures is triggered by the syntactic structure building process. 31 There are at least two different ways of characterizing those configurations: Firstly, following Kratzer (1996, 2003), they could be understood as the contributions of some functional categories to the computation of SF. One consequence of this assumption is that the introduction of argument positions is directly linked to the generation of syntactic structure. Secondly, the configurations can be seen as special SF-operators of logical type shifting (or coercion).32 Such operators do not have syntactic correlates. Rather, they are only inserted in SF composition if a logical type conflict would emerge otherwise. The O-expression in (34) illustrates how such a SF-operator could look like. (34)

ARGacc:

OPOxOe. Tacc(x, e) & P(e)

ARGacc shifts one-place predicates into two-place predicates. As is indicated by the index, in doing so the operator introduces the argument position that can be occupied by the SF of an accusative NP. Tacc acts as an SF-parameter whose domain of values comprises those predicates of thematic relations that are possible for arguments in that syntactic position. Although the pragmatic conditions for the fixation of that sort of parameters have still to be specified, it appears that, for instance, the predicate constant TH(eme) is a suitable value for Tacc whereas the predicate constant AG(ent) is not such one. In analogy to ARGacc SFoperators like ARGnom or ARGdat are assumed to account for the introduction of further argument positions.

31

32

Krifka (1989, 1992) suggests that the predicate constants representing the thematic relations should be part of the meaning of the argument NPs. To avoid unacceptable combinations, Krifka assumes that verbs and their complements possess thematic features in their syntactic characterizations. Only if these features are compatible the combination is allowed. But this means that genuine conceptual information is transferred to the syntax and that is an undesirable result. Furthermore, the relations of participation can vary. In such cases Krifka's solution forces the assumption of more than one meaning for one and the same NP, what cannot be motivated independently. Cf. the proposal by Marten (2002). For a more general framework of logical type shifting see e.g. Partee (1987, 1995).

Verb Meaning: How much Semantics is in the Lexicon?

61

In (35) we give a simplified SF derivation to demonstrate the establishing of thematic argument structure. (35) a. Hans öffnete die Tür. (= 33c) b. öffnen: V, [D/NPacc][E/NPdat]/NPnom; Oe. OPEN(e)

« « ARGacc: OPOxOe. Tacc(x, e) & P(e) « öffnen: V, /NPacc [E/NPdat]/NPnom; OxOe. Tacc(x, e) & OPEN(e) « «die Tür: NPacc; def_DOOR « die Tür öffnen: VP, [E/NPdat]/NPnom; Oe. Tacc(def_DOOR, e) & OPEN(e) « « ARGnom: OPOxOe. Tnom(x, e) & P(e) « die Tür öffnen: VP, /NPnom; OxOe. Tnom(x, e) & Tacc(def_DOOR,e) & OPEN(e) « «Hans: NPnom; hans « Hans die Tür öffnen: S; Oe. Tnom(hans, e) & Tacc(def_DOOR, e) & OPEN(e)

As shown in (35b), the occurrence of the accusative NP die Tür requires a type shifting of the lexical SF of öffnen. The application of ARGacc results in a shifted SF of the verb that now disposes of a thematic argument position for the SF of die Tür. Thus, we get the composed SF of the VP die Tür öffnen, which is again a one-place predicate over eventualities. Subsequently, in order to meet also the requirement of the nominative NP this SF is shifted by means of ARGnom. In a final step, the SF of Hans occupies the argument position arising from that operation. Although the procedure sketched seems rather simple to realize it is surely not adequate with respect to syntactic considerations. Therefore, we are convinced that the first option of establishing the thematic argument structure, namely, via functional categories has to be preferred. The main difficulty in pursuing this option, however, is whether internal argument positions can be introduced in the same manner as Kratzer (1996, 2003) has proposed it for the external one. A short look at the data reveals that, for instance, the influence of the direct object (or accusative) NPs on the interpretation of the respective structures is manifold. Whereas the direct objects of verbs of creation and consumption force a telic interpretation of the eventuality predicate, i.e. the predicate denotes an attribute of events if they have a quantized refer-

62

Johannes Dölling & Tatjana Heyde-Zybatow

ence (see (36b)), there is no such interpretational effect with verbs of transportation amongst others (see (36c)).33 (36) a. b. c.

Peter aß zehn Minuten lang/*in zehn Minuten Brot. Peter ate for ten minutes/*in ten minutes bread. Peter aß *zehn Minuten lang/in zehn Minuten eine Scheibe Brot. Peter ate *for ten minutes/in ten minutes a slice of bread. Peter schob zehn Minuten lang/*in zehn Minuten einen Wagen. Peter pushed for ten minutes/*in ten minutes a cart.

Borer (2005) proposes that the direct object NPs have different positions in the syntactic hierarchy depending on their influence on the eventuality structure. This means that a NP like das Hemd has to be introduced in the specifier position of the aspect category in a sentence like (37a). Unlike that, the position of das Hemd in sentence (37b) will be the specifier of a default category only serving the identification of participants of the described situation. (37) a. b.

Anna bügelte das Hemd in 5 Minuten. Anna ironed the shirt in 5 minutes. Anna bügelte das Hemd 5 Minuten lang. Anna ironed the shirt for 5 minutes.

But notice that it is unclear in which position the NP has to be localized if there is no temporal modification clarifying the aspectual interpretation. The problem to be resolved is whether direct objects – in opposition to the introduction of the external arguments via the voice category – generally disallow a uniform treatment with respect to their syntactic position. This problem is linked to Kratzer's (2003) criticism that a separation of the internal arguments from the verbs would lead to an overgeneralization with regard to the generation of transitive structures. In addition to the difficulties concerning the treatment of direct objects and their introduction into the syntactic construction, the sentences in (38) show that there are also a number of questions with respect to the analysis of indirect objects and the differentiation between them and adjuncts. (38) a. b.

33

Peter schrieb dem Mann einen Brief. Peter wrote the mandat a letter. Martin brach seinem Bruder einen Zeh. Martin broke his brotherdat a toe.

Cf. also the sentences in (3) and (4).

Verb Meaning: How much Semantics is in the Lexicon?

c. d.

63

?

Gunnar fand seiner Frau den Handschuh. Gunnar found his wifedat the glove. *Margit erreichte dem Kind den Bahnhof. *Margit reached the childdat the station. ?

One question is whether dative NPs like those occurring in (38a) – (38c) are arguments of verbal expressions at all.34 Obviously, the proposal of Marten (2002) to treat any verbal adjunct as argument provides no real solution, because it does not offer an explanation why then not all verbs allow such an extension of their argument structure as the unacceptability of sentence (35d) shows. The preceding discussion has indicated that the problem with the status of the internal arguments of verbal expressions is a rather intricate one. What has to be clarified in future research is how, for instance, the information about object positions is connected to the appropriate functional category to result in an acceptable SF of the respective sentence. Our first step in direction to an answer is the assumption of a division of labour: thematic argument positions are not represented in the lexicon, but have a reflex in the syntactic component of the lexical entry of verb.35

7 Meaning Variation and Parameter-fixed Structure of Verbs So far we have explored what the lexical SF of a verb looks like and how it is inserted in the compositional SF derivation. In the following we want to outline how the contextual specification of the SF of an utterance works and, particularly, what kind of structure is achieved at the level of PFS. For illustrating, consider again the sentence in (39a) along with its SF and PFS. (39) a. b. c.

Hans öffnete die Tür. (= 33c) SF: e [Tnom(hans, e) & Tacc(def_DOOR, e) & OPEN(e)] PFS: e [AG(hans, e) & TH(def_DOOR, e) & OPENl1(e)]

As is argued above, the PFS of an expression arises from its SF by fixing the SF-parameters with a suitable value. Accordingly, the structure in (39c) differs from that one in (39b) only in it that the free predicate 34

35

The traditional analysis would claim that the dative NP in (38a) is the indirect object of the verb, whereas the dative NPs in (38b) and (38c) are so-called free (unselected) datives. For a similar assumption cf. Ramchand (2006).

64

Johannes Dölling & Tatjana Heyde-Zybatow

variables Tnom, Tacc and OPEN are substituted by the predicate constants AG, TH and OPENl136, respectively. Traditionally, a structure like the PFS in (39c) would be viewed as the semantic representation that the sentence immediately acquires in the process of grammatical computation of meaning. Similarly, the PFS in (40), which represents the particular reading of the verb involved in (39a), could be assumed as one of the lexical semantic representations of öffnen. (40) PFS: OyOxOe. AG(x, e) & TH(y, e) & OPENl1(e)

But such an understanding disregards the fact that both structures are already the result of a pragmatic enrichment of the SF of expressions. It is in the process of PFS derivation that the formal meaning of the respective expressions is contextually specified to an individual meaning variant. We suppose that the mapping from SF to PFS is established with the help of particular pragmatic conditions of fixation of SF-parameters. In (41) we have picked some of those conditions that take part in the PFS derivation concerned with. (41) a. b. c.

Tnom Ÿ AG Tacc Ÿ TH OPEN Ÿ OPENl1

It is obvious that numerous elements of world knowledge have to be consulted for the decision which values are suitable for the fixation of the parameters in the respective context. Thus, to achieve the PFS in (39c) and, with it, that in (40) we need to be acquainted with the conceptual conditions that determine the relationship between the various predicate constants occurring in the structures. In (42) three meaning postulates that express conditions relevant for the interpretation of the verb öffnen in sentence (39a) are given. (42) a. b. c.

36

… e [OPENl1(e) o x AG(x, e)] … e [OPENl1(e) o x [DOOR(x) & TH(x, e)]] … e [OPENl1(e) o e'e''xy [e = e' † e'' & AG(x, e') & CAUSE(e', e'') & TH(y, e'') & s [BECOME(e'', s) & OPENl1(s) & TH(y, s)]]]

According to (ii) of PRAG in (32) the constant OPENl1 is a member of the domain that is assigned to the parameter OPEN if the verb öffnen is used transitively.

Verb Meaning: How much Semantics is in the Lexicon?

65

The axioms in (42a) and (42b) formulate some restrictions on the application of OPENl1 in regard to the kind of objects being participants in the respective events. In addition, the axiom in (42c), where † is the operator of summation (cf. e.g. Link 1983 or Krifka 1989, 1992), characterizes such events of opening as including two subevents and resulting in a state of being open in a certain way. Notice that although a CAUSE-BECOME-structure is used, in contrast to the strategy of meaning decomposition, this structure forms only a necessary but not a sufficient condition for the application of OPENl1. The relations represented in (42) on their part call for general axioms of conceptual ontology like those in (43). (43) a. b. c.

… es [BECOME(e, s) o RESULT(s, e) › e'e''[e' = e'' † e & CAUSE(e'', e) & RESULT(s, e')]] … es [RESULT(s, e) o e is a subquestion of against the empty context. If Q is a Wh-question and Q' is a yes-no question for P about one a  A: Q' = , then Q' is a subquestion of Q in the empty context. (29) Remnant question: A question R is the remnant question to a question Q in context C iff R is a subquestion to Q in the empty context, and a complete answer to R in context C is a complete answer to Q in C. (30) Two partial questions Q1, Q2 are equivalent to Q in context C iff any complete answer A to Q1 in context C, and complete answer B to Q2 in context C+A together constitute a complete answer to Q. (In other words: the discourse A. B. is a complete answer to Q in C.)

88

Regine Eckardt

Next, we need to take care of the fact that the structure of a question answer discourse will be represented as a tree structure, whereas real conversation follows the linear order of time. We need to map utterances in a tree to a linear order, and we need to explicate the epistemic background for each utterance in a tree: all that the interlocutors have learned by utterances that precede a specific turn in the tree. (31) Linearization of utterances in a tree: Let QAD be a ordered tree (i.e. the daughter nodes of each node are ordered), and assume that some nodes in that tree are annotated with utterances u. The tree then defines a linear order on the utterances as follows: x If an utterance u in the tree dominates an utterance u' in the tree, then u occurs before u'. x If an utterance u is a left sister node of an utterance u' in QAD, then u occurs before u'. x If an utterance u is dominated by a node r(u) and u' is dominated by r(u') in the QAD-tree, and if r(u) is left to r(u'), then u occurs before u' in D. (32) Local context of node x in the QAD-tree (with utterances): Let C0 be the context before the QAD started. Let u1, u2, ... un be the utterances before x, in their actual order in the tree (i.e. ui is before uj iff i [p1 & p2]] (variable induced at Stage I) qp [q ALT(p) { q o ™ p] (selection principle) q [q o ™p2: [p1 & q] >> [p1 & p2]] (27b applied to 27a) /p1/ = t x, y {parents'}[ill'(x) at t Ÿ nurse' (y, x) at t] (assumption) /p2/ = t x, y {parents'}[nurse'(y, x) at t Ÿ Ÿ™go out to work' (y) at t] (assumption) /p3/ = ™ go out to work'(mom') at t (=q) (27a-e applied to p1 & p2 of (24b))

The assumptions (27d,e) are the premises due to which q, instantiated by /p3/ in (27f), meets the preference condition "[p1 & q] >> [p1 & p2]" in (27a). The reading of (24b) derived along the lines of (27) may be paraphrased as in (28), which makes the concession more explicit: (28) Mein Vater ist krank 'my dad is ill' p1 If my dad is ill, one would expect my mom not to go out to work /p1 Ÿ q, where q o ™ p2/ aber 'but' & Meine Mutter geht arbeiten 'my mom goes out to work' p2

What aber contributes to the reading of (24b) is to conjoin p1 and p2 under the inferred condition /p1 Ÿ q, where q o ™p2/ according to which p1 & ™p2 would come closer to what we expect than p1 & p217 thereby meeting the condition "[p1 & q] >> [p1 & p2]". Note the structural conditions controlling the derivation of the "concessive" reading of (24b) at Stage II: (i) the assumptions (27d,e) are restricted to x, y {parents'} by DC; (ii) the conclusion (27f), which yields the suitable instance for q, is determined by the wide Focus of C2, hence the alternative set is q ALT (p2). Finally, the selection principle (27b) constrains the search for q by imposing on it the condition q o ™p2, cf. (27c, 27f, 28). The selection principle formulated in (27b) is a by-product and/or building block of our claim that the search for q proceeds from Structural context to DC to World Knowledge. The principle guarantees that the Structural context (i.e. C1, C2) is maxi16

17

The relation symbolised by ' p Ÿ q' is not to be confused with logical implication 'p o q'. The former is to be interpreted as "given our everyday experience, p is expected to have the consequence q". Thus, (28) captures all versions of what is called 'contra-expectation but' in the literature.

226

Ewald Lang and Marcela Adamíková

mally exploited in order to minimize information that has to be supplied from World Knowledge. A final remark on (25)-(28) might be in order. The derivation of the "contrastive" reading of (24a) is straightforward, not the least due to a FBS which is prosodically unambiguous. (24b), however, leaves us with a problem. Let us look at it once more: (24) b.

DC: Was machen denn deine Eltern? 'How are your parents doing?' CONCESSIVE

L*H H* H*L L% [Mein Vater [ist ernsthaft krank]F [aber meine Mutter geht ARbeiten]FU]

We are sure that C2 in (24b) has a wide (all-new) Focus such that C2 does not yield a direct answer to DC (unlike C2 in (24a)) but rather a continuation of C1. However, we are still uncertain about the (possibly conspiring) prosodic parameters that bring about this effect. Thus, the intonation contour assigned to (24b) above is in need of further exploration. The compensatory effect of marked prosody comes out most clearly with the connector und 'and' as shown in (29) below. The semantic contribution of und 'and' is just p1 & p2, it lacks the "adversative" component q [q ALT (p2) : [p1 & q] >> [p1 & p2]] that is part of the lexical content of aber, hingegen, dennoch etc. Syntactically, und occurs in the Null-Position only and does not impinge on the FBS of the conjuncts it links, cf. §5. In view of this, we may expect prosody to be the decisive factor in determining the FBS, and hence the reading, of the given CC. The expectation is borne out as illustrated by (29a–c): (29) DC: Was machen denn deine Eltern? 'How are your parents doing?' CONTRASTIVE a. L*H L*H H% [[[Mein Vater]T [ist ernsthaft KRANK]FIP] L*H H*L L% [und [meine Mutter]T [geht ARbeiten]FIP]U] CONCESSIVE b. L*H H* nH*L L% [Mein Vater [ist ernsthaft KRANK]F [und meine Mutter geht ARbeiten]FU] c.

CAUSAL

L*H nH*L L% [Mein Vater [ist ernsthaft KRANK]F [und meine MUtter geht ARbeiten]FU]

In (29a), the parallel FBS of the conjuncts make C1 and C2 into a list of partial, hence "contrastive", answers to DC. The absence of "adversa-

The lexical content of connectors and its interplay with intonation

227

tivity" would allow for C1 and C2 to change in order or to be rendered in a factoring construction, e.g. Mein Vater und meine Mutter sind krank bzw. gehen arbeiten 'My dad and my mom are ill and go out to work, respectively', without changing their interpretation w.r.t. DC. In (29b,c), the FBS of C2 differs from that of C1 in such a way that what results is an implicative reading, cf. §3.1 above. The prosodic means causing this effect is (i) the occurrence in C2 of what Ladd (1996: 280 ff.) calls "raised H peak" (nH*L), which, we claim, indicates a sort of Contrast Focus; (ii) the particular constituent in C2 marked that way. According to Ladd, raised H accent peaks are "extrinsic to the tonal system". Their use is characterized as expressing surprise or irritation. Recast within our framework, this implies that a raised H peak does not indicate a structurally determined Focus but one that is induced by, and hence to be reconstructed from, the context. In view of the DC given, the information needed to account for (29b,c) has to be supplied from World Knowledge. In (29b), where the raised H peak is on the VP (wide Focus), the inferences which lead to the "concessive" reading are the same as those in (27a–f), although they differ in the way they are triggered. In (24b), the inferences are induced by the lexical content of aber (thus, in (24b) there is no need for a raised H peak in C2); in (29b), however, they are induced by the "surprise" expressed by nH*L and are controlled by the wide Focus thus marked. Finally, let us look at (29c), where the raised H peak is on the subject of C2 (narrow Focus). Along the lines of what has been presented so far, the derivation of the "causal" reading of (29c) proceeds as in (30), where (30a–d) cover Stage I and (30e-i) cover Stage II: (30) a. b. c.

d. e. f. g. h. i.

p1 = ill' (dad') at t0, p2 = go out to work' (mom') at t0 (assertions) p1 & p2 (conjunction) (F-value of C2, Op.xt < t0 [x  {{ALT (mom') ˆ {parents'}} & p = go out to work' (x) at t] nH*L on the subject, alternative set restricted by DC) p3 = tx [x  {parents'} & go out to work' (x) at t] (VP de-accented, presupposition) /p2/ = tx [ill' (x) at t Ÿ ™ go out to work' (x) at t] (assumption from world knowledge) (from p1, /p2/) /p3/ = ™ go out to work'(dad') at t0 (from /p1/, /p3/) /p4/ = go out to work'(mom') at t0 p2 { /p4/ p1 Ÿ p2; p1, ?p2 (causal relation)

228

Ewald Lang and Marcela Adamíková

The direct effects of the raised H peak on meine Mutter in C2 are noted in (30c,d). The only assumption to be supplied from World Knowledge is (30e), the rest is derivable by familiar inferencing. The causal interpretation of (29c) results from p1 & p2 being enriched by (30c-g) in such a way that p2 is not just conjoined to p1 but coincides with a conclusion that is deducible from p1 via (30h, i).

7 Conclusions Focussing on the role of co-ordinating and subordinating connectors, the paper has presented some sets of data to illustrate the relation between the grammatical properties of certain connectors and the prosodic make-up of the constructions they occur in. Drawing on a selection of "adversative" connectors, we attempted to substantiate the following claim: what these connectors lack in specified lexical content might be compensated by syntax and/or prosody. The discussion of how the compensation phenomena can be accounted for by conditions on the syntax/semantics and the semantics/prosody interfaces may be summarized as follows: (1) It is the syntactic position of the connector outside of vs. within the conjunct C2 which determines the FBS of the construction, not the other way round. (2) The default case is that syntactically (a-)symmetric conjuncts are associated with (a-)symmetric prosodic patterns and FBSs. (3) Syntactically symmetric conjuncts may be de-parallelized by asymmetric FBSs that are licensed by inferences from the context. (4) The notion "context" is to be divided into three domains, viz. (i) Structural context, (ii) Discourse context, (iii) World Knowledge. The most urgent problem we are left with is to pinpoint the prosodic parameters involved in the compensation phenomena under review and then to combine them into intonational contours in a compositional way (as suggested by e.g. Pierrehumbert & Hirschberg 1990).

References Adamíková, M. (2003): Sentential Negation and Focus in Slovak Adversative Constructions. In: P. Kosta et al. (eds): Investigations into Formal Slavic Linguistics. Frankfurt am Main: Peter Lang, 157–168. Adamíková, M. (2004a): Kontrast oder Korrektur? Prosodische Disambiguierung bei negationshaltigen Adversativ-Konstruktionen in den Westslavinen. Leipzig (=Linguistische Arbeitsberichte 82).

The lexical content of connectors and its interplay with intonation

229

Adamíková, M. (2004b): Operatorenskopus und Fokus-Hintergrund-Gliederung: ihre prosodische Kennzeichnung in den Westslavinen. Linguistische Arbeitsberichte 81, 53–70. Adamíková, M. (2005): Das Inventar adversativer Konnektoren in den Slavinen. In: M. Bayer, M. Betsch & R. Zimny (eds): Beiträge der Europäischen Slavistischen Linguistik (POLYSLAV) 8. (Die Welt der Slaven. Sammelbände XX), München, 1–10. Adamíková, M. & D. Fehrmann (2001): Prosodische Differenzierung von Kontrast- und Korrektur-Konstruktionen im Slovakischen und Polnischen. Linguistische Arbeitsberichte 77, 81–112. Blühdorn, H., E. Breindl & U. H. Waßner (eds): Brücken schlagen. Grundlagen der Konnektorensemantik. Berlin/New York: de Gruyter. Breindl, E. (2004): Relationsbedeutung und Konnektorbedeutung: Additivität, Adversativität und Konzessivität. In: H. Blühdorn, E. Breindl & U. H. Waßner (eds): Brücken schlagen. Grundlagen der Konnektorensemantik. Berlin/New York: de Gruyter, 225–253. Büring, D. (1997): The Meaning of Topic and Focus. The 59th Street Bridge Accent. London/New York: Routledge. Fehrmann, D. (2004): Prosodic Distinction of Contrast and Correction. Readings of Polish Adversative Coordinate Structures. In: A. Steube (ed.): Information Structure. Theoretical and Empirical Aspects. (= Language, Context, and Cognition I). Berlin/New York: de Gruyter, 307–343. Féry, C. (1993): German Intonational Patterns. (=Linguistische Arbeiten 285). Tübingen: Niemeyer. Féry, C. (2000/2002): Phonologie des Deutschen. Eine optimalitätstheoretische Einführung. Vol. 2, Potsdam. Féry, C. & K. Hartmann (2005): The Focus and Prosodic Structure of German Right Node Raising and Gapping. Linguistic Review 22, 67–114. Frey, W. (2003): Syntactic conditions on adjunct classes. In: E. Lang et al. (eds): Modifying Adjuncts. 130-209. Berlin/New York: de Gruyter. Frey, W. (2004): The grammar/pragmatics interface and the German prefield. Sprache & Pragmatik 52, 1–39. HdK = R. Pasch, U. Brauße, E. Breindl & U. H. Waßner (2003): Handbuch der deutschen Konnektoren. Linguistische Grundlagen der Beschreibung und syntaktische Merkmale der deutschen Satzverknüpfer (Konjunktionen, Satzadverbien und Partikeln). Berlin/New York: de Gruyter. de Hoop, H. & H. de Swart (2004) (eds): Contrast in Discourse. (=Journal of Semantics 21,2). Oxford UP. Jacobs, J. (1991): Negation. In: A. von Stechow & D. Wunderlich (eds): Semantik/Semantics. (=HSK 6). Berlin/New York: de Gruyter, 560–596. Ladd, D. R. (1996): Intonational Phonology. (=Cambridge Studies in Linguistics 79). Cambridge UP.

230

Ewald Lang and Marcela Adamíková

Lang, E. (2000): Adversative Connectors on Distinct Levels of Discourse: a Reexamination of Eve Sweetser's Three-level Approach. In: I. Couper-Kuhlen et al. (eds): Cause – Condition – Concession – Contrast. Cognitive and Discourse Perspectives, Berlin/New York: de Gruyter, 235–256 Lang, E. (2001): Kontrastiv vs. implikativ I: Interpretationseffekte intonatorischer Distinktionen bei Koordination. Linguistische Arbeitsberichte 77, 113–138. Lang, E. (2002a): Kontrastiv vs. implikativ II: Interpretationseffekte einer intonatorischen Distinktion bei elliptischen Sprichwörtern. Linguistische Arbeitsberichte 79, 187–212. Lang, E. (2002b): Die Wortart »Konjunktion«. In: D. A. Cruse et al. (eds): Lexicology/Lexikologyie. (= HSK 17.1), Berlin/New York: de Gruyter, 634–641. Lang, E. (2004): Schnittstellen bei der Konnektoren-Beschreibung. In: H. Blühdorn et al. (eds): Brücken schlagen. Zur Semantik der Konnektoren. BerlinNY: de Gruyter, 45–92. Lang, E. (in prep.): Contrasts in Grammar: a survey of their roles, sources, and ways of realization. Ms. Lang, E. & C. Umbach (2002) Kontrast in der Grammatik: spezifischen Realisierungen und übergreifender Konnex. Linguistische Arbeitsberichte 79, 145–186. Lohnstein, H. (2004): Variable und invariante Strukturmerkmale von Satzkonnektoren. In: H. Blühdorn, E. Breindl & U. H. Waßner (eds): Brücken schlagen. Grundlagen der Konnektorensemantik. Berlin/New York: de Gruyter, 137–160. Pierrehumbert, J., & J. Hirschberg (1990): The Meaning of Intonational Contours in the Interpretation of Discourse. In: Ph. Cohen et al. (eds.): Intentions in Communication, Cambridge MA: MIT, 271–311. Stede, M. (2004): Kontrast im Diskurs. In: H. Blühdorn, E. Breindl & U. H. Waßner (eds): Brücken schlagen. Grundlagen der Konnektorensemantik. Berlin/New York: de Gruyter, 255–285. Steube, A. (2001): Correction by Contrastive Focus. Theoretical Linguistics 27. 2/3, 215–249. Sweetser, E. E. (1990): From Etymology to Pragmatics. Cambridge: Cambridge UP. Umbach, C. (2004): On the Notion of Contrast in Information Structure and Discourse Structure. In: de Hoop, H. & H. de Swart (2004) (eds): Contrast in Discourse. (=Journal of Semantics 21,2), Oxford UP, 155–175. Umbach, C. (2005): Contrast and information structure: a focus-based analysis of but. Linguistics 43:1, 207–232. Umbach, C., I. Mleinek, C. Lehmann, Th. Weskott, K. Alter & A. Steube (2004): Intonational Patterns in Contrast and Concession. In: A. Steube (ed.): Information Structure. Theoretical and Empirical Aspects. (= Language, Context, and Cognition I), Berlin-NY: de Gruyter, 277–306.

Eva Hajiþová, Petr Sgall & KateĜina Veselá (Prague)

Contextual Boundness and Contrast in the Prague Dependency Treebank 1 Language system and the layer of cognition It has been richly documented in the development of linguistics that among the fundamental requirements there is the necessity to distinguish between (i) the system of language (F. de Saussure's langue as a system of oppositions, N. Chomsky's linguistic competence) and (ii) the cognitive layer, or the domain of ontological content, which up to now has not found any systematic descriptive framework; cf. Chomsky's distinction between problems (which mainly concern language) and mysteries (concerning above all cognition). What is beyond language, may still remain very mysterious for us as linguists; on the other hand, the system of language can be described by methods that are at our disposal, although linguists differ in preferring specific methods and ways of their application. In our view, it is important to look for explicata necessary for a theoretical handling of a presystemic notion X, rather than to ask "what does X mean?" (where X may stand for givenness, contrast, and so on). The distinction between layers (i) and (ii) can be helpful in looking for such explicata. Among the points that have gained a relatively high consensus, there is the awareness that the notions of given and new information belong to the domain of cognition. Their unclearness may be illustrated by differences such as those between knowing a person personally, by name, by face, from her/his writings, and so on, and by given meaning "available for the hearer in her/his memory" – but: to what degree? One of the aspects of such a classification can be specified (in part) as a scale of the degrees of salience in discourse. It may be assumed that

232

Eva Hajiþová, Petr Sgall & KateĜina Veselá

more than one pair of explicata will be necessary for handling these presystemic notions. On the other hand, the topic (T) and the focus (F) of a sentence belong to the language system, or, more exactly, to the systems of individual languages, and have their place in their descriptions. There are reasons to understand the opposition of contextually bound (CB) and non-bound (NB) items as underlying that of T and F, and thus also as constituting a part of the language system. Another phenomenon belonging to TFA, and thus to the sentence structure, is the communicative dynamism (CD), the degrees of which, first characterized by J. Firbas, are partly determined by contextual boundness, in that every CB daughter node is less dynamic than its head, and every CB sister node is less dynamic than every its NB sister. We can compare this situation to the distinction between linguistic and cognitive patterns in better known subdomains. Let's take as an example the linguistic (language specific) category of number in English. The prototypical situation, distinguishing singular and plural, contrasts with non-prototypical cases such as those of trousers or scissors, with which the difference between the cognitive singularity and plurality is not directly rendered by a difference between the word forms; similar non-prototypical cases concerning gender can be found in English with the names of countries or ships, in German with nouns such as das Weib (a neuter noun with the meaning 'woman'), and so on. In the sequel, we distinguish in a similar way between contrast as a phenomenon from the cognitive layer and contrastive items as present in the language system. It can then be discussed, on the one hand, whether contrast (in the domain of cognition) always is determined as a choice from a set of alternatives (whether such a set is a necessary and/or sufficient condition for contrast, be it explicit or just determined by discourse strategies in the sense of Büring 2003), and, on the other hand, what are appropriate operational criteria for the identification of contrastive items in a sentence. The differences between the language system and the domain of cognitive content may be illustrated by sentence (1a), in which a. the pronoun jeho 'him' occurs in its strong form, typical for contrast or for focus, although (similarly as s ní 'with her') as an anaphoric pronoun it refers to an item that is 'given' in the sense of being easily identifiable by the hearer; linguistically, the word order of the Czech sentence (pronounced with the prototypical intonation) determines the fronted pronoun as CB and the pronoun occupying the rightmost position as NB, which may be tested by criteria con-

Contextual Boundness and Contrast in the Prague Dependency Treebank

233

cerning the scope of negation in such sentences or by the question test and by possible continuations; b. it would require further discussion whether the referent of jeho is being chosen among a set of alternatives, cf. the possible occurrence of the alternative sentence (1b) in the present context; however, the strong form of the pronoun determines its occurrence as that of a contrastive item, in which (1a) differs from (1b), with ho 'him' being the corresponding weak form, typical for non-contrastive CB items. (1)

a.

b.

(Milena nás seznámila se svým bratrem.) Jeho jsme pozvali do Prahy, a do Brna jsme jeli s ní. (Milena made us acquainted with her brother.) Him we invited to Prague, and to Brno we went with her. Pozvali jsme ho do Prahy, a do Brna jsme jeli s ní. We invited him to Prague, and to Brno we went with her.

A major issue in language description is that of markedness. This concept has been exploited and characterized by R. Jakobson and others in the Prague School. It is crucial to distinguish between unmarked (primary, prototypical, although not always most frequent) items of all levels as opposed to their marked (secondary, peripheral) counterparts, which only occur under specific contextual or other conditions. In our view, CB items, as well as contrastive ones (among which we only pay attention in this paper to those which are CB) are marked.

2 Contextual boundness in the Prague Dependency Treebank (PDT) 2.1 The basic nature of contextual boundness The interactivity of communication is reflected not only in the patterning of discourse, but even in the structure of a sentence as a type, as a unit of the system of language. An elementary way of its reflection may be seen in the opposition of contextual boundness, i.e. of CB and NB lexical items.1 This opposition concerns every lexical occurrence in 1

In marked cases, the opposition of CB/NB concerns even morphological values: e.g. in I have not yet written it, but I WILL put it down, the morphological value of Future (which we consider in FGD to constitute a grammatical, rather than lexical element) carries the intonation centre and thus is marked as NB (a narrow focus), differing from the CB lexical item put.

234

Eva Hajiþová, Petr Sgall & KateĜina Veselá

the sentence, i.e. every node in its structural tree. Topic and focus can be specified, in the prototypical case, as CB and NB items in the upper subtree, respectively, together with their dependents. Contextual boundness thus is the basis of the information structure of the sentence, i.e. of its topic-focus articulation (TFA). In the Functional Generative Description (FGD), elaborated by the Charles University research group of theoretical linguistics since the 1960s, we consider the level of tectogrammatics as constituting the interface level, i.e. the level of underlying sentence structure, which may be understood as a starting point of the semantic(-pragmatic) interpretation of sentences. The proper place for contextual boundness and for TFA in the language structure can then be found on this level. In other words, the information structure of the sentence, connecting the sentence to its cognitive content as well as to its context, is seen as one of the fundamental aspects of sentence syntax. Using a terminology that comes closer to that of Minimalism, we may say that in the computational system underlying the derivation of structural descriptions, each item chosen from the lexicon is assigned the value of NB (prototypical) or CB (marked) already in the first part of the derivation, preceding the spell-out. Along with conditions determined by the valency of the main verb (and by that of other head words in the sentence), it has to be ensured that every CB item precedes its head and also all of its NB sister nodes. Thus the semantically relevant differences in TFA can be reflected as those expressed by differences in the morphemic and phonemic patterns of the sentence (surface word order, sentence prosody, particles or constructions such as clefting).2 Another issue concerns the relationships between contextual boundness (or givenness) and contrast. According to Steube (2001, 2003) and others, contrast (as well as contrastive items viewed as a grammatically marked category) may concern given as well as known information. We assume that in the domain of cognition, contrastive information may be understood as new in the sense that it is chosen from a set of alternatives, but as given in the sense that the set of alternatives is (with or without a strict and complete specification) at hand for the addressee, it has been activated enough either by the preceding co-text or by the 2

In Sgall et al. (1986: 224), Hajiþová et al. (1998) and elsewhere, we illustrate the semantic relevance of the topic-focus articulation using examples such as German is spoken in AUSTRIA vs. GERMAN is spoken in Austria, out of which only the latter can be used as a true assertion in an unmarked context (in which Germany has not been excluded).

Contextual Boundness and Contrast in the Prague Dependency Treebank

235

situation of the discourse, by the conditions of the given culture, by the domain discussed, etc. Discussing issues of the language system itself, i.e. contextual boundness as the reflection of cognitive givenness in language, we prefer not to deny the existence of CB contrastive items (contrastive topic, CT, which either constitutes the whole of topic, or a part of it). In this we agree with A. Steube, and to a certain degree also with Büring (2003), who considers the given-new distinction to be orthogonal to the function of CT The detailed analysis of the concept of givenness by Schwarzschild (1999), unfortunately, characterizes contrast as new, not distinguishing between the (typically falling or rising-falling) intonation center of the sentence and the (typically rising) "phrasal," contrastive stress (cf. Pierrehumbert & Hirschberg's L+H*, R. Jackendoff's B contour; for Czech, see Veselá et al. 2003). While the intonation centre (falling for typical declarative sentences, rising for questions and for utterances signalling a continuation or just leaving open space for it, often written with "…" at their end) carries the focus of the sentence, the rising phrasal stress carries contrastive topic. Thus, it is not exact to say that both of these accents are connected with "prominence" and contain "new information". We prefer to admit the possibility of a (cognitive) case of contrast being given, and understand it as a question opened for further discussion whether in the language under examination contrastive CB or NB items are linguistically (grammatically, systematically) distinguished from non-contrastive ones. A further corroboration of such a view can be gained if the aspects of sentence structure directly connected with the interactive nature of language are taken into consideration, i.e. if the TFA of the sentence is understood as based on the relation of aboutness. This is reflected by the interpretation of the scope of negation: if the underlying pattern of the sentence is viewed as containing the scheme "F about T" with a positive declarative sentence, and "non-F about T" with its negative counterpart, then the appurtenance of a CT to the topic of the sentence can be supported by its position outside the scope of negation. An illustration may be found in our characterization of Büring's example (35), enriched here as for the intonation centre of the question Q and reproduced as (2). The word female, dependent on pop stars, which was (a part of) the F in Q, can only be CT, rather than F, in the answer, instantiating a change of T; it carries the rising contrastive stress (which we denote by italics), rather than the intonation center: as Büring states, the answer A is excluded, but our A' (similarly as Büring's (34a)) is not, and in its negation B the topic the female pop stars is not negated: B says that about the female pop stars it holds that they didn't wear caftans.

236

Eva Hajiþová, Petr Sgall & KateĜina Veselá

(2) Q: A: A': B:

What did the POP STARS wear? # The FEMALEF pop stars wore CAFTANSF. The female pop stars wore CAFTANS. The female pop stars didn't wear CAFTANS.

The givenness of an utterance as a whole is a phenomenon from another layer than the givenness of a referring expression: while the former includes issues such as the truth conditions and the modalities of the utterance, the latter concerns the activation of the referent. Such factors corroborate our decision not to follow Schwarzschild's (1999) argumentation too closely and to understand CT as grammatically CB and possibly referring either to new or to given cognitive information. In any case, intricate issues of the layer of cognition are connected with open questions such as those of the number of different kinds and/or degrees of "givenness" (cf. e.g. I have met him, but I do not know his name vs. I have met him, but now I am not able to recall his name or I have not met him, but I know him as the author of several papers, and so on). Our investigations corroborate the view that in Czech (as well as in English, German and many other languages; see also Lahousse 2003, as for French) CB items often are contrastive, as we illustrate in Sect. 3.2 below by examples from the Prague Dependency Treebank. 2.2 Contextual boundness and TFA in PDT The opposition of NB vs. CB can be handled as a parameter of the nodes of a tectogrammatical (underlying syntactic) representation (TR), more exactly, as two values of an index at the label of every node.3 A CB node precedes its mother node and its NB sisters, and a NB node follows them (exceptions concern the focusing operators and 'proxy focus', see Hajiþová et al., 1998: 134ff). In Czech and in English, in unmarked cases, the main verb (V) and those of its direct dependents that on the surface (i.e. in the morphemic representation and in the outer shape of a sentence) follow it belong to F, and the items preceding V are parts of T. In marked cases, V can be CB, i.e. in T, or (a part of) F may precede V on the surface; usually the intonation center 3

Only coordinated constructions and apposition require networks of more than two dimensions; the linearized form of the trees, characterized in the previous writings quoted above, makes it possible to describe also different combinations of dependency and coordination or apposition in a highly perspicuous form, the pattern of which is fundamentally rather simple, comparable to that of propositional calculus.

Contextual Boundness and Contrast in the Prague Dependency Treebank

237

(sentence stress) then marks F, occupying a marked position. The dependents of nouns are primarily NB. In the prototypical case, NB items are included in F, and CB items belong to T; in peripheral cases, a deeper embedded item may be CB (although it belongs to F, since it depends on a member of F different from V), and an embedded NB item may belong to T. The Prague Dependency Treebank (PDT) is annotated on three levels (automatically for the level of morphemics, and manually, using a partially automated procedure on the other two levels): i. ii.

the morphemic level, the tectogrammatical level (the TRs of the sentences), i.e. the underlying structure of the sentence, the two main hierarchies of which are lexical valency (determining the obligatory and optional dependants of individual head words) and TFA, and iii. the analytical layer, which we understand as a technical device, absent in a theoretically oriented description, but helping to handle the transition between the two theoretically substantiated levels, i.e. corresponding in a sense to "surface syntactic" annotation. The analytical layer of annotation captures some of the main features of the TRs (especially the edges of the dependency tree), but the left-toright order of its nodes corresponds to the surface (morphemic) word order, rather than to the degrees of communicative dynamism.

3 Contrastive topic and projectivity 3.1 Projectivity of the TRs and deviations in the surface word order We work with the assumption that the TRs of Czech sentences meet the condition of projectivity, which can be briefly formulated as follows: For every triple of nodes a, b, c, if a depends on b, and c is placed either to the left of a and to the right of b or to the right of a and to the left of b, then c is subordinated to b (where 'subordinated' refers to the transitive closure of 'depends'). This is a rather strong restriction, which has to be further discussed. We assume that in prototypical cases the condition is met and that cases appearing as deviations may be described as differences between underlying and surface (morphemic) word order. The high degree of variation of the Czech word order might then be understood as reflecting not only the underlying degrees of communicative dynamism (go-

238

Eva Hajiþová, Petr Sgall & KateĜina Veselá

ing from T proper to F proper), but also certain deviations from it. Some of these can be compared to the wh-movement in English, or to constructions such as a larger town than Boston. Along with these cases, conditioned by grammatical context, there are several further groups. For Czech, the following classification seems to be appropriate (for more details, see Hajiþová et al. 2004). Three groups are to be distinguished, in which the (surface) word order does not correspond to the left-to-right order of the nodes of the TR and in some cases causes the analytical trees to deviate from projectivity: (A) combinations of lexical units with function words (especially with auxiliaries and with prepositions), which in the TRs are represented by single nodes, so that a deviation from the projectivity of the TR does not come into question here; (B) syntagmas divided, under the surface word order, into a NB part and a CB part, the latter being fronted (transferred to the left) in the surface word order;4 (C) phrasemes, consisting of more than one surface word, which eventually are to be treated as not containing a dependency relation in the TRs (each of them is to be specified either by a single node of the TR, or by a specific relation, different from syntactic dependency). Since the issues of phrasemes still require a deeper analysis of their patterns, for the time being, the specific problems of the constructions of type (B) constitute the main task; it is necessary to formulate and check the contextual conditions determining both their possible occurrences and the word order positions of their parts on the analytical as well as on the tectogrammatical levels. This is necessary, before a solution of the main task can be looked for, i.e. the formulation of rules describing the differences between communicative dynamism and surface word order. 3.2 Illustrations Let us now illustrate the group (B) deviations from projectivity on the analytical level by examples taken from the Prague Dependency Tree4

Note that the morphemic representation of a sentence has the form of a string of symbols, rather than that of a tree, so that at this level the condition of projectivity is absent.

Contextual Boundness and Contrast in the Prague Dependency Treebank

239

bank, adding short remarks, which concern a comparison of the means of expression of TFA and contrast in Czech and in English. In Czech, a contrastive CB item has quite a strong tendency to stand in the initial position in the surface word order, no matter how deep it is embedded in the underlying structure of a sentence. In sentences such as (3) this does not lead to a surface deviation of projectivity, since (the head of) the fronted sentence part depends directly on the main verb. In English, a corresponding word order variation is impossible, and the contrast may just be expressed here by a rising pitch on the Patient (object). (3)

(Nenadálou finanþní krizi musela podnikatelka Ĝešit… NČco málo zaplatila i z osobních pĤjþek poskytnutých pĜíbuznými a pĜáteli. The unexpected financial crisis had to be solved by the manager… A little she paid even from personal loans from her relatives and friends.) Hlavní pomoc ale nacházela v dalších obchodních aktivitách. Lit. Main help however she-found in further commercial activities E.: However, she found the main help in further commercial activities

Also in cases such as (4) the surface word order in Czech can be understood to correspond to CD, since such a collocation as the complex predicate projevit zájem o 'to show interest in' can be understood as constituting a phraseme (a complex predicate), which in the TR should eventually be represented by a single node in the position of the main verb. Thus the fact that the noun group zájem o pĜevod... 'interest in the transfer of...' is divided here is just a matter of the surface form; in the TR the item corresponding to o pĜevod 'in the transfer' depends on the complex predicate, rather than on the noun zájem. In E., here also, the contrast may just be expressed by a specific pitch. (4)

O pĜevod dalších desítek tisícĤ družstevních bytĤ do Lit.: In transfer of-further tens of-thousands of-cooperative flats to soukromého vlastnictví zcela urþitČ projeví zájem þlenové private property quite certainly will-show interest members bytových družstev. of-flat cooperatives. E.: Interest in the transfer of further tens of thousands cooperative flats to private property quite certainly will be shown by members of apartment cooperatives.

A genuine example of a split noun group constituting a surface deviation of projectivity in Czech, i.e. a difference between word order and CD, can be seen in (5), if the NB verb zaþínají 'they-start' is understood as depending on the contrastive CB jedno 'one' (as an adjunct express-

240

Eva Hajiþová, Petr Sgall & KateĜina Veselá

ing identity, cf. e.g. the town Pardubice). The syntactic structure of the English equivalent sentence differs, and also here a marked variant of sentence prosody can be expected, which is the case also in the following examples. (5)

Jedno mají americké Vánoce a prezidentské volby Lit.: one have American Christmas and presidential elections spoleþné: zaþínají þím dál, tím dĜíve. (in-)common: they-start the longer the earlier E.: One thing is common to the American Christmas and to the presidential elections: the longer, they start earlier.

A similar case is the split of the noun group smutné konstatování 'sad statement' in (6), in which the contrastive CB noun is fronted, leaving its NB adjunct as focus proper in its position after the verb: (6)

(…postup do Ligy mistrĤ si Pražané opravdu nezasloužili. …the Praguians actually did not deserve an advance to the Champions League.) Konstatování je to smutné a velmi bolestné pro celý þeský fotbal. Lit.: statement is that sad and very painful for whole Czech football. E.: This is a sad and very painful statement for the whole Czech football.

Other examples of similar deviations concern dislocated dependents of infinitives in sentences in which a quasi-modal predicate occurs between a noun group (with or without a preposition) and an infinitive depending on the predicate. It is a question open to further discussion whether such a split of the two verb forms can occur without any contrast in Czech. It has been observed by L. UhlíĜová (1987: 94) and other specialists in Czech that this word-order position is spreading with such quasi-modal verbs as nelze 'it-is-impossible-to' in (7), se snaží 'theyattempt' in (8) and umožĖuje 'enables' in (9), which behave similarly as simple modals such as muset 'must' or mít 'may', which in FGP are classified as function words (so that their position in the surface word order is not directly relevant for the TRs; the task is just to describe the position of all of them as similar to that of a clitic). (7)

(V pĜípadČ asociace jde o sdružení firem v daném oboru, které chce garantovat serioznost vĤþi zákazníkĤm. In the case of an association we face an union of firms in the given domain, which is prepared to warrant serious behavior to its customers.) O monopolu tedy nelze hovoĜit. Lit.: Of monopoly thus it-is-impossible to-speak E.: Thus, one cannot speak of a monopoly.

Contextual Boundness and Contrast in the Prague Dependency Treebank

(8)

(9)

241

(Ale popsaný postup v žádném pĜípadČ nenutí výrobce ke zlepšení kvality, protože – However, the described procedure in any case does not make the producers to raise quality since –) finanþní ztrátu za zmetky se snaží pĜenést na nás – prodejce. Lit.: financial loss-Acc for spoilage Refl attempt to-transfer on us – the sellers E.: They attempt to transfer the financial loss for spoilage to us – the sellers. Milionový poplatek za vydání osvČdþení, které umožĖuje vést Lit.: million tax for issuing certificate which enables to-lead E.: Collecting the million-crowns tax for issuing a certificate that allows lékárnu, zakázalo vybírat Ministerstvo pro hospodáĜskou soutČž. Lit.: pharmacy forbade to-collect Ministry for Economic Competition E.: To lead a pharmacy was forbidden by the Ministry for Economic Competition.

In our corpus, this sentence starts a text, so that no preceding co-text is present; the understanding of the Patient (direct object) as contrastive is corroborated by its fronting; in the leftmost position a weak form of a pronoun would not be acceptable, and a contrastive rising pitch can well be placed on this utterance part if the sentence is pronounced).

4 Conclusions Distinguishing between the layer of cognition and the language system, we can see the opposition of given and new information (with their different degrees and kinds) and contrastiveness as belonging to the former layer, and to assign the notions of contextually bound and nonbound items, as well as those of topic and focus, to the latter, to which also the difference between contrastive and non-contrastive items belongs. Functional Generative Description works with a strong hypothesis offered for discussion and for further checking, according to which the tectogrammatical representations (TRs) of unmarked sentences are projective rooted trees and the deviations from projectivity (as well as many other marked cases and exceptions) are viewed as differences between underlying and morphemic structures. The surface word order is handled as belonging to the morphemic level, on which the representation of the sentence has the shape of a string rather than of a tree. A specification of the transition between the two levels covering the marked cases in which the surface word order differs from the left-to-

242

Eva Hajiþová, Petr Sgall & KateĜina Veselá

right order of the nodes of the TR (from the degrees of communicative dynamism) can include movement rules (basically, the problem is similar to that of discontinuous constituents in other descriptive frameworks). In the Prague Dependency Treebank, the texts of which are annotated on three levels, we have found substantiation for assuming that contrastive CB items often are fronted, which is one of the causes of such surface deviations from the projective word order, as illustrated by exx. (5)–(8) above. Even though there are also cases of such deviations without contrastive items, deviations of all kinds are determined by contextual restrictions (definable by lists, including e.g. a list of quasimodal predicates), by specific indices in node labels (contrast) and by specific behavior of certain items (phrasemes and others). We are convinced that such an approach to the description of language can lead to a perspicuous view of sentence structure, the patterning of which may be characterized as close to elementary logic (propositional calculus), thus reflecting its proximity to general human intellectual capacities. This might help to understand the easiness of language acquisition by children.

References Büring D. (2003): On D-trees, beans, and B-accents. Linguistics and Philosophy 26, 511–545. Hajiþová E., J. Havelka, P. Sgall, K. Veselá & D. Zeman (2004): Issues of projectivity in the Prague Dependency Treebank. Prague Bulletin of Mathematical Linguistics 81, 5–22. Hajiþová E., B. H. Partee & P. Sgall (1998): Topic-Focus Articulation,Tripartite Structures, and semantic content. Dordrecht:Kluwer. Hajiþová E. & P. Sgall (2004): Degrees of contrast and the Topic-Focus Articulation. In: A. Steube (ed.): Information structure – Theoretical and empirical aspects. Berlin – New York: Walter de Gruyter, 1–13. Lahousse K. (2003): La complexité de la notiopn de topique et l'inversion du sujet nominal. In: Charolles, M. & S. Prévost (eds): Adverbiaux et topiques. Travaux de linguistique 47, Brussels: Duculot, 111–136. Schwarzschild, R. (1999): Givenness, AvoidF and other constraints on the placement of accent Natural Language Semantics 7, 141–177. Sgall P., E. Hajiþová & J. Panevová (1986): The meaning of the sentence in its semantic and pragmatic aspects. Dordrecht: Reidel – Prague: Academia. Steube, A. (2001): Grammatik und Pragmatik von Hutkonturen. In: Steube, A. & C. Umbach (eds): Kontrast: lexikalisch, semantisch, phonologisch. Linguistische Arbeitsberichte Nr.77, 7–29.

Contextual Boundness and Contrast in the Prague Dependency Treebank

243

Steube, A. (2003): Bridge contours in German assertive main clauses. Folia linguistica 37, 163–190. UhlíĜová, L. (1987): Knížka o slovosledu. Prague:Academia. Veselá K., N. Peterek & E. Hajiþová (2003): Some observations on contrastive topic in Czech spontaneous speech. Prague Bulletin of Mathematical Linguistics 79–80, 5–22.

Andreas Späth (Leipzig) On the Semantic Foundations of the Contrastive Focus Within a Lexicalist Approach 1 Introduction The focus of new information and the contrastive focus have different prosodic, syntactic and semantic properties. The two accent types show differences in their F0-contour and in the timing of the pitch peaks (see Mehlhorn 2002: 48–52, Toepel & Alter 2004: 227–229). While the focus of new information in German only occurs in situ or in the sentence initial position, there are no syntactic restrictions on the realisation of contrastive focus (see Drubig 1998b, Steube 2003). Semantically, the focus marks the positive or negative case relative to the proposition of the sentence (see Löbner 1990: 169, Roberts 1996). The contrastive focus induces the additional information that the referent of the contrastively focused constituent is part of a set of alternative elements (see Dietrich 1990, Büring 1997, Drubig 1998a; also Rooth 1985, 1992 where focused constituents generally are associated with an alternative set). So, the generation of alternative sets is not a semantic property but rather a pragmatic function. It is the result of the conceptual processing of an uttered sentence containing a contrastively focused constituent. Pragmatically, the contrastive focus can be used to correct an assumption of the hearer. However, the correction of a referent that fulfils the proposition is possible only if the set of alternative elements is conceptually accessible in the given discourse context or by means of encyclopaedic knowledge. Especially question-answer contexts make clear the association of the contrastive focus with an alternative set. From a pragmatic point of view, the utterance of an interrogative sentence generates a set of propositions in possible worlds. Hamblin (1973) pointed out that "a question sets up a choice situation between a set of propositions, namely those propositions that count as answers to it" (ibid.; see also Ginzburg 1996: 390). Thus, every question defines a set of alternative proposi-

246

Andreas Späth

tions, i.e. an answer-set (q). Every element of this set is a valid option for an answer to the question: (1)

answer-set ("Who likes Bill") = Op [y (p = LIKES (y, bill))] = [LIKES (j, b), LIKES (m, b), LIKES (c, b), ...] (cf. Ginzburg ibid.)

Following (1), a valid option for the external argument in a possible world is the set of the values {j, m, ..., c  Oy.LIKES(y, bill)}. This set is an option only under the condition that the elements {j, m, ..., c} are presupposed or at least conceptually accessible in a given index of evaluation. From this it follows that what we have are individual constants, which constitute a set of alternative elements. The crucial point of the usage of contrastive focus is the generation of possible alternative referents in the given context of discourse. This issue is the precondition for the pragmatic function of correction by contrastive focus. The question arises, how the meaning of contrastive focus can be semantically represented and compositionally integrated into the sentence meaning. Authors who follow the theoretical framework of Rooth (1985) divide a sentence into its ordinary meaning and its alternatives (see also Krifka 2001: 293): (2) i. expression: [JOHN]F ordinary meaning: john alternatives: ALT(john) = {john, george, jim, ...} ii. expression: [MARY]F ordinary meaning: mary alternatives: ALT (mary) = (mary, sandra , jane, ...}

[JOHN]F loves Mary. love(mary)(john) {love(m)(x) | x  ALT(j)} = {love(mary)(john), love(mary)(george), ...} John loves [Mary]F love(mary)(john) {loves (x)(sandra) | | x  ALT(mary)} = {loves(mary)(john), loves(sandra)(john), ...}

"For meanings D, ALT(D) stands for the alternatives of D. The alternatives can either be taken broadly (e.g. ALT(D) = all objects of the same type as D; this is the option that Rooth 1992 chooses), or restrictively (ALT(D) = the set of relevant alternatives in a particular context); they must contain at least two elements" (ibid. 293). If it is the case that the ordinary meaning represents the intensional and truth conditional sentence meaning, both sentences should have identical truth conditions. And thus, both sentences should be usable in identical contexts. But,

On the Semantic Foundations of the Contrastive Focus

247

this is not the case. Under contrastive focus sentence (2i) can be successfully uttered only if there exists at least one alternative element that can replace the value of the external argument John and vice versa for the internal argument in example (2ii). If the semantic meaning of the contrastive focus corresponds to the induced set of alternatives, a serious problem arises: The compositional sentence meaning is taken to be a context free semantic representation. However, the alternative set is a representation of constants that are given in the shared knowledge base of the interlocutors, i.e. the representation of an alternative set contains extra-linguistic information. Such a representational form is inconsistent with the fundamental requirement of Two-Level Semantics, which requires that the semantic representation of a sentence contains only information that is part of the linguistic knowledge. Only utterance meaning can be enriched by non-linguistic information (see Wunderlich 1991). The same problem arises in the framework of the Structured Meaning Approach (e.g. Jacobs 1983, von Stechow 1985/89, Rooth 1985, 1996). Here, the focus-background structure corresponds to an expression of a lambda-calculus: (3)

a. b.

[JOHN]F loves Mary. Ox John loves [Mary]F Oy

The lambda-abstraction on the focused constituent takes into account that we are dealing with a structured proposition that denotes a set of possible alternatives. On the other side, in the logical schemes of (3) the focus semantic value is given and we can replace the O-abstracted argument by the expressions john and mary, respectively. Thus, the two representations are logically equivalent and represent the same proposition: loves (john, mary). Again, we receive identical conditions for the usage of a sentence in a context, i.e. the context changing potential of the sentence is not visible in the semantic representation. The specific information that provides the different prosodic structures in (3) is cancelled. In the following sections of this paper I will make an attempt at filling the gap between the generation of alternative sets (being part of extra-linguistic knowledge) and the ordinary meaning of a sentence. The goal is to insert into the compositional sentence meaning that linguistic information which triggers the restrictions of possible contexts in which that sentence can be uttered depending on the constituent that bears the feature of contrastive focus. Note that I do not question the correctness of the association of contrastive focus with alternative sets.

248

Andreas Späth

However, I will argue that an alternative set is the result of an inferential operation that is triggered by certain information in the sentence meaning. That is, I doubt that the ordinary meaning adequately represents the complete sentence meaning and its contextual potential. Therefore, I will enrich the semantic representation of the sentence meaning by exactly those triggers that generate alternative sets on the level of the conceptual system (CS). The theoretical framework of this paper is based on the assumptions of Two-Level Semantics. The Semantic Form of a sentence contains only linguistic knowledge (see Bierwisch 1979, 1987, 1988; Wunderlich 1991, Lang 1994). Following Wunderlich (1991: 40), the notion of sentence meaning comprises both truth conditions and context conditions, in so far both conditions are a compositional part of the sentence.1 The analysis is based on a strong lexicalist approach according to which the lexicon provides completely specified units that enter into the syntactic derivation. The interface between syntax and semantics is defined by the argument structure of an expression. The lambda-prefix of an expression determines its argument structure. During the syntactic derivation and the semantic composition, all lambda-abstracted arguments have to be bound. In this respect, I make use of the lambda-calculus in a way that differs from that of the Structured Meaning Approach, for instance.

2 Extensionality of alternative sets The question arises whether the generation of alternative sets belongs to the truth conditions of a sentence that contains a contrastive focus. This seems not to be the case. Sentence (4) illustrates that the elements of an alternative set must be extra-linguistic entities. The given set is a part of the shared knowledge bases of the interlocutors: (4)

i. ii.

1

Anna liebt [PETER]CF Anna loves [PETER]CF ALT(Peter) = {Peter, Otto, Franz, Wilbert, ...}

Wunderlich (1991) discusses the notion of context condition on the basis of indexical expressions. Since a definite DP presupposes the unique existence of a referent, the notion of context conditions can be applied to the informational status of a constituent as well. Analogously to indexical expressions, the information structure of a sentence contains requirements on the structure of the context of utterance. In this sense, information structure restricts the set of possible contexts in which a sentence can be used.

On the Semantic Foundations of the Contrastive Focus

249

Since the sentence meaning is a context-free semantic representation, the expression (4ii) cannot be a part of the intensional Semantic Form of the sentence (4i), because the alternative set contains individual constants. However, the sentence meaning does contain variables. It suggests itself that non-explicitly mentioned alternative elements occur as unbounded individual variables, i.e. as semantic parameters that are interpreted on the conceptual system as constants. iii.

ALT (Peter) = {Peter, z, r, y, ...}

After mapping the Semantic Form of the sentence onto its context, the referential value of the variables can be determined. However, the alternative set in (4iii) contains both extensional and intensional expressions. Furthermore, on the level of the grammatically determined sentence meaning, the extension of the contrastively focused constituent cannot be the argument of the function ALT. The contrastive constituent that evokes an alternative set is a DP-structure. The focus exponent is a part of the nominal head. Only the extension of the DP, i.e. its referent can be the argument of the ALT-function. The focused constituent has to be interpreted relative to a given discourse model: (5)

Anna liebt [DP ‡ [D den] [NP LEHRER]]CF Anna loves [DP ‡ [D the] [NP teacher]]CF i. ALT (eden LehrerfM,g) = {der Lehrer, der Direktor, der Hausmeister, ...} ii. ALT (eden LehrerfM,g) = {Peter, Otto, Franz, Wilbert, ...}

The alternative sets can be restricted relative to different domains. The elements of the set (5i) belong to the situation 'school'. The set contains referents that can be named by the role they play in a school, i.e. the teacher, the director, ..., the concierge. The alternative set in (5ii) contains male proper names that can be assigned to the bearer of a role. So, the elements of both sets can be identical referents. It seems to be appropriate to treat the ALT-function as a result of an inferential operation on the conceptual system. Thus, the ALT-function is not part of the linguistic knowledge. Rather the linguistic structure of a contrastive constituent contains some linguistic information that triggers the conceptual generation of plausible alternative elements.

250

Andreas Späth

3 Grammatical integration of the contrastive focus 3.1 Minimal semantic requirements on contrastive focusing Since the contrastive focusing generates a set of alternative elements, the positive case of contrast rules out the possibility that the proposition is fulfilled by an element of the given alternative set that is distinct from the referent of the contrastively focused constituent (6a). Otherwise, the proposition of the sentence is not true: (6)

a. b.

Otto trägt [ein HEMD]CF (und keinen Pullover). Otto wears [a SHIRT]CF (and not a jumper). Otto trägt [kein HEMD]CF (sondern einen Pullover). Otto does not wear [a SHIRT]CF (but a jumper).

In the case of negative contrast (6b), the reverse holds true. Unlike sentence negation, a sentence that contains a negative contrastively focused constituent asserts an affirmative statement of affairs. At the same time, the negative case of contrast rules out that the referent of the contrasted constituent fulfils the proposition. The proposition is fulfilled only by some element of the associated alternative set. Let us first look at the positive contrastive focus. The focus expands the meaning of a constituent in such a manner that only one entity – in the case of (6a) the object shirt – can fulfil the proposition, but not an entity whose properties R are distinct from the referent of the focused constituent. In this regard, it is very important to point out that an alternative property is subject to restrictions that result from the argument's thematic role and the syntactic function of a contrastively focused constituent. In addition, lexical selection restrictions determine the relevant properties of R.2 In a first attempt, we enrich the meaning of a contrastively focused DP by the partial meaning [™R (y)]. This information excludes the validity of a proposition with respect to an individual that has the property

2

The German verb tragen in (6) is a polysemous expression. By the usage of the verb in (6), only pieces of clothing can be contrasted with one another. However, the following sentence is, in principle, a possible utterance: i. Otto trägt eine HOSE, und keine Verantwortung/keinen Koffer etc. Otto wears TROUSERS, but he does not bear responsibility/a suitcase etc. A meaningful utterance of the sentence is possible only in very specific contexts. The present paper does not consider these cases of polysemy and contrast. The more so since the truth conditions of the correction only result from the conceptual determination of the respective polysemous variants of the verb.

On the Semantic Foundations of the Contrastive Focus

251

R (7i):3 Similarly to non-contrastive DPs, a contrastively focused DP triggers an existential presupposition (7ii). At the same time, it holds that there exists an individual z bearing the property R, which does not fulfil the proposition (7iii). From this constellation there results the existence of an alternative set containing at least two elements (7iv): (7)

i. ii. iii. iv.

DP-SF (ad interimum): presupposition: excluded alternative: alternative set:

OQ [y [[[shirt (y)]š[™R (y)]]š[Q (y)]]] y [shirt (y)] z [R (z)] ALT (y) = {y, z, ... }

The question arises how we can establish the existence of an alternative element z. The negation on [R (y)] has to result in the inferred existence of an individual whose properties are distinct from the referent of y. At the same time, this individual has to be a plausible alternative element. This problem goes back to the nature of negation and existential statements. Firstly, for an unbound variable, its interpretation holds under existential closure on the level of the conceptual system: (8)

R (y) { y [R (y)]

Secondly, with respect to the equivalence in (8), we have to consider the scope of negation. External negation affects the whole expression, i.e. a proposition. So, the existential statement of the referent of y can be cancelled. However, a variable of semantic type e serves as the referential value of a function and determines the truth-value of an expression of the semantic type t. The structural constellation of external negation does not hold for the semantics of natural language sentences (see Späth & Trautwein 2004).4 A positive or a negative statement is true if (9) holds: 3

4

We deal with DPs as quantifiers. The differentiation of DPs as individual terms e and as quantified NPs in dependence on their referential potential gives rise to several undesirable consequences. For instance, the lexicon provides two different operators for the binding of the referential argument of the nominal predicate. If so, the referential status of a DP, i.e. reference to an individual or reference to a set of individuals, is already determined in the lexicon. This proceeding is in contradiction to the fact that in many cases we can determine the semantic type of a DP only by means of conceptual interpretation (see Späth 2006 for details). Already in the earlier works of Hajiþová (1973) and Givón (1978) but also Roberts (1996) it has been pointed out that external negation is excluded for natural language sentences. This assumption holds especially for categorical sentences, which have a sentence topic. The topic is part of the presupposition of a sentence and has moved into a syntactic position beyond the scopal access of the operator of sentence negation. Therefore, the existential statement of the topic expression cannot be suspended. In the case of thetic sentences, the statement of an event, that instantiates the proposition, is

252 (9)

Andreas Späth

i. ii.

R (y) is true, iff eyfM  eRfM ™R (y) is true, iff eyfM  eRfM

The partial meaning of the contrastive focus has to make sure that the expression ™[R (y)] results in the presupposition that there exists an alternative individual which is not identical with the referent of the syntactic constituent that is contrastively stressed. For this reason, we have to provide two variables whose reference is distinct from each other. Thus, we introduce an additional piece of information into the semantic representation of the contrastively focused constituent: if the referent of y does not fulfil the property R, then a referent z exists that fulfils the property R: (10) Truth condition of contrast: Oy [[P (y)]š[™R (y) o R (z)]]

The definition in (10) triggers the generation of an alternative set containing at least two elements. The alternative set on CS is achieved by existential instantiation. That means, the existential binding of an individual variable corresponds to an expression that is stated to be true. Therefore, in the given index of evaluation, the variable can be replaced by an individual constant: (11) i. ii.

equivalence: R (y) { y [R (y)] existential instantiation: y [R (y)] ? R (w)

For existential instantiation, the following preconditions hold: "If an existentially quantified statement is true, there is at least one assignment to its variable which provides an instantiation for the matrix. Therefore we can infer from the truth (x)I(x) that I(w) for some constant w interpreted by an object in the universe of discourse. In general some instantiations of the matrix may be false, because the object assigned to the variable is not a true instantiation, and other assignments provide true instantiations" (Partee, ter Meulen & Wall 1990: 157). The referential value of the variable w has not necessarily to refer to a concrete object. It fixes the range of individuals that are possible referents. Thus, we do not look at the whole universe of discourse, but only at the subset of those objects that are true instantiations of the matrix in question (cf.

preserved from sentence negation: e [™[ ... e ... ]]. Otherwise, the sentence would not be an informative linguistic statement (for details see Späth 2006: 157–163).

On the Semantic Foundations of the Contrastive Focus

253

ibid.). In this way, we can restrict the set of alternative elements with respect to the context and the encyclopaedic knowledge, respectively. With regard to the semantic implementation of the statement [R (z)], we have to modify the truth conditions of (10). The partial meaning, i.e. [R (z)], gives rise to the assumption that the proposition is not fulfilled by the argument value of z, but by that entity that is the argument of the contrastively focused predicate: (12) Semantic template for contrastive focus: Op [[p]š[™R ( ... y[CF] ... ) o R (z)]]

The semantic template (12) is a functional lexical entry and combines with the contrasted entry in the lexicon. In the following, we can apply (12) to the contrastively focused DP in (6a). The result is the DPmeaning (13): (13) OQ [y [[shirt (y)]š[™R (y) o R (z)]]š[Q (y)]]]

With respect to the partial truth condition of the DP, sentence (6a) is true, if the individual y fulfils the property shirt and if the following requirement holds: if the referent of y does not have the property R, then an object z exists that fulfils the property R. In a next step, the unbound variable z can be interpreted relative to the given discourse model. Its existential presupposition results from the equivalence given in (8). The semantic template for contrastive focus is an intensional expression and gives rise to the presupposition of at least two elements (shirt, pullover, ..., t-shirt). However, the existence of an alternative element does not hold absolutely. The presupposition of an alternative object can be rejected: (14) A: B:

Otto trägt heute ja [ein HEMD]CF Otto wears a SHIRT today. Es hat auch niemand etwas anderes erwartet. Nobody expected anything else.

The utterance of (14A) leads to the implicit existential presupposition of an alternative object that possibly fulfils the proposition. The utterance of (14B) suspends that presupposition. Until now we discussed cases of positive contrast. For negative contrast, we have to draw a clear dividing line between contrastive negation and sentence negation. Sentence negation occurs in a fixed syntactic position, i.e. the polarity phrase PolP. The projection of PolP

254

Andreas Späth

immediately dominates the VP and hosts the feature of sentence negation (see Haftka 1994). Contrastive negation adjoins to that phrase which contains the contrastive focus constituent, i.e. contrastive negation (CN) is an adjunct of the focus phrase.5 (15) Es hat [nichtCN [PETER]]i[CF] [PolP[+neg] [VP ti keine Einladung erhalten]]] expl. aux [notCN [PETER]]i[CF] [PolP[+neg] [VP ti no invitation received]]]

It was shown above, that the contrastive focus enriches the sentence meaning by the information that a proposition is fulfilled, but not by the referent of the contrastively stressed constituent: (16) a. b.

Otto trägt kein HEMDCF (sondern ... ). Otto wears not/doesn't wear a SHIRTCF, (but ... ). Otto trägt nicht das HEMDCF (sondern ... ). Otto wears not/doesn't wear the SHIRTCF, (but ... ).

Neither sentence (16a) or (16b) expresses a negative state of affairs. Rather, the sentences assert that the person Otto wears a piece of clothing, but it is not a shirt (16a) or a certain, contextually given shirt (16b). Thus, the referent of the object-DP does not correspond to the referential value of the internal argument of the verbal predicate. The contrastive negation reverses the direction of contrasting. Now, the sentence is true if the referent does not have the property of the argument expression (for our examples in (16)). In addition, it fulfils the alternative property R. If that is the case, an alternative element z must exist that bears the property R. (17) i. ii.

5

6

positive contrast: Op [#[[p]š[™R (y)]]o R (z)]6 negative contrast: a. Op [™[[p]š[™R (y)]] o R (z)] b. { Op [[™[p]š[R (y)]] o R (z)]

For the reason that the focus exponent does not necessarily hold for the complete syntactic phrase, Krifka (1998) distinguishes between the focus phrase (FP) and the focus constituent (FC): i. [DP [the [NP [N' [AP GREEN][N shirt]]]]] That means for example (i): The focus phrase corresponds to the DP, but the focus constituent is the AP. The sign '#' marks the position where in the reverse case the negation operates. The scopal material of '#' is subject to information that can be rejected as an illocutionary act of correction.

On the Semantic Foundations of the Contrastive Focus

255

In (17iia), the negative operator replaces the operator of position in (17i). The expression in (17iia) is logically equivalent to (17iib). Now we can apply the contrastive negation to the object-DP in example (16a): (18) OQ [y [™[[shirt (y)]š[™R (y)]] o R (z)]š[Q (y)]]] { OQ [y [[™[shirt (y)]š[R (y)]] o R (z)]š[Q (y)]]]

Until now a proposal for a purely semantic and context free account for contrastive focus was outlined. We exemplified the issue by means of contrastive focus on the head of a DP. In the following section, we will analyse constituents that modify the DP-structure and further contrastively focused constituents inside and outside the VP. 3.2 A proposal for lexical integration of the contrastive focus meaning 3.2.1 Focus phrase versus focus constituent The evaluation of the possible contrasting of referents does not result in a homogenous pattern of alternative elements. For example, Rooth (1985, 1992) up to Krifka (2002) mainly discussed alternative sets that are triggered by proper names and definite descriptions. This account gives the impression that an alternative element replaces a referent described by the whole DP. In contrast, Büring (1997) and Krifka (1998) analysed sentences, where parts of a DP – for example adjectival modifiers – establish an alternative set. Furthermore, complex functions – for example partial descriptions of an event – can be subject to contrast. The following examples outline the cases of contrasted (partial) constituents which will be discussed in the following sections, in particular: (19) a. b. c.

d. e.

Otto trägt [DP ein [NP HEMD]CF], (und keinen Pullover). Otto wears [DP a [NP SHIRT]CF], (but not a pullover). Otto trägt [DP ein [AP GRÜNES]CF [NP Hemd]] ... Otto wears [DP a [AP GREEN]CF [NP shirt]] ... Otto hat [DP ‡ [D DEN]CF [NP Apfel]] gegessen (und nicht diesen). Otto has [DP ‡ [D THIS]CF [NP apple]] eaten (but not that one). Otto hat den Apfel [GEGESSEN]CF Otto has the apple [EATEN]CF Otto trug [GESTERN]CF ein Hemd. Otto wore [YESTERDAY]CF a shirt *(und hat die Milch nicht getrunken. *(but he didn't drink the milk).

256

Andreas Späth

f.

g.

Otto hat das Hemd ANgezogen (und nicht ausgezogen). Otto put the shirt ON, (but he didn't take it off). Otto hat [VP [DP die [SUPPE]]i [PosP ‡ [VP ti gegessen]]]CF (und nicht die Milch getrunken). Otto has [VP [DP the [SOUP]]i [PosP ‡ [VP ti eaten]]]CF (but he didn't drink the milk).

Following Krifka (1998: 94), in the examples (19a–c) the DP is the focus phrase. In (19b,c), only parts of the argument are subject to the contrastive focus. These parts constitute the focus constituent, i.e. the adjective modifier in (19b), the alleged determiner in (19c) or the verbal participle (19d). Furthermore, we will discuss possible alternative elements of a temporal adverbial (19e) and the case of the contrast on parts of the semantic lexical meaning structure of a verb (19f). Finally, we will look at some problems of narrow and wide scope of contrast (19g). 3.2.2 Lexicalist integration of the contrastive focus meaning Syntactically based models of focus act on the assumption that the focus feature is freely assigned to a syntactic constituent of a sentence. That means, at first we derive the syntactic structure of a sentence, then we assign the focus feature (Rosengren 1993, Jacobs 1992, see also Eckardt 1996 for details). Example (20a) – taken from Steube (2001: 14) – indeed shows that the contrastive focus feature can be assigned to lexical components as a prefix that can be separated from its lexical root. At the same time, the component can be moved into a position that is dislocated from the position of the finite verb. It is questionable whether we are dealing with an elliptic realisation or with a copy of a partial constituent (see 20b):7 (20) a. b.

7

dass /AN Peter tatSÄCHlich das Licht gemacht hat und /AUS LEna. that /ON Peter perhaps really the light switched aux and /OUT LEna. that Peter really has switched the light on and Lena switched it out. *dass /ANgemacht Peter vielleicht tatSÄCHlich das Licht hat.

In German, the position of the accent is not necessarily identical with the lexical accent. The phenomenon of metalinguistic correction supports this claim (see Steube 2001 for details): i. Man sagt TomAte und nicht TOmate. One says tomAto, but not TOmato.

On the Semantic Foundations of the Contrastive Focus

257

The question arises whether the contrastive focus has a lexical counterpart in terms of a functional lexical entry. Languages that realise focusing by morphological means may serve as the starting point for this claim.8 In some European languages we can observe some instances of lexical realisation of contrastive focus, for example the use of strong pronouns: (21) a. b.

Daáem mu ksiąĪkĊ. I gave him the/a BOOK. Daáem JEMUCF ksiąĪkĊ. I gave HIM the/a book.

(Polish)

Bosnian-Croatian-Serbian (BCS) signals the VERUM-focus by the strong form of the copula or the auxiliary: (22) a. b.

I to je bilo pitanje ravnopravnosti. And it was a question of equal rights. I to JESTECF bilo pitanje ravnopravnosti. And it WAS a question of equal rights.

(BCS)

The selection of lexical units depending on the feature of contrastive focus shows that the assignment of the focus feature obviously takes place within the lexicon. It is not a post-lexical process that happens after the generation of the syntactic surface structure. This phenomenon points to a lexicalist account of contrastive focus, the more so since the contrastive focus enriches the meaning of a syntactic constituent. In the case of contrastive focus on a nominal predicate, the meaning of the contrastive focus is added to the basic entry of the noun. The pre-lexically assigned feature for contrast [CF] triggers this operation that takes place in the lexicon:9 (23) i. ii.

8

9

Nominal predicate: Ox[CF] [shirt (x)] CF-template: Op [[[p]š[™R (y)]] o R (z)]

Schauber (1978) as well as Vallduví & Engdahl (1996) present a number of languages that have a focus morpheme. For details on processes of the lexicon see Bierwisch & Schreuder (1992). For the lexicon, the authors differentiate basic entries from functional entries. Functional entries comprise e.g. affixes, but also rules of word formation.

258

Andreas Späth

First, the expressions (23i,ii) combine via functional composition: 10 iii.

Op [[[p]š[™R ( ... y[CF] ... )]] o R (z)] (Ox[CF] [shirt (x)]) { Ox [Op [[[p]š[™R ( ... y[CF] ... )]]o R (z)] (Ox[CF] [shirt (x)]) (x)] { Ox [[[shirt (x)]š[™R (x)]] o R (z)]

The result (23iii) is an update of the basic entry (23i), which is entered into its syntactic basic position, i.e. the head of NP in the present case: DP OQ [x [[[[shirt (x)]š[™R (x)]] o R (z)]š[Q (x)]]]

(24) SpecDP ‡

D0 ein OP OQ [x [[P (x)]š[Q (x)]]]

D' NP SpecNP ‡

N'

N0 HEMDCF Ox [[[shirt (x)]š[™R (x)]] o R (z)]

In the case of an indefinite DP, the sentence asserts that its proposition is fulfilled if the referent of the argument x has the property of the nominal predicate. In the case of a definite DP, the determiner enriches the partial truth conditions of the DP by the condition of unique existence of the DP-referent. 3.2.3 Adjectival modifiers as focus constituents Next, we will analyse contrasted adjectival modifiers of nominal heads. In these constructions, the focus phrase is the DP, but the focus constituent corresponds to the head of the adjectival phrase. With respect to the occurrence of adjectival modifiers within a DP, we have to take into account the following variants of focus constituents: (25) a.

10

Otto trägt [ein grünes [HEMD]FC]DP (und keinen grünen Pulli). Otto wears [a green [SHIRT]FC]DP (and not a green jumper).

The rule of functional composition allows the combination of two functions that differ in their number of argument places. The subordinate function transmits at least one argument place to the main function (see also Steedman 2000): i. P (Q) { Oyn ... Oy1 [P (Q (yn) ... (y1))] where P  D/E, Q  ( ... (E/J1... )/Jn, yi  Ji (Zimmermann 1988: 163)

On the Semantic Foundations of the Contrastive Focus

b. c.

259

Otto trägt [ein [GRÜNES]FC Hemd]DP (und kein weißes Hemd). Otto wears [a [GREEN]FC shirt]DP (and not a white shirt). Otto trägt [ein [grünes HEMD]FC]DP (und kein weißes T-Shirt).11 Otto wears [a [green SHIRT]FC]DP (and not a white jumper).

In the case of contrast on the head of the NP (25a) the semantic operations of (23) apply to the nominal predicate. In a next step, the adjectival modifier combines with the contrastively focused noun: (26) i. ii.

Semantic Form of N': Ox [[[[shirt (x)]š[™R (x)]] o R (z)]:[green (x)]] Restriction on alternatives of the DP-referent: ALT (x) = {x, ..., z} where  e greenf M

The property described by the modifier is not subject to the contrast. The contrastive meaning is restricted to the descriptive content of N0. That means, that all alternative elements have to fulfil the property 'green'. In (25b), the focus constituent only corresponds to the adjectival modifier. In contrast to (25a), the contrastive meaning does not affect the property of the nominal predicate. The modifying adjectival predicate has to be enriched by the template of contrast: (27) i. ii. iii.

Basic lexical entry of the adjective: Ox[CF] [green (x)] Semantic template of contrast: Op [[[p]š[™R ( ... x[CF] ... )]] o R (z)] Updated entry (resulting from functional composition): Ox [[[green (x)]š[™R (x)]] o R (z)]

The updated lexical entry of the adjectival predicate enters into its syntactic base position, i.e. into the head of AP that is an adjunct of N':12

11

12

There are possible contexts for modified nominal heads as a complex focus constituent: i. A: Ich werde Otto in dem Gewimmel sicherlich an seinem weißen Pulli erkennen. In the mass of people, I surely will recognize Otto by his white jumper. B: Aber Otto trägt ein [grünes HEMD] (und keinen [weißen Pulli]). But Otto wears a [green SHIRT] (and not a [white jumper]). Zimmermann (1992) discusses in detail the rule for modification and its application in compositional semantics: i. Semantic template for modification: OQ2 OQ1 Oxi [[Q1 (xi)]:[Q2 (xi)]] The semantic template (i) is a functional entry of the lexicon, as well. The basic entry of the adjective (27i) is not specified for its grammatical realisation, i.e. it can take part in word formation or it can occur as a modifier. In the latter case, rule (i) applies to the semantic integration of the adjectival predicate.

260

Andreas Späth

(30) i.

N' Ox [[shirt (x)]:[[[green (x)]š[™R (x)]] o R (z)]]

AP GRÜnesCF Ox [[[green (x)]š[™R (x)]] o R (z)]

ii.

N' N0 Hemd Ox [shirt (x)]

Restriction on alternatives of the DP: ALT (x) = {x, ..., z} where  eshirtf M

In (28i) the adjectival modifier is enriched by the meaning of the contrastive focus. Unlike (25a), the property of the nominal predicate is not affected by the contrastive meaning. The alternative elements have to belong to the set of those objects that fulfil the extension of the adjectival predicate.13 Example (25c) is a case of a complex focus constituent. The focus domain comprises the concatenation of the noun and the adjective. In contrast to (25a), the condition for the identification of the referent consists in a syntactically complex expression. The nominal head contains the focus exponent. However, the elements of the alternative set are not restricted to the meaning of the nominal predicate. The same holds for languages that allow postposition of the adjectival modifier: (29) Jarek nosi [páaszcz przeciwdeszCZOwy]FC, a nie wiaTRÓWkĊ. Jarek wears a [coat RAINproof]FC, but not a WINDcheater. Jarek wears a RAINcoat, but not a windcheater.

(Polish)

The focus is realised on the right most syllable that can be accentuated within the focus constituent. Also, in the case of (29), the alternative set does not consist of properties that are possible alternatives with respect to the focused adjective. There is no given set of coats from which we can single out some elements with distinct properties. Therefore, in the case of (25c) we cannot assume that the focus domain is semantically underspecified. Thus, we have to find a solution that excludes the generation of truth conditions which are identical to those of (25a). Unlike (25a,b), the focus does not only concern the descriptive content of the nominal head or the descriptive content of the head of AP, but the concatenation of both constituents. That means, the focus relates to the syntactic projection level N'. For semantic combination, it holds that the focus applies a modified structure that forms a complex expression, i.e. 13

Umbach (2001) discusses in detail the problem of the discourse-pragmatic restriction of alternatives that result from adjectival modification of a noun.

On the Semantic Foundations of the Contrastive Focus

261

the focus domain comprises the semantic scheme (template) of modification (see also footnote 12). In a final step, the adjectival predicate can saturate the argument position of Q1 (30iii): (30) i. ii. iii.

Semantic scheme of modification: OQ2 OQ1 Oxi [[Q1 (xi)]:[Q2 (xi)]] Application of the CF-template to (i) by functional composition: Op [[[p]š[™R (x[CF])]] o R (z)] (OQ1 OQ2 Oxi [[Q1 (xi)]:[Q2 (xi)]]) { OQ2 OQ1 Oxi [[[Q1 (xi)]:[Q2 (xi)]]š[™R (x[CF])]]o R (z)]] SF der AP: OQ2 OQ1 Oxi [[[Q1 (xi)]:[Q2 (xi)]]š[™R (x[CF])]]o R (z)]] (Oy [green (y)]) { OQ1 Oy [[[Q1 (y)]:[green (y)]]š[™R (y)]]o R (z)]]

The resulting expression can be inserted into the syntactic base position of the modifier. The scope of the contrastive focus in (25c) results from the semantic composition up to N': (31)

N' AP grünes

N'

OQ1 Oy [[[Q1 (y)]:[green (y)]]š[™R (y)]]o R (z)]]

N0 HEMD Oy [shirt (y)]

Oy [[[shirt (y)]:[green (y)]]š[™R (y)]] o R (z)]]

The result of the compositional syntax-semantics mapping at the level of N' provides the information that the denotation of the complex expression grünes Hemd 'green shirt' is a part of a communicatively relevant set of alternatives. The indefinite or definite determiner binds the referential argument of the nominal predicate. In this way, the referential potential of the modified DP is narrowed down. 3.3.3 Contrast on alleged determiners As example (19c) shows, the contrastive focus can be assigned to an expression that seems to be a determiner. The question arises, whether or not it is really the determiner that is contrastively focused. If so, what is an alternative element with respect to the definite or indefinite deter-

262

Andreas Späth

miner?14 It is a contentious issue whether the definite and the indefinite article stand in are in opposition to each other. However, according to Löbner (1990: 43) this only holds for the definite article. If so, the focus on the definite article has no alternative element. A closer look at the usage of focused determiners shows that we are dealing with is a focused definite determiner used as a demonstrative: (32) Otto trägt DAS/DIESES Hemd (und nicht jenes). Otto is wearing THIS shirt (but not that).

The critical property that the focus marks is the descriptive content of the demonstrative. The existence of at least two objects is presupposed. The correction operates on the deictic reference to one element of an alternative set with identical properties. So, the contrastive focus concerns the specific meaning of the demonstrative which asserts the proximity of the object: (33) i. ii.

Syntax of the DP: [DP [D [+def] [NP [AP dieses] [N' [N Hemd]]]] SF of the DP: OP OQ [[!x [[shirt (x)]:[prox (x < z)]]š[Q (x)]]]

The SF (33ii) states that the truth of the sentence depends on the proximal object x, but not on the distal object z. The semantic integration of the specific truth conditions in (33ii) results in the assertion that there is a distal object z, but z does not fulfil the proposition of the sentence. For the semantic composition it is important that the nuclear scope of the quantifier is not affected by the contrast. The problem disappears if we analyse a demonstrative as a modifier of N0. The head of the DP hosts an abstract feature for inherent definiteness (33i). The CF-template applies to the basic lexical entry of the demonstratives (34). (34) a.

14

/dies-/: i. basic entry: Oy[+def] [prox (y) < prox (z)] ii. updated entry by the CF-template: Oy[+def] [[[prox (y) < prox (z)]]š[™R (x)]] o R (z)]]

There are occurrences where the accent really falls on the determiner. These cases seem to be subject to metalinguistic correction (see footnote 7) which is not the subject of this paper: i. Das ist nicht (nur) EIN Hit, sondern DER Hit. This isn't (only) A hit, but THE hit.

263

On the Semantic Foundations of the Contrastive Focus

b.

/jen-/: i. basic entry: Oy[+def] [prox (y) > prox (z)] ii. updated entry by the CF-template: Oy[+def] [[[prox (y) > prox (z)]]š[™R (x)]] o R (z)]]

The result of the lexical processes in (34) is an updated entry that we can insert in its syntactic basic position. The focused demonstrative combines with the nominal head. Finally the determiner in D0 binds the referential argument: (35)

DP SpecDP ‡

D'

D0[+def] OP OQ [!y [[P (y)]š[Q (y)]]] SpecNP ‡

NP N'

AP Oy[+def] [[[prox (y) < prox (z)]]š[™R (y)]] o R (z)]]

N'

N0 Oy [shirt (y)] Oy [[shirt (y)]:[[[prox (y) < prox (z)]]š[™R (y)]] o R (z)]]] OQ [!y [[[shirt (y)]:[[[prox (y) < prox (z)]]š[™R (y)]] o R (z)]]]š[Q (y)]]]

The result of the semantic composition of the DP asserts that the proposition is fulfilled, if the referent of y has the extension of the nominal predicate of the restrictor and if the proximity of the object y is locally precedent to an object z. The truth of the sentence depends on the precondition that the referent y does not have the alternative property R. If so, there is an object z that has the property R. Thus, the directional local deixis triggers the existence of two alternative objects. In German, the number of possible alternative sets is restricted to the deictic points that are expressed by the demonstratives dieser 'this' and jener 'that'. However, there are languages that can express more than two levels of positional deixis (see Berger 1994).

264

Andreas Späth

3.3.4 An example for contrast on sentential modifiers In the next section I will look at the semantic contribution of contrastively focused sentential adverbs. The analysis exemplifies the problem by means of a temporal adverbial. It will become clear that the interpretation of the contrastive focus again is based on the lexical analysis of the expression at hand. The problem can be outlined by the following example: (36) Hans kam GESternCF Hans came YESterdayCF

According to Bierwisch (2003: 138) we can assume the following lexical entry for the temporal adverbial gestern 'yesterday': (37) Lexical entry for gestern (Bierwisch ibid.): /gestern/: [+V, +N] Ox [yesterday x] where [yesterday x] =DEF [[T x Ž z]&[u Ž y]&[day z]&[day y]&[z )( y]] with T x = the time interval assigned to x, u = the utterance time, z )( y = z directly precedes y.

The question arises what the contrasted part of the decomposed semantic structure of the adverbial in (37) is. The contrast obviously concerns the time interval that is assigned to the variable x. Here, the variable x belongs to the ontological domain of events, i.e. an event is embedded into the temporal span of z. Thus, the contrast concerns the linguistic information that the event instantiates a proposition whose temporal localization is embedded in the interval [day z]. The interval z directly precedes the temporal span of the interval [day y]. The contrast on the temporal adverbial yesterday refers to the instantiation of the proposition with respect to that temporal interval that provides the lexical meaning of the adverbial: (38) i. ii. iii.

Contrast template: Op [[[p]š[™R (x[CF])]] o R (z)] Lexical entry of the adverbial: Oe [[T e Ž z] & [u Ž y] & [day z] & [day y] & [z )( y]] Result after functional composition of (i,ii): Oe [[[[TeŽz] & [uŽy] & [day z] & [day y] & [z )( y]]š[™Re]] o R (z)]

The truth conditions of the temporal adverbial determine the semantic sort of the alternative elements. The alternatives result from the reference to the temporal interval [day (z)] and from the meaning of the contrastive focus. The negation of the predicate in the second conjunct

On the Semantic Foundations of the Contrastive Focus

265

rules out that the proposition is fulfilled if the event that instantiates the proposition, is embedded into the span of another temporal interval than z, where z directly precedes the interval [day y]. 3.3.5 An example for contrast on a lexical compound Example (19f) in section 3.2.1 showed that contrastive focus can also affect lexical compounds, i.e. in the given case the prefix of a verb, where only the prefix meaning is affected by the contrast, not the entire verbal meaning. The examples in (39) illustrate this phenomenon: (39) a. b.

Otto hat das Hemd \AUSgezogen. Otto has taken OFF the shirt. Anna wollte in den Zug \EINsteigen. Anna was going to get INTO the train.

The grammatical analysis of prefixes is similar to the analysis of prepositions. However, the prefix is part of the word structure. The composition of the root of a lexical item with a prefix has to take place in the lexicon during the processes of word formation. However, the contrastive template applies to the lexical meaning of the prefix. The question arises on what descriptive content the contrast operates in examples such as (39a): (40) /anprefix/: Oe [[ ... e ... x ... ]:[fin [loc (x) at loc (z)]]

In the given example, the contrast does not affect the complete verb meaning, but only the partial meaning that an object x moves on a local path to the point of the localisation of z. Thus, the contrast on the prefix rules out the movement in the opposite direction of the path:15 (41) Nicht /AN hat er das Hemd gezogen, (sondern AUS). not /ON aux he the shirt put, (but OFF). He didn't put the shirt on, but rather took it off. 15

The opposition of the Geman verbs anziehen and ausziehen results from the fact that the prefixes differ in the directionality of the path that they describe. However, this is a language specific phenomenon. In English, one can contrast the prefixes (put) on and (take) off, although the lexical roots of the verbs are not identical. In contrast, if the sentence is translated into Polish, the opposition is expressed by completely different lexems: wáoĪyü (koszuáĊ) 'put on (a shirt)' vs. 'zdjąü (koszuáĊ) 'take off (a shirt)'. This phenomenon shows that the generation of alternative elements is not only subject to extralinguistic knowledge, but it can be triggered by the grammatically determined meaning of a linguistic expression.

266

Andreas Späth

In the lexicon, the contrastive template combines with the meaning structure of the prefix. The resulting expression combines with the root of the verbal predicate: (42) i. ii.

Lexical entry of the verbal prefix: Oe [[ ... e ... x ... ]:[fin [loc (x) at loc (z)]] CF-template: a. positive: Op [[[p]š[™R ( ... x[CF] ... )]] o R (z)] b. negative: Op [™[[p]š[™R (y)]] o R (z)] iii. lexical update of (i): a. positive: Oe [[[[ ... e ... x ... ]:[fin [loc (x) at loc (z)]]š[™R (e)]] o R (z)] b. negative: Oe [™[[[ ... e ... x ... ]:[fin [loc (x) at loc (z)]]š[™R (y)]] o R (z)] { Oe [[™[[ ... e ... x ... ]:[fin [loc (x) at loc (z)]]š[R (y)]] o R (z)]

The elements of the alternative set result from the possible directional variations. In the example at hand, we are dealing with alternative referents that are embedded in a local path of. 3.3.6 An example for contrast on the verbal meaning Within the verbal complex, the focus can be assigned not only to the actual verb but also to its auxiliary components. Furthermore, a modal verb can contain the focus exponent. The contrast on the verbal meaning has to be distinguished from VERUM-focus. But, in present tense a contrastively focused verb shows ambiguity in German: (43) a. b. c.

Otto BÜgelt das Hemd. Otto is ironing the shirt. Otto hat das Hemd gebügelt. Otto has the shirt ironed. Otto HAT das Hemd gebügelt. Otto HAS the shirt ironed.

Following Brandt et al. (1992), the VERUM-focus operates on the statement of an event that instantiates the proposition of the sentence, i.e. the VERUM-focus operates on the scale of 'true – false' and concerns the polarity of an utterance. Thus, the VERUM-focus has to operate on that constituent of the verbal complex which bears the event role. Only a finite verbal form can fulfil this function, since the participle

On the Semantic Foundations of the Contrastive Focus

267

does not have the grammatical features of a verb (see Zimmermann 1999a: 129–130, Späth 2006: 257–249).16 Sentence (43b) differs from (43a,c) in that the contrast only affects the descriptive content of the verbal predicate. The contrast marks whether the event is an instance of the proposition or not, depending on the positive or negative contrast: (44) Verbal predicate: Oy Ox Oe[CF] [e inst [iron (x, y)]] i. Positive contrast: Oy Ox Oe [[[e inst [iron (x, y)]]š[™R (e)]] o R (z)] ii. Negative contrast: Oy Ox Oe [™[[e inst [bügeln (x, y)]]š[™R (e)]] o R (z)] iii. ALT (e) = e inst {iron, wash, ... }

It is not the case, that the contrast necessarily concerns the verbal predicate only. There is an ambiguity between the narrow focus on the verbal predicate and the wide focus that comprises an argument noun, for instance.17 One the one hand, the descriptive content of the event results from the compositional meaning of the proposition that consists of a verbal predicate and its arguments. On the other hand, sentences are subject to information structure. Thus, an argument DP can move out of the VP, for instance if it belongs to the background of the sentence like the topic-DP Otto in (45): (45)

a. b. c.

Otto hat das Hemd nicht gebügelt. Otto hat das HEMD nicht gebügelt. Otto hat nicht das HEMD gebügelt, sondern den Tisch abgeräumt.

In sentence (45a) the descriptive content of the verbal predicate is subject to the contrastive focus and corresponds to the focus phrase. In (45b) the focus phrase is the object-DP. In both cases there is narrow focus. However, as example (45c) and its context show, the focus phrase can be a complex expression that consists of the concatenation of the verbal predicate and an argument-DP. The sentence contrasts a complex property that holds with respect to the sentence topic Otto. In this case there is wide focus. Although the focus constituent is the ob16

17

Sentences like (43a,c) are ambiguous again. The contrastive focus does not necessarily concern the polarity. It can also contrast, e.g., the information provided by tense. However, such cases of contrast on grammatical categories cannot be considered in this paper. We are dealig with a phenomenon similar to that where the focus domain is a concatenation of the nominal head and an adjectival modifier (as shown in 3.2.3., example (25c).

268

Andreas Späth

ject-DP, the contrast concerns the descriptive content that results from the composition of all syntactic constituents in VP. In the given example (45c), the contrast has to operate on the concatenation of the objectDP and the verbal predicate: (46) i. ii.

Concatenation of object-DP and verbal predicate: Ox Oe [!y [[shirt (y)]š[[iron (x, y)] (e)]]] Application of the CF-template to the complex function (i): Ox Oe [™[[!y [[shirt (y)]š[[iron (x, y)] (e)]]]š[™R (e)]] o R (z)]

The complex function that is subject to the contrast cannot be semantically decomposed. This is apparent in examples with fixed lexically units where the noun does not refer to an extra-linguistic object:18 (47)

Otto hat nicht die Konsequenzen gezogen, Otto aux not the consequences draw Otto didn't take the necessary steps, a. sondern den Vorfall einfach ignoriert. but he simply ignored incident. b. */?sondern den Hut. but he took his hat off.

That this is a case of wide focus on a complex function which describes a (partial) state of affairs and that this function cannot be decomposed becomes clear if we move one constituent of the VP into sentence initial position. The topicalisation of the object-DP is possible only if the DP is a referential expression that refers to a discourse referent (see also Fanselow 2003): (48) a.

b.

18

Was hat Peter gemacht? What did Peter do? i. [Das HEMD gebügelt]i hatk erj tj ti tk [the shirt ironed-participle]i auxk hej tj ti tk ii. ?[Das HEMD]i hatk erj tj ti gebügelt tk ?[ the shirt]i auxk hej tj ti ironed-participle tk Wie hat Otto reagiert? How did Otto react? i. [Die KONSEQUENZEN gezogen]i hatk erj tj ti tk [the necessary STEPS token-participle]i auxk hej tj ti tk ii. ?[Die KONSEQUENZEN]i hatk erj tj ti gezogen tk ?[ the necessary STEPS]i auxk he ti token-participle tk

The German lexical units die Konsequenzen ziehen 'to take the necessary steps' and den Hut ziehen 'to take one's hat off' have the same verbal base, but they differ in their object-DPs.

On the Semantic Foundations of the Contrastive Focus

269

Similarly to the contrast on the adjectival modification of a noun, where the focus affects the logical scheme of a modified nominal structure, in (45c), the focus affects the compositional pattern of the concatenation of a verbal predicate with an argument-DP. The compositional combination is achieved by the rule of semantic type shift:19 (49)

Type shift rule: OP Oƒ Oxn-1 ... Ox1 [ƒ (Oxn [P (x1, ..., xn)])] (Zimmermann 1999b: 51)

The combinatory rule of semantic type shift is again a functional entry in the lexicon. Since the contrast concerns the syntactic constellation that corresponds to the semantic combinatory rule of type shift, the CFtemplate specifies that rule (50i). In the next step the modified template for type shift applies to the argument structure of the verb (50ii). The result of (50ii) is an expression that can be inserted into the base position of the verb. Finally, the type shifted semantic form of the verb combines with the DP of the semantic type (50iii): (50) i. ii.

Application of the CF-template to (49): OP Oƒ Oxn-1 ... Ox1 [[[ƒ (Oxn [P (x1, ..., xn)])]š[™R ( ... x[CF] ... )]] o R (z)] Application of (i) to the argument structure of the verb: OP Oƒ Oxn-1 ... Ox1 [[[ƒ (Oxn [P (x1, ..., xn)])]š[™R ( ... x[CF] ... )]] o R (z)] (Oy Ox Oe [[iron (x, y)] (e)]) { Oƒ Ox Oe [[[ƒ (Oy [[[iron (x, y)] (e)])])]š[™R ( ... x[CF] ... )]] o R (z)]

iii. Compositional integration of the object-DP:

ti

VP Ox Oe [[[OQ[!y[[shirt(y)]š[[[iron (x, y)](e)]]])]š š[™R(e)]]oR (z)] V'

DP V0 OQ [!y [[shirt (y)]š[Q (y)]]]  Oƒ Ox Oe [[[ƒ (Oy [[[iron (x, y)] (e)])])]š[™R (x[CF])]] o R (z)]  —.ARL]Topic wahrscheinlich nach PO\len reist I think that Karl probably to Poland goes und dass [—OTto]Topic sicherlich einige Wochen in ITA\lien and that Otto certainly some weeks in Italy verbringen wird. spend will b. #Ich denke, dass wahrscheinlich >—.ARL]Non-Topic nach PO\len reist und dass sicherlich [—OTto]Non-Topic einige Wochen in ITA\lien verbringen wird. (15) Wer macht denn die Reise mit? Who is coming on the trip Gestern hat anscheinend —.ARL InterESSE\ bekundet und heute yesterday has apparently Karl interest expressed and today hat erfreulicherweise —OTto fest ZU\gesagt. has luckily Otto firmly promised

In both (14) and (15), the answering sentences contain phrases with the fall-rise intonation. The questions in (14) trigger the effect that these phrases figure as aboutness topics of the answering sentences. According to (4), topics occurring in the middle field have to be positioned in front of a sentence adverbial. This explains the difference in acceptability between (14a) and (14b). In contrast, the question in (15) does not induce topical status for the phrase with the fall-rise intona-

336

Werner Frey

tion. Therefore, in the answer in (15), this phrase does not have to be positioned in front of the sentence adverbial. Example (14) disconfirms McNally's conclusion that the notion of 'aboutness topic' is irrelevant for the analysis of sentences containing phrases with the fall-rise intonation. Rather, (14) and (15) together show that the topical status of a phrase is independent of the fall-rise intonation. Given the independence of the two markings and given (4), we expect that a phrase which, by its very nature, cannot be a sentence topic but which is assigned the fall-rise intonation cannot appear in the topic position. This expectation is confirmed: (16) a.

b.

??Heute wird fast —JEder sicherlich mindestens EIN\ Gemälde today will nearly everyone certainly at least one painting buy kaufen. Heute wird sicherlich fast —JEder mindestens EIN\ Gemälde kaufen.

4 Some further contextual effects of sentence topics We have seen an example in which the context of a sentence forces a certain element to be designated as a topic of that sentence: In (14), a question which contains a topical plural expression denoting a set demands that phrases denoting members of this set be treated as topics. In this section, it will be investigated whether there are further examples showing the interaction of sentence topics with the context. We may start by considering cataphoric pronouns again. In (7a) above, the cataphoric pronoun and the co-referential phrase appeared in the same sentence but a cataphoric pronoun and the co-referential phrase may also appear in separate sentences. Here too, we find that a cataphoric pronoun has to relate to a sentence topic: (17) a. b.

Er1 war zufrieden. Heute wurde Koch1 erwartungsgemäß nominiert. he was satisfied today was Koch as expected nominated *Er1 war zufrieden. Heute wurde erwartungsgemäß Koch1 nominiert.

Thus, (17a) constitutes further evidence that a sentence topic may have effects outside of its sentence. Let us now consider some examples exhibiting different discourse relations to see whether they are sensitive to the designation of sentence topics. The following discussion will not be systematic. Rather, the discourse relations are more or less randomly chosen.

Some contextual effects of aboutness topics in German

337

Example (18) is an instance of the discourse relation Elaboration.1 The first sentence is elaborated by the sequence of the following sentences: (18) Den ganzen gestrigen Tag hat Emil1 überraschenderweise zu Hause verbracht. all yesterday long has Emil fortunately at home spent Zuerst hat er1 die Küche gereinigt. Dann hat er1 das Kinderzimmer geordnet. first has he the kitchen cleaned then has he the children's room tidied a. #Dann war Mittagessenszeit und da hat [Maria]Topic glücklicherweise then was lunch time and then has Maria fortunately [dem tapferen Mann]1Non-Topic eine Linsensuppe gegeben. the brave man a lentil soup given b. Dann war es Mittagessenszeit und da hat [Maria]Topic [dem tapferen Mann]1Topic glücklicherweise eine Linsensuppe gegeben.

In the sentence to be elaborated, the phrase Emil figures as a topic. The elaborating sentences contain phrases which are co-referential with Emil. (18a, b) show that it is preferred to keep the topical status of Emil constant. The following constraint seems to hold: (19) If a sentence E to be elaborated contains a topical phrase Į and an elaborating sentence Ei contains a phrase ȕ which is co-referential with Į or denotes a member or a subset of the denotation of Į, then ȕ is a topic in Ei.

Intuitively, (19) makes sense: If the sentence which will be elaborated is especially concerned with a certain referent, then the elaborating sentences will be too. Thus, if the referent in question is explicitly referred to in an elaborating sentence, the referring phrase has to be a sentence topic. The next example also illustrates (19): (20) Auf der Sitzung werden sich die führenden CDU-Politiker on the meeting will REFL the leading CDU politicians höchstwahrscheinlich auf einen Kandidaten einigen. most likely on a candidate agree a. Laut FAZ wird Schäuble mit Sicherheit für Merkel plädieren. according to FAZ will Schäuble definitely for Merkel argue b. #Laut FAZ wird mit Sicherheit Schäuble für Merkel plädieren.

1

For our purposes the following rough characterization of Elaboration is sufficient: Elaboration is a subordinating discourse relation whose Nucleus, the elaborated statement, describes something about which its Satellites, the subsequent elaborating statements, give more specific information (cf. e.g. Mann & Thompson 1988). I conjecture that in a broader sense question-answer pairs can also be seen as instances of Elaboration.

338

Werner Frey

The term die führenden CDU-Politiker is a topic and so is the instantiating term Schäuble in the elaborating sentence. In a Narration, a sequence of events is reported.2 The discourse participants thereby implicitly characterize an event of a more general type which comprises the singular events reported in the Narration (cf. e.g. Asher 2004). The question we may ask is whether the designation of sentence topics has any effects on this undertaking. Let us look at the following examples: (21) a.

b. c.

Erst hat den Paul überraschenderweise Maria beleidigt. Dann Maria insulted. Then First has theACC Paul surprisingly hat den Fritz leider die Frau Professor gekränkt. hurt has theACC Fritz unfortunately the professor Erst hat Maria überraschenderweise den Paul beleidigt. Dann hat die Frau Professor leider den Fritz gekränkt. Erst hat Maria den Paul überraschenderweise beleidigt. Dann hat die Frau Professor den Fritz leider gekränkt.

It seems that sentence topics do have effects on the content of the 'super-event'. According to the intuition of my informants as well as my own intuition, (21a) is interpreted as reporting the terrible things that happened to Paul and Fritz. In contrast, (21b) is more likely to be interpreted as reporting the terrible things that Maria and the professor did. Finally, (21c) is concerned with the terrible things that Maria and the professor did to Paul and Fritz, respectively. These intuitions support the following constraint on Narration: (22) The sentence topics of the different sentences of a Narration figure in the event which is implicitly reported by means of the Narration.

Let us a now consider an instance of the relation Contrast3 between referents: (23) Wie haben Otto und Maria den Antrag aufgenommen? how have Otto and Maria the proposal taken up Zunächst hat Otto zum Glück dafür argumentiert. at first has Otto fortunately for it argued 2

3

Narration is a coordinating discourse relation, whose Nuclei, the statements which constitute the Narration, present a succession relationship between events (cf. e.g. Mann & Thompson 1988). Contrast is a coordinating discourse relation, which has not more than two Nuclei, namely, the statements which constitute the Contrast relation. The Nuclei present eventualities as being similar in many ways but contrasting in ways the speaker wants to point out.

Some contextual effects of aboutness topics in German

a. b.

339

Danach hat Maria leider Zweifel ausgedrückt. after this has Maria unfortunately doubts expressed #Danach hat leider Maria Zweifel ausgedrückt.

The text is intended to express a contrast between Otto and Maria. In (23), the term Otto is established as a sentence topic. (23a,b) demonstrate that the contrasting element Maria has to be topical too. Note that the relation Contrast does not demand the designation of topical phrases: (24) Der Antrag wurde von unseren Vorsitzenden leidenschaftlich diskutiert. the proposal was by our chairpersons vehemently discussed Zunächst hat zu meiner Überraschung Otto dafür argumentiert. at first has to my surprise Otto for it argued Danach hat erwartungsgemäß Maria Zweifel ausgedrückt. after this has as expected Maria doubts expressed

According to my intuition, the contrast relation expressed in (24) is not the one between Otto and Maria but the one between the events expressed by the second and the third clause, respectively4. Let us now look at the discourse relation Result5. Kehler (2002: 20) gives an example like the following: (25) Koch1 ist ein Politiker; man muss diesem Mann1 misstrauen. Koch is a politician you must this man distrust

The rule that must be presupposed in order to establish the coherence of (25) is something like (26) (cf. Kehler 2002, 21): (26) Being a politician implies being not trustable.

In (27), a sentential adverbial is added in order to make clear what is topical in the second clause: (27) a. b.

4

5

Koch1 ist ein Politiker; man muss diesem Mann1 leider misstrauen. Koch is a politician you must this man unfortunately distrust *Koch1 ist ein Politiker; man muss leider diesem Mann1 misstrauen

See e.g. Kehler (2002: 16) for the differentiation between contrast between referents and contrast between eventualities. Result is a subordinating discourse relation whose Nucleus presents an eventuality that could cause the eventuality described in the Satellite. The presentation of the Nucleus is more central to the speaker's purposes than the presentation of the Satellite (cf. Mann & Thompson 1988).

340

Werner Frey

In (25), the term Koch is very likely to be a topic in the first sentence6. In order to interpret the two sentences in (25) as standing in the Result relation, one has to rely on the general rule given in (26). It seems that we want to treat all the coreferring phrases which instantiate such a general rule as sentence topics. This has been done in (27a) but not in (27b). Thus, if we want to interpret the two sentences as being related by Result, only (27a) sounds perfect. (27b) is not understood as exhibiting Result, rather this text appears to be incomplete, requiring enrichment by additional material in order to become part of a coherent text. Let me stress once more that it is not the anaphoric status of diesem Mann which triggers the preposing. An anaphoric expression per se does not have to be preposed in front of a sentence adverbial, cf.: (28) a.

b.

Maria konnte heute endlich mit Guido1 sprechen. Bislang war sie Maria could today finally to Guido talk so far has she leider diesem Mann1 nie begegnet. unfortunately this man never met Dort drüben läuft Emil1. Maria hat erstaunlicherweise diesen over there walks Emil Maria has amazingly this Angeber1 vor kurzem geheiratet. show-off recently married

Let us now consider an example of Violated Expectation7: (29) Koch1 ist ein Politiker, aber man kann diesem Mann1 vertrauen. Koch is a politician but you can that man trust

The rule which licences the interpretation of (29) as a instance of Violated Expectation is the same as the rule which is presupposed to make the example (25) coherent. If we now add a sentence adverbial we get the following judgements: (30) a.

6

7

Koch1 ist ein Politiker, aber man kann diesem Mann1 zum Glück Koch is a politician but you can that man fortunately vertrauen. trust

The subject of a predicative clause is a topic, cf.: (i) a. Laut Maria ist Hans wahrscheinlich eine Genie according to Maria is Hans probably a genius b. ??Laut Maria ist wahrscheinlich Hans eine Genie This is the term Kehler (2002: 21) uses. Violated Expectation can be characterized as a subordinating discourse relation whose Nucleus presents an eventuality that could cause the eventuality described by the negation of the proposition presented in the Satellite.

Some contextual effects of aboutness topics in German

b.

341

*Koch1 ist ein Politiker, aber man kann zum Glück diesem Mann1 vertrauen.

Again, we find that the co-referential phrases which instantiate the two parts of the presupposed general rule should be sentence topics. If we make the discourse relations connecting the sentences in (27) and (30) explicit by conjoining them with the relevant conjunctions, we find: (31) a.

b.

Koch1 ist ein Politiker, deshalb muss man leider diesem Mann1 Koch is a politician therefore must you unfortunately this man misstrauen. distrust Koch1 ist ein Politiker, trotzdem kann man zum Glück diesem Mann1 Koch is a politician even so can you fortunately this man vertrauen. trust

To perceive the texts in (31) as coherent, it is no longer necessary to rely on (26). Rather, the conjunctions reveal in which discourse relation the sentences in (31a) and (31b), respectively, stand. Thus, for the question whether the texts in (31) are judged as true, (26) might be relevant, but not for the establishment of text coherence. The examples in (31) show that, under such circumstances, it is not necessary to designate the co-referential terms as topics in the two sentences standing in the 'cause-effect' relation8. To sum up this section, different data were presented which show that sentence topics do have contextual effects. For example, the referent of a topic in a sentence to be elaborated cannot be picked up by a non-topic in an elaborating sentence. A contrasting topical phrase wants to be contrasted with a phrase which is also a topic; and a topic occurring in a Narration becomes part of the description of the event which subsumes the different events reported in the Narration. Furthermore, instantiations of the 'cause-effect' discourse relations which rely on a general rule demand that the phrases which instantiate this general rule have to be sentence topics.

8

Kehler (2002) subsumes Result and Violated Expectation and other discourse relations under the label 'cause-effect' relations.

342

Werner Frey

5 Two putative effects of sentence topics I would like to make a comment on a very concrete proposal about the role sentence topics play for the coherence of texts. Although this is an old proposal I think it still makes sense to consider it. This proposal overestimates the role of sentence topics. The reason is, I believe, that the different functions of preposing constructions were less clear at the time this proposal was made than nowadays. Reinhart (1980) states the following condition: (32) Two sentences are connected only if they are referentially linked or linked by a semantic connector. Two sentences are referentially linked only if there is an aboutness topic or a scenesetting expression9 in the second sentence which is referentially controlled by an element of the first sentence.

At least for German, this condition is too strong. The following two sentences constitute a coherent sequence: (33) Peter1 ist ein netter Kerl. Laut Maria mag wahrscheinlich Peter is a nice guy. According to Maria like probably fast j eder [diesen wackeren Schwaben]1Non-Topic almost everyone this good Swabian

There is no semantic connector between the two sentences and there is no scene-setting expression in the second sentence which would be semantically controlled by an element of the first sentence. There is an anaphoric element in the second sentence but given (4), it is not a topic. Thus, according to (32) the discourse should be incoherent, contrary to fact. Which reasoning lead Reinhart (1980) to propose (32)? The following data are important: (34) a. b.

??Dan is a nice fellow. To her parties Rosa always invites him. ??This book is very boring. As for Rosa, she couldn't finish it.

In addition, the following assumptions are crucial for Reinhart's claim in (32): (i) Every clause contains at most one sentence topic; (ii) Topicalization, Left Dislocation and the 'as-for'-construction mark a sentence topic. Now, Reinhart (1980) reasons as follows: Given (ii), the 9

Reinhart (1980: 173) defines a scene-setting expression as "that part of the sentence that specifies the spatial and temporal framework for the event reported in the sentence, or the particular world about the state of affairs in which the sentence asserts something.”

Some contextual effects of aboutness topics in German

343

topics of the second sentences of (34a,b) are to her parties and Rosa respectively. Given (i), there can be no other topics in these sentences. Furthermore, these sentences do not contain any scene-setting expressions. Thus, the examples in (34) are bad because the topics of the second sentences are not referentially controlled by elements of the first sentences. Both cited assumptions of Reinhart (1980) are problematic. First, Rizzi (1997) and many others (including Frey 2004) assume that a clause may contain more than one topic. Second, Prince (1998) and others (including Frey 2005) do not assume that Topicalization and Left Dislocation necessarily mark a sentence topic. However, even if we accept both assumptions, there are reasons other than those given by Reinhart why the texts in (34) are bad. In Prince (1998), it is shown that the preposed phrase in Topicalization is anaphoric, i.e., it has to be linked to a phrase in the preceding context. This requirement is not fulfilled in (34a), and this fact already explains why (34a) is bad. In Frey (2005), it is argued that Left Dislocation in English and 'Freies Thema' in German induce a point of discontinuity in the discourse structure, i.e., they have the effect that the current discourse unit is closed and that a new one is started. In all likelihood, the same holds for the English 'as for'-construction. In addition, the DP inside the preceding 'as-for'-phrase has to pick up a referent which was already introduced into the discourse. Neither requirement is fulfilled in (34b), therefore (34b) is bad. Thus, to account for examples like (34), the problematic condition (32) is not necessary. Rather, the deviance of (34) is due to the violation of certain properties of Topicalization and the 'as for' construction. Let us now come to another instance of a presumed extrasentential effect of topics, which turns out to be not real. In theories of anaphoric relationships, it is a standard assumption that the more reduced an anaphoric expression is, the more salient its antecedent has to be (e.g. Ariel 1990, Gundel et al. 1993). It follows, for example, that a pronoun relates to a phrase with highest salience, i.e., to a phrase which denotes the most salient referent. The semantically most reduced referring expression refers to the most prominent referent at that point of the discourse. The question, then, is what makes a referent salient. Many authors assume that topics refer to salient referents (e.g. Gundel et al. 1993, Lambrecht 1994, Erteschik-Shir 1997, Strube & Hahn 1999).10 Given 10

The different authors use different notions of 'topic', but usually the concept of 'aboutness' is part of these notions.

344

Werner Frey

that, according to (4), topics in German are unambiguously designated, it is easy to check whether in fact pronouns prefer topics as antecedents. The answer is negative (cf. Hoffman 1998 for a similar conclusion with respect to Turkish): (35) a.

b.

Gestern hat überraschenderweise [den Paul]1Non-Topic der the yesterday has unexpectedly theACC Paul Direktor2 getroffen. Er*1/2 hat sich sehr darüber gefreut. director met he was REFL very about it pleased Gestern hat [den Paul]1Topic überraschenderweise der Direktor2 getroffen. Er*1/2 hat sich sehr darüber gefreut.

On the basis of our topic criterion in (4), (35a,b) show that irrespective of whether or not the object is a topic, it is the subject to which the pronoun relates.11 This observation can be controlled with the help of a German left peripheral construction which is standardly assumed to designate a topic (cf. e.g. Lambrecht 1994, Jacobs 2001, Frey 2005) – the so-called Linksversetzung illustrated in (36) (den is a resumptive pronoun which enters into a chain-relation with the preposed phrase): (36) Den PaulTopic, den mag theACC Paul RP likes

jeder. everyone

In (37) it is checked whether the dislocated phrase of a Linksversetzung is the preferred antecedent of a following pronoun: (37) Gestern war Institutsversammlung. [Den Paul]1Topic, den hat RP has yesterday was the departmental meeting theACC Paul erfreulicherweise der Direktor2 gelobt. Er*1/2 war sehr zufrieden. fortunately the director praised he was highly satisfied

The intuition is again pretty clear. In (37), it is not the topical object but the non-topical subject which figures as the antecedent of the pronoun. The following examples also show that preceding subjects are more salient for anaphor resolution than are preceding topical objects:

11

Independently of the question of which role topics play for anaphor resolution, the data in (35) disprove Rambow's (1993) claim that the surface order of the main constituents from left to right in the middle field corresponds to their descending salience for pronoun resolution.

Some contextual effects of aboutness topics in German

(38) A:

B: A:

345

Wir planen eine Party für den 50. Geburtstag von Peter. Wen sollen wir einladen? We are planing to have a party for Peter's 50th birthday. Who should we invite? Was ist mit Max? What about Max? der Peter2 nicht. i. Laut Maria mag den Max1 leider according to Maria likes theACC Max unfortunately theNOM Peter not ii. Den Max1, den mag leider der Peter2 nicht. Max sicher nicht dabei haben. a. Daher will er2 den therefore wants he theACC Max certainly not to be there b. *Daher will Peter ihn1 sicher nicht dabei haben.

The examples (38a,b) show that the pronoun refers to the subject of the preceding clause. In view of the data in (35), (37) and (38), it seems that pronouns do not want to relate to a topic of the preceding clause but, rather, to the subject, be it topical or not. Thus, the reason for the view that the most salient referent for anaphor resolution is the one denoted by a topic might be because, contrary to our assumptions, subjects are taken by many authors to be the canonical sentence topics. Note, however, the following example pointed out by Manfred Krifka: (39) Peters großes Haus hat ein schönes Dach1. Es1 ist grün. Peter's big house has a pretty roof it is green

The pronoun es is interpreted as co-referring with the object of the preceding clause. This object is not topical12. Rather, it denotes the referent which is introduced into the scene by this very construction. (39) shows that in this construction, it is the referent of the object that is more salient for pronoun resolution than that of the subject. However, again, it is not the case that the resolution of the pronominal reference is to an entity denoted by a topic. Thus, all the above examples demonstrate that it is not the phrase which tells us what the sentence is about which figures prominently in the linking of sentences by means of pronominal relationships.

12

This is shown by the ungrammaticality of (i): (i) *weil Peters zukünftiges Haus ein schönes Dach selbstverständlich haben wird. since Peter's house-to-be a pretty roof as a matter of course have will

346

Werner Frey

6 Conclusion In the literature claims can be found to the effect that sentence topics play a significant role for establishing the coherence of a text. In this squib, a small piece of evidence for this claim was contributed. It was argued that, given an unambiguous criterion for whether or not a phrase is an aboutness topic in German, it can be shown that there are textual configurations which are only coherent if certain phrases are designated as topics. Furthermore, it was argued that aboutness topics have an influence on the construction of the implicit theme which summarizes the content of a sequence of sentences. On the other hand, one field was addressed in which the contextual role of sentence topics seems to have been occasionally overrated in the literature. It was shown that, in German, a phrase's status as an aboutness topic does not increase its saliency for pronoun resolution. Although at this point no theoretical systematisation of the data was given, I think that the observations presented in this squib demonstrate that if one considers textual phenomena, the notion of an aboutness topic cannot be dispensed with, and that it is worthwhile to study the role aboutness topics can play for the coherence of texts.

References Ariel, M. (1990): Accessing noun-phrase antecedents. Routledge: London. Asher, N. (2004): Discourse topic. Theoretical Linguistics 30.2/3, 163–201. Büring, D. (1997): The Meaning of Topic and Focus – The 59th Street Bridge Accent. London: Routledge. Erteschik-Shir, N. (1997): The Dynamics of Focus Structure. Cambridge: University Press. Frey, W. (2003): Syntactic conditions on adjunct classes. In: Lang, E. et al. (eds): Modifying Adjuncts. Berlin: de Gruyter, 163–209. Frey, W. (2004): A medial topic position for German. Linguistische Berichte 198, 153–190. Frey, W. (2005): Pragmatic properties of certain German and English left peripheral constructions. Linguistics 43.1, 89–129. Gundel, J. (1988): Universals of topic-comment structure. In: M. Hammond et al.: Studies in syntactic typology. Amsterdam: John Benjamins, 209–239. Gundel, J., N. Hedberg & R. Zacharski (1993): Cognitive status and the form of referring expressions in discourse. Language 69, 274–307.

Some contextual effects of aboutness topics in German

347

Hoffman, B. (1998): Word Order, Information Structure, and Centering in Turkish. In: Walker, M., A. Joshi & E. Prince (eds.): Centering Theory in Discourse. Oxford: Clarendon Press, 251–271. Jacobs, J. (1997): I-Topikalisierung. Linguistische Berichte 168, 91–133. Jacobs, J. (2001): The Dimensions of Topic-Comment. Linguistics 39/4, 641–681. Kehler, A. (2002): Coherence, Reference and the Theory of Grammar. CSLI Publications. Kehler, A. (2004): Discourse topic, sentence topic, and coherence. Theoretical Linguistics 30.2/3, 227–240. Kuno, S. (1972): Functional Sentence Perspective: A Case Study from Japanese and English. Linguistic Inquiry 3, 269–336. Lambrecht, K. (1994): Information Structure and Sentence Form. Cambridge: CUP. McNally, L. (1998): On Recent Formal Analyses of Topic. In: Jonathan Ginzburg et al. (eds.): The Tbilisi Symposium on Language, Logic, and Computation: Selected Papers, CSLI Publications, Stanford, CA, 147–160. Mann, W. & S. Thompson (1988): Rhetorical Structure Theory: Toward a Functional Theory of Text Organization. Text 8(3), 243–281. Molnár, V. (1991): Das TOPIK im Deutschen und Ungarischen. Stockholm: Almquist & Wiksell International. (= Lunder germanistische Forschungen 58). Müller, G. & W. Sternefeld (1993): Improper Movement & Unambiguous Binding. Linguistic Inquiry 24, 461–507. Oberlander, J. (2004): On the reduction of discourse topic. Theoretical Linguistics 30.2/3, 213–225. Potts, C. (2005): The Logic of Conventional Implicature. Oxford: Oxford University Press. Portner, P. (2005): Instructions for Interpretation as Separate Performatives. In: K. Schwabe & S. Winkler (eds): Information Structure and Grammar. John Benjamins. Portner, P. & K. Yabushita (1998): The Semantics and Pragmatics of Topic Phrases. Linguistics and Philosophy 21, 117–157. Rambow, O. (1993): Pragmatic Aspects of Scrambling and Topicalization in German: A Centering Approach. Ms. Reinhart, T. (1980): Conditions for Text Coherence. Poetics Today 1, 161–180. Reinhart, T. (1981): Pragmatics and Linguistics: An Analysis of Sentence Topics. Philosophica 27, 53–94. Reinhart, T. (1995): Interface Strategies. OTS Working Papers, Utrecht University. Rizzi, L. (1997): The Fine Structure of the Left Periphery. In: L. Haegeman (ed.): Elements of Grammar. Handbook in Generative Syntax. Dordrecht: Kluwer, 281–337.

348

Werner Frey

Roberts, C. (1996): Information Structure in Discourse: Towards an Integrated Theory of Pragmatics. In: J.-H. Yoon & A. Kathol (eds): Ohio State University Working Papers in Linguistics, Vol. 49: Papers in Semantics, 91– 136. Steube, A. (2000): Ein kognitionswissenschaftlich basiertes Modell für die Informationsstrukturierung. In: J. Bayer & Chr. Römer (eds.): Von der Philologie zur Grammatiktheorie: Peter Suchsland zum 65. Geburtstag. Tübingen: Niemeyer, 213–238. Steube, A. & A. Späth (2002): Semantik, Informationsstruktur und grammatische Modularität. Linguistische Arbeitsberichte 79, 235–254. Strube, M. & U. Hahn (1999): Functional Centering: Grounding Referential Coherence in Information Structure. Computational Linguistics 25.3, 309–344. Vallduví, E. & R. Zacharski (1994): Accenting Phenomena, Association with Focus, and the Recursivness of Focus-Ground. Technical Report RP–49. HCRC, University of Edinburgh. Vallduví, E. & E. Engdahl (1996): The Linguistic Realization of Information Packaging. Linguistics 34, 459–519.

Ingolf Max (Leipzig)

Interfaces, (Non-)Compositionality, Information Structure and Presupposition1 1 Introduction Thanks to Anita Steube I learned a lot about information structure and its relevance for the analysis of phenomena connected with negation and presupposition. This article gives me the opportunity for representing my basic methodological position and to compare it with Anita's approach. Information structure is linked with the whole complex of interfaces between prosody, syntax, semantics and pragmatics. If we consider a simple sentence like "The child slept" we have to ask why the first two of the following four question-answer-pairs are acceptable but the other two not: (1)

a. b. (1') a. b.

WHO slept? The CHILD slept. WHO slept? *The child SLEPT.

(2)

a. b. (2') a. b.

What DID the child? The child SLEPT. What DID the child? *The CHILD slept.

According to the information structure account, the explanation goes like this: Any question and any answer (in general: any sentence) has at least a focus-background-structure.2 In unmarked cases the focus-domain of a sentence should contain the main accent of this sentence – indicated here by placing the relevant syllable in capitals. Focus indicates the new information within a simple sentence. Background marks the given/old information. Taking this information structure into account, we 1

2

I would like to thank Ryan Young and Marko Malink for inspiring discussions and improving the English of this paper. A more complex model of information structure includes the pair topic-comment and sometimes also the pair theme-rheme. In this article I restrict myself on the indicated pair background-focus.

350

Ingolf Max

are forced to consider factors beyond pure syntax and semantics. We need some information about the prosody of the considered sentence and/or about the context beyond the sentence itself. Many authors call this methodological step a pragmatic one. For example, the question (1a) asks for a person. The "slept" is given with respect to possible answers. We say that this old information is in the background. With respect to a coherent continuation of the discourse any constituent of the background should remain in this position. For that reason we have a mismatch between (1a) and (2b). The accent on "slept" indicates that "SLEPT" is part of the new information which is in conflict with the question. The opposite is the case with the pair (1b) and (2a). Here "child" is already mentioned and cannot be part of the new information. The interesting question is now: How can we formally explain both this mismatch between and the congruency of questions and answers in such pairs? Does the information structure as a pragmatically (sentenceexternally) triggered phenomenon influence the semantics of simple sentences? And if so, how does it work? Is this process a compositional or a non-compositional one? What is an appropriate theoretical background to answer these questions? We will begin with some preliminaries regarding Steube's approach and the notions compositionality and presupposition (section 2). The notion compositionality is closely related with the notion extensionality in classical and many-valued logics and also with the notion intensionality in modal theories. In section 3 we explain very briefly these connections and introduce a variant of non-compositionality. It is then shown, with respect to a four-valued system, that we can formalize cases of non-compositionality within an extensional system. We therefore cannot assume that an extensional system is unable to formalize phenomena which seem to be non-compositional (section 4). We will next show how information-structured sentences can be explicated within a two-dimensional framework. This approach explains the mismatch between information-structured and the congruency of questions and their answers. Furthermore, it allows us to differentiate between potential background presuppositions induced by information structure and real background presuppositions with respect to a negation test using a negation operation which is sensitive to the background domain (section 5). Section 6 links section 5 with section 4 by showing that our new negation behaves non-compositionally in an interesting, formally precise way. Within the two-dimensional framework our negation is the exact syntactic counterpart of one of our 4-valued operators, but makes its internal structure transparent. The scope of this negation is the whole sentence but the realized focus depends on the background domain of

Interfaces, (Non-)Compositionality, Information Structure and Presupposition

351

the sentence argument. The equivalent operating negation, when applied to truth-functionally equivalent sentences, yields negated sentences which differ in their truth-conditions. In the last section a demonstration is given that potential background presuppositions are not necessarily potential lexical presuppositions. This is a paradigmatic case to show that we need a more complex model to deal with questions regarding (non-)compositionality and the internal structure of interfaces.

2 Some preliminaries 2.1 Steube's approach: the focus-background-distinction and compositionality Steube (in prep.) uses two types of focus markers to represent the focus-background-distinction of more or less complex sentences: a. b.

(3)

[X]F indicates that X is a focus domain which can be the whole sentence (full-focused sentence) or a non-empty subpart of that sentence. Because of possible movements not any constituent within the focus domain is a focus constituent. If a constituent with the subindex "i" is moved outside the F-Domain we get traces of it indicated by "ti". Most of the constituents are marked by +/íF-indices. Exceptions are attitudinal and verificational adverbs which mark the border between the background and the focus domain. Let us take a more complex example: Die Kinderi warenk an dem Tagj wirklich [nicht tj ti in der SCHUle tk. ]F íF +F íF +F íF íF +F +F

All these indices are pragmatically determined by the context, i.e. by factors which are external relative to the sentence under consideration. For the moved constituents it is sufficient to mark the traces in the focus domain (cf. 3). "Conceptual structure transmits them into the grammatical modules when the conceptual units are lexicalised and the lexicon has defined the combinatorial potential of the lexical units. In the modules of grammar the pragmatic categories are grammatically realized: -

by word order in syntax by the meaning of background and focus constituents in semantics

352 -

Ingolf Max

by the accent patterns in prosody" (Steube in prep.: 2, the 2004version).

Steube assumes that the +/íF-indices allow semantic structure to be mapped on syntactic surface structure (or vice versa) in a 1:1 fashion. She uses an extended notion of Semantic Form (SF) which represents both the truth conditions of a sentence and its conditions of use. I will later show why I prefer the use of ordered pairs with well-defined internal dependencies to represent such a kind of SF. The one-to-one correspondence between syntactic surface structure and semantic structure ensures that the computation of the SF can differ in accordance with the F-markers but also that it is compositional with respect to a lambda-categorical derivation of the SF. Let us take one of Steube's examples to demonstrate the influence of the information structure on the SF: (4)

a. b. c.

(5)

a. b. c.

Warum bist du so nervös? Why are you so nervous? PEteri riefj [nicht ti an-tj.]F +F +F +F Peter has not called. SF: s Neg [!x [[[PERSON, x] & [MALE, x]] & [NAME, x, Peter]] & [s INST [CALL, x]]] Was ist mit Peter? What is about Peter? Peteri riefj [nicht ti AN-tj.]F +F íF +F Peter did not call. SF: !x [[[[PERSON, x] & [MALE, x]] & [NAME, x, Peter]] &  s Neg [s INST [CALL, x]]]

The difference in meaning between (4b) and (5b) "expresses the different scopes of negation and says that the else syntactically and prosodically identical sentences are part of different contexts" (Steube in prep.). By relying on the basis of information-structurally indexed surface structure and respecting the scope of the semantic operators we have a general strategy to build up semantic composition. 2.2 Compositionality Many, many formulations of the so-called "principle of compositionality" (C) exist in the literature. Here is one relatively weak characterization: "The meaning of a complex expression is determined by its structure

Interfaces, (Non-)Compositionality, Information Structure and Presupposition

353

and the meaning of its constituents" (Szabó 2001: section 1.1.). This formulation immediately confronts us with the syntax-semantics-interface: The syntax alone seems to be responsible for such notions as "structure", "complex expression" and "constituents". However the semantics has to tell us how we should explain the "meaning" of syntactical constructions. Compositionality can be variously understood as an inherent part of a theory, as a property of a theory, as a hypothesis about natural and/or formal languages, as a guiding principle in formulating new theories, as a criterion to evaluate existing theories etc. It is a concept that for a long time has been used by many authors without them making precise what they mean by it. Nowadays there is a discussion on a meta-level. Without discussing specific theories researchers try to find proofs for very abstract properties of compositionality (cp. Hodges 2001). Let us mention only one subtle attempt to trivialize compositionality due to Zadrozny (1994): Let S be a set of strings generated from an arbitrary alphabet via concatenation. Let m be a meaning function which assigns the members of an arbitrary set of meanings M to the members of S. Then we can construct a new meaning function P such that for all s, t  S P(s, t) P(s)P(t) and P(s)(s) m(s). This shows that it is possible to convert an arbitrary meaning function into a compositional one as long as we replace the old meanings with new ones from which they are – in principle – uniformly recoverable. This result depends on no constraint on how complex a calculation might be to determine the value P (s)(s) . The discussion on Zadrozny's result – e.g. by Kazmi & Pelletier (1998), Westerståhl (1998) and Dever (1999) – makes it clear that it is no trivial task to devise an adequate meaning assignment for S. Synonymy according to P and synonymy according to m do not coincide. Therefore, it is questionable whether P is as adequate as m, and whether the values of P are values in the same sense as the values of m. In this article we will take the opposite approach: we will explore a specific definition of non-compositionality in a particular system to see what we can learn about compositionality in general from it. Information structure is a very interesting source to look for non-compositional but nevertheless highly systematic accounts of meaning. In Steube's information structure approach we have already been confronted with the influence of pragmatic/external factors on the meaning of sentences. From an empirical point of view it seems therefore to be reasonable to look for non-compositional but theoretically fruitful and empirically adequate explications of SF. Below we will specify a general formulation of non-compositionality within several logical systems to show conditions under which we get interesting cases of non-compositionality.

354

Ingolf Max

2.3 Presupposition "The theory of presupposition is concerned with a component of meaning that appears to be distinguishable from ordinary truth-conditional entailments" (von Fintel 2003). In this vague formulation we are immediately confronted with the difficult subject of an adequate form of representing presuppositions. What does it mean to be "distinguishable from ordinary truth-conditional entailments?" Should we compute presuppositions of simple/complex sentences independent of the computation of the truth-conditions or not (presupposition projection)? How can presuppositions influence the truth-condition and vice versa? The situation is complex not only with respect to the projection of presuppositions and the interaction of presuppositions with the truthconditions but also with respect to the sources of these potential presuppositions. I call presuppositions with respect to their sources "potential" because they have to pass some invariance test successfully to become "real" presuppositions. Here is a disputable list of different potential presuppositions: x potential presuppositions triggered by a lexical item o definites and other noun phrases o factive verbs (be aware of, find, regret) o additive particles (too, again) o change-of-phase predicates (quit, stop, start, begin) o phase particles (already, still) x potential presuppositions as what the speaker(s) take(s) for granted connected with grammatical constructions o clefts o that-constructions x potential presuppositions induced by information structure (background constituents) of simple and complex sentences x potential presuppositions related with sentence connectors o connectors like in spite of, after, before, because etc.) The problem of presuppositions projection consists in computing the presuppositions of complex expressions given the presuppositions of its constituents. Presupposition projection reproduces the problem of compositionality known from computing truth-values of simple/and complex sentences. Traditional discussions of presupposition phenomena connected with lexical triggers are settled in the syntax-semantics-interface. Early approaches on context-presupposition (e.g. what a speaker takes for granted) and contemporary models of information structure

Interfaces, (Non-)Compositionality, Information Structure and Presupposition

355

extend the interface to a syntax-semantics-pragmatics-interface. I intend to show that our multi-dimensional approach is able to deal with this complex type of interface, yields the expressive power to differentiate between several types of potential presuppositions and allows us to formalize the necessary negation test for real presuppositions.

3 (Non-)Compositionality, extensionality and intensionality Usually we take bivalence and extensionality as two fundamental principles of classical logic. Furthermore we are in general convinced that classical logic is throughout compositional. If we leave the frame of classical logic by omitting at least one of the two principles we have to not only explain negatively what it means for a logic to be non-bivalent (e.g. we think that a one-valued logic is of no interest for us) or non-extensional; but we have also to characterize positively what it means for a logic to be many-valued or intensional. These two principles are strongly related with compositionality and so it is their reformulation. We can rephrase compositionality syntactically as well as semantically using the notion of substitution of one expression by another expression in a context C which leaves the context unchanged: If A = B, then CƔA = CƔB or

A CxA

m(A) B or m(C x A) CxB

m(B) .3 m(C x B)

This is realized that by the following versions of extensionality in classical logic: (E1) A (A { B) Š (I1A { I1B) or (A { B) A (I1A { I1B)

[syntactic formulation]

(E2) B (A { B) Š (I A { I B) or (A { B) B (I A { I B)

[semantic formulation]

1

1

1

1

The style of reading these formulas is this: If A and B are materially equivalent – have the same truth value – then putting the unary functor I1 in front of both A and B keeps this equivalence. The whole construction is a theorem ( A ) or a tautology ( B ) and you can use any of the four classical unary functors, e.g. classical negation ™: A (A { B) Š (™A { ™B). Here the context condition "CƔ" is realized by a unary functor inter3

The last version with a negated second line is used e.g. by Kazmi & Pelletier (1998) to express non-compositionality. "m(X)" represents "the meaning of X". The symbol " x " indicates a syntactic concatenation. We will reinterpret this form in several ways.

356

Ingolf Max

preted as a function. The identity sign "=" is interpreted as the material conditional " { ". The extension to n-placed functors/functions In looks then as follows: (E3) A (A1 { B1 ) š ! š (A n { Bn ) Š (In A1 ! A n { In B1 ! Bn ) (E4) B (A1 { B1 ) š ! š (A n { Bn ) Š (In A1 ! A n { In B1 ! Bn )

[syntactic formulation] [semantic formulation]

Now youn can take any of the functors of the form "In" with 1 d n d 22 . If we consider only one position in an arbitrary context "C" we can take the representation form used by Carnap: (E5) A (A { B) Š (! A! { ! B!)

Here the context "C" is indicated by the dots. We call contexts which fulfill the above listed E-conditions "extensional". Extensionality is a special case of compositionality. From a classical point of view non-extensionality is a violation of compositionality. If we look at modal words like can interpreted as modal operators they do not fulfill the requirement of extensionality: G (A { B) Š (—A {—B). We need a new paradigm to recover compositionality. Carnap's idea was to use not materially equivalent but logically equivalent expressions as input: (I1)

AA{B A I1A { I1B

(syntactically)

or

BA{B B I1A { I1B

(semantically).

With respect to this criterion of intensionality operators which are not extensional but intensional in the indicated sense form intensional contexts. We are able to compute the meaning of expressions like —A if we know the meaning of "A" and add a new basic notion to our semantics, such as "possible worlds" or "situations". The generalization of this situation for n-placed operators is: A A1 { B1 #

A A n { Bn (I2) A In A n ! Bn { In B1 ! Bn

In this new paradigm we can reformulate the intensional operator In as a compositional one.

Interfaces, (Non-)Compositionality, Information Structure and Presupposition

357

We have already shown that extensionality is strongly connected with compositionality. The vast majority of researchers would expect that any extensional logic is compositional. That would mean that any n-valued logic with n ! 2 is compositional. We will challenge this assumption by constructing a 4-valued logic. Let us take non-compositionality in the following forms: (NC)

A B CxA z CxB

or

BAlB . H I1A l I1B

If we want to represent non-compositionality in a 4-valued logic, the question is how we should represent "=" / "l" and " C x " /I1. Here we have several choices: How should we represent cases of non-compositionality in that 4-valued system? There are relevant differences between classical logic and a new 4-valued logic L4: Classical logic has only two truth-values: 1 and 0. L4 has the values 1, 2, 3 and 4. The definition of classical validity uses only the value 1. The definition of validity of L4 uses the notion of designated values which can be one, two or three of the four possible values. Classical logic has only one symbol for our familiar material equivalence. L4 offers several choices for "l". In the next paragraphs I show how we can formulate non-compositionality in a 4-valued logic. Then I give a logically equivalent two-dimensional reformulation of that logic. Finally I show that this reformulation is also a useful explication of a domain of information structure called background and how we can understand background constituents as a special kind of presuppositions. The general aim of my argument is to show that non-compositionality is not our enemy. There is an interesting analogy between extensionality and non-extensionality/intensionality. If we have a precise notion of compositionality with respect to a theory of logic (compositionality in the air means nothing) we can precisely formulate cases of non-compositionality if there is enough room for this in our formal theory. We can then formulate the conditions under which we can recover compositionality. The analogy to an intensional system is that we can say under which conditions (e.g. taking only one possible world) an intensional context yields an extensional one.

358

Ingolf Max

4 Non-compositionality and extensionality: A puzzling case in a 4-valued logic Let L4 be our 4-valued logic and K {1, 2,3, 4} be the set of its values. Let D {1, 2} be the set of designated values of L4 and let f be an assignment of values to the atomic formulas of L4 and let V be a – recursively defined – valuation. A formula A is true if its value is an element of the set of designated values: V (A, f ) is true if and only if V (A, f )  D. A formula A is valid if and only if for all assignments f A takes a designated value: B L4 A iff f V (A, f )  D. Now we consider the unary functors " † ", " … ", " : " and the binary functor " l ". The value tables are: A 1 2 3 4

†A

…A

:A

AlB

1 4 3 4

1 3 4 4

3 4 1 4

1 2 3 4

1 1 2 3 4

2 2 1 4 3

3 3 4 1 2

4 4 3 2 1

The principle of non-compositionality can be reformulated as follows: B L4 A l B . H L4 I 1A l I 1B

"A" can be any formula of L4, e.g. " † p " with "p" being a propositional variable of L4. "B" is " …p " and the context-creating operator" I1 " is our L4-operator " : ". Then we are confronted with the following situation: B L4 † p l …p , i.e. H L4 : † p l : … B

† 1 4 3 4

p 1 2 3 4

l

…

1 2 2 1

1 3 4 4

p 1 2 3 4

:

†

3 4 1 4

1 4 3 4

p 1 2 3 4

l 1 4 4 1

: 3 1 4 4

… 1 3 4 4

p 1 2 3 4

Why can this happen? "† p" and "… p" can be logically equivalent because there are situations with:

Interfaces, (Non-)Compositionality, Information Structure and Presupposition

359

V († p, f ) z V (…p, f ) but V († p l …p, f )  D.

That means that L4 is extensional with respect to "V" but not completely compositional. In any functionally complete 4-valued logic we can define " † ", " … ", " : ", " l " and formulate the indicated situation. To get a fully compositional logic back we have to consider a functionally incomplete 4-valued logic or put some other restrictions on it or the formulation of compositionality: x x x

We are not allowed to use an equivalence relation like "l" to formulate (non-)compositionality. We are allowed to use "l" together with the restriction that D {1} . We accept only an equivalence relation with V (A œ B, f )  D iff V (A, f ) V (B, f ) in order to express the identity between "A" and "B".

Any of these steps is a step towards classical logic! Furthermore, it is possible to formulate cases of non-compositionality in a precise way within an extensional framework. Many researchers seem to be convinced that loosing compositionality means loosing the chance of a theoretical formulation of the relevant data. But who would defend the position that L4 is not a logic, or not a formal theory? The above postulated value-tables do not tell us anything about the micro-structure of the situation. Therefore, it is not easy to see how this apparently strange situation can be useful for explaining empirical data. Let us begin with a reformulation of the value-table using pairs of "classical" values 1 and 0. We propose the following biunique translations: 1 œ 11

2 œ 10

3 œ 01

4 œ 00 .

The notation "k1" with k,l  {1,0} is an abbreviation of ordered pairs of the form  k,l ! . Our value tables are: A 11 10 01 00

†A

…A

:A

AlB

11 00 01 00

11 01 00 00

01 00 11 00

11 10 01 00

11 11 10 01 00

10 10 11 00 01

01 01 00 11 10

00 00 01 10 11

360

Ingolf Max

Let us take the case of " †A " with respect to "A": We observe that the left column of " †A " is exactly the conjunction of the two columns of "A"; whereas the second columns of " †A " and "A" are the same. To make these relations more transparent we use pairs of expressions in our language. Let "A" be the pair "  A1 , A 2 ! ". When we can compute our functors " † ", " … ", " : ", "l" as follows: A  A1 ,A 2 ! 11 10 01 00

†A  A1 š A 2 ,A 2 ! 11 00 01 00

AlB  A1 { B1 ,A 2 { B2 ! 11 10 01 00

…A  A1 š A 2 ,A1 ! 11 01 00 00

11

10

01

00

11 10 01 00

10 11 00 01

01 00 11 10

00 01 10 11

:A  ™A1 š A 2 ,A 2 ! 01 00 11 00

Please note that the functors within the ordered pairs (" ™ ", "š", "{") are the classical connectives. From a compositional point of view the step toward pairs of expression seems to be dangerous. Within "  A1 , A 2 ! " there is no functional connection between "A1" and "A2". The same holds for all the other pairs. Up to now we need the ordered pairs only to show that in contrast to the atomic values of L4 these pairs tell us something about the microstructure of these values in a semantic or syntactic way. But there is more. Let's come to some puzzling questions regarding the information structure of simple sentences.

5 The two dimensional representation of information-structured sentences 5.1 General two-dimensional form of a simple sentence S ASS (= FOC š PBP) º Oxn ! Oxn ª« »¼ : PBP ¬

Interfaces, (Non-)Compositionality, Information Structure and Presupposition

361

Let be a placeholder that marks the position for negation, correction etc. (Cp. Steube's position after attitudinal/verificational adverbials). ASS represents the assertion/proper (ordinary) meaning of a sentence. This coincides with the matrix of the semantic form SF given in a language of first-order logic. FOC represents the focus – i.e. a sub-conjunction of ASS. PBP represents the potential background presupposition/meaning component induced by the information structure of the sentence which is characterized by the background B. PBP is only potential because calling PBP "real" means that PBP has to pass well-defined invariance test using negation and/or subjunction, possibility etc. (cp. Geurts 1999).

We assume a Russellian style of representing (definite) NPs by unary predicates. In cases where the VP of S is part of PBP the expression of the form Fmn 1 x1 ! x n is part of the PBP. Both ASS and PBP have a conjunctive form. "X is part of ASS/PBP" means that X is a conjunct of ASS/PBP. Let F11 ,!, Fm1 be the matrix predications expressed by (definite) NPs. Expressions of the form Fmn 1 represent the VP-predications. The superscript i of Fji indicates the number of the arguments of Fji . The subscript j (1 d j d m  1 ) acts as a counter which numbers the oci currences of Fj in ASS. Let A i be any of the conjuncts of ASS with A i # Fi1 x k if 1 d i d m and A m 1 # Fmn 1 x1 ! x n . Then we get the following more specific form: ª F1x š ! š Fm1 x n š Fmn 1x1 ! x n º Ox n ! Ox1 … « 1 1 » A k1 š ! š A kl ¬ ¼ n d m , 1 d k1 ,!,k l d m  1 , 1 d l  m  1 .

x m+1 tells us that ASS contains m+1 conjuncts. x k j with 1 d j d l and 1 d k j d m  1 indicate that each constituent (conjunct) of A can be a possible part of PBP (a possible B-constituent). The whole PBP represents the background of the sentence. x k l d m  1 tells us that A k1 can be Fmn 1 , i.e. the VP can be part of PBP. x Because l is not allowed to be m  1 the sentence cannot exclusively consist of B-constituents. x 1 d k j says that the sentence S has at least one B-constituent, i.e. a non-empty B-domain. In other words, S is a categorical sentence.

362

Ingolf Max

In the case of categorical sentences we get: ASS B BPP

but: PBP H ASS

(extensional reading)

That means that PBP can be a candidate for a logical presupposition. If we consider our example (S) we have to put the whole assertion in a conjunctive style into the first dimension ASS and the background domain as a sub-conjunction into the second dimension PBP: (6)

[LIna]F [ate a cookie]B.

The two-dimensional representation of this sentence is (6')

LINA(x) š COOKIE(y) š ATE(x, y) º y!x ª« »¼ COOKIE(y) š ATE(x, y) ¬

º is called a matrix. D1: An expression of the form ª« ASS ¬ PBP »¼ ª LINA(x) š COOKIE(y) š ATE(x, y) º The matrix of (6') is then « ». COOKIE(y) š ATE(x, y) ¬ ¼ D2 Let CASS be the set of atomic conjuncts – formulas of the form Fn x1 ! x n or ™Fn x1 ! x n .

D3 Let CPBP be the set of atomic conjuncts of BPB. In the case of categorical sentences we get: CPBP  CASS

(intensional reading)

CPBP z CASS

(Full backgrounded sentences are excluded.)

In the case of thetic sentences we get CPBP constant "T": (7)

What happened?

(7')

CHILD(x) š SLEPT(x) º [The child slept.]F ª« T ¬ ¼»

‡ represented by the truth-

We can introduce a B-operator which has a one-dimensional information-structured expression as its input and gives a two-dimensional expression as output. It yields that by copying the BPB into the second dimension:

Interfaces, (Non-)Compositionality, Information Structure and Presupposition

363

FOC š BPBº . B(FOC š BPB) Ÿ ª« BPB »¼ ¬

If we restrict our examples to very simple sentences with infinite verbs the matrix will be of this form: ª A š Bº or ª A š Bº «¬ A »¼ «¬ B »¼

in accordance whether the NP or the VP is a background constituent. Now we observe an interesting analogy to our puzzling case of noncompositionality within our four-valued logic L4: The functors † and † can be interpreted as specific versions of our B-operator. Starting with a classical conjunction they produce in accordance with the information structure a new expression which respects the focus-background distinction in a two-dimensional way. 5.2 Congruency of/mismatch between questions and answers 5.2.1 Simple/narrow focus (8a) Which girl ate a cookie? ª WHICH(x) š GIRL(x) š COOKIE(x) š ATE(x, y) º GIRL(x) š COOKIE(x) š ATE(x, y) ¬« ¼» (8b) [LIna]F [ate a cookie]B LINA(x) š GIRL(x) š COOKIE(x) š ATE(x, y) º (8b' ª« GIRL(x) š COOKIE(x) š ATE(x, y) ¬ ¼» Mismatch regarding (8a) (8c) [Lina ate]F [a COOkie]B LINA(x) š GIRL(x) š COOKIE(x) š ATE(x, y) º (8c') ª« LINA(x) š GIRL(x) š ATE(x, y) ¬ ¼»

PBP

ASS

CQuestion z CAnswer

(8d) [NO girl]F [ate a COOkie]B GIRL(x) š COOKIE(x) š ATE(x, y) º (8d') ™(x) ª« ¬GIRL(x) š COOKIE(x) š ATE(x, y) »¼

= PBP-congruent answer!

364

Ingolf Max

5.2.2 Multiple foci (9a) Which woman saw which man? WHICH(x) š WOMAN(x) š WHICH(y) š MAN(x) š SAW(x, y) º (9a') ª« »¼ WOMAN(x) š MAN(x) š SAW(x, y) ¬ (9b) [MAry]F [saw]B [JOHN]F MARY(x) š WOMAN(x) š SAW(y) š MAN(x) š SAW(x, y) º (9b') ª« »¼ WOMAN(x) š MAN(x) š SAW(x, y) ¬ = PBP-congruent answer!

5.3 Potential background presuppositions (BPBs) as real background presuppositions BP with respect to a negation test Presuppositions can be induced by several types of facts (cp. section 2.3): (i) syntactically: lexical items (proper names, verbs, phase particles etc.), lexical phrases (DPs, VPs etc.), grammatical constructions (cleft-sentences) (ii) pragmatically: background constituents (iii) (matrix clause)-that-(embedded clause)-construction (iv) complex sentences consisting of two simple sentences using a connector like and, or, if-then, regardless-of, before, after, because-of and many more. The result of such an induction process is in any case only a potential presupposition. Having a real presupposition depends on passing some tests which show the invariance of the potential presupposition. The negation test is a prominent test used to decide whether the potential presupposition is a real one. With respect to our two-dimensional schema the question arises how an appropriate negation should look like. In our two-dimensional system any operator will be characterized by a reduction rule which in a unique way allows to eliminate expressions prefixed by this operation. Starting with any arbitrary formula which can contain many two-dimensional sub-expressions we reach finally a formula which is nothing else than a pair of classical formulas. Let us define some useful notions:

Interfaces, (Non-)Compositionality, Information Structure and Presupposition

x

365

Necessitation  :

A D4: ª« º»  J ¬B ¼

df

(i) A B C, and (ii) B B C .

A, B and C are arbitrary formulas of classical first-order logic. A B B indicates that B classically follows from A / A semantically entails B. It is trivial that in our two-dimensional model any information-structured sentence necessitates its background = PBP: ª FOC š PBP º  PBP PBP ¬« ¼» because of (i) FOC š PBP B PBP , and (ii) PBP B PBP .

The interesting question is now which negation we can use to test whether our PBPs are real presuppositions. We would like to have a uniform negation. And we would like to know in which sense our negation is a compositional or a non-compositional one. Here is my proposal:4 x

PBP-preserving negation ™FOC :

R ™FOC

A ™A š B º ™FOC ª« º» Ÿ ª« . B ¬ ¼ ¬ B »¼

™FOC is a complex operation which demands the following: -

Copy the IS-induced part of the meaning of a sentence S – the background B (= PBP) – to the first dimension. Negate the proper meaning of S (= ASS = PBP š FOC ) completely (= classical negation of it): ™ASS ( ™(BPB š FOC) ). Create a new conjunction of the products of the first two steps in the first dimension: ™ASS š PBP . Keep the second dimension (the background of S = PBP) for further use in the ongoing discourse – e.g. for rejecting PBP in the future. FOC š PBP º ™FOC ª« PBP ¬ ¼»

4

ª ™(FOC š PBP) š PBP º PBP ¬« ¼»

ª ™FOC š PBP º PBP ¬« ¼»

For more details about this negation and reasons for its adoption see Max (2004a).

366

Ingolf Max

The PBP-preserving negation " ™FOC " of a sentence S necessitates its PBP -

FOC š PBP º ™FOC ª« »¼  PBP, PBP ¬

because of (i) ™FOC š PBP B PBP , and (ii) PBP B PBP . In accordance with the explication by van Fraassen (1968) we can define a presupposition relation as necessitation relation between the sentences/the negations of this sentence and another expression which is then a real presupposition: x R6

The presupposition relation 6 : ªD º «¬E »¼ 6 J

df

D D (i) «ª »º  J , and (ii) ™FOC «ª »º  J . Because of E E ¬ ¼ ¬ ¼

ª FOC š PBP º 6 PBP, PBP ¬« ¼»

we get the result that PBPs are real background presuppositions BPs in this framework. Both the unnegated representation and the ™FOC negated representation of an information-structured sentence necessitates its PBP! 5.4 Some examples (8b) [LIna]F [ate a cookie]B ª LINA(x) š COOKIE(x) š ATE(x, y) º 6 (8P) ¬« COOKIE(x) š ATE(x, y) ¼» 6 (x)(y)(COOKIE(x) š ATE(X,Y)) (9b) [MAry]F [saw]B [JOHN]F ª MARY(x) š WOMAN(x) š SAW(y) š MAN(x) š SAW(x, y) º 6 (9P) ¬« WOMAN(x) š MAN(x) š SAW(x, y) ¼» 6 (x)(y)(WOMAN(x) š MAN(x) š SAW(x, y)) (9P') 6 (x)(y)SAW(x, y) (9b-) not 6 (x)(y)(WOMAN(x) š JOHN(y) š MAN(y) š SAW(x, y)) not 6 (x)(y)(MARY(x) š WOMAN(x) š MAN(y) š SAW(x, y))

Interfaces, (Non-)Compositionality, Information Structure and Presupposition

367

6 The non-compositionality of the PBP-preserving negation If we compare our functor " : " (  ™A1 š A 2 , A 2 ! ) of the 4-valued logic L4 with the PBP-preserving negation " ™FOC " we can immediately observe an interesting analogy. The structure produced by " : " = ª ™A1 š A 2 º looks very similar to that produced by "™ " = ª ™A š Bº . FOC ¬« B ¼» ¬« A 2 ¼» Let's go back to our example (1)

The child slept.

We already know that this sentence has several readings which depend on its information structure: (1b) [The CHILD]F[ slept]B : (2b) [The child]B[ SLEPT]F : (3)

[The child slept]F

ªCHILD(x) š SLEPT(x) º «¬ »¼ SLEPT(x) ªCHILD(x) š SLEPT(x) º «¬ »¼ CHILD(x) ªCHILD(x) š SLEPT(x) º T ¬« ¼»

(1b), (2b) and (3) coincide in their assertion ASS. Their first dimensions are identical. But they differ in their potential background presuppositions PBP in the second dimension. We can say that these versions have the same truth-functional meaning or that they are semantically identical. Nevertheless their information structure is different. From a pragmatic point of view the sentence "The child slept" has three different interpretations. We can express this by using two different notions of validity and a kind of classical equivalence relation: Rl

ª A1 º l ª A 2 º Ÿ ª A1 { A 2 º . «¬ B1 »¼ «¬ B2 »¼ «¬ B1 { B2 »¼

This equivalence relation coincides with the equivalence relation of our system L4. A D5: D ª« º» ¬B¼ A D6: C ª« º» ¬B¼

df

df


B A and B B.


B A.

368

Ingolf Max

The notion of validity " C " defined in D6 is characterized only by using the classical notion of validity " B ". This holds for every notion of validity introduced so far. Furthermore, " C " corresponds to the notion of validity " B L4 " with respect to the set of designated values D {1, 2}. 5 Now it is easy to show that the IS-variants of "The child slept" coincide semantically but not pragmatically: C ª« ¬

CHILD(x) š SLEPT(x) º ª CHILD(x) š SLEPT(x) º »¼ l «¬ »¼ SLEPT(x) CHILD(x)

^

`

B (CHILD(x) š SLEPT(x) { CHILD(x) š SLEPT(x)) SLEPT(x) { CHILD(x)

C ª« ¬

CHILD(x) š SLEPT(x) º ª CHILD(x) š SLEPT(x) º »¼ l «¬ »¼ SLEPT(x) T

^

`

B (CHILD(x) š SLEPT(x) { CHILD(x) š SLEPT(x)) . But SLEPT(x) { T

CHILD(x) š SLEPT(x) º ª CHILD(x) š SLEPT(x) º not D ª« »¼ l «¬ »¼ SLEPT(x) CHILD(x) ¬ because of

^

`

B (CHILD(x) š SLEPT(x) { CHILD(x) š SLEPT(x)) etc. H SLEPT(x) { CHILD(x)

After putting our negation PBP-preserving negation "™FOC" in front of these IS-variants of "The child slept" the situation changes. In spite of the fact that it keeps the background constituents this negation is sensitive to the background domain. It uses it to compute the new assertion, i.e. the new truth conditions of the negated sentence: CHILD(x) š SLEPT(x) º ª CHILD(x) š SLEPT(x) º because I ™FOC ª« »¼ l ™FOC «¬ »¼ SLEPT(x) T ¬ of

^

`

H (™CHILD(x) š SLEPT(x) { CHILD(x) š ™SLEPT(x)) . H SLEPT(x) { CHILD(x)

Otherwise our negation behaves classically if we iterate it and use an information-structured argument:

5

It can be shown that a full 4-valued system can be equivalently represented using a twodimensional syntactic counterpart. Cp. Max (1996).

Interfaces, (Non-)Compositionality, Information Structure and Presupposition

369

ASSº ASSº A A D ™FOC™FOC ª« l ª« , but not in general D ™FOC™FOC ª« º» l ª« º» . ¬ PBP »¼ ¬ PBP »¼ ¬B¼ ¬B¼

Let us summarize the non-compositionality of our PBP-preserving negation: CHILD(x) š SLEPT(x) º ª CHILD(x) š SLEPT(x) º C ª« »¼ l «¬ »¼ SLEPT(x) CHILD(x) ¬ . CHILD(x) š SLEPT(x) º ª CHILD(x) š SLEPT(x) º I ™FOC ª« l ™ FOC »¼ «¬ »¼ SLEPT(x) CHILD(x) ¬

This negation behaves non-compositionally in the defined sense when the PBP-components of one and the same assertion differ in at least one conjunct! But nevertheless we have an exact characterization of that negation. We can compute the meaning of a complex expression using the meaning of the sub-expressions and operations which are characterized by a new type of reduction rules. The inspiring part consists in the interaction of the two dimensions, i.e. the interaction of a semantic part of the meaning and a pragmatically induced part of meaning which can influence the semantic part in computing the new two-dimensional expression.

7 Potential background presuppositions are not necessarily potential lexical presuppositions Let us go back to our starting example again: (1)

The child slept.

(definite reading)

Many authors think that the expression "the child" in subject position triggers an existence presupposition which can be paraphrased as (1P) There is exactly one child.

For Frege (1892) such an existence presupposition of the proper name "the child" is a tacit assumption of our discourse and there is no need to formalize it. The tacit assumption is already fixed in the frame of classical logic: Every proper name has to have exactly one object to which it refers. Russell (1905) tells us that the expression "the child" is a definite description because it contains a predication of "being a child".

370

Ingolf Max

This allows to explicate the presupposition using a battery of formal means: existential quantifier, classical conjunction, material implication/equivalence, identity and predication: (1')

(x)(CHILD(x) š (y)(CHILD(y) Š y

x) š SLEPT(x)).

We will call this approach the linear-conjunctive approach. There is only one dimension and the classical conjunction is the tool used to connect the assertion and the presupposition of our sentence (1). But this means that presuppositions are on the same level as assertions and they can influence the truth value of a sentence independently of its information structure. In the spirit of our two-dimensional approach it is clear that assertion and presupposition should be represented on distinct levels, i.e. in different dimensions. Our matrix should look the following way: ªCHILD(x) š (y)(CHILD(y) Š y x) š SLEPT(x) º . «¬ »¼ CHILD(x) š (y)(CHILD(y) Š y x)

In order to characterize the whole form we need a new characterization of the existential quantifier which respects both dimensions regarding the assertions ASS: R š

A (i)(A š B) º . š (i) ª« º» Ÿ ª« ¬B ¼ ¬ (i)B »¼

A š (i) ª« º»  (i)B . ¬B ¼

Here "i" is a meta-variable for any individual variable which can occur both in A and in B. Let us now have a look at the interaction between the two-dimensional existential quantifier and the PBP-preserving negation: FOC š PBP º š (i)™FOC ª« »¼ PBP ¬

™FOC š PBP º š (i) ª« »¼ PBP ¬

FOC š PBP º š (i)™FOC ª« »¼ PBP ¬

ª š (i)((™FOC š PBP) š PBP) º « » š (i)PBP ¬ ¼

FOC š PBP º š (i)™FOC ª« »¼ PBP ¬

ª š (i)(™FOC š PBP) º « » š (i)PBP ¬ ¼

FOC š PBP º ªš (i)™FOC š š (i)PBP) º š (i)™FOC ª« C » PBP š (i)PBP ¬ ¼» «¬ ¼

Interfaces, (Non-)Compositionality, Information Structure and Presupposition

371

FOC š PBP º š š (i)™FOC ª« »¼   (i)PBP and finally PBP ¬ FOC š PBP º š š (i)™FOC ª« »¼ 6  (i)PBP . PBP ¬

The interaction of both operators characterized by appropriate reduction rules let the potential existential presupposition with respect to "i" go through and makes it a real existential presupposition. But we need to be careful; this applies only to the existential presuppositions induced by the information structure of the sentence. I am convinced that some of the trouble in discussing existential presuppositions goes back to neglecting the differences between existential presuppositions triggered internally by lexical items like DPs and background presuppositions externally induced by the information structure. If "the child" is interpreted as a DP and additionally in the background then there seems to be no need to differentiate these two kinds of presuppositions: (2a) What DID the child? (2b) [The child]B [SLEPT]F. CHILD(x) š (y)(CHILD(y) Š y x) š SLEPT(x) º š (x) ª« »¼ 6 . CHILD(x) š (y)(CHILD(y) Š y x) ¬ 6 (!)(x)CHILD(x)

It can be computed that this sentence presupposes the existence of (exactly) one object in the given discourse that is a child. Maybe this paradigm explains why many authors put potential background presuppositions (PBPs) and potential existential presuppositions (PEPs) into one and the same class of presuppositions. Maybe these presuppositions could also be called "topic-induced presuppositions". But what should we do if our sentence gets another information structure: (1a) WHO slept?

In this case it is difficult to see a potential existential presupposition. Possible correct answers can be: NObody/SOMEbody/THE x/ALL x slept. (1b) [The CHILD]F [slept]B.

372

Ingolf Max

Like in our analysis above we have a potential background presupposition: SLEPT(x). If somebody assumes that "the child" triggers lexically the existence of exactly one object independent of the information structure we have to differentiate between potential background presupposition (PBP) and potential existential presupposition (PEP). CHILD(x) š (y)(CHILD(y) Š y x) š SLEPT(x) º š (x) ª« »¼ 6 CHILD(x) š (y)(CHILD(y) Š y x) ¬ 6 (!)(x)CHILD(x) CHILD(x) š (y)(CHILD(y) Š y š (x) ª« SLEPT(x) ¬ 6 (!)(x)SLEPT(x)

x) š SLEPT(x) º »¼ 6

These two representations agree regarding their assertion ASS. But they have different presuppositions. Again, the negation "™FOC" does not preserve the equality of the assertion. If we put PBP und PEP in one and the same dimension and allow both of them to have some influence on the first dimension (the truth conditions) it is not surprising that our negation behaves again noncompositionally. In order to avoid mixing up these two kinds of presuppositions we can accept both kinds of presuppositions and put them in two different dimensions: Assertion = ASS ª º « Potential presupposition 1 = PP1» . « Potential presupposition 2 = PP2» ¬ ¼

If we follow this line and accept other kinds of presuppositions – e.g. associated with connectors – we can generalize this frame to Assertion = ASS ª º « Potential presupposition 1 = PP1» « ». # « Potential presupposition n = PPn » ¬ ¼

The corresponding notions are:

R

š n

ª A1 º ª (i)(A1 š ! š A n ) º «A2 » « » (i)A 2 n (i) « » Ÿ « » # # «A » « » (i)A  n ¬ n¼ ¬ ¼ š

Interfaces, (Non-)Compositionality, Information Structure and Presupposition

R™

n FOC

™

n FOC

ª A1 º ª ™A1 š A 2 š ! š A n ) º «A2 » « » A2 « # » Ÿ « » # «A » « » An ¬ n¼ ¬ ¼

ª A1 º

D7: « # »  n C «A » ¬ n¼

373

df

A1 B C,!,A n B C.

(strong necessitation)

Our example can get the following three-dimensional SF: (1b) [The CHILD]F [slept]B ªCHILD(x) š (y)(CHILD(y) Š y x) š SLEPT(x) º » 6 ??? (1b*) 3š (x) « SLEPT(x) « » CHILD(x) š (y)(CHILD(y) Š y x) ¬ ¼

In the sense of the strong necessitation above defined this expression has no non-trivial presupposition. Now we have a case in which potential presuppositions of two different kinds block each other so that they do not become real presuppositions. The negation test is not successful for them. If we like to have them as presuppositions in a restricted sense we have to be careful and restrict the notion of necessitation that we use too: ªCHILD(x) š (y)(CHILD(y) Š y x) š SLEPT(x) º » 3n (x) « SLEPT(x) « » CHILD(x) š  (y)(CHILD(y) Š y x) ¬ ¼ 1,2 (x)SLEPT(x) / 1,3 (!)(x)CHILD(x).

A final example: Iterated information structure and conditionals We take as an example a conditional which contains two who-questions and then a possible answer. (10a) (10b)

WHO is happy if WHO is coming?

[JOHN]F [is happy]B if [MAry]F [is coming]B

The matrix of this sentence looks like the following: ª MARY(y) š COMING(y) º o ª JOHN(x) š HAPPY(x) º . «¬ »¼ «¬ »¼ COMING(y) HAPPY(x)

374

Ingolf Max

If we assume a conjunctive projection of presuppositions we get ª MARY(y) š COMING(y) Š JOHN(x) š HAPPY(x) º . «¬ »¼ COMING(y) š HAPPY(x)

Of course this conjunctive projection is only an ad hoc offer. The point I want to show is what happens if the whole conditional has its own information structure. Let us assume that somebody likes to explain why somebody is happy: (10c) [[JOHN]F [is happy]B ]B if [[MAry]F [is coming]B ]F .

The superscripts indicate the information structure of the whole conditional at the sentence level. Now we have three background domains: The internal background of the consequent (PBPS2), the internal background of the antecedent (PBPS1) and the whole consequent as the background of the conditional (S1): ª ASS [= (S2 Š S1)]º « PBPS2 š PBPS1 » « » S1 ¬ ¼

ª MARY(x) š COMING(x) Š JOHN(y) š HAPPY(y) º « » COMING(x) š HAPPY(y) « » JOHN(y) š HAPPY(y) ¬ ¼

8 Conclusions and outlook Steube assumes that the semantic form SF includes the truth-conditions (semantics) as well as its conditions of use (pragmatics). Negated sentences with different information structures get different semantic forms (SFs). The position of the negation in that SF depends on the order of the steps of composition. In our approach any sentence gets a multi-dimensional interpretation which reflects the truth-conditions and the background domain as a selected/empty part of that truth-conditions. Our negation is a sentence negation in the sense that the whole sentence, the whole multi-dimensional SF is in its scope. The complex reduction rule which characterizes its behavior regarding multi-dimensional SFs yields the appropriate narrow scope which coincides with the focus domain. Both explications of SF are motivated by considering the relevance of pragmatic factors and its direct influence on the semantics. There are several strategies to handle (potential) presuppositions. The classical model by Frege (1892) offers no syntactic tool to expli-

Interfaces, (Non-)Compositionality, Information Structure and Presupposition

375

cate (existential) presuppositions. Russell's proposal (1905) represents definite descriptions within the classical framework using a linear and conjunctive model. This approach does not reflect the different status of assertions and presuppositions in discourse and has to buy an ambiguity of negation. Horn (1989) pleas for splitting the models: Assertions are characterized in a purely semantic way. No ambiguity of negation is necessary. Presuppositions are explained pragmatically by using and developing Grice's theory of implicatures. Therefore he gets an external version of the semantics-pragmatics interface. We accept the distinct levels of assertion and presupposition and use different syntactic dimensions to represent them. The so-called SF is reconstructed as a multi-dimensional expression which contains the semantic as well as the pragmatic aspects of utterances. Therefore we get an internal version of this interface. If we are confronted with reliable cases of non-compositionality we can handle them in a precise and fruitful manner. We are convinced that an accurate separation of at least the following cases of increasing complexity of expressions and their context with respect to (non-)compositionality is very helpful from a methodological point of view: (A1) sentence-internal: level of simple sentences with respect to isolated lexical presuppositions: e.g. DPs and VPs as presupposition triggers (cp. Sauerland in prep.); (A2) sentence-internal: level of simple sentences with respect to the interaction between lexical presuppositions: e.g. VPs and phase particles (cp. Zybatow & Malink 2003); (B1) sentence-external: level of simple sentences and narrow context: e.g. information structure (cp. Max 2004a); (B2) sentence-external: level of complex sentences: e.g. sentence coordination (cp. Max 2004b); (B3) sentence-external twofold: coordination together with information structure6 and (C) complex text structures. Our multi-dimensional approach will be challenged by all of these cases. This indicates interesting directions of further research. 6

Take as one illuminative case our "final example". We get even more complex cases if we assume (a) that the causal German connective da triggers lexically the factuality of the simple sentence q in constructions of the form "p, da q", (b) "p" and "q" contain lexical presupposition triggers, (c) "p" and "q" have their own information structures and (d) the whole construction has its own global information structure.

376

Ingolf Max

References Carnap, R. (1947): Meaning and Necessity. A Study in Semantics and Modal Logic. Chicago: The University of Chicago Press. Dever, M. (1999): Compositionality as methodology. Linguistics and Philosophy 22, 311–326. Fintel, K. von (2003): Pragmatics: notes on presupposition. Lecture Notes. Frege, G. (1892): Über Sinn und Bedeutung. Zeitschrift für Philosophie und philosophische Kritik, NF 100, 25–50. Geurts, B. (1999): Presuppositions and Pronouns. London: Elsevier. Hodges, W. (1998): Compositionality is not the problem. Logic and Logical Philosophy 6, 7–33. Hodges, W. (2001): Formal features of compositionality. Journal of Logic, Language, and Information 10, 7–28. Horn, L. R. (1989): A Natural History of Negation. Chicago: The University of Chicago Press. Kazmi, A. & F. J. Pelletier (1998): Is compositionality formally vacuous? Linguistics and Philosophy 23, 629–633. Max, I. (1996): External, restricted external and internal negations in a two-dimensional logic. In: H. Wansing (ed.): Negation. A Notion in Focus (= Perspectives in Analytical Philosophy 7), Berlin & New York: de Gruyter, 59–85. Max, I. (2004a): Information structure – two-dimensionally explicated. In: A. Steube (ed.) Information Structure. Theoretical and Empirical Aspects (= Language, Context, and Cognition, Vol. 1), Berlin & New York: de Gruyter, 97–123. Max, I. (2004b): Assertion und Präsupposition. Zur Semantik und Pragmatik von Konnektoren. In: H. Blühdorn, E. Breindl & U. H. Waßner (eds): Brücken schlagen – Grundlagen der Konnektorensemantik (= Linguistik – Impulse & Tendenzen, Vol. 5), Berlin & New York: de Gruyter, 93–121. Russell, B. (1905): On denoting. Mind 14, N.S., 479–493. Sauerland, U. (in prep.): Implicated presuppositions. In: A. Steube (ed The discourse potential of underspecified structures. Language, Context & Cognition. Berlin/New York: de Gruyter. Steube, A. (2006): The influence of operators on the interpretation of DPs and PPs in German information structure. In: S. Winkler & V. Molnár (eds): The architecture of focus. Berlin & New York: de Gruyter, 489–516. Steube, A. (in prep.): German negation and information structure. Ms. Szabó, Z. G. (2005): Compositionality. In: E. N. Zalta (ed.): The Stanford Encyclopedia of Philosophy. http://plato.stanford.edu/archives/spr2005/entries/ compositionality/ Van Fraassen, B. C. (1968): Presupposition, implication, and self-reference. Journal of Philosophy 65.

Interfaces, (Non-)Compositionality, Information Structure and Presupposition

377

Westerståhl, D. (1998): On mathematical proofs of the vacuity of compositionality. Linguistics and Philosophy 21, 635–643. Zadrozny, W. (1994): From compositional to systematic semantics. Linguistics and Philosophy 17, 329–342. Zybatow, T. & M. Malink (2003): Verbklassen und Phasenpartikeln. In: M. Weisgerber (ed.): Proceedings of the Conference "sub7 – Sinn und Bedeutung". Arbeitspapier Nr. 114, FB Sprachwissenschaft, Universität Konstanz, Germany; http://ling.uni-konstanz.de/pages/conferences/sub7/proceedings/download/ sub7_zybatowMalink.pdf, 337–351.

Subject Index

A aboutness 151, 235, 330, 332 aboutness topic see topic accent 83, 89, 278–281, 286– 287, 290–292 accent pattern 89, 352 accent peaks 207, 227–228, 245 activity 7–8, 23, 47–49, 184– 185 alternative set 81–82, 224–225, 233–234, 245, 247–250, 252, 255, 260, ambiguity 7–8, 57, 173, 176, 266–267, 375, anaphoric expression 100, 111, 147, 150, 232, 330, 333, 340, 343 animacy constraint see contraint argument 11, 47, 52–55, 58, 63, 68, 97, 101, 104, 106–108, 112, 144, 150, 160, 169, 183, 186, argument position 35, 42–43, 51–54, 58, 60, 61, 63, 69, 261 argument structure 25, 35, 45, 49, 51–52, 54, 59–61, 165, 183, 248, 269

external argument 52–53, 55, 62, 144–146, 150, 246– 247 internal argument 52–55, 61– 63, 104, 150, 247, 254 referential argument 145, 149, 151, 261, 263 articulation 10, 23 articulation and perception 1– 2, 6, 10, topic-focus articulation 234 assertoric illocutionary force 98, 102, 106, 110

B background 81, 109, 144, 148– 149, 151, 194, 203–204, 211, 267, 278, 316, 325, 349, 350– 351, 357, 361–369, 371, 374, background presupposition 350–351, 361, 364, 366, 371– 372 boundary 12, 20, 287

C case 36, 157, 161, 163, 165– 166–167, 170, 172–176, 192, 194, 299, 301–302, 304–314, 318–325

380

Subject Index

case assignment 158, 176, 192 case checking 165, 166 case feature 151, 153, 164, 166, 171–172, 176 case filter 167 cataphoric pronoun 332–333, 336 causative 25, 35–37, 47, 54–55, 66–67 cleft sentence 141–147, 153, 234 coherence 329, 341–342, 346 common ground 97, 100, 102, 108, 112, 210 communicative dynamism 232, 237–238, 242 complement clauses 98, 105, 107 compositionality 3, 43, 50, 68, 222, 350–354, 357, 359, computational system 17, 19, 30, 234 conceptual 3, 5, 6, 9, 11, 28, 33, 34, 41, 44, 46, 54, 56–58, 64–65, 97, 100–102, 109– 110, 112, 245, 246, 249, 251, 273, 351 conceptual representations 17, 102, 110 conceptual primitives 44–45 conceptual structure 2, 18, 44, 46, 102, 351 constraint 12, 13, 59, 159, 281, 299, 300–304, 325–326, 338, 353 animacy constraint 299, 301, 310–312, 315, 325–326 dative constraint 299, 301, 303, 305, 309–312, 316, 321, 324–326

definiteness constraint 299, 301, 316, 318, 320–321, 324–326 contextual specification 34, 47, 56–57, 63–64, 69, 205, contextually 34, 44, 57, 59, 69, 170, 221, 254 contextually bound 232, 241 contextually non-bound 232, 241 contrast 83, 93, 200, 205, 208– 210, 212, 214–215, 217, 224, 227, 231–235, 239, 240, 245, 250, 252, 256, 259, 261, 264, 265, 267, 269, 339, 341 contrastive 206, 224, 226, 232, 235, 241, 249, 272 contrastive focus 209, 227, 245–248, 250, 252–257, 260–262, 267, 272 contrastive reading 207, 216 contrastive meaning see meaning contrastive negation 253– 255, 270–272 contrastive topic 83, 204, 206, 219, 223, 235 copy movement 157, 159, 172, 176

D dass-clauses 98, 102–103, 105– 106, 112 dative constraint see contraint declarative 97, 98, 99–103, 105–108, 110, 112, 118, 235, 300, declarative clauses 100–101

Subject Index

declarative root clauses 97– 100, 103, 105–106, 108, 110, 112, decomposition 26, 34, 44, 46– 48, 54–56, 58–59, 65, 68–69, definite description see description definiteness 28, 262, 304, 316– 325 definiteness constraint see contraint description 19, 103, 232, 234, 341 definite description 19, 137, 144–145, 255, 334, 369, 375, indefinite description 19, 66, 91, 157, 161, 166–168, 172, 258, 316, 318, 320– 321, 324–325, 333 descriptive meaning see meaning discourse 79, 81–88, 90, 92, 93, 199 , 231–233, 235, 252, 329–330, 333, 342–343, 350, 365, 369, 371, 375 discourse context 84, 87, 199, 200, 202–203, 210, 212, 220–221, 223–224, 245 discourse referent 81, 91, 100, 111–112, 268, 330, discourse relation 336–337, 338, 341 discourse topic 83–86, 329, 334–335 ditransitive verbs 306, 315, 321 dossiers 27–28, D-tree 83, 86–89

381

E Event-Related Potentials (ERPs) 278, 280–281, 286, 290, 292, 306 exclamative 109, 116–118, 123, 134 extensionality 79, 249, 350, 355–359, 362

F F0-contour 213–214, 216, 217, 245, 277, 284, 288, factive predicate 97, 99, 103, 105, 110–111 fall-rise intonation 334–336 familiarity 21, 333–334 focus 89, 92, 99, 103, 109, 144, 145, 148, 150–152, 194, 203– 204, 213, 217, 232, 240, 245– 247, 268–272, 278, 280, 299– 301, 303, 324, 332, 350–351, 361, 374 focus constituent 151, 254, 255, 256, 258–260, 267, 270–271 focus exponent 249, 254, 260, 266 focus of new information 194, 217, 219, 231, 245, 278–279, 350 focus particle 272, 280–282, 286–287, 290–293 focus phrase 103, 254, 256, 258, 267, 271 focus-background-structure 203, 247, 278, 349, 351, 363 Full Interpretation 4, 6, 9

382

Subject Index

fundamental frequency see F0contour

K

G

knowledge 2, 9, 34, 39, 43, 45, 68–69, 100, 102, 118, 221– 228, 245, 247–249, 253, 330

givenness 83, 86–87, 127, 221, 231, 235–236, 241, 245– 247, 278, 371

I I-Language 3–7, 10, 17, 26–27 illocution 101, 102, 106, 108, 110, 118, 149 illocutionary force 98, 100, 106 illocutionary functor 106, 108, 112, illocutionary operator 97, 101, 106, 149, 151 indefinite description see description infinitive construction 182, 192, 194–195 intensified wh-phrase 116, 134 intensionality 246, 350, 356, 357 intentional representation 3, 17 Interface 1–6, 9–10, 14–15, 17, 23, 24, 27, 46, 101, 106, 211, 228, 234, 248, 349, 351, 353– 355, 375 interrogative 98, 99, 101, 103, 105, 107–108, 115–121, 129, 131, 132, 135–136 interrogative clause 98, 100, 141–143, 146-148, 245 interrogative complement 98, 107–108, 115 intonational phrase 287, 292 intransitive use 36–37, 49

L left dislocation 151, 342–343 left periphery 142–143, 146– 147, 150–151 lexical 4, 7, 16–17, 26–28, 33–34, 38, 40–48, 51–61, 66, 68–69, 107, 117, 130, 144, 157, 176, 183, 188, 201–211, 220–222, 226–228, 233, 238, 250, 253, 256–257, 259, 262– 268, 350–354, 364, 371–372, 375 lexical case 165, 173, 176 lexical meaning 33–34, 38, 43–45, 48, 50–51, 53, 58, 221, 256, 264, 265 lexical semantics 33, 40, 42, 51, 68, 98 lexical semantic structure 183, 188, 190 linear order 10–12, 13, 81, 88, 314 linearization 88, 303, 305, 310, 315, 327 Logical Form 3, 4, 18, 143

M matrix predicate 97–101, 103– 109, 111–112 meaning 2–3, 33, 40, 42, 43, 46, 47, 50, 54, 57, 64, 79, 84, 86, 92, 100, 118, 142, 144–

Subject Index

145, 147–151, 183, 200, 202, 221–222, 232, 246, 247, 250, 352–256, 361, 367, 369 meaning of utterance 33, 51, 57, 247 meaning postulate 40, 42, 64, 67–68 meaning variants 34, 36–38, 40, 42–45, 48, 50, 52, 55, 60, 66–69 ordinary meaning 101, 106, 110, 112, 246–248 primary meaning 34, 45, 47, 50, 69 utterance meaning see meaning of utterance mental 3, 18–24, 26, 175 mental model 18–21, 24, 26 mental organization 2, 17, 19, 23 mental representation 20, 22, 329–331 mention-some answers 122– 123, 126, 129–130 middle field 135, 144–145, 205, 218, 331–333, 335, 144– 145, 205, 218, 331 Minimalism 3–4, 68, 164, 234 Minimalist Program see Minimalism morphemic level 237, 241

N negation 20, 85, 121, 131, 172, 184, 193, 195, 208–210, 213, 233, 235, 250–251, 253–255, 264, 270–272, 349–353, 355, 361, 364–370, 372–375 new information see focus of new information

383

nominalization182–183, 185, 188, 190, non-compositionality 350, 353, 357–359, 363, 369 non-propositional situation 103, 106, 108, 112–113

O ontological 44, 231, 265 ordinary meaning see meaning

P parameter 57–60, 63–64, 66, 69, 213, 215, 216, 226, 249 particle 77, 92–93, 234, 272, 280–282, 286–287, 290– 293354, 364, 375 perception 1–2, 6, 10, 17, 211, 214–215, 282 Phonetic Form 3, 22, 28, 141 pied piping 157, 158, 165, 168, 172, pitch accent 278–281, 285– 287, 290, 292, 297 polar interrogative 115–116, 119–121, 129–132, 138 polysemy 33–34, 40, 68 prefield 331–333 proper name 28–29, 249, 255, 364, 369 prepositional phrase (PP) 39, 45, 157–160, 163–164, 166– 176, 183, 186 presupposition 88, 108, 117, 120, 123–124, 128–132, 138, 251–252 primary meaning see meaning primitives 16, 44–45, 48, 55 projectivity 237–239, 241

384

Subject Index

proposition 13, 17, 19, 46, 48, 77, 85–86, 97–100, 103, 105, 106–113 propositional content 97, 99, 106, 109, propositional predicates 97, 100–109, 112 propositional situation 98– 106, 108–113 prosodic pattern 84, 206, 211, 213–215, 228, 277, 293, pseudocleft 141–143, 146–148, 150–153

Q qualifying elements 27, 29, 30 question under debate 83–87, 89, 92 question-answer 82–83, 143, 245, 278–279, 280

R referential argument see argument relative clause 141–144, 146, 278, 281 resumptive pronoun 146–148, 150–151, 344 right periphery 141–142, 152 root clause 97–99, 103–106, 108, 110–112, 151–152

S scope of negation 233, 235, 251, 270–271

Semantic Form 2, 5, 20, 25, 34, 46, 56–57, 110, 145, 248, 249, 259, 269, 352, 361, 374 sentence internal topic see topic sentence negation 208, 213, 250, 253–254, 270, 374 sentential adverbial 151, 169– 170, 172, 256, 264, 331–332, 339, 340, 361 situation 6–7, 9–10, 13–14, 16– 18, 20, 38, 43, 48, 62, 97– 100, 103–105, 112, 145, 245, 249, 354, 359 situational argument see argument solitaire 98, 103, 105–106, 108–109, 112 systematic element 26, 29–30

T tectogrammatics 234, 237, 238, 241 temporal particles see particles text coherence 329, 339, 341– 342, 346 thematic roles 40, 301, 311 topic 82, 84, 142, 147–148, 150–152, 234–235, 241, 267– 269, 329–330, 333, 336, 343– 344 aboutness topic 329–333, 335–336, 342, 346 discourse topic 83–86, 329, 334–335 sentence internal topic 267, 329–330, 333–343, 345– 346 topic position 269, 333, 336 topicalization 151, 342–343 topichood 329, 331, 333–334

Subject Index

Two-Level Model 46, 47, 57, 247–248

U underspecification 7, 8, 34, 44, 46, 57, 69, 170, 202, 205, 208, 260 utterance meaning see meaning

V verb meaning see meaning verbalization 9, 27 VERUM-focus 257, 266 volitional predicate 97, 99, 103, 109, 113

385

W weakly exhaustive answer 120, 122–123, 128 wh-clauses 134, 142–143 wh-interrogatives 89–90, 115– 116, 120, 130, 132–133, 135– 138, 168 word order 205, 232, 234, 237–242, 299–304, 306, 309– 310, 315–317, 320–321, 325, 351 world knowledge 34, 39, 43, 64, 68–69, 221–228