The Oxford Handbook of Information Structure 0199642672, 9780199642670

Citation preview

Introduction

Oxford Handbooks Online Introduction   Caroline Féry and Shinichiro Ishihara The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara Print Publication Date: Jul 2016 Subject: Linguistics, Morphology and Syntax Online Publication Date: Oct 2015 DOI: 10.1093/oxfordhb/9780199642670.013.43

Abstract and Keywords This chapter is an introduction to the Oxford Handbook of Information Structure. After introducing the goals of the handbook and a brief history of information structure studies, this introductory chapter provides the definitions of the key notions of information structure (focus, givenness, and topic) in Krifka (2008), which were used as the baseline for all the chapters of the handbook. The Handbook is organized in four parts: an overview of various linguistic theories of information structure in semantics, pragmatics, morpho-syntax and phonology (Part I); case studies of various information structurerelated topics (Part II); the state of the art in the study of information structure from experimental perspectives (Part III); and chapters discussing information structure in selected languages or language families (Part IV). The introduction contains short summaries of each of the chapters in the Handbook. Keywords: information structure, focus, givenness topic, information packaging, Common Ground

1.1 Goals INFORMATION structure (IS) refers to the structuring of sentences (information packaging in Chafe’s 1976 terms) in different kinds of information blocks. IS is not only directly related to some of the central disciplines of linguistics (semantics, pragmatics, syntax, morphology, and prosody), but also to some of the extra-linguistic aspects such as interlocutors’ psychological perception of the world. This Handbook contains forty chapters discussing various theories and issues on IS, and aims to comprehensively cover the state of the art in one volume. It is designed as a guide

Page 1 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Newcastle; date: 27 December 2018

Introduction to the theoretical and practical aspects related to IS, surveying what researchers have achieved so far, as well as raising outstanding questions that still need to be investigated. By bringing together this diversity of questions and this diversity of approaches, we hope to encourage our readers to explore different avenues of research in the future. The volume is intended for a wide audience: graduate students, faculty, and researchers in all disciplines of linguistics who are interested in information structure and its effect on grammar as well as meeting the needs of linguists of all theoretical persuasions at graduate level and above. It will also be useful for cognitive psychologists, computational scientists, philologists, and philosophers. Studies of IS face various kinds of challenges, which are rooted in one striking aspect of IS: its diversity. A first challenge is the diversity, or abundance, of theories and definitions of IS-related notions. There are countless definitions of basic IS notions and related theories that have been proposed in the literature. Many researchers use the same terminology to refer to different notions, or different terms are applied to the same concept. Choosing which definition to adopt in one’s own study can be a first obstacle for starting work on IS-related issues. This volume aims to bring some clarity to the terminological confusion typical of new concepts and ideas. While there is an emerging consensus about many terms and notions related to information structure, some clarifications are still badly needed. In order to avoid unnecessary repetition of terminological confusion, (p. 2) all the chapters in this Handbook adopt a uniform set of definitions, as summarized by Krifka (2008), as a starting point. Therefore, each chapter does not define these basic notions in depth. Only deviations from Krifka’s definitions are discussed by the authors. Krifka’s definitions of IS notions are summarized in this introduction (Section 1.3). A second challenge is the diversity of IS grammatical reflexes. IS is expressed in different ways in different languages, and even within a single language, there is diversity of means. Focus (see Section 1.3.2 for the exact definition adopted in this Handbook), for example, can be expressed by phonetic/phonological means (pitch accents, metrical prominence, prosodic phrasing, pitch range expansion/compression, lengthening/ shortening, etc.), by morphological/syntactic means (morphological marking, syntactic movement of focused material, word order manipulation of non-focal material, or specific focus-constructions such as clefts), and by semantic/pragmatic means (focus-sensitive operators and discourse particles, manipulation of pragmatic implicatures, and conversational maxims). There is no doubt that the great number of grammatical means is one of the major causes of the diversity of IS notions and definitions mentioned above. Furthermore, while the notions of IS refer to the formal and communicative aspects of language for the expression of information structural roles, IS is also closely related to psychological perception of the world and of the minds of the participants of the conversation. The notions of IS may therefore denote the extra-linguistic, cognitive, or mental states of referents, actions, locations, and temporality as well (see Kuno 1972; Chafe 1976; Prince 1981; Lambrecht 1994, and many others for this dichotomy).

Page 2 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Newcastle; date: 27 December 2018

Introduction A third aspect that makes IS study challenging is the methodological diversity found in studies of IS. In most (if not all) cases, linguistic and extra-linguistic factors of IS are not simply playing their roles independently. Rather, they are intricately interwoven. In order to disentangle substantial knots of relevant factors and fully understand the nature of ISrelated phenomena, linguists need to approach IS-related linguistic phenomena from various interdisciplinary perspectives. A wide range of novel and interdisciplinary approaches, though certainly a desirable direction for the advance of a scientific field, may become an obstacle when one needs to survey the relevant literature in unfamiliar subfields. One of the aims of this handbook is to facilitate interdisciplinary investigation by introducing some of the methodological developments that have emerged in recent years, especially those that take experimental approaches. A fourth challenge is the cross-linguistic diversity. Typological investigation is an essential part of IS studies. This Handbook introduces selected languages and language families to illustrate the breadth of cross-linguistic variation of IS expressions as well as variation within each language(-family). The chapters in this volume are grouped into four parts, each of which addresses one of the challenges mentioned above. The thematic organization within each part of the volume reflects some of the principal fields of research and application in IS. As such, the volume can be read by focusing on specific aspects and parts. Also, each chapter contains cross-references to other chapters, so that the related discussions can be easily found within the volume. This introductory chapter presents a theoretical background, including a short history of IS studies (Section 1.2), the definitions of IS-related notions adopted throughout the volume (Section 1.3), and brief summaries of each chapter (Section 1.4). (p. 3)

1.2 A short history of information structure The point of departure for research on IS starts with Mathesius (1975), who published at the beginning of the twentieth century and founded the Prague Linguistic Circle. Mathesius is often credited as the father of modern IS, see for instance Lambrecht (1994).1 Mathesius replaced the psycholinguistic terms ‘psychological subjects’ and ‘psychological predicates’ of von der Gabelentz (1869) with the notions of ‘theme’ and ‘rheme’, see also Daneš (1974a, 1974b) and Steedman (2000) for this terminology. According to Mathesius, theme is what the sentence is about, and rheme is what is being said about the theme. These terms easily translate into topic for theme and focus for rheme. Firbas (1964, 1966) developed the ideas of the Prague school further, and integrated them in a theory of dynamic communication: the theme is ‘the sentence element (or elements) carrying the lowest degree(s) of C[ommunicative] D[ynamism] within the sentence’, and the rheme is the important part of the sentence, that ‘pushes the communication forward’ (1964: 272). Halliday (1967–68) first used the term Page 3 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Newcastle; date: 27 December 2018

Introduction information structure. Focus is what is ‘not being recoverable from the preceding discourse’ (1967–68: 204). ‘The newness may lie in the speech function, or it may be a matter of contrast with what has been said before or what might be expected’ (1967–68: 206). Chafe (1976: 30) defined given (or old) information as ‘that knowledge which the speaker assumes to be in the consciousness of the addressee at the time of the utterance’ and new information as ‘what the speaker assumes he is introducing into the addressee’s consciousness by what he says’. Both Prince (1981) and Lambrecht (1994) were especially interested in the characterization of Chafe’s notion of givenness and proposed elaborate hierarchies of givenness. Another branch of information structure was initiated by Marty (1918). He was inspired by the philosopher Brentano (1874) who discussed the categorical and thetic judgement types. A typical categorical judgement consists of a subject–predicate structure. First an entity is named and second a statement is made about it. Thetic judgements, by contrast, express an event, a state, or a situation, and are thus simpler than categorical ones. Kuroda (1972) revived this distinction by observing that they straightforwardly apply to Japanese expressions containing ga and wa respectively. In the last decades, the interest in information structure has grown immensely, (p. 4) and this has lead to an extraordinary evolution of linguistic themes and ideas. Many authors have been crucial for the advancement of information structural theories, including von Stechow (1981), Jacobs (1983), Rooth (1985, 1992), Rochemont (1986), Erteschik-Shir (1997), Vallduví (1992), Steedman (2000) to cite only a few names, as can be gathered from the chapters of the volume.

1.3 Definitions As a starting point, all the chapters in this handbook adopt the definitions of IS notions proposed by Krifka (2008). This section briefly summarizes Krifka’s definitions of the basic IS notions.

1.3.1 Information packaging and Common Ground Following Stalnaker (1974) and Reinhart (1981), Krifka (2008: 243) claims that information structure notions should be grounded in theories of how communication works: ‘The basic notions of Information Structure (IS), such as Focus, Topic and Givenness, are not simple observational terms. As scientific notions, they are rooted in theory, in this case, in theories of how communication works.’ To start the discussion of IS notions, Krifka (2008) follows Chafe’s (1976) approach to IS by adopting the idea that IS should be regarded as the matter of information packaging, that is, how ‘the speaker accommodates his speech to temporary states of the addressee’s mind, rather than to the long-term knowledge of the addressee’ (1976: 28). He discusses various ‘statuses’ of nouns (or noun phrases)2 that are related to how the information is transmitted between

Page 4 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Newcastle; date: 27 December 2018

Introduction participants of a discourse, in other words, how the content of the utterance is packaged by the speaker and sent to the addressee: I have been using the term packaging to refer to the kind of phenomena at issue here, with the idea that they have to do primarily with how the message is sent and only secondarily with the message itself, just as the packaging of toothpaste can affect sales in partial independence of the quality of the toothpaste inside. (Chafe 1976: 28) According to Chafe, the speaker packages the information to be sent to the addressee according to the knowledge that he assumes is shared by the addressee. This shared knowledge is often called Common Ground (CG), a term proposed by Stalnaker (1974, 2002). The CG forms the background of a conversation—the information that is mutually (p. 5) known (or believed) to be shared by speaker and addressee—and to which new information is added. The idea is that a discourse proceeds in such a way that each utterance by the participants of the discourse updates the content of the CG. As a result, the CG is continuously modified in communication. The notion of CG also allows us to make a distinction between ‘presuppositions, as requirements for the input CG’, and ‘assertions or the proffered content, as the proposed change in the output CG’ (see Krifka 2008: 245). Although Krifka agrees with Chafe in that IS deals with the way the message is delivered, Krifka further points out that IS not only deals with how to deliver the message, but also affects the content of the message itself. He therefore makes a distinction between what the content of the current CG is (CG Content), and how these contents should be developed in terms of the relevance to the current discourse (CG Management). This distinction allows us to ‘associate those aspects of IS that have truth-conditional impact with CG content, and those which relate to the pragmatic use of expressions with CG management’ (Krifka 2008: 246).3

Page 5 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Newcastle; date: 27 December 2018

Introduction

1.3.2 Focus The definition of focus is based on the theory of alternative semantics of focus proposed by Rooth (1985, 1992, chapter 2 of this volume). Focus assigned to a linguistic expression α always indicates that there are alternatives to α relevant in the current discourse. Putting it differently, anything that does not indicate any alternatives to α should not be called focus. This analysis allows various ways of focus marking, as there are various ways to signal the presence of alternatives, for example pitch accents, word order, specific syntactic constructions like clefts, etc. Krifka (2008: 247) defines focus as follows: (1)

Focus may be used to influence either the CG content or the CG management. Krifka (2008) distinguishes between the two in terms of semantic vs pragmatic uses of focus. The semantic use of focus affects the truth-conditional aspects of the discourse, while the pragmatic uses of focus regulates how the CG of the discourse is to be updated by imposing pragmatic requirements on the discourse to fulfil the communicative needs of the discourse participants. For example, focus sensitive operators (see Beck, chapter 12 of this volume), such as only, also, and even, are always associated with focus to influence the truth-condition of the sentence in which they appear. The sentence John only introduced Mary to Sue may have different truth-conditions depending on the location of focus, which is indicated by pitch accents. In the context where John introduced Mary (p. 6) to Sue and Bill (and didn’t introduce anyone to anyone else), the sentence is true if the pitch accent is on Mary, but false if the pitch accent is on Sue. The pragmatic use of focus regulates the direction to which the discourse is developing. A first example is the so-called question–answer congruence. A wh-question sentence requires a congruent answer sentence, thus adding a particular type of information to the current CG. In other words, a question ‘changes the current CG in such a way as to indicate the communicative goal of the questioner’ (Krifka 2008: 250). The answer sentence fulfils this communicative goal by expressing the required information to be added to the CG as the focus of the sentence. Other pragmatic uses are correction and confirmation of information, highlighting of parallel information, and delimitation of the utterance (cf. definition of contrastive topics below). Further discussions of CG management and related notions include ‘question under discussion’ (Velleman and Beaver, chapter 5), ‘(non-)at-issue’-ness (Horn, chapter 6), and ‘presupposition’ (Sæbø, chapter 7).

Page 6 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Newcastle; date: 27 December 2018

Introduction

1.3.3 Givenness The status of referents can be new (inactive at the point of their introduction into the discourse) or given (active in the consciousness of the interlocutors). According to Clark and Haviland (1977), given information is ‘information [the speaker] believes the listener already knows and accepts as true’, and new is ‘information [the speaker] believes the listener does not yet know’. Givenness is divided into text-givenness (previously mentioned in the discourse) and context-givenness (contextually salient). Using the notion of CG, Krifka defines givenness as in (2). (2)

This definition allows two different interpretations of givenness: givenness may be either a categorical feature (given vs not given, i.e. new), or a scale that expresses the degree of discourse salience, following two lines of theories of givenness (e.g. Schwarzschild 1999 for the former and Prince 1981, Gundel et al. 1993, Chafe 1976, and Lambrecht 1994 for the latter). Givenness may be part of the lexical information (as in pronouns, clitics, and definite articles), or arbitrarily assigned to linguistic expressions by means of various grammatical devices (such as deaccentuation, word order, and deletion). See chapter 3 by Rochemont for further discussion.

Page 7 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Newcastle; date: 27 December 2018

Introduction

1.3.4 Topic The notion of topic is related to the way information is stored in human memory and organized in communication. Krifka describes topic as follows: ‘ … topic is the entity that a speaker identifies about which the information, the comment, is given. This (p. 7) presupposes that information in human communication and memory is organized in a certain way so that it can be said to be “about” something’ (Krifka 2008: 265). He adopts the following definition of topic (often referred to as aboutness topic), which makes use of the notion of CG. (3)

This definition follows the proposal by Reinhart (1981), who uses the organization of a library catalogue as a metaphor for how topics and comments are related to the CG. The CG (for which she uses the term ‘context set’) is organized like a subject-oriented library catalogue, in which book entries (propositions stored in the CG) are organized according to their subjects (topics). A topic is like a subject in the catalogue, according to which book entries are collected in a single file card. Each time a new book entry (a new proposition) is added to the catalogue (the current CG), the topic specifies the file card to which the book entry is to be added.4 (But see also Roberts 2011 and Büring, chapter 4 of this volume, for the difficulties of defining topic.) Based on this definition, Krifka also discusses the interaction between topic and focus to define contrastive topics (see also Büring, chapter 4 of this volume). He claims that a contrastive topic contains a focus, which induces a set of alternatives within a contrastive topic and indicates the presence of other topics relevant for the current CG. The presence of alternatives indicates that there are other topics and their comments that may be added to the CG. In other words, a contrastive focus can imply the presence of further information to be added to the CG.5

1.4 Organization of the Handbook As mentioned in Section 1.1, this Handbook is divided into four parts, each reflecting a type of diversity found in IS studies: Part I Theories of information structure, Part II Current issues on information structure, Part III Experimental approaches to information structure, and Part IV Language studies on information structure. In order to comprehensively review each thread of research that seemed important for a coverage of all aspects of IS, we selected what we thought were major topics and fundamental issues (p. 8) and assigned them a chapter each. In planning the specific contents and structure Page 8 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Newcastle; date: 27 December 2018

Introduction of the volume, our goals were to gather a wide range of perspectives, across subfields, languages, and disciplines, and to highlight the complementarity of approaches and backgrounds. Inevitably, however, theoretical approaches and empirical facts are interwoven, and are often addressed in more than only one chapter. And conversely, some important aspects of IS may not be fully represented in this volume, which is also inevitable to a certain extent given the breadth of the range of IS-related phenomena. When some topic or phenomenon could not be included in a chapter due to space limitation, however, authors were asked to provide references to relevant literature. In the remainder of this section, a general overview of each chapter is provided.

1.4.1 Part I: Theories of information structure Part I comprises an overview of various linguistic theories of information structure in semantics, pragmatics, morphosyntax, and phonology. The first three chapters of this section provide a comprehensive overview of the basic notions of information structure: focus, givenness, and topic. These first chapters take a semantic perspective, and lay the ground for the following chapters. In chapter 2 (Alternative semantics), Mats Rooth defends the idea that ‘the semantics of the language makes available, in addition to an ordinary semantic value (which is a proposition in the case of a clause), a set of eligible alternatives, which for a clause is a set of propositions. The set of alternatives is a “focus semantic value”, or “alternative semantic value” ’. After reviewing empirical applications of the alternative semantic theory of focus, he explicates how focus is interpreted and how alternatives are composed in this theory. He also discusses an alternative theory by Schwarzschild (1999) as well as Büring’s (1997) extension of the theory to contrastive topics. But he refutes the idea that the notion of ‘focus’ can be assigned a single definition. There may be ‘several different theoretical notions of focus’, depending on the language and the phenomenon under discussion. According to Michael Rochemont’s chapter 3 (Givenness), the notion of givenness, which he discusses in relation to the notion of salience, is related to the notion of CG management rather than that of CG content (see Krifka 2008 and above for this distinction). He examines deaccenting in English and comes to the conclusion that ‘failure to deaccent when to do so would be consistent with speaker’s actual communicative intent misleads interlocutors to think the speaker must have some other intent’. This can be illustrated with Lakoff’s (1972) famous sentence John called Mary a Republican and then SHE insulted HIM. If insulted is not deaccented, then it may be assumed that the speaker does not consider that calling somebody a Republican is insulting. And he proposes that even under a precise definition of givenness, focus cannot be entirely eliminated. Daniel Büring’s chapter 4 ((Contrastive) Topics) specifies an analysis of contrastive topics as an extension of alternative semantics for focus: ‘CT [contrastive topic] marking results in a set of alternative propositions which are explicitly not used for exhaustification’. A sentence containing a contrastive topic and a focus must not be a complete answer to the question under (p. 9) discussion. Büring

Page 9 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Newcastle; date: 27 December 2018

Introduction also makes a plea for eliminating non-contrastive topics, which according to him, do not consist in a coherent type of linguistic phenomena. Chapters 5 to 7 introduce theories of information structure in pragmatics and discourse analysis. In chapter 5 (Question-based models of information structure), Leah Velleman and David Beaver present a Question Under Discussion model of discourse and information, following a proposal by Roberts (2012). This model uniformly treats focus as a pragmatic notion, and explains its function by developing discourse structures based on questions implicitly or explicitly indicated by focus. They discuss the notion of ‘relevance’ of utterances and of ‘congruence’ of answers to questions in assessing a model in terms of a range of constraints that relate IS to discourse structure. Chapter 6 by Laurence Horn (Information structure and the landscape of (non-)at-issue meaning) explores issues around the border between semantic entailment and pragmatic implicature, or between CG content and CG management, from the perspective of at-issueness. Among other things, Horn shows that exhaustivity in structural focus constructions like clefts (as in It was a pizza that Mary ate) belongs to a non-at-issue component of meaning, and rejects a semantic treatment of exhaustivity. In chapter 7 (Information structure and presupposition), Kjell Johan Sæbø examines the notion of presupposition in detail. [I]‌ did the dishes presupposes that someone did the dishes. He points out three areas where the notion of presupposition interacts with IS and discusses each in detail: (i) presupposition triggered by focus operators under Rooth’s alternative semantic theory of focus; (ii) presupposition as the background of focus; and (iii) relation to discourse structure, to which both IS and presupposition are sensitive. Presupposition is understood as the conditions that the CG must meet in order to be updated with the sentence. It is shown that a theory which does not directly involve an existential presupposition, but which creates the potential, in the form of the alternative set, for a general process to generate a defeasible presupposition is to be preferred over theories that create presuppositions as the complement of focus. The following chapters discuss specific theories dealing with syntactic theories of information structure and the syntax–phonology interface. Chapter 8 by Enoch Aboh (Information structure: A cartographic perspective) surveys the cartographic line of syntactic approaches to information structure, as proposed by Rizzi (1997). This theory claims that information-structure-sensitive notions (e.g. topic, focus) are encoded by means of discourse markers that trigger various constituent displacement rules. He illustrates his approach with numerous examples, among others taken from Gbe languages of the Kwa family that have topic and focus markers, in bold face in the following example: Náwè lɔ́ yà gbákún étɔ̀n wɛˋ   é ɖè ‘As for the woman, she took off HER HAT’. In chapter 9 (Nuclear stress and information structure), Maria Luisa Zubizarreta compares the predictions of the nuclear stress for Germanic and Romance languages, the latter kind of languages having a rigid rightmost stress pattern, as in Compró el libro JUAN ‘Juan bought the book.’ Moreover, she proposes that the unmarked pattern under wide-focus condition, originates in a different way from narrow focus. Chapter 10 by Karlos Arregi (Focus projection theories) is dedicated to a comparison between two theories (‘Default Prosody’ and ‘F-projection’ approaches) of how focus projection arises. Page 10 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Newcastle; date: 27 December 2018

Introduction In the (p. 10) first model, it is a consequence of general and default rules of grammar, and in the second model, it needs to be implemented by means of rules. Arregi shows that the first model makes better predictions in general and illustrates with Basque data. Chapter 11 by Vieri Samek-Lodovici (Constraint conflict and information structure) provides an overview of conflict-based analyses, specifically Optimality Theory (OT), of information structure. It centres on focalization of Italian post-verbal constituents, as in Parleranno tutti. ‘Everybody will speak.’ No. Non parlerà [NESSUNO]F. ‘No. Nobody will speak.’ And it shows that OT provides the tools for accounting for typological variations.

Page 11 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Newcastle; date: 27 December 2018

Introduction

1.4.2 Part II: Current issues on information structure Part II introduces case studies of various IS-related topics necessitating interface approaches. The direct reference to IS in grammar, as a result of its inclusion in a theory of grammar, has changed our views about what is possible and what is marked in linguistic structures, and it has shown that modules of grammar have to be more connected with each other than has been assumed until now. The need to bring multidisciplinary approaches to bear on critical questions is the motor for many fruitful collaborations across research specialties for issues addressing IS. However, we only start to understand why languages are so diverse in the way they implement IS. In the first 10 chapters of this part (chapters 12–21), various semantics and morphosyntactic issues are discussed in relation to IS. In chapter 12 (Focus-sensitive operators), Sigrid Beck studies the role of focus sensitive operators as expressions that operate on alternatives. She agrees with Rooth (1992, chapter 2 of this volume), who proposes that it is the squiggle operator ~ that is the alternative evaluating operator, rather than quantifiers like only, even, and always, as in Derk only saw a raven. Moreover, and this is also reminiscent of Rooth’s chapter, she examines further expressions that operate on alternatives like those involved in question formation and negative polarity item licensing. In chapter 13 (Quantification and information structure), Manfred Krifka surveys data and theoretical models that concern the interaction between IS and quantification. He first discusses a type of quantification expressed by adverbial quantifiers, generic sentences, or modal operators (A-quantification) and suggests that adverbial quantification (always, usually) is focus sensitive. In the second part of the chapter, quantification expressed by nominal quantifiers (Every black die is loaded) and determiners (D-quantification) are investigated and it is shown that these quantifiers, whose restrictors are usually fixed rigidly by syntax, can interact with IS to exhibit exceptional cases. He also discusses cases where some quantifiers are explicitly focused or topicalized, depending on their meaning. In chapter 14 (Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar), Sophie Repp shows that the notion of contrast is a difficult one. In the sentence Pete went to Rome, Marc went to London, not only the alternatives that the contrastive constituents evoke, but also the discourse relations that connect the discourse segments containing the contrastive constituents, are subjected to detailed analysis for their effects on grammar (prosody, (p. 11) morphosyntax). In chapter 15 (Verum focus), Horst Lohnstein examines Höhle’s (1988, 1992) notion of verum focus ‘emphasizing the expression of truth of a proposition’, as in Carl DID feed the dog. After presenting several approaches to verum focus in the literature, Lohnstein examines the role of sentence mood and discourse situation in verum focus constructions. He claims that verum focus is focus on sentence mood, and that it is used to reduce the alternatives of (verbal) behaviour characterized by the functions of the sentence mood (e.g. stopping disputations about the issue that is verumfocused). In chapter 16 (Predicate focus), Malte Zimmermann studies focus on the verbal element and on functional elements in the extended verbal projection (Peter kicked the cat. No, Peter PETTEDF the cat.) The chapter contains a cross-linguistic overview of the Page 12 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Newcastle; date: 27 December 2018

Introduction grammatical strategies available for marking predicate focus. It investigates asymmetries in the realization of predicate as opposed to term focus by means of obligatory marking, grammatical strategy, and complexity. In chapter 17 (Information structure and discourse particles), Patrick Grosz studies the relationship between IS and German discourse particles, such as ja in examples like weil man ja arbeitet/weil ja wer arbeitet. ‘because one/someone is working as you know’. These particles contribute to the CG management, and are not necessarily truth conditional, as one can see from the translation. The chapter evaluates the thesis that discourse particles separate clauses into rhematic and thematic information and that they are focus sensitive. Grosz also addresses ‘relational’ discourse particles (e.g. doch and schon) operating on a contextually salient alternative proposition. Chapter 18 (Ellipsis and information structure) by Susanne Winkler investigates the role of information structure in ellipsis. VP-Ellipsis is illustrated in the following sentence where the striked out part is elided: Anna promised to play the piano but she DIDN’T play the piano. According to Winkler, syntactic and information-structural theories interact in accounting for the licensing of the different types of elliptical phenomena. Moreover, information structure (especially givenness and focus) and discourse factors influence the form and the interpretation of ellipsis. In chapter 19 (Word order and information structure), Ad Neeleman and Hans van de Koot investigate the phenomenon of allegedly free word order in a typological perspective and show that IS is a powerful motor for word order changes. Using data from Dutch, they establish four generalizations (Given-before-new, A-relatedness, No derived topic–focus mismatch, and No focus resumption) that capture the relation between IS and word order. Chapter 20 (Dislocations and information structure) by Luis López examines dislocations, as illustrated by the following Spanish example: Los cubiertos, ya los he puesto sobre la mesa. ‘I already put the silverware on the table.’ He separates the dislocations in H-type (hanging topics, left dislocations, weakly connected to the main clause) and D-type (left and right dislocations, contrastive, strongly connected to the main clause) and shows that they have different information structural properties. In chapter 21 (Discourseconfigurationality), Balázs Surányi examines the property shared by many languages that topic and focus are associated with particular phrase structure configurations. He shows that discourse-configurationality and (non-)configurationality are mutually independent properties, and illustrates their effect and variation in numerous languages. For example, the word order of non-configurational languages (p. 12) may be governed by non-IS factors such as person/animacy (Fijian), evidentiality (Quechua), and so on. The next three chapters are dedicated to phonological analyses at their interface with information structure. They investigate the theoretical approaches that have been developed to deal with metrical and prosodic prominence in European languages. In chapter 22 (On the expression of focus in the metrical grid and in the prosodic hierarchy), Sara Myrberg and Tomas Riad address the relation between prosodic hierarchy and metrical grid structure that can change as a result of information structure. They compare English, a lexical stress language, and Swedish, a pitch accent language, and investigate nested focus in both languages, as in the so-called Superman sentences of Neeleman and Szendrői (Johnny was reading Superman to some kid) and Second Page 13 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Newcastle; date: 27 December 2018

Introduction Occurrence Focus (SOF). In chapter 23 (Focus, intonation and tonal height), Hubert Truckenbrodt is interested in the tonal effect induced by focus (F-marking in his account) and givenness. He concentrates on German and Mandarin Chinese. Besides showing the effect of focus prominence, a principle positing that focus attracts the strongest stress in the sentence, his main point is to show that focus immediately changes the height of accents without changing first the reference lines of prosodic phrases. He uses dual focus in Chinese for illustrations. In chapter 24 (Second occurrence focus), Stefan Baumann reviews the effect of Second Occurrence Focus (SOF) on the intonational correlates of focus and givenness. In the sentence A: Everyone knew that Mary only eats [VEgetable]FOF. B: If even PAUL knew that Mary only eats [vegetables]SOF, then he should have suggested a different restaurant, the expression vegetables in the second sentence is SOF because it is a focus associated with a focus sensitive operator only, and it appears for the second time in the discourse. Baumann surveys the theoretical questions (semantic and phonological) this phenomenon raises, as well as various analyses in the literature. In the last chapter of Part II (chapter 25 Information structure and language change), Regine Eckardt and Augustin Speyer examine the effect of IS on language change. They propose that the range of focus sensitive particles, the focus related syntactic patterns, and the alternative-based constructions can change with time, but that the invariant semantics of focus do not change. They illustrate their proposal with the contrast between V2 movement in Germanic languages, and its loss between Middle English and Early Modern English.

1.4.3 Part III: Experimental approaches to information structure Part III is an overview of the state of the art in the study of information structure from different experimental perspectives. It introduces studies from various sub-disciplines of linguistics that shed a new light on this research. Experimental perspectives have been another motor for theoretical advances. The eye-opening effect of experimental method for the sake of investigating possible structures has highlighted the complexity of linguistic behaviour with respect to IS. In a methodological shift, processing and (p. 13) cognition have entered the scene only recently, but these advances are progressively and radically changing the way we deal with theoretical issues. In chapter 26 (Information structure and language comprehension: Insights from psycholinguistics), Elsi Kaiser examines how psycholinguistics has studied comprehension of IS correlates, especially on how prosodic and syntactic cues are processed in real time. She reviews several experimental methods to this effect: timebased measures, attention-based measures, and off-line methods. Chapter 27 (Information structure and production planning) by Michael Wagner is complementary to chapter 26, as it concentrates on the production of IS cues from a psycholinguistic perspective. He distinguishes two views on the effects of IS: production planning models see IS effects as a consequence of contextual salience affecting the speaker’s lexical selection or Page 14 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Newcastle; date: 27 December 2018

Introduction grammatical functions assignment at the functional level, while linguistic models treat IS functions as part of grammatically encoded information which is processed incrementally in production. Chapter 28 (Information structure in first language acquisition) by Barbara Höhle, Frauke Berger, and Antje Sauermann, reviews the literature on the acquisition of linguistic means related to IS, especially production and the comprehension of accentuation, word order, and the effect of focus particles. In chapter 29 (Towards a neurobiology of information structure), Ina Bornkessel-Schlesewsky and Petra Schumacher investigate the neurophysiological and neuroanatomical correlates of IS notions and relate them to higher-order cognitive processing, like prediction and mental modelling, attention orientation, memory, and inferencing. The last chapter in Part III, chapter 30 (Corpus linguistics and information structure research) by Anke Lüdeling, Julia Ritz, Manfred Stede, and Amir Zeldes passes review of existing linguistic corpora for the study of IS. The authors show that there are two sensitive aspects for corpora: the design of a corpus that can serve as a basis for qualitative or quantitative studies and the problem of annotation.

Page 15 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Newcastle; date: 27 December 2018

Introduction

1.4.4 Part IV: Language studies on information structure Part IV groups chapters on different languages or language families. Each of the chapters examines how the different IS roles are reflected in the grammar of the particular languages (or language families). Most chapters compare the role of syntax, prosody, and morphology for the expression of topic, givenness, and focus, but they also address specific cases of IS when they use surprising or non-standard ways of expression. This part can be understood as test cases for the preceding chapters. It reveals the diversity of the means used by different languages. In chapter 31 (Syntactic and prosodic reflexes of information structure in Germanic) Gisbert Fanselow shows that despite evident differences among Germanic languages, common tendencies can be identified, like the preference for prosodic prominence in the final part of the sentence, and the fact that focus can be realized in situ, in which case it is realized with the highest prominence in its domain. However, some Germanic languages change word order as a function of IS more easily than others. In chapter 32 (Syntactic and prosodic effects of information structure in Romance) Giuliano Bocci and Cecilia Poletto review Romance languages in a cartographic approach. Different syntactic positions can have different IS roles. Dislocations are also investigated, as they are typical for Romance languages. Specific tonal analyses are proposed for Italian, which have been shown to realize information structural roles of constituents. Chapter 33 (Discourse functions: The case of Hungarian) by Katalin É.Kiss discusses the IS in the Hungarian sentence from a syntactic, semantic, and pragmatic perspective. She envisages the topic as the logical subject of the sentence, binding an empty argument in the main clause. Focus also gets a syntactic analysis, as it is necessarily derived by movement to a special syntactic position. Furthermore, it expresses exhaustive identification. In chapter 34 (Information structure in Modern Greek), Stavros Skopeteas reviews three properties of Greek IS: variable word order, free focal accent placement, and clitic doubling of DP arguments. It is shown that Greek is both similar and different from Germanic languages. Both left- and right-dislocations play a major role in the expression of IS in Greek. In chapter 35 (Information structure in Slavic) Katja Jasinskaja reviews the grammatical reflexes of IS in Slavic languages. She shows the effect of intonation and syntax for different IS roles. She shows the role of different tonal patterns in Russian, and also reflects on the role of full forms, clitics, and zeroes for the expression of givenness. Particles are treated in a separate section. She emphasizes that, even if Russian is better studied than the other Slavic languages, differences among languages abound and need more study. In chapter 36 (Topic and focus marking in Chinese), Yiya Chen, Peppina Po-lun Lee, and Haihua Pan show that, even though syntax and prosody compete for the expression of IS, they are usually complementary rather than mutually exclusive in Mandarin Chinese. Prosodic effects are pervasive. There are numerous syntactic and morphological means to identify a focus. The notion of topic is traditionally associated with the notion of subject, and Mandarin is a topic-prominent language. Chapter 37 (Information structure in Japanese) by Satoshi Tomioka, is dedicated to the expression of IS in Japanese. It is shown that Japanese is both similar (p. 14)

Page 16 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Newcastle; date: 27 December 2018

Introduction and different from English in several subtle ways: prosody and syntax are used in similar ways in both languages. However, the use of cleft sentences and of markers like wa is typical for Japanese, and express different kinds of IS roles. Tomioka also illustrates recursivity of information structure with Japanese examples. Chapter 38 (Information structure in Asia: Yongning Na (Sino-Tibetan) and Vietnamese (Austroasiatic)) by Alexis Michaud and Marc Brunelle, describes two understudied Asian languages, Na and Vietnamese, and emphasizes the diversity of means used for the expression of IS in these languages, which they describe as typical for the diversity among languages in general. Beside emphatic stress and word order changes, Na has a wealth of discourse particles used for the expression of topic and focus. Moreover, it also uses intonational and phrasing means for expressing givenness. Vietnamese, by contrast, has much fewer IS particles, but it extensively uses intonational means. For givenness, ellipsis is a common strategy. In chapter 39 (Information structure in Bantu) Laura Downing and Larry M. Hyman emphasize the diversity that Bantu languages present in their grammatical means for expressing focus. They are particularly interested in the prosodic, syntactic, and morphological reflexes, and suggest that information structure does not need special (p. 15) syntactic positions or special markers in prosody and morphology. This suggestion is illustrated among others with disjoint and conjoint verb allomorphs (Meeussen 1959), but also with word order and dislocations, all operations that are not exclusively used for IS. Other similar examples are nominal cases expressed by tones, as well as metatony, which typically have different functions, one of them being IS. The final chapter of the volume, by Vadim Kimmelman and Roland Pfau, chapter 40 (Information structure in sign languages) describes how IS is expressed in Sign Languages, that is, in languages using the visual-gestural modality of signal transmission, as opposed to the oral-auditory modality of spoken languages. The authors show that there are striking similarities in how these languages transmit IS. For instance, all languages use prosodic and syntactic means, and none of them uses morphological means. Non-manual marking (like raised brows) is commonly used.

Acknowledgements This Handbook is the result of a project which benefited from various kinds of support from many people. Without their help, this book would have never become a reality. We would like to thank everyone who has helped us. First of all, the authors of the chapters kindly agreed to our invitation, and contributed great chapters which deepen our understanding of IS and further stimulate our interests in the topic. Another group of researchers who contributed enormously to this volume and certainly deserve acknowledgement is the large group of anonymous reviewers of the chapters. We sincerely appreciate their cooperation.

Page 17 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Newcastle; date: 27 December 2018

Introduction This project originally started while both of us were still involved in the Collaborative Research Center (SFB632) ‘Information Structure’ at the University of Potsdam and the Humboldt University in Berlin. However, the major part of the endeavour took place at Goethe University Frankfurt am Main after we moved there. We would like to thank our colleagues at both institutes for their support and countless discussions. In May 2012, we organized a small workshop with some of the authors for the typological part of the Handbook (Part IV). We would like to thank the participants for stimulating discussions. We also thank Klaus von Heusinger and Ede Zimmermann for their useful advice in the early planning phase. Many thanks also go to our student assistants Julia Biskupek, Kerstin Czerner, Laura Hirtler, Stefanie Krkeljas, Alexandra Lowles, Melanie Lüttin, Johannes Messerschmidt, Katrin Müller, Katharina Schöning, Dominik Thiele, and Johannes Wolf for their diligent help with style, formatting, reference check, and indexing. Finally, we would also like to thank Oxford University Press (especially John Davey, Molly Davis, Karen Morgan, Julia Steer, Vicki Sunter, and Lauren Konopko) for giving us the opportunity to pursue this project, and their constant and generous support throughout. The people of Newgen also efficiently contributed to the elaboration of the final product.

Notes: (1) In chapter 33, É. Kiss also cites Sámuel Brassai (1860), a Hungarian linguist, who was interested in some aspects of word orders from the point of view of their givenness and newness (for which he used different terms). However, his work was never well-known outside of Hungary. (2) Chafe (1967) limits his discussion to noun phrases. So does Reinhart (1981), in her discussion of aboutness topics (see Section 1.3.4). But see, for example, chapters 15 by Lohnstein (on verum focus) and 16 by Zimmermann (on predicate focus) for cases where non-nominal elements are focused. (3) See Velleman and Beaver (chapter 5) for a concise overview of this distinction, and Horn (chapter 6) for cases that lie around the border of this distinction. (4) Morphosyntactically, topics may be presented via various strategies: specific syntactic position(s) (see, for example, Aboh, chapter 8, Surányi, chapter 21, É. Kiss, chapter 33, Chen, Lee and Pan, chapter 36, for relevant discussion), dislocations (López, chapter 20, Poletto and Bocci, chapter 32, Skopeteas, chapter 34), morphological marking (Tomioka, chapter 37, Michaud and Brunelle, chapter 38), or certain word order configurations (Neeleman and van de Koot, chapter 19, Fanselow, chapter 31). (5) But see Repp (chapter 14) for further discussion on the notion of ‘contrast’. Also, see Krifka for two related notions: frame setting and delimitation.

Page 18 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Newcastle; date: 27 December 2018

Introduction

Caroline Féry

Caroline Féry is a Professor of Phonology at the Goethe University Frankfurt. Her research is in phonology and theory of grammar with a special focus on intonation and prosody, as well as the interface with information structure. She is the author of articles in journals such as Natural Language and Linguistic Theory, Lingua, and The Linguistic Review, and is co-editor, with Malte Zimmermann, of Information Structure: Theoretical, Typological, and Experimental Perspectives (OUP, 2010). Shinichiro Ishihara

Shinichiro Ishihara is Senior Lecturer at the Centre for Languages and Literature at Lund University, having previously held positions at Goethe University Frankfurt, the University of Stuttgart, and the University of Potsdam. His research focuses on the syntax-prosody interface and its relation to information structure in Japanese and other languages. His work has appeared in international journals such as Lingua and Syntax and in edited volumes published by OUP and Mouton de Gruyter.

Page 19 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Newcastle; date: 27 December 2018

Alternative Semantics

Oxford Handbooks Online Alternative Semantics   Mats Rooth The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara Print Publication Date: Jul 2016 Subject: Linguistics, Semantics Online Publication Date: Feb 2016 DOI: 10.1093/oxfordhb/9780199642670.013.19

Abstract and Keywords This chapter presents the semantics and pragmatics of prosodic focus in alternative semantics. Half a dozen examples are given of empirical phenomena that are to be covered by the theory. Then a syntax marking the locus, scope, and antecedent for focus is introduced. The syntax is interpreted semantically and pragmatically by a presupposition involving alternatives. The alternative sets that are used in the definition are computed compositionally using a recursive definition. Alternatives are also employed in the semantics of questions, and this ties in with the phenomenon of question-answer congruence, where the position of focus in an answer matches questioned positions in the question. A different semantic interpretation for focus is entailment semantics, which uses a generalized entailment condition in place of a condition involving alternatives. The semantic and pragmatic interpretation for contrastive topic uses an additional layer of alternatives. Independent of focus, alternatives are deployed in the semantics of disjunction and of negative polarity items. Keywords: prosody, focus, questions, alternative semantics, contrastive topic, presupposition

2.1 Introduction ALTERNATIVE semantics is a semantic framework that finds application in the analysis of questions, focus, disjunction, negative polarity, presupposition triggering, and implicature. The unifying idea is one of semantic, pragmatic, or discourse-structural operations or constraints referring to ‘alternative’ phrasal meanings. This chapter presents the analysis in alternative semantics of prosodic focus. Some of the other applications are sketched in the last section.

Page 1 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics In English, German, Japanese, Korean, and many other languages, there are constructions, discourse configurations, and pragmatic interpretations that show a phonology and phonetics of prominence, and where a common semantic-pragmatic element related to alternatives and/or redundancy can be identified. We begin with examples of these.1 Scalar some. The existential determiner some is frequently used with a limiting implication. In example (1a), the speaker suggests that only some people can be easily eliminated. Sentence (1b) functions as a correction or hedge, admitting that the group that are supporting Mr Valentine do not necessarily include the public or all the public. In these utterances, which were made on sports talk shows, the word some is noticeably prominent. In sentence (1c) the word some was destressed, and here a limiting implication along the lines of ‘some but not many’ would not fit in, because it would undercut the positive message.2 (p. 20)

(1)

Questions and answers. When questions are paired with clausal answers, constituents in the answer that correspond to the WH position in the question are prominent. Thus in (2), the vertical pairings are well-formed question–answer pairs, and the diagonal ones are ill-formed. (2)

The correlation, which was discussed as early as Paul (1880), is called question–answer congruence. Descriptively, answer A is congruent to a question Q if and only if substituting wh-phrases for the focused phrase (or phrases) in A and then performing morphosyntactic adjustments such as wh-movement and do-support can result in Q. Comparatives.

Page 2 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics The four utterances in (3) embed the comparative clause than I did. The first two were spoken with prominence on the subject pronoun I, and the second two were spoken with the subject pronoun destressed. Utterances of this form are covered by a simple descriptive generalization: the subject pronoun is prominent if reference has shifted from the subject position of the main clause to the subject position of the than clause, and is destressed when reference is constant (Howell 2011).3

(3)

Korean indeterminates. Korean indeterminate morphemes such as nugu (‘who’, ‘someone’) are ambiguous between wh- and existential readings. The wh-reading is reported to have prosodic characteristics of focus, with pitch boosted on the (p. 21) indeterminate morpheme, and reduced pitch following. Sentence (4), which is quoted from Yun (2013), is ambiguous between the three readings in (5). Boosting of prosody on the indeterminate followed by reduced prosody for the rest of the sentence favours a wh-question reading.4 (4)

(5)

Substitution instances for quantifiers. This scenario comes up frequently in sports talk shows. In a discussion of a specific player, a generalization is stated that is understood to imply an application to the player as a substitution instance. When the phrase any player is used as the quantifier in this discourse configuration, the determiner any is markedly prominent. See the utterances transcribed in (6).

Page 3 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics (6)

Lists. List-structured phrases show quasi-predictable prominence patterns. Example (7) is a transcript of a listing of some radio station call signs and their home towns, as spoken by Scott Hollis, a DJ. The call signs were spoken as sequences of four-letter names, and are transcribed using capitals to mark prominence. Prominence falls on the letters where the current call sign differs from the previous one.5 (p. 22)

(7)

Contrasting antecedents. In (8), the prominence on the subject in the second sentence can be seen as motivated by contrast with the first sentence.6 (8) Farmer sentences. Farmer sentences are sentences with sentence-internal contrasting antecedents, where the contrast is at the nominal level rather than a clausal one, and which can show a dual, symmetric expression of focus. In (9) the first word in Canadian farmer is more prominent than the second word, even though the phrase has default prominence on the right. Optionally, prominence can be shifted in American farmer as well. (9) Accommodated contrasts. In many cases, a prominence shift seems to be motivated by a contrast with something that is not overt in the discourse. Example (10) is a statement by DJ Hollis at the end of a weekly programme. The word next was prominent. The statement does not require an overt antecedent along the lines of the statement that the DJ is here this week with three hours of great jazz. But the speaker intends for his listeners to accommodate this contrasting antecedent.

Page 4 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics (10)

Association with only. John introduced Bill and Tom to Sue, and there were no other introductions. In these circumstances, sentence (11a) is false, while (11b) is true. Contexts like this one where focus has an influence on truth value (or other aspects of compositional semantics, such as semantic presupposition) are known as association with focus contexts.7

(11)

Reasons. Dretske (1972) pointed out that focus has a truth-conditional effect in counterfactuals, descriptions of reasons, and further contexts that seem to involve underlying counterfactual reasoning. Assuming the situation in (12), in the sentences of (13) we observe variation in truth value that is conditioned by the location of focus, just as in the sentences with only. Dretske emphasized that such examples show that one has to (p. 23)

pay attention to focus in compositional semantics, not just at the discourse and pragmatic levels.8 (12)

Page 5 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics (13)

Multiple focus. Most of the types of examples discussed above work also with two or more focused phrases, rather than just one. In (14) we see two focused phrases, with a preceding contrasting antecedent sentence. Example (15) has two focused phrases in an answer to a question with two wh-phrases. Our definition of question–answer congruence already allowed for multiple Fs. (14)

(15)

This concludes the catalogue of examples. All of them, at least under many accounts, are instances of the same phenomenon of ‘focus’. This is essentially a grammaticallymediated correlation between a phonology-phonetics of prominence, and semanticpragmatic factors that are hypothesized to be common to the constructions and discourse configurations. (p. 24)

We turn now to a scheme for annotating three parameters of variation in focus:

the phrasal location of a focus, the scope of focus, and antecedents for focus. Jackendoff (1972) introduced the strategy and grammatical hypothesis of localizing focus on surface syntactic phrases, using a focus feature ±F. Usually only the positive value is indicated, using a subscript. When a phrase is F-marked, there is a prominence realizing the focus in the phonological realization of the phrase.9 The dialogues (16)–(17) motivate the notion of scope. Questions are numbered using subscripts, and various phrases that do or do not satisfy question–answer congruence relative to the question are marked with a brace. Congruence is annotated as ~k for the phrase being congruent to question k, and ≁ k for the phrase not being congruent to question k. The pair (16a,b) is the standard case of question–answer congruence. Statement (16c) is another response to the question (16a), also with focus on Justin. Although (16c) is not really an answer to (16a), we can assume that the focus is motivated by congruence to the question. While the entire sentence does not satisfy congruence Page 6 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics (pair (16a,d)), the embedded sentential subject does satisfy congruence with respect to the question (pair (16a,c)). In (17), the question is changed, with a resulting switch in the phrases that satisfy congruence. This time the entire sentence does satisfy congruence with respect to the question (pair (17a,c)), and the embedded sentential subject does not (pair (17a,b)).

(16)

(17)

In this way congruence provides motivation for the hypothesis that the scope of the focus in the dialogue (16a,c) is the embedded sentential subject, and in the dialogue (17a,c), the entire response. Remarkably, the scopes that result from considering congruence agree with a prosodic notion of scope. In the dialogue (16a,c), the embedded subject Justin is prominent, but it does not outrank the following predicate surprise me (p. 25) in prominence—the latter bears the nuclear accent of the sentence. On the other hand, in dialogue (17a,c), Justin does bear the nuclear accent. We can hypothesize that in the first dialogue, the phonological domain of prominence for the embedded subject Justin is the embedded clause, while in the second dialogue, the domain of prominence for Justin is the matrix clause. If so, phonological domains of prominence agree with the scopes that result from the assumption that congruence with respect to the question is to be satisfied. This phonological-semantic isomorphy is the strongest argument for the locus/scope/ antecedent grammar of focus. At this point we have arrived at the notation from Rooth (1992). The scope of F is marked with an operator ~k in surface trees, where k is the index of an antecedent with respect to which congruence is satisfied. The indices have the status of semantic indices, that is, indices which correspond to variables in semantics, and/or to discourse referents in a discourse representation. While in many cases antecedents correspond to an overt phrase, they can also be accommodated, as in the nextF week example.

Page 7 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics The significance of the scope of focus was realized relatively late. In Jackendoff (1972), it seems to be assumed that the semantic scope of focus in our sense is always the matrix sentence. In agreement with this, in the phonology, F-marked phrases take the matrix sentence as their phonological domain, because they bear a special stress feature that is not demoted in the application of cyclic stress rules in the system of Chomsky and Halle (1968). Rooth (1992) discussed non-maximal scope in farmer sentences, using the representation in (18). The isomorphy argument was developed in Truckenbrodt (1995), referring to farmer sentences.10 His point about (18) is that while Canadian is maximally prominent in its host nominal, it is not the location for the nuclear accent, which falls on joke. (18)

Notice that in (18), the notion of congruence has been generalized. The antecedent [an American farmer] is hypothesized to be congruent to the host phrase [a CanadianF farmer], but it is not assumed (or is not necessarily assumed) that the antecedent contributes a question, either directly or indirectly. A couple of notes are in order about the status of the notation and examples introduced above. The syntactic locus/scope/antecedent notation embodies a grammatical hypothesis, the adequacy of which is not taken for granted. The hypothesis has to be spelled out, notably by articulating the semantic/pragmatic and phonological interpretation for the syntax, and it has to tested against evidence and compared to competitors. The examples, in addition to orienting the reader, are intended as an ostensive definition of focus, or of a certain kind of focus. In this it is not assumed prior to analysis that these examples have the same underlying nature. If they do not, then they do not belong in the same theoretical box, and we should countenance several different theoretical notions of focus, or apply different terminology. Sections 2 and 4 of this chapter review analyses which do succeed in identifying a shared deep commonality in the constructions and configurations listed above. (p. 26)

By the way, not every construction in every language that shows question–answer congruence is necessarily an instance of the kind of focus under discussion here. The arguments reviewed in É. Kiss (2010 and this volume) indicate that the structural or movement focus found in Hungarian has a distinct semantics from English-type prosodic focus, and a distinct distribution. Yet movement focus is used in default answers in Hungarian, and shows question–answer congruence. In general, we should resist giving any kind of substantive definition of focus prior to analysis, referring either to question– answer congruence, the evocation of alternatives, or a broadly information-theoretic notion of the focus being unpredictable relative to the rest of the material in the scope of the focus. The fundamental problem with starting in this way is that it prejudges the issue

Page 8 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics of what the optimal theoretical account is. Interesting terms in scientific theories do not have non-ostensive definitions that are independent of theories.

2.2 Semantic interpretation Consider (19), where the clause [ϕ ~ k] embeds a focused phrase and is indexed to a preceding contrasting clause. We assume a system of interpretation where clauses semantically contribute propositions, for example as constructed in possible worlds semantics. Suppose discover is a two-place function from individuals to propositions. Then the antecedent in (19) denotes the proposition discover(n,c), and the host clause for the focus denotes the proposition discover(l,c). Informally, the antecedent proposition is an alternative to the host proposition that can be obtained by making a ‘substitution’ in the position of the focused phrase. This is the core idea of alternative semantics: a legitimate antecedent for focus denotes an alternative to the scope of the focus, or as we will see in a moment, a set of alternatives.

(19)

The notion of making substitutions in the focus positions of propositions need not be taken literally. For one thing, propositions as constructed in possible words semantics (p. 27) do not have positions—they are unstructured sets of worlds. For another, some way of tracking the focus positions from the syntax to the semantics is needed. For now, we will just assume that the semantics of the language makes available, in addition to an ordinary semantic value (which is a proposition in the case of a clause), a set of eligible alternatives, which for a clause is a set of propositions. The set of alternatives is a ‘focus semantic value’, or ‘alternative semantic value’. The focus semantic value for ϕ in (19) can be expressed formally as in (20), using set abstraction.11 It is the set of propositions that can be obtained by plugging in some individual for y in the open proposition-naming term discover(y,c). Hamblin (1973) introduced a useful informal way of naming such alternative sets. We say that the alternative set is the set of propositions of the form ‘y discovered calculus’. (20)

Page 9 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics Now we are ready to formalize this simple idea: a phrase of the form ϕ ~ k is associated with the constraint that the antecedent k is an alternative to the semantic object contributed by ϕ. The set of eligible alternatives is ⟦ϕ⟧f, so we require that the semantic element k be an element of the alternative set. Rule (21) says in addition that k should be different from the ordinary semantic value ⟦ϕ⟧o . (21)

The terminology ‘the semantic element k’ is explained in the same way as with other varieties of indexing, such as indices on traces and pronouns. In a standard formulation of a static semantics, values for variables are given by assignment functions, so the semantic object is g(k), where g is the assignment function. It is natural to think of the index k as a discourse referent in a discourse representation, which may be projected from a syntactic index, but may also be constructed. The framework should make available discourse referents of all types, including propositions. So the picture is that (19a) sets up a propositional discourse referent 2, which is used as an antecedent in checking the focus constraint for the second sentence. So while ϕ ~ k is a piece of syntax, the focus constraint is checked semantically. The licensing condition (21) covers cases where the antecedent has the same type as the host phrase for the focus. When the host phrase is a clause and has the propositional type, the contrasting object is a proposition. When the host phrase denotes a generalized quantifier as in the farmer sentence (9), the antecedent is also a generalized quantifier. This does not work when the antecedent is a question and the host phrase is a declarative answer, (p. 28) because questions and statements have different semantic types. This brings up the connection between alternative semantics for focus and the alternative semantics for questions that was proposed in Hamblin (1973). Hamblin’s semantics can be viewed as being motivated by the principle that any viable semantics for questions must be capable of characterizing what counts as an answer. One way of meeting this constraint is to take the semantics of a question to be a set of propositions, the set of atomic answers to the question, independent of the truth of the answers. So the semantic value of the question (22a) is the set of propositions of the form ‘y invented calculus’, where y is a person. This is nearly the same as the focus semantic value of the answer (22b). This suggests generalizing the congruence condition, so that the antecedent can be a set of alternatives, rather than a single alternative. Rule (23) is a way of stating this. The idea is that since the alternative propositions in the semantics of the question are restricted to people, while the alternatives in the focus semantic value are unrestricted, the antecedent is a proper subset of the focus semantic value. Rooth (1992) also included the condition that the ordinary semantic value of the scope of the focus is an element of the antecedent, and that the antecedent has cardinality of at least two. These conditions, which in a way correspond to the distinctness condition in part (i) of (23), are included in part (ii).

Page 10 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics

(22) (23)

A different way of going is to set up the semantics so that the focus semantic value itself gets restricted. Suppose that in generating alternatives for (22b), only people are substituted for Leibniz, so that we get the set of alternatives of the form ‘y invented calculus’, where y is a person. Then the focus semantic value of the answer would match the semantics of the question exactly.

2.3 Composing alternatives The interpretation principle for the focus scope configuration [ϕ ~ k] refers to the alternative set ⟦ϕ⟧f for the phrase ϕ. This section looks at how the alternative set is derived. Hamblin (1973) and Rooth (1985) suggested a recursive strategy: there are alternatives ‘all the way down’, and the alternatives propagate up the tree. Figure 2.1 shows a binary-branching tree for sentence (24). This is a multiple-focus example, where alternatives are generated from the focused subject ArethaF, and the focused object ClydeF. In the tree, each node is annotated with its ordinary semantics, and below that, an alternative set. To control the size of the sets, we assume that there (p. 29) are just three entities in the model, namely entity a (Aretha), entity b (Bertha), and entity c (Clyde). At the top, we see the ordinary semantics introduce(a,b,c), and an alternative set that contains alternatives such as introduce(c,b,a). There are nine alternatives, resulting from multiplying a choice among three in the subject position by a choice among three in the object position. This is the set of propositions of the form ‘y1 introduced y2 to Clyde’ in this simple model. Looking at the bottom of the tree, the alternative set for the focused subject is {a,b,c}, the set of individuals in the model. The same goes for the focused object. This is how alternatives are ‘launched’: the alternative set for a focused phrase (whether it is a terminal or not) is the set of semantic objects that match the ordinary semantic value of the phrase in type. Non-focused terminals give a trivial alternative set, namely the unit set (singleton set) of the ordinary semantic value of the phrase. Since ⟦Clyde⟧ is c and the phrase is not focused, its focus semantic value is the unit set {c}.

Page 11 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics

Click to view larger Figure 2.1 Recursively computed alternatives in a syntactic tree. Each node is annotated with its ordinary semantics, and below that a focus alternative set. The domain of individuals is {a, b, c}.

(24)

If a complex phrase is focused, it goes by the rule already given—its alternative set is the set of semantic objects matching the ordinary semantics of the phrase in type. So for (p. 30) instance, the alternative set for a focused VP is the set of all properties. Alternative sets for non-focused complex phrases are derived from the alternative sets for their children. Suppose we have a complex phrase with children α and β. Say α is semantically the function, so that the ordinary semantics is formed as ⟦α⟧(⟦β⟧). If we pick any element f of ⟦α⟧f, and any element y of ⟦β⟧f, they can also be combined as function and argument. The alternative set for [αβ] is the set of all semantic objects f(y) that can be formed in this way. Example (25) illustrates this for the phrase [introduce BerthaF], where the verb is the function. There is one choice for f in ⟦introduce⟧f, and there are three choices for y in ⟦BerthaF⟧f. From these three semantic objects can be formed as f(a), as shown in (25).

Page 12 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics

(25)

Generalizing a bit, let F be the semantic rule that is used to combine ⟦α⟧ with ⟦β⟧ in the phrase [αβ]. Then ⟦αβ⟧f is defined to be the set of all semantic objects that can be formed as F(x,y), where x is an element of ⟦α⟧f and y is an element of ⟦β⟧f.12 In this case ⟦introduce BerthaF⟧f is obtained as the image of the rightward function-application operator fRFA acting on

.

Another way of defining alternatives is to introduce variables in the F positions that have the same type as the ordinary semantic value (Kratzer 1991; Wold 1996). For this purpose a separate family of variables is used. Focus semantic values have the same type as ordinary semantic values, and have variables in the focus positions. The table in (26) illustrates this for the example in Figure 2.1, using the u variables with subscripts for the special focus variables. In this system, there are no alternatives at recursive levels. Instead, alternatives are introduced in defining the ~ operator, by making substitutions for the special focus variables. (26)

Either of these approaches provides a workable solution to the problem of defining focus alternative sets, in the service of making available the semantic objects that are used in the definition of the ~ operator. (p. 31)

Hamblin applied alternative semantics to questions. This works in the same way as alternative semantics for focus in the recursive part. There is an issue though about how to treat the restrictive property of the wh-phrase, which interacts in a subtle way with question–answer congruence. Consider the examples in (27), where the wh-pronoun has a lexical restriction to persons, the wh-phrase what dog has a restriction to dogs, and the Page 13 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics wh-phrase what Thai restaurants has a restriction to Thai restaurants. One option is to equate the alternative set for the wh-phrase with the extension of the restriction, so that ⟦who⟧f is the set of people.

(27)

In this passage, Hamblin describes this ‘restricting’ option, argues against it, and proposes an alternative ‘conjunctive’ option. We would like to think that the phrase what dog could be treated as an interrogative proper name denoting the set of dogs, and that what dog walks with Mary has as answers just the set [of propositions of the form] ‘x walks with Mary’ where x is the name of a dog. But the composition of the set of dogs does not necessarily remain constant from universe to universe: in some universes Rover may be a horse, and Mary herself a dog. I have taken the attitude that when someone answers what dog walks with Mary with Rover he states not merely that Rover walks with Mary but also implicitly that Rover is a dog, and hence that he states the conjunction. [Hamblin 1973: 51, with adjustments in quotation styles.] The proposal then is that the alternative set for (27a) is the set of propositions of the form ‘x is a person and x is awake’, and that the alternative set for (27b) is the set of propositions of the form ‘x is a dog and x walks with Mary’. This can be formulated by defining the alternative sets for who and what at the generalized quantifier level as in (28).

(28)

The conjunctive interpretation is not compatible with the theory of question–answer congruence from Section 2. The representation (29) is licensed if the alternative value for the question is a subset of the alternative value for the answer. In any realistic model, it is not, because for instance ‘Justin is human and walks with Mary’ is an element of the alternative value for the question, but not an element of ⟦JustinF walks with Mary⟧f, which is the set of propositions of the form ‘x walks with Mary’. (p. 32)

(29)

Treatments of question–answer congruence in alternative semantics have assumed the restrictive strategy in the semantics of questions (Rooth 1992). But what about Hamblin’s point that there is no stable set of propositions of the form ‘x walks with Mary’, where x is a human, because the sets of humans are different in different possible worlds? Relatedly, Page 14 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics there is a worry that the focus constraint is trivialized in certain cases. In a world where there are no nuclear engineers, the denotation of the question (30a) is the empty set. Therefore the subset constraint coming from the focus interpretation in (30b) is trivial. This is not quite a problem, because the oddity of the pair (30a,b) can be attributed to the logic and pragmatics of the question–answer relation, rather than anything about focus. In any viable account of the question–answer relation, the proposition contributed by (30b) is not an answer to the question (30a). So to rule out this question–answer dialogue, it is not necessary to appeal to the focus presupposition. But if we fix this by adjusting the response as in (30c) (perhaps the cat being on the mat would prevent any dancing), focus in the cat-clause is still not licensed.

(30)

The right move here is to accept that question alternative sets are world-dependent. This is explicit in the analysis of indeterminate pronouns in Kratzer and Shimoyama (2002). They give the semantics (31) for the Japanese indeterminate pronoun dare, which has a use as a wh-pronoun. Here the alternative set of individuals contributed by dare varies from world to world, since there is reference to the world of evaluation w in the restrictive clause of the set abstraction. (31) Using the same definition for who results in the denotation given in (32a) for the question in (29a). Here I have switched to a notation for propositions that has explicit reference to worlds. λυ.walkwith(υ,y,m) is the set of worlds υ in which y walks with Mary; this was earlier written walkwith(y,m), without any commitment to modelling propositions as sets of worlds. The set (32a) is world sensitive, because there is reference to the world of evaluation w in the restriction of the set abstraction. (32)

(32b) gives the focus denotation for the answer in the same notation. This set is not world sensitive. Nevertheless for any world w, the question alternative set (32a) is a subset of the (p. 33) focus alternative set (32b). Applying the constraint in each world is exactly what falls out if the congruence constraint is modelled as a presupposition. For comparison, consider the semantics of the second sentence (33) in an extensional semantic model of presuppostion. The sentence intuitively presupposes that Kim is male. This is captured in an extensional model by the assignment of the third (undefined) truth value when Kim is not male in the world of evaluation. While this semantics of presupposition is extensional, in the pragmatics, a common ground model of Page 15 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics presupposition can be applied, by checking the presupposition in each world that is an element of the context set of the common ground. This captures the fact that the discourse is perceived as taking for granted that Kim is male. (33)

So, if it is formalized semantically as a distributive presupposition that places a constraint on a world, it falls out of the pragmatic interpretation of such presuppositions that the constraint gets imposed in all worlds of the context set. This addresses the worries about the question alternative set not being stable, and about the focus constraint being trivial in some cases. Summing up, to fit alternative semantics for focus in with a generally Hamblin-like alternative semantics for questions, the restricting version of Hamblin semantics for questions has to be used. Hamblin’s official conjoining option does not result in values that satisfy the presupposition of the focus reduncancy operator in configurations of question–answer congruence.

2.4 Entailment semantics According to Section 2, the presupposition of the focus-scope operator is satisfied when the indexed antecedent is an alternative to the semantic value of the argument of the operator, or a set of alternatives. This section looks at a different licensing condition due to Schwarzschild (1999). This says that the semantics of the antecedent should entail a semantic object derived from the focus semantic value of the scope of the operator, in a certain generalized sense.13 Consider the example repeated in (34). The first step is to derive the focus closure from ϕ. This is a proposition with existentially quantified variables in the position of each free F. Formally, it can be obtained as the (p. 34) union of the focus semantic value, ∪⟦ϕ⟧f. Next, entailment is checked between the antecedent and this closure. In this case, the antecedent proposition ‘Newton invented calculus’ does entail the closure ‘some entity invented calculus’, and the representation is licensed. (34) So, in cases where both the antecedent and ϕ have propositional type, the presupposition introduced by the configuration [ϕ ~ k] is that the antecedent proposition entails the focus closure for ϕ. Look now at (35), where for the sake of argument we assume that the focus on Canadian takes scope at NP. This constituent has the property type, rather than a propositional one. Here it is proposed that the property type is lowered to the propositional one, by existentially quantifying the argument. The same type adjustment is

Page 16 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics performed for the antecedent. The result in this case is the proposition ‘there are American farmers’ for the antecedent and for the scope (applying also focus closure) ‘there are farmers’. The former entails the latter, and the constraint is satisfied. (35) Neatly, type shifting works out even in question–answer congruence, where the antecedent is a question, providing that Karttunen’s semantics for questions is used, where the question alternative sets are restricted to true propositions (Karttunen 1977). In (36), the existential closure of the antecedent relative to a base world w is the proposition that there is some proposition of the form ‘y invented calculus’ that is true in w such that y is a person in w. This is equivalent to ‘some person invented calculus’, while the focus closure for the scope is ‘some entity invented calculus’. Since these stand in the relation of entailment, the indexed redundancy operator is licensed. (36) So far we have not seen any cases where entailment licensing and alternative licensing give different results. A simple one is where the antecedent is existentially quantified (Rooth 2005). Since somebody eating the cake entails some entity eating it (the latter is the focus closure), the representation (37) is licensed. But if we try alternative licensing, the proposition ‘somebody ate the cake’ is neither a proposition of the form ‘y ate the cake’, nor a set of propositions of that form.14 (37) For an argument the other way, we can exploit the fact that focus and existential closure of the scope result in weak existentially quantified propositions, so it can be too easy for an arbitrary antecedent to entail them. In (38b), there is focus on the noun cat, with sentence scope. In (38c), there is focus within the subject generalized quantifier, and taking scope at it. In both cases, the closure procedure results in trivial propositions. (39a) is the focus closure obtained in (38b). This is necessarily true. Here is the reason. Every entity that tasted the flounder fillets tasted the flounder fillets. Therefore the property of tasting the flounder fillets serves as a witness for the existential quantifier ∃P. Since (39a) is necessarily true, it is entailed by any antecedent, and the licensing condition for the pair (38a,b) is satisfied. This is a bad result. In the pair (38a,c), argument closure as well as focus closure apply, because the scope is a generalized quantifier, rather than a proposition. This results in the closure (39b), which by similar reasoning is trivially satisfiable. So the representation (38a,c) is licensed, again a bad result. (p. 35)

(38)

Page 17 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics

(39)

Summing up, generalized entailment is another proposal for the presupposition of ~k. Relative to the alternatives presupposition from Section 2, it has some advantages and some disadvantages. There is a need for deeper investigation of this issue.

2.5 Topic alternatives This section looks at an extension of alternative semantics that posits a higher level of alternatives, consisting essentially of alternatives to alternative-sets. This was proposed in Büring (1997) as a way of theorizing about the discourse pragmatics of contrastive topics. As discussed in Büring (this volume) and Velleman and Beaver (this volume), discourses involving multiple wh-questions and sub-questions can trigger answers that combine focus with an additional prominent element, called the contrastive topic. In sentence (40), focus on Manny correlates with the wh-position in the immediately preceding question. Contrastive topic on the subject is somehow triggered by the complex discourse context. Example (41) illustrates that the pattern can be reversed, with focus followed by contrastive topic (Jackendoff 1972). Contrastive topic and focus strike many speakers as different prosodically, and experimental work has shown that they can be pronounced in ways that differ consistently in pitch contour (Liberman and Pierrehumbert 1984). Current accounts that use the phonological tone model from Pierrehumbert (1980) transcribe the topic constituent L+H* within (p. 36) a phrase with L–H% boundary tones, and the focus as H* within a phrase with L–L% boundary tones.15

(40)

(41)

In (40) and (41), the answers are congruent with the questions they answer, if CT is ignored. Therefore it can be hypothesized that the CT feature does not affect focus semantic values, and that congruence is represented with indexing to the question as before. Suppose the extra contribution of contrastive topic is conceptualized as a contrasting question, which is also represented with an index. The contrasting question stands in a relation of substitution contrast with the focus antecedent, with substitution in the CT position. This leads to representations like (42) and (43), with an additional argument of the focus redundancy operator. In both cases, the question with index 4,

Page 18 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics namely ‘who did Keisha come with’ can be obtained from the question with index 2, namely ‘who did Anna come with’, by making a substitution in the CT position.

(42)

(43)

In each of these representations, one antecedent is overt, and the other accommodated. In dialogue (42), we can conceive of the question with index 4 as a question which remains open, under the general discourse topic of who came with whom. In dialogue (43), the respondent switches the discourse topic from the question of who Keisha came with to the question of who Anna came with. The focus antecedent is accommodated, while the contrastive topic antecedent is the overt question. Adding CT adds another dimension to the compositional problem. Büring (1997) followed the recursive strategy by defining an additional alternative semantic value ⟦⋅⟧t. The focus semantic value of [AnnaCT came with MannyF] can be conceptualized as the Hamblin question ‘Anna came with what entity’. The topic semantic value is then (p. 37) obtained by making substitutions in the CT position, yielding the set of Hamblin questions of the form ‘z came with what entity’, where z is an individual. (44) gives recursive clauses defining topic semantic values.16 The third line defines the topic semantic value for a topic-marked phrase as the set of singleton sets of elements of the semantic domain for the phrase. In our example, this generates questions with individual substitutes for the topic-marked phrase subject, for instance ‘Keisha came with what entity’ and ‘Hannah came with what entity’. The second line defines the topic semantic value of an F-marked phrase as the unit set of the focus semantic value. This has the effect of making elements of the topic semantic value look like the focus semantic value in F positions. For instance, the element ‘Hannah came with what entity’ of the topic semantic value has variation in the position of the object of with, just like the focus semantic value ‘Anna came with what entity’.

Page 19 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics (44)

The two-place redundancy operator can now be defined as in (45). j is the index of the local question. As before, this question is presupposed to be a subset of the focus semantic value. k is the index of the contrasting question. The contrasting question is selected from the topic semantic value, which can be thought of as the set of potential contrasting questions. The subset condition is included to allow for contextual narrowing of the contrasting question, just as for the congruent question. (45)

This formulation allows possible contrasting question antecendents to be characterized semantically, just like focus antecedents, and licenses the representations (42) and (43)17. This assumes that the presupposition associated with CT is satisfied by the presence in the discourse representation of a question that satisfies the semantic constraints (p. 38) on the antecedent. Much of the literature, however, works with more specific constraints which say that question antecedents have to be in specific positions in a tree-structured model of discourse state. See Roberts (2012), Büring (2003 and this volume).18 This is a good place to talk about the discourse pragmatics that goes with the locus/ scope/antecedent representation for focus from Section 2.2, and the semantic interpretation for it from Sections 2.2 and 2.4.19 Those interpretations took the form of presuppositional contraints on the antecedent k in the configuration [ϕ ~ k]. A constraint is placed on the semantic object k: it is required to be an alternative to ⟦ϕ⟧ in the sense defined by the focus semantic value, a set of alternatives, or in the formulation from Section 2.4, to entail the focus closure for ϕ in a generalized sense of entailment. Eligible antecedents for k are discourse referents that have the right semantic type, and which satisfy the focus constraint. Pragmatically, a speaker who uses a sentence containing focus and ϕ ~ k signals an intent to assume a discourse representation containing an antecedent that is identified by indexing, which should be salient in the same way that antecedents for pronouns and anaphoric definite descriptions are salient. Discourse Page 20 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics referents that provide antecedents for focus can be projected from syntactic phrases. They can also be constructed, as exemplified by the accommodated question antecedents marked with parentheses in (42) and (43).

Page 21 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics

2.6 More applications of alternative semantics The discussion so far has looked at the application of alternative semantics to intonational focus, with a short consideration of alternative semantics for questions. The same or a similar framework is applied to other phenomena in semantics and pragmatics. Aloni (2003), Simons (2005), and Alonso-Ovalle (2006) have presented analyses in which natural language disjunctions contribute alternatives. The analysis in Alonso-Ovalle (2009) of conditionals such as (46a) is representative. It is argued that combining a boolean semantics for or with a minimal-change semantics for counterfactuals (Stalnaker 1968; Lewis 1973) gives bad results, because on reasonable assumptions about what worlds count as ‘normal’ or similar to our own, worlds where we have good (p. 39) weather are much more normal than ones where the sun grows cold. With this background assumption, sentence (46a) comes out as equivalent to (46b). Instead it seems to be equivalent to (46c). (46)

Alonso-Ovalle analyses these data using the hypothesis that the material inside the ifclause, instead of contributing a proposition, contributes the set of propositions containing ‘we have good weather this summer’ and ‘the sun grows cold’. Then a quantificational semantic rule for the way that if combines with the clause headed by would is stated, which has the effect of using the elements of the alternative set independently as restrictions for would, in order to obtain a reading equivalent to (46c). Krifka (1995) introduced the hypothesis that ‘weak’ negative polarity items such as any introduce alternatives into the semantic derivation. Negative polarity items (NPIs) occur in restricted environments, characterized by Ladusaw (1979) as downward entailing environments. These include the environments (47a–c), but not the hash-marked environments (47d–e), where the sentence is possible only with a generic interpretation. The alternatives generated by [any NP] are those obtained from sub-properties of the property contributed by the NP. For the case in (47), these are sub-properties of ⟦cookie⟧, for instance the property of being a square cookie, the property of being a poisonous cookie, and the property of being a non-square cookie. These alternatives propagate in the way described in this chapter. On the assumption that any is semantically an existential quantifier, (48a) is the ordinary semantics of the clause (47e). In addition we get the set of alternatives of the form described in (47b).

Page 22 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics

(47)

(48)

The remaining part of the theory is a principle that restricts the distribution of NPIs by referring to alternative sets. For Krifka, this is a principle of scalar assertion, which says that all alternatives that are not entailed by the ordinary semantics are false. For (47e) this turns out to be a contradictory requirement, which is hypothesized to be responsible for the oddity of the sentence. If Justin ate some cookies, then either he ate some (p. 40) square cookies, or he ate some non-square cookies. So these two alternatives (neither of which is entailed by the ordinary semantics) could not both be false. On the other hand, looking at (47b), nobody eating cookies is perfectly compatible with nobody eating square cookies and nobody eating non-square cookies.20 These applications of alternative semantics are formally similar to alternative semantics for focus, in that they involve alternatives that are launched from specific positions, propagated compositionally, and then interpreted by certain operators. Currently there is not enough understanding of how these systems of alternatives relate to each other in the grammatical system. One issue is that disjunction and weak NPIs usually show default prosody or are de-stressed, so that it is difficult to analyse them as systematically involving focus of the prosodic kind.

Notes: (1) Audio recordings are included in Rooth (2015). (2) These examples are drawn from the study in Chereches (2014) of the pragmatics and acoustics of about two hundred tokens of some from sports talk shows. (3) Howell looked at several hundred examples of this form from online sources. Listeners who were asked to naively classify prominence listening just to the three words than I did behaved in a way that is consistent with the reference criterion about 90 per cent of the time. Machine learning classifiers trained to make the reference-shift decision based on acoustic features in than I did, such as duration, formant spread, and pitch are able to make the reference-shift decision with about 90 per cent accuracy from the acoustic features.

Page 23 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics (4) In a perception experiment, Yun presented listeners with tokens of such sentences with manipulated pitch contours, and asked them to evaluate fit with parapraph contexts that favoured one reading or the other. The wh-reading correlated with the focus-like prosody of boosting and subsequent reduction. (5) This is obvious in [wsQg], where the prominence is non-final. Probably [wskg fm] is a unit with default prominence to the right, so that even in the first call sign, prominence has shifted to the left. In some cases the town names also sound like they have extra prominence. (6) See Repp (this volume) on phenomena and analysis of this kind of use of focus. (7) See Beck (this volume) on the analysis of association with focus effects for only and similar operators. (8) (12)–(13) are a version of an example of Dretske’s, and are given in this modified form in Rooth (1999). (9) Thus the F feature and its scope marker are interpreted phonologically as well as semantically. The phonological interpretation in terms of prominence in a metrical grid that is presented in Myrberg and Riad (this volume) is compatible with the representational scheme introduced here. (10) The analysis of focus in Chomsky (1971) referred to representations where the scope of focus is represented by what amounts to movement and bound variables, corresponding to LF movement in subsequent theory. This creates the potential for submaximal scope, because movement can be to a sub-maximal level, such as an embedded clause. But sub-maximal scope was not discussed. Jacobs (1983), von Stechow (1991), and Rooth (1985) gave semantically oriented accounts that generate sub-maximal focus scopes, while hardly talking about examples with embedded scope. (11) The ‘destructuring’ set abstraction notation {discover(y,c)|yϵDe} is potentially ambiguous, because one has to know whether y is allowed to vary, or is held constant. In this chapter, all free variables before the bar are allowed to vary over the combinations of values that satisfy the constraint after the bar. (12) Mathematically, this is an image construction (Hamblin 1973: fn. 8). In general the image of a function h on a subset Z of its domain is the set of values h(z) that can be obtained by applying h to an element z of Z. In (25), ⟦[αβ]⟧f is the image of the rightward function application operator on the cross product ⟦α⟧f × ⟦β⟧f.

Page 24 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics (13) Schwarzschild’s proposal was cast in a different way from the framework under discussion here. See Rochemont (this volume) for a presentation of this Givenness framework. A couple of aspects of Schwarzschild’s proposal are independent of the entailment condition. The most important of these is the hypothesis that the redundancy operator need not scope over F. In (34), the pronoun it is destressed, and this is attributed to the pronoun being ‘given’ relative to its antecedent calculus, in the same sense that the scope of an F is given or redundant with respect to its antecedent. Under the scheme presented here, the pronoun is represented as [it4 ~ 4], with the operator ~4 not scoping over F. (14) This counterexample is undone if existential quantifiers contribute alternative sets. See Section 2.6. Also, it is somewhat plausible to posit an accommodated question ‘who ate the cake’ following the first sentence. This would serve as the antecedent. (15) Among studies concerned with the semantics interface for prosodic phonology, see Steedman (2000, 2014), Büring (2003), and Constant (2012). (16) Section 3.3.1 of Büring (1997) gives a definition along these lines that also deals with free variables and lambda. (17) Another feature of recent work is the hypothesis that both the local and contrasting questions are drawn from the topic semantic value, using the fact that by construction, the focus semantic value is an element of the topic semantic value (Büring 2003). This move is motivated by a desire to characterize possible antecedent questions in a maximally general way. (18) An alternative to the recursive dual-layer system of alternatives is given by Constant (2012), who works from a representation with a CT head that is realized by the contrastive topic boundary tones. It has two arguments, each of which embeds F constituents that are interpreted in alternative semantics in the normal way. In the structure in (i), both the contrastive topic and the focus are analysed as bearing F features. The distinction between the two is encoded in the semantics of CT, which manipulates focus semantic values. ((i)) (19) Sæbø (this volume) articulates a discourse-pragmatic model along these lines. (20) Subsequent literature has suggested that the scalar exclusive operator is a phonetically null operator in the compositional structure, similar to only (Chierchia 2013).

Mats Rooth

Page 25 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Alternative Semantics Mats Rooth is Professor in the Department of Linguistics and the Faculty of Computing and information at Cornell University. He does research in computational linguistics and natural language semantics. He has worked on statistical models of parsing and the lexicon, the semantics of focus, and on related phenomena such as ellipsis and presupposition. In addition to these, he is currently working on finite state optimality theory and web harvesting of intonational data.

Page 26 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Uppsala Universitetsbibliotek; date: 23 October 2018

Givenness

Oxford Handbooks Online Givenness   Michael Rochemont The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara Print Publication Date: Jul 2016 Subject: Linguistics, Semantics Online Publication Date: Sep 2015 DOI: 10.1093/oxfordhb/9780199642670.013.18

Abstract and Keywords Distinguishing between two forms of givenness status, known and salient, this chapter investigates the latter, using deaccenting as a probe into the nature of salience-based givenness. A presuppositional account of salience-based givenness is presented, based on entailment and coreference. Other putative semantic relations claimed to underlie givenness-based deaccenting are shown to be inadequate. The question whether givenness can be reduced to focus is considered, with motivation provided for distinguishing among given, focused, and discourse new. It is seen that the distribution of accenting and deaccenting in English is only partly a function of context, with speaker’s communicative intent playing a critical and unpredictable role. Keywords: givenness, deaccenting, focus, presupposition, discourse new, communicative intent

3.1 Introduction GIVENNESS has many forms and functions.1 In all its uses, givenness ties some linguistic property of an expression, its syntactic form or position or its prosody, to the informational or cognitive status of its denotation as already present in the discourse model in some sense. Differing uses reflect the differing ways in which an expression or its denotation may be said to be already present. Consider, for instance, the two selfcontained mini-discourses in (1), where (1a) is followed by either of (1b,c) (Chafe 1976: 41).

Page 1 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness (1)

Proper nouns, such as those in (1a), usually identify entities that the speaker takes to be already known to the other participants in the conversation. Such assumed familiarity is critical to the speaker’s successful communication of the content of the proposition to her audience. Generally, the felicitous use of names and definite descriptions (the beach, the beer) requires that the entities they designate be familiar or at least uniquely identifiable. For the participants of the two discourses in (1), John and Mary must be ‘given’ in this sense, as must the beach and the beer. The expression the beer in the second conjunct of (1b) uniquely identifies a referent through ‘bridging’ (Clark 1977) to the picnic supplies introduced in the first conjunct: the beer is understood as one of the supplies just mentioned, given the interlocutors’ general knowledge of what sorts of supplies (p. 42) picnic supplies may consist of. This sense of givenness (familiar, identifiable) is a distinguishing property of definite noun phrases generally, and various types of (in)definiteness marking signal the varying degrees to which, or manner in which, the referent of a particular noun phrase may be reliably tied to the background of commonly held knowledge of individuals, events and meanings shared by the participants in a specific discourse (e.g. Prince 1981; Ariel 1990; Gundel et al. 1993). But there is also a different sense of givenness exemplified in (1). As Chafe notes for parallel examples, there is a contrast in pronunciation between the second conjuncts of (1b,c). These sentences, although segmentally identical, form a minimal pair: in (b) beer is intonationally prominent (pitch accented), while in (c) beer is deaccented (it shows a complete lack of pitch prominence). Patently, what makes deaccenting possible in the second conjunct in (c) is the prior mention of beer in the first conjunct.2 This requirement for a situationally salient antecedent for the deaccented expression also reflects a form of givenness. But since both instances of the beer in (1b,c) are definite, deaccenting must reflect a different notion of givenness than that which marks the use of a non-pronominal definite noun phrase. In particular, no deaccenting of the beer is possible in the second conjunct in (b) despite the ready accessibility of a uniquely identified referent. One way in which shared knowledge is represented is through Common Ground (CG), a notion originally proposed by Robert Stalnaker in the early 1970s. CG is the set of propositions and entities that are shared among interlocutors in a conversation and whose content is updated as the conversation proceeds. Krifka (2008) distinguishes CG Content, construed in this way, and CG Management.3 CG Content records and updates the shared knowledge and beliefs of interlocutors, while CG Management is concerned with the immediate and temporary informational needs and communicative goals of interlocutors, governing how CG Content should develop. Adapting Krifka’s distinction between CG Content and Management, a plausible hypothesis is that it is the participants’ shared beliefs about givenness in CG Content that license the use of the Page 2 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness beer in (1b,c), but it is CG Management and beliefs about givenness in the conversation itself (that the beer referred to is salient or conspicuous in (1c) in a way that it is not in (1b)) that condition the possibility for deaccenting. From this perspective deaccenting is a device of CG Management that is dependent on the situational (hence temporary) salience of an antecedent, rather than on CG Content. It is this CG Management notion (p. 43) of givenness that will concern me in this chapter. To distinguish salience-based givenness from other possible forms of givenness, I will refer to the former as Givenness (with a capital G).4 The view that deaccenting is conditioned by Givenness as an element of CG Management rather than Content finds support in consideration of the deaccenting of the presupposed clausal complements of factive predicates. Such presuppositions define propositions that for utterances that invoke them to be felicitous must be known to the interlocutors and not simply salient—they must form part of CG Content. The following examples from Wagner (2012a) illustrate.5 (Capitals mark accented words and underscoring marks deaccented strings.)

(2)

(3)

6

(4)

The clausal complements to the factive predicate realize in (2)–(3) encode the presupposition that it was too cold to go swimming, but deaccenting is possible only when this proposition is Given. The presupposed truth of the embedded proposition is neither (p. 44) necessary (4) nor sufficient (2)–(3) for deaccenting. If deaccenting is conditioned by salience, a notion of CG Management, then the relative salience of a proposition is unrelated to its accepted truth, this latter an issue of CG Content. A similar demonstration is provided by the examples below from Kratzer (2004) (citing Christopher Potts).

Page 3 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness (5)

Use of the operator too in the second conjunct of (5a) imposes a presupposition (∃x [x attended the meeting]) that cannot be satisfied by the first conjunct, which does not assert that Ed attended the meeting. But as the felicity of deaccenting in (5b) shows, the failure of assertion of this embedded proposition (and its corresponding update in CG Content) is irrelevant to its salience for purposes of deaccenting. Examples (2)–(5) show a distinction between presupposition as commonly construed and salience. Dryer (1996) rightly objects that such examples do not yet show that salience (his activation) is not a form of presupposition that may be distinct from the more widely recognized presupposition that derives from a lexical factive or additive element, for instance. It remains possible that deaccenting presupposes not that the relevant proposition is accepted/true, but instead that it has a local legitimate antecedent in the ongoing conversation/context. (Dryer refers to this form of presupposition as metapresupposition.) And indeed, the prevailing assumption among analysts seems to be that Givenness is presuppositional in nature, although it is not quite clear how. After all, Givenness projects across environments that block presupposition projection in other cases: (5) shows us that the metapresupposition of a Givenness antecedent projects even when the closely similar (if not identical) presupposition that the additive particle too introduces does not. Presuppositions like the existential presupposition introduced by too in (5a) are generally modelled in CG as conditions on the input CG that must be satisfied for the utterances that invoke them to be able to update the CG. In terms of Krifka’s distinction of types of CG function, such presuppositions are based in CG Content. In contrast, Dryer’s metapresupposition would be cast in terms of CG Management, governing the temporary nature of salience-based presupposition in ongoing discourse. Givenness conditioned deaccenting can then be seen as presuppositional in the latter sense only—it expresses a presupposition of a situationally salient antecedent of a particular sort, an effect of CG Management rather than a condition on CG Content. In what follows, I will use deaccenting as a probe into the precise nature of Givenness. Before proceeding, three caveats about deaccenting are in order. First, it is important to observe that while Givenness is necessary for deaccenting, it is not sufficient. In each response in (6), the italicized phrase is Given in virtue of its mention in the question, but deaccenting it would be distinctly odd in this context. In general, deaccenting is overruled by considerations of focus (Rooth 1992, this volume), regardless of the Givenness (p. 45) status of a candidate expression. (We return to the relation between focus and Givenness in Section 3.3.3.)

Page 4 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness

(6)

Second, as (7) shows, deaccenting of a salient expression, when possible, is mandatory only when the expression follows a nuclear accent, or might itself bear the nuclear accent were it not given. In pre-nuclear position, expressions whose denotations are Given (and not focused) may be deaccented, but need not be (see Horne 1990, 1991; Gussenhoven 1999).

(7)

The student from her class, if it is understood as coreferential with the student introduced in the context sentence, may be either accented or deaccented in (7a), whereas it seems it must be deaccented in (7b). Given this difference, I adopt the practice of (re-)fashioning examples that test deaccenting so that relevant candidates consistently appear in post-nuclear position. Third, I will deal here only with the Givenness of denotations and/or (the mental representations of) their referents. Nevertheless, it has been known since at least Williams (1981) that sometimes it is the form of an expression and not its meaning that acts as Given for the purposes of deaccenting. Examples (8) and (9) below from Williams (1981) and Wagner (2012b), respectively, illustrate. (8)

(9)

Although the informational content is the same in both examples in (9), (9a) tolerates an accent on them whereas this same pronunciation with an accent on public is distinctly odd in (9b). It is evident from Williams’ example (8) that it is the repetition of form rather than content that triggers the effect. Wagner labels this ‘a givenness illusion’ and reports on experimental results confirming this effect. The illusion (Wagner 2012b: 1443) is due to the possibility that ‘… accenting phonologically given material is avoided, even at the cost of not marking a contrast’. I will not be concerned in this chapter with such givenness illusions. The reader is referred to Williams (1997) and Wagner (2012b) for proposals regarding such cases.

Page 5 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness The question we turn to next is, what are the precise conditions that define salience based Givenness? In (1c), for instance, the expression some beer that serves as antecedent for the deaccented beer is segmentally identical to it. But, as shown below, not all cases of deaccenting are conditioned by such identity. We are left again with the question why beer counts as salient in (1c) but not in (1b). In Section 3.3.2 we examine the semantic relations that qualify an expression as salient for the purposes of Givenness, using deaccenting as a diagnostic. (p. 46)

3.2 Semantic relations 3.2.1 Entailment and coreference Schwarzschild (1999) posits the relevance to Givenness of two core semantic relations: coreference (10) and entailment (11)–(13) (see also Rochemont 1986). (10) (11)

(12)

(13)

The deaccented proper noun, pronominal and epithet objects in (10) are all understood as coreferential to the proper noun possessor of the subject noun phrase. In (11), each deaccented constituent is entailed (in terms of set inclusion) by its italicized antecedent. In (12), entailment predicts the potential for deaccenting the hypernym on the basis of its hyponym. Where entailment fails to hold in (13) (that is, from hypernym to hyponym), deaccenting is illicit. In Schwarzschild’s proposal, entailment is strictly a relation between propositions that is facilitated by a mechanism of type-lifting existential closure (∃-type shifting) to raise non-propositional denotations apart from e to the type of propositions, for calculating Givenness. An utterance U is then Given if (i) U is entailed (modulo ∃-type shifting) by a valid antecedent A in the discourse, or (ii) U is of type e and U and A corefer.7 For example, in (12a) animal is deaccented because ∃x [animal (x)] (p. 47) is entailed under ∃-Closure by the mention of gorilla in the first clause. Importantly, the relevant entailments are not necessarily logical entailments of the discourse, but rather entailments under ∃-type shifting from what is salient in prior discourse. (See also (19).)

Page 6 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness Many accounts of Givenness (e.g. Dryer 1996) propose a three-way distinction between active, non-active, and semi-active (or accessible) denotations (like the beer in (1b)). A denotation is said to be accessible if it bears a pragmatic (e.g. functional or meronymous) relation to a locally prior denotation. For instance in (14), comparable to (1b), the driver is said to be accessible due to the prior mention of the bus. (14) But notice that the possibility for deaccenting shows that the accessible referent denoted by the driver is not Given in the preferred sense needed for deaccenting: driver in (14) is not possibly deaccented despite being accessible.8 This is predicted if deaccenting is licensed specifically by entailment. More generally, the relation between meronym and holonym, unlike the relation of hyponym to hypernym, is not a relation of entailment. It is thus predicted that deaccenting is not generally licensed by meronymy or holonomy, as the following examples confirm. (15)

(16)

Notably, (15)–(16) improve when the main stress is shifted to the deaccented phrase, whose denotation is evidently not salient despite being accessible. Deaccenting under entailment or coreference can give rise to inferences about speaker’s meaning, as in this famous example from Lakoff (1972). (17) If insulted can only be deaccented under entailment, then calling someone a Republican must entail insulting her. Here the entailment is not logical, but speaker defined (van (p. 48) Deemter 1994) and inferentially available to interlocutors who may not otherwise share it: if insulted is deaccented then its meaning must be Given, hence entailed by a prior (effectively type-compatible) antecedent, for which there is but one candidate in (17).9 Inferencing from coreference conditioned Givenness is also possible, as in (18).10 Page 7 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness (18)

In each case, if the underlined phrase is deaccented, then it is understood to co-refer with the italicized antecedent; if the phrase is instead accented, it may not co-refer (though in (18d,e), it may specify a subset). Van Deemter (1994) argues for expanding the notion of anaphor to allow the deaccented expression in such cases to be anaphorically dependent on the antecedent (see also Williams 1997, 2012 for a different such proposal). It would seem odd, however, to characterize (18c) for instance as a case where the deaccented proper noun is anaphorically dependent on the prior descriptive noun phrase. The inference instead seems to be that both expressions refer to the same entity.

3.2.2 Other putative relations for Givenness The conclusion that Givenness is a function of entailment/coreference is consistent with the conclusion that Givenness is to be distinguished from topic-hood (Halliday 1967; Krifka 2008; Kučerovà and Neeleman 2012).11 I want now to argue that several other notions that have been appealed to for Givenness are similarly inappropriate. For instance, appeal is sometimes made to Ariel’s (1990) notion of accessibility as the conditioning factor in deaccenting (e.g. Reinhart 2006). But accessibility in Ariel’s use functions like the givenness of the Givenness Hierarchy (Gundel et al. 1993): degree of accessibility is (p. 49) correlated with the likelihood or possibility of appearance of a particular form of referring expression, like the contrast between pronouns and full referential descriptions. Deaccenting is overall insensitive to distinctions among referring expressions in these terms. This perspective belies a widespread claim in the literature that definite noun phrases are consistently Given and indefinite noun phrases consistently lack Givenness. In contrast, like other constituents, definite noun phrases are typically accented when discourse new and deaccented when Given (modulo the usual exemptions from deaccenting), as is evident already from Prince’s (1981) discussion of her category of Unused noun phrases (see also Brown 1983; Bosch 1988; Terken and Hirschberg 1994; Umbach 2001). Moreover, it is readily possible for an indefinite noun phrase to be deaccented, as in (19). (19)

Page 8 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness There is no hot dog introduced by the conditional clause of (19) that serves as a referential antecedent for the deaccented second instance. Compare (20a), where the underscored indefinite noun phrase in the second clause is deaccented in virtue of the entailment licensing antecedent a dog in the first clause. (20)

In (20a) two new discourse referents are introduced—Peter’s dog and my pet. Evidently even a deaccented indefinite noun phrase (a pet) can introduce a new discourse referent. In addition, in arguably enthymematic fashion, subsequent references to my pet (the underscored phrases in (20b)) can yet be deaccented, even though each introduces new information about this discourse referent. I conclude that neither definiteness nor indefiniteness are uniquely or predictably associated with accenting or deaccenting independent of salience. Analysts also sometimes use the term D-linked to refer to Given in the intended sense. But D-linking (Pesetsky 1987), to the extent that is it well defined, is independent of Givenness. In Pesetsky’s analysis, which-headed phrases in wh-questions are D-linked, as indicated by their exemption in (21c) from Superiority effects (21b). As Rochemont (2013b) observes in relation to (22), D-linking cannot be co-extensive with Givenness, since which phrases are uniformly D-linked while still expressing possible distinctions in Givenness via deaccenting.

(21)

(22)

(p. 50)

(23)

Page 9 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness Although judgements are subtle, the (b) examples in (23) are equally as grammatical as (21c) and in contrast to (21b), showing conclusively that which-phrases are D-linked for Superiority regardless of the Givenness distinctions among them. D-linking must therefore be independent of Givenness. Some analysts use the terms predictable, repeated, or unimportant to describe the type of Givenness that is relevant to deaccenting.12 But these terms are ill-chosen for that function and, as a result, entirely misleading. Examples (17), (18), and (20) show that deaccenting is not restricted to repeated expressions nor to expressions that are presented as unimportant. Predictability is a more widespread term that is nevertheless as readily ill-chosen. First, predictability too encounters the problems faced by repetition and unimportance as predictors of felicitous deaccenting. For instance, in (17), (18), and (20) the deaccented expressions are not only arguably important but also unpredictable in that they add potentially new information about their antecedents. Moreover, predictability, when not confused with probability, is more often than not determined only post hoc when applied to deaccenting, even with expressions that are repeated. (24)

It is certainly not predictable in advance of the utterance of the second sentence in (24) that Mary did something to John or if Mary kissed anyone, she kissed John: she might have kissed Bill, the cat, or even the couch. And finally, when predictability does seem to have some predictive force, it is not generally associated with deaccenting, as in (25). The word beak in (25a) is fully predictable, but it is not possibly deaccented (without adding further context), as shown by the contrast in (25b). (25)

I conclude that neither predictability, nor repetition, nor unimportance sufficiently characterize Givenness. In this section I have ruled out definiteness, D-linking, repetition, importance, and predictability as factors that categorically condition deaccenting. I conclude that Givenness is solely a function of coreference or entailment. (p. 51)

3.2.3 Givenness accommodation We have seen in example (17), repeated below, that deaccenting by a speaker can support an inference on the part of the addressee. Let us now consider exactly how.

Page 10 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness (17) In deaccenting insulted, speaker B presents its denotation as Given, a presentation that is not obviously true in (17). Rather, the addressee is led to infer that the speaker believes that to call someone a Republican is to insult him/her. What precisely is the source of this inference? The inference follows from the anaphoric Givenness presupposition introduced by deaccenting, which indicates the need for a salient antecedent that entails ∀x, y [x calls y a Republican→x insults y]. In this case B is invoking accommodation, classically construed as a means of introducing an informative presupposition into a conversation (see Stalnaker 2002; von Fintel 2008 for recent discussion). B appeals to A to recognize that there is a situationally salient antecedent for the deaccented expression and to accommodate a proposition that allows that antecedent to entail the deaccented phrase under Existential Closure. An appeal to accommodation is subject to the restriction that it must be accepted ‘quietly and without fuss’ (von Fintel 2008). This restriction characterizes the difference between (17) and (26). (26) Most speakers would not easily agree to accommodation of the proposition that is needed to render deaccenting felicitous in (26): ∀x, y [x insults y →x calls y a Republican]. Accommodation gives a natural account of such cases as (17)/(26) and also of parallel cases of accommodation of coreference as in (18), and (20). The naturalness restriction on accommodation, that it be accepted ‘quietly and without fuss’, emphasizes the inherently pragmatic nature of the process, thus readily characterizing its culturally and contextually sensitive variability. Accommodation displays other characteristic restrictions as well, related to this naturalness requirement. First, it is optional, in the sense that speakers are not required to use accommodation wherever possible. Rather, speakers may use locutions that directly alter the context, introducing the relevant meanings that would satisfy presupposition without the need for accommodation. Thus, except in cases where there is limited risk, speakers are typically conservative in their appeal to accommodation, for the reason that its use poses the risk that the speaker’s contribution may be rejected by interlocutors who refuse to accept the appeal quietly and without fuss. A related restriction is that accommodation is sensitive to the specific intimacy of the interlocutors: in many cases greater (p. 52) intimacy implies less risk on the part of the speaker, modulo the specific content to be accommodated. The view that deaccenting may give rise to the need to accommodate in order to satisfy a presupposition of salience leads to a re-evaluation of cases which have otherwise often been characterized in strictly grammatical terms or in terms of notions like predictability or unimportance, such as so-called thetic sentences (27), Bolinger-style deaccenting (28), or discourse new though unaccented indexical locative and temporal expressions (29).13 In our terms, the problem presented by such cases is this: if α must be Given to be

Page 11 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness deaccented, then α that is not Given (i.e. new) must be accented; but the literature claims that all of (27)–(29) can be used in contexts where the underscored constituents are both new and accentless. (27)

(28)

(29)

Most accounts of such cases seek non-Givenness based sources for the lack of an expected accent. Among the grammatically based alternatives, one widely favoured proposal links theticity, for example, to the unaccusative/unergative contrast (but see (27 e,f,i)). Another links theticity to the stative/eventive distinction (Your EYES are red/BLUE, but see (27g)). But if the only possible source of the lack of an expected accent is deaccenting and deaccenting is solely dependent on Givenness as analysed here, then these cases must have a Givenness-based source. Accommodation related to Givenness may provide an analysis consistent with this conclusion. One leading indicator is that accommodation is a pragmatic mechanism, with the potential flexibility to characterize the range of examples in terms other than predictability or unimportance. Accommodation is also not susceptible to the inherent flaw in the grammatically (p. 53) driven accounts, that they are simultaneously too weak and too strong, as Rochemont (2013b) shows. If the problem posed by (27)–(29) is fundamentally pragmatic, consider how the characteristic conditions on accommodation noted above give insight into an account of such cases. For one thing, judgements of acceptability vary as a function of levels of intimacy between interlocutors: (29a) would be judged less acceptable if it could not be taken as Given without comment that events regularly occur on the local subway system, or where there is no local subway for example. With thetic sentences it has been Page 12 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness observed that in many cases deaccenting the predicate is optional. Since making appeal to accommodation can be a risky business, it only appears mandatory in cases where the risk is completely minimal. Compare (27a,c) with (27b,d,e,g) and the alternatives in (27f,h,i). Alternative pragmatic accounts that claim that the underscored phrases in (27)– (29) are either predictable or unimportant, while plausible for some examples ((27h,i) for instance), face severe difficulties with others: the unaccented predicate in (27e) is not necessarily predictable or unimportant, and that in (27d) is most certainly neither. Importantly, as both Stalnaker (1998) and von Fintel (2008) emphasize with accommodation generally, the adjustment to CG that accommodation demands is made only after the utterance that invokes the presupposition and before the update to CG content. Thus, the context that Givenness appeals to is not the virtually null context that precedes these discourse initiating utterances but the discourse initial context that includes these utterances (hence the contrasts between examples in (27h,i)). Predictability/unimportance are often considered appealing for such cases, but as already argued, these pragmatic notions fail to provide a general account. This view of Givenness as a presuppositional device of CG Management contrasts with the way Givenness is often seen. A more traditional view is that when a speaker deaccents a string this leads to a recoverability problem for the addressee: recover a meaning that is sufficiently activated. Whether activation is sufficient is usually tied to limits of memory and attention: what is the span of time and/or distance between utterances that may suffice to allow a string to be seen as activated? The CG Management view of deaccenting presents the addressee’s problem differently: which CGs are consistent with the presuppositional requirement of an antecedent for a constituent presented as Given? Here there is no issue of a timespan beyond which Givenness deaccenting is illicit under an intended interpretation. Rather, the addressee must select a CG that is compatible with the speaker’s presentation. The only consideration is whether the speaker is successful in appealing to the addressee to construct an appropriate CG. Dryer’s (1996: 501) discussion of an example due originally to Chafe (1976) illustrates. Adapting this example slightly, imagine that Sherlock Holmes, sitting quietly at the desk in his study, suddenly says to Watson, who is reading a newspaper, ‘The BUTLER killed the Kingsdale widow!’ As Dryer observes, one way to view this exchange is that Holmes is appealing to Watson to recover a context in which the relevant constituent (the VP) is Given, when the present context obviously does not qualify. Dryer refers to this process as ‘activation accommodation’ on the model of presupposition accommodation. As noted for presupposition accommodation generally, (p. 54) Holmes’ appeal succeeds only to the extent that the accommodation is made ‘quietly and without fuss’ (von Fintel 2008). If Watson is unwilling or unable to accommodate a CG with an appropriate antecedent, Holmes’ appeal to accommodation will fail. But there is no issue of how long it has been since the expression has been uttered, as some absolute bound on persistence of activation. While it may be possible to determine some statistical measure of persistence of activation for certain purposes, such a measure is but a poor reflection of the actual state of affairs. This is that whether a particular use of deaccenting succeeds or fails will be strictly a function of the Page 13 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness addressees’ willingness or ability to construe under accommodation an antecedent for the deaccented expression in the current CG or to accommodate a different local CG which supplies an appropriate antecedent.

3.3 Given, new, and focused

Page 14 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness So far in the discussion we have considered only cases where entire constituents are Given under entailment or coreference, under Schwarzschild’s preliminary definition of Givenness. But Schwarzschild’s ultimate proposal provides for cases where a phrasal constituent is only partly Given. He delivers an informal procedure in which constituents are checked for Givenness recursively, Given constituents are unmarked, and non-Given constituents are F-marked under an operation of Existential F-closure which replaces each F-marked constituent from within a potentially Given constituent with a variable, existentially closing the result to check for antecedence, modulo the need for ∃-type shifting. In order to avoid over-liberal use of F-marking in context, F-marking is limited by an economy measure (Avoid F) to just what is necessary to satisfy Givenness for any particular constituent.14 The final elements of Schwarzschild’s informal analysis are given in (30).15 (30)

Consider the following illustrations of how this procedure allows for partly Given constituents. (p. 55)

(31) (32) While John bought a red convertible is not GIVEN in (31B), the nominal, predicate, and sentential constituents are GIVEN if red is F-marked and the resulting constituents undergo Existential F-Closure (and ∃-type shifting as needed): the ensuing denotations for the noun phrase (∃x ∃Y [x is a Y convertible]), verb phrase (∃x ∃Y [x bought a Y convertible]), and sentence (∃Y [John bought a Y convertible]) are all entailed by the discourse prior John bought a blue convertible. It is also possible in (31) to consider the VP to be F-marked, since it is Given from A’s statement that ∃P[P (John)]. But VP cannot be F-marked here: Avoid F rules out a representation in which both VP and red are Fmarked, since F-marking just the latter suffices to satisfy GIVENness (30). (The VP alone cannot be F-marked since red convertible is not Given.) In (32B), though John is GIVEN, kissed John is not. John is thus F-marked despite being GIVEN, in service of satisfaction of GIVENness for the containing verb phrase and sentence which are otherwise not GIVEN.

Page 15 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness In this section, I will discuss three problematic aspects of this analysis. First, the analysis fails to capture focus sensitivity effects when all constituents in a specific domain are GIVEN (Beaver and Clark 2008). In fact, AVOID F rules out any possible F-marking in such cases. Second, Avoid F also proves overly restrictive in some cases where rhetorical relations play a role (Kehler 2005). Third, deaccenting of one constituent under Givenness often results in accenting some other constituent. Wagner (2006: 298) presents data that argue for a modification to Givenness such that ‘[m]‌arking a constituent x as given introduces the presupposition that there is an alternative y′ to its sister y, such that the constituent [y′ x] is given’. Here Givenness of the deaccenting candidate does not suffice to license deaccenting; rather, deaccenting is dependent on the ability of the candidate’s structural sister to express a focus of contrast.

3.3.1 Focus sensitivity In Schwarzschild’s system lack of F-marking is interpreted (by GIVENness), but Fmarking has no particular interpretation. A constituent that is F-marked may be GIVEN or not. Nevertheless, if an F-marked constituent is GIVEN, it is F-marked in service of a higher prerogative only: the GIVENness of a containing constituent. While this consequence correctly characterizes many cases of contrastive focus, it fails in cases (p. 56) where the GIVENness of a containing constituent is not at stake. Consider the following example from Beaver and Clark’s (2008) discussion of this specific problem.

(33)

Because the emotive predicate glad in (33) is a focus sensitive expression, the interpretations of these examples differ: (33a) expresses the students’ preference for Brady, rather than someone else, to teach semantics, whereas (33b) expresses their preference for Brady to teach semantics rather than some other subject. But in Schwarzschild’s system not only are each of the elements of the clausal complement to the predicate in both examples GIVEN, the clausal complement itself is GIVEN and hence grammatically unmarked and predicted to be deaccented, as in (33c). AVOID F rules out any further internal F-marking within these clausal complements, and so does not license F-marking of Brady in (33a) or semantics in (33b). But Brady and semantics are not deaccented though Given, and so must be F-marked in their respective sentences. Evidently, F-marking must distinguish between focus-as-alternatives (as in cases of focus sensitivity) and focus-as-new (in cases assessed by GIVENness directly). There are several revisions to Schwarzschild’s analysis on offer: see especially Féry and SamekLodovici (2006), Krifka (2006), Kratzer and Selkirk (2007), Beaver and Clark (2008),

Page 16 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness Selkirk (2008), Beaver and Velleman (2011), Katz and Selkirk (2011). Most of these proposals elect to reserve F-marking for focus-as-alternatives, marking either Given or new constituents distinctly while adapting elements of Schwarzschild’s analysis in (30). Other facts point to this same need to distinguish focus-as-alternatives and discourse new. Selkirk (2008) and Katz and Selkirk (2011) argue that when focused and new constituents co-occur in the domain of a focus sensitive operator (see Beck, this volume), an undifferentiated notion of focus does not suffice to distinguish the range of interpretations available. Consider the following example from Rochemont (2013b). (34)

In the context indicated, the utterance He only introduced Bill to Sue is multiply ambiguous, as the possible continuations in (34) indicate: only may associate with just Bill (34a), with just Sue (34b), with both Bill and Sue (34c), or with the VP (34d). (p. 57) But in Schwarzschild’s analysis all of these constituents are F-marked (35), since none are GIVEN. (35)

(36)

Representation (35) does not distinguish among the various ambiguities evident in (34). On the other hand, if focused and new are distinguished by F-marking and new is unmarked, then the various interpretations in (34) can be represented distinctly as in (36), one representation corresponding to each of the interpretations invoked in (34a–d) respectively. A similar argument presents itself from it cleft sentences, which are widely acknowledged to be syncategorematic focus sensitive contexts.16 A cleft partitions a sentence into focus (the pivot) and background (suitably adjusted to replace the understood position of the focus within the cleft clause with a variable), as in (37).

Page 17 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness (37)

If focus bears the same relation to background as it does to Givenness, then the background in an it cleft must be Given. But while the background in an it cleft may be Given, and very often is, it need not be (for recent discussion, see Hedberg 2012). (38)

For B’s reply to be felicitous, the context set that precedes the utterance should include at least a shared proposition to the effect that Mary lost something on the cruise. Notwithstanding that this condition is satisfied (for instance, both speakers know that the fact that Mary lost something on the cruise is part of the context set), the cleft clause in sentence (38B) will be deaccented only in case the proposition that Mary lost something on the cruise is already Given in the conversational context, and not otherwise. Since this latter condition is not satisfied in the mini-discourse in (38), deaccenting of the cleft clause in (38B) is disfavoured, pending further amendment to the context. (p. 58) Plainly, the cleft clause that marks the background in an it-cleft need not be Given for the cleft to be felicitous.17 Accordingly, some distinction between background and Given is mandated —focus cannot be the complement to both background and Givenness.18 The need to distinguish focused and new extends also to prosody. Neeleman and Szendrői (2004) and Féry and Samek-Lodovici (2006) together give several arguments against Fmarking as a focus-marking device in grammar, but what they implicitly argue is that for prosodic purposes discourse new and focus must be distinguished and F-marking, if used, must be restricted to focus (Selkirk 2008; Rochemont 2013b). In each case, the argument is based on showing that whenever new constituents compete with focused constituents for prosodic prominence, focus inevitably wins. These arguments are strongly supported by the experimental results reported in Katz and Selkirk (2011), who show that in sentences containing focus sensitive only and in which both new and focused constituents appear, focus prominence shows greater amplitude, duration, and intensity than new prominence, even in cases where a new constituent linearly follows a focused constituent. A further prosodic argument (see Reich 2012) for the need to distinguish between focused and new stems from consideration of cases of Second Occurrence Focus (SOF— Baumann, this volume), illustrated in the following example adapted by Beaver et al. (2007) from Partee (1999). (39)

Page 18 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness The pitch accented Focus vegetables in (39A) is Given with its containing clause in (39B), and it then comes as no surprise that this clause, including vegetables, is deaccented. What is surprising is that vegetables continues to function interpretively as a Focus in association with only even though it does not bear the expected pitch accent (as measured through f0 excursion in particular) that usually marks a Focus prosodically in English. Beaver et al. (2007) show that though it fails to show significant f0 excursion, the SOF vegetables in (39B) nevertheless continues to bear phrasal prominence as indicated by measures of intensity and duration. These results are interpreted to mean that focus in English is associated with some degree of prosodic prominence even when it is deaccented. It cannot then be the case that focus is simply the complement of Given if Givenness conditions deaccenting, since in SOF we see a variety of focus that is evidently deaccented.19

(p. 59)

3.3.2 Rhetorical effects

Schwarzschild (1999: 165) observes that Givenness marking through deaccenting is sensitive to rhetorical relations. In (40a) it is the Givenness of the matrix clause (John borrowed the book) that is central to the interpretation; in (40b) it is the givenness of the embedded clause (Max purchased the book).

(40)

In either case, the relevant proposition (x borrowed it, Max x’d it) is GIVEN, and there is no problem with either GIVENness or AVOID F. But this is not always the case. Rhetorical effects are possible when GIVENness is satisfied but AVOID F is not. Kehler (2005) compares Schwarzschild’s (1999: 170) examples in (41) with the patently equivalent examples in (42) (although I have slightly modified the contexts): the accent pattern in (a) is preferred in (41) but infelicitous in (42), and that in (b) is preferred in (42) and is infelicitous in (41).20

(41)

Page 19 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness (42)

Nevertheless in Schwarzschild’s proposal,21 only the F-markings and pronunciations in the (a) examples are predicted: AVOID F uniformly prohibits the F-marking and corresponding pronunciations of the (b) examples. Kehler explains that the contrast between (p. 60) (41) and (42) is due to a difference in rhetorical relations: the felicitous accent pattern in (41) reflects Contrast or Parallel (which are in the Resemblance category of coherence relations—see Culicover and Jackendoff 2012) while that in (42) reflects Contiguity. Despite the evident relevance of rhetorical relations to the choice of accent pattern in (41)/(42), the approach that Schwarzschild advocates for (40) will not extend to these cases without modification or elimination of AVOID F, as Kehler suggests. A possible solution to this difficulty utilizes the revision proposed already in Section 3.1. If F-marking is restricted to focus-as-alternatives and discourse new constituents remain unmarked (one of several possible alternative revisions), then representations (41a)/(42a) remain unchanged, but (41b)/(42b) are revised to (43a,b).

(43)

Examples (41a)/(42a) mark the dual contrasting alternatives reading consistent with Contrast/Resemblance that is far more natural in the context of (41) than (42). The accent pattern and representations in (43), on the other hand, are consonant with the simple rhetorical Contiguity that is most natural to (42) in particular. The implicit question in the contexts of both (41) and (42) asks only what happened next—the differing felicity of patterns in the target utterances reflects the relative suitability of invoking rhetorical Contrast in (41) vs (42). Importantly, ‘relative suitability’ is at least partly a function of the speaker’s rhetorical intent, the intentional use of accents to indicate either Contrast or Contiguity as the intended rhetorical relation.

3.3.3 Local alternatives It has been commonly observed that deaccenting of one constituent under GIVENness often results in the accenting of some other constituent, the latter usually proposed to be a metrical or syntactic sister of the former (Ladd 1980, 2008; Culicover and Rochemont 1983; Selkirk 1984; Williams 1997, 2012; Wagner 2006, 2012a). In some cases, the shifted accent must mark a focus of contrast or deaccenting is barred, as in the following example from Wagner (2006).

Page 20 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness (44)

Although convertible is GIVEN in (44a–c), it is only acceptably deaccented in (a). Wagner argues that what makes deaccenting possible in (a) but not (b) is that cheap in (p. 61) (a) is an acceptable alternative to high-end as a focus of contrast, but red in (b) is not. GIVENness as formulated in (30) is not sufficient to capture this contrast between (44a,b) since in both cases convertible is equally GIVEN. Wagner argues for a reformulation of Givenness as Relative Givenness such that ‘[m]‌arking a constituent x as given introduces the presupposition that there is an alternative y′ to its sister y, such that the constituent [y′ x] is given’. An observation that I think relevant here is that there is a further alternative in the context provided in (44). This is the continuation in (45). (45) Since the object is the locus of contrast in alternatives that are called for by the explicit question in (44) (I wonder what he brought), the object cannot be fully deaccented even if Given (cf. #He BROUGHT a convertible). Here we see again the relevance of rhetorical relations in speakers’ use of deaccenting: if the speaker wishes to draw a contrast with high-end convertibles she may (44a), but need not (45). It nevertheless remains true, as Wagner claims, that deaccenting seems dependent in such cases on whether the deaccenting candidate’s sister can plausibly mark a focus of contrast (Repp, this volume).22 Moreover, as Wagner observes, while in some cases the shifted accent must mark a focus of contrast (44); in others it seems it need not (46).23 (46)

Here the structural sister of the deaccented John does not mark a contextually licensed focus of contrast—although discourse new, kissed in (46) is not contrasted with other contextually given relations that hold between Mary and John. Wagner’s (2006) solution is to posit LF raising of arguments, which allows the sister requirement on contrasting alternatives to be satisfied. Relative Givenness is satisfied after raising if the raised argument is Given and its sister, the residual sentence from which it is extracted, marks a weak contextually licensed contrast, ‘weak’ in the sense that any discourse prior proposition that contains the Given argument can count as a valid alternative.

Page 21 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness Wagner (2012a) modifies the Relative Givenness account of local alternatives to require local excluded alternatives (LEA), whereby the negation of the existential closure of the antecedent must be entailed by universal closure of the constituent under evaluation (akin to Schwarzschild’s existential F-closure), effectively enforcing a partition of the set that unifies the two subsets so defined, such that one subset excludes the other. LF raising a Noun Phrase argument a, yields a proposition level constituent [a b] and givenness in such cases is determined by applying a two place exhaustive operator that requires there to be ‘… a salient constituent [a′ b] in the context such that the exhaustive operator applied to a and b excludes [a′ b]’ (Wagner 2012a: 125). These proposals successfully cover (46) and related examples (such as those in footnote 23). They also provide a means to address the focus sensitivity data in (33) from Beaver and Clark (2008), through accommodation of a form consistent with our earlier proposals. Roughly, the deaccenting of Brady in (33a) gives rise to a requirement for a salient antecedent of the form ∃x[x teaches semantics] that is excluded by Brady teaches semantics. A similar account in terms of accommodation is possible for (33b). As Wagner observes, ‘… a context in which an alternative statement is excluded may well be the default type of context that is accommodated if one hears a sentence with marked prominence out of the blue’ (Wagner 2012a: 126). Although these cases are not out of the blue, they do represent a choice among alternative presentations of the sentence in the context indicated. (Compare the third alternative (33c)). Notably, a speaker’s choice of what to deaccent as Given relative to some accented alternative is not driven by context but must be compatible with context (as licensed by Givenness), as the contrast between (33a–c) makes clear. (p. 62)

In contrast to Schwarzschild’s analysis, Wagner’s LEA proposal introduces alternatives into the calculation of Givenness, so that both Focus-as-new and Focus-as-alternatives might be seen to complement Givenness. It remains unclear whether such a programme can succeed and so eliminate reference to Focus completely. Problems remain that have no clear solution, including how to provide for evidence of a prosodic contrast between New and Focused and how to capture SOF.24 In the absence of a proposal on these issues, Focus and Givenness do not yet appear wholly susceptible to unified analysis.

3.4 Conclusion This chapter has explored Givenness as expressed through Givenness marking in English in the form of deaccenting. It has been argued that the data support a coreference/ entailment based approach to Givenness, in contrast to other conceivable (p. 63) formulations. Moreover, Givenness is expressible as a form of presupposition so long as a distinction is drawn between two distinct types of beliefs in Common Ground (beliefs about the subject matter of a conversation and beliefs about the conversation itself (Stalnaker 2002)) or between two distinct components of CG, CG Content, and CG

Page 22 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness Management (Krifka 2008). This view gives rise to a novel account of some longstanding problems stemming from the lack of accenting in English all-new utterances. In evaluating the influential proposal of Schwarzschild (1999), we have also seen that Givenness presents necessary though not always sufficient contextual conditions on deaccenting—speakers deaccent in accord with their communicative intents in regard at least to evoking focus alternatives and expressing rhetorical relations, not because the context strictly speaking requires them to. Apparent mandatory deaccenting, when it arises, is best seen from this perspective in the following way: failure to deaccent when to do so would be consistent with a speaker’s actual communicative intent misleads interlocutors to think the speaker must have some other intent: deaccenting is mandatory when it is unavoidably consistent with a specific rhetorical intent. With regard to the question of how focus and givenness are related, it was argued that while Givenness (specifically under Wagner’s 2012a formulation) comes close to eliminating Focus, it does not do so completely. Some evidence still supports a three-way contrast between Given, focused, and new.25

Notes: (1) This chapter has benefited from comments and suggestions by the editors and two reviewers, as well as participants in seminars at UBC. I am grateful also to Peter Culicover, Patrick Littell, Valéria Molnár, Hotze Rullmann, and Susanne Winkler for comments on earlier drafts. (2) Cruttenden (2006: 314) cites Walker (1781) and Bell and Bell (1879) for early recognition of this fact about English. (3) Krifka is not alone in distinguishing elements of CG. Stalnaker (2002: 708) distinguishes two sorts of beliefs that are encoded in the CG as a conversation proceeds: ‘… beliefs about the subject matter of the conversation’ and ‘ … beliefs about the conversation itself’. Stalnaker’s distinction mimics the distinction drawn by Dryer (1996) between presupposition and metapresupposition and also parallels Kripke’s (2009) distinction between active and passive context. Further, these distinctions are implicated in the taxonomy of projective content of Tonhauser et al. (2013). None of these proposals are equivalent, however, and I will not decide among them here. For concreteness I cast the discussion in terms of Krifka’s proposal. (4) This notion of Givenness corresponds to Prince’s (1981) givennesss. Prince’s givennessk more closely relates to givenness in CG Content. The former is recast in Prince (1992) as Discourse Old and the latter as Hearer Old. Many other terms have been used to name or include this sense of givenness: among others, old, activated (Chafe 1974, 1976; Lambrecht 1994; Dryer 1996; Beaver and Clark 2008), contextually bound (Sgall et al. 1973), salient (Prince 1981; Ward and Birner 2011), c-construable (Culicover and Rochemont 1983; Rochemont 1986), presupposed (Chomsky 1971; Jackendoff 1972;

Page 23 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness Zubizarreta 1998), ground/tail (Vallduví 1992; Vallduví and Engdahl 1996), predictable (Bolinger 1972; Kuno 1972), topic (Gundel 1988, 2003; Erteschik-Shir 1997, 2007). (5) See Schmerling (1976), Allerton (1978), and Rochemont (1986) for parallel argument. The non-reduction of Givenness to presupposition is discussed by many other authors, including Prince (1981), Tancredi (1992), Dryer (1996), Büring (2004), Jacobs (2004), Gundel and Fretheim (2004), Kratzer (2004), Abusch (2008). In contrast, numerous authors propose to reduce Givenness to presupposition; see Zubizarreta (1998), Guerts and van der Sandt (2004), and Sauerland (2005) for different recent proposals to this effect. See Sæbø (this volume), Velleman and Beaver (this volume), and the collection of papers in Theoretical Linguistics 30.1 (2004) for extended discussion. (6) This example may be improved in this context by the use on cold of the HLH accent that is normally used to mark a Contrastive Topic (Büring, this volume). (7) Is it possible to reduce coreference based salience to entailment? This could be done by type shifting e type denotations to the type of predicates (Partee 1987) and subjecting them to ∃-type shifting. Hotze Rullmann (p.c.) suggests two further possibilities: treating individuals as generalized quantifiers or reformulating entailment as a relation between denotations of any type (van Benthem 1986). The failure of definite pronouns to bear an accent except when contrastive may be due not to coreference with a salient antecedent but to the failure of functional categories (analysing pronouns as Ds) to be parsed in the mapping of syntax to phonology unless specifically stressed. (8) This does not rule out the possibility that accessible denotations might yet find distinct prosodic expression, as claimed by Baumann (2006), Baumann and Grice (2005). (9) It is thought that deductive inference under entailment from what is asserted can also motivate deaccenting. In (i) such inference seems to be sufficient for deaccenting, but does not so readily license pronominalization. ((i))

(10) These examples are modelled on or borrowed from Ladd (1980), van Deemter (1994), Umbach (2001), Büring (2007), Baumann and Riester (2012). (11) For analyses in terms of topic-hood see Gundel (1988, 2003), (Lambrecht 1994), Erteschik-Shir (1997, 2007), for example. (12) Deaccenting is sometimes lumped together with ‘prosodic reduction’, but it may be that prosodic reduction is common with repeated or predictable expressions regardless of their status as accented or deaccented (e.g. Jurafsky et al. 2000).

Page 24 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness (13) For complete references and fuller discussion of the claims in the remainder of this section, see Rochemont (2013b). (14) An alternative to minimizing F-marking is maximizing Givenness, as proposed in Truckenbrodt (1995). This option is pursued in different ways by Williams (1997) (DOAP— Don’t overlook anaphoric possibilities), Büring (2012) (MaxAna—Maximize Anaphoricity), and Sauerland (2005), Wagner (2006), and Kučerovà (2007) (Maximize Presupposition— Heim 1991). (15) Schwarzschild’s formal analysis uses designated assignment functions rather than Existential F-closure. (16) A reviewer objects that the pivot of a cleft is not a focus except by assumption. The assumption seems warranted by the transparent alternative semantics of the pivot/ background relation, and is rendered yet more plausible by the lack of existentiality and exhaustiveness implicatures in the clefts of some languages, where clefts serve solely a focus function (e.g. Zerbian 2006; Koch 2008). (17) This argument is particularly compelling in the case of Informative-Presuppositional clefts (Prince 1978), which may be used to initiate a discourse (e.g. as the lead sentence of a newspaper article) with virtually no part of the utterance Given. ((i)) (18) Williams (1997) proposes nested topic/focus configurations to account for such cases, while still introducing a distinction between prosodically and structurally identified focus in clefts. (19) This argument presupposes that a Givenness-based account of SOF is possible, a position argued against in Büring (2015). See Beaver and Velleman (2011) for discussion. (20) Rochemont (2013b) observes that (41b) is indeed felicitous if the accent pattern is understood to mark Contiguity rather than Resemblance. The option for both patterns is made clearer in the modified example in (i).

((i))

Many speakers agree that both responses are possible in (i), but the responses differ in that (b) marks a dual contrast between cited/dissed and Mary/Sue, whereas (a) marks only a contiguous event (see 43a). (21) To guarantee a systematic relation between F-marking and accenting, Schwarzschild restricts required accents to privileged F-marked constituents, those that are FOCmarked under (i).

Page 25 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness ((i))

(22) Büring (2012) treats contrast in terms of the notion of issue: an issue is plausible if it partitions the worlds in the context set into complementary sets, each compatible with only one of the alternatives in the issue. (23) Ladd (1980) proposes default accent to characterize the cases in which no contrastive interpretation ensues from deaccenting, although Godjevac (2006) and Rochemont (2013a) argue that the most widely cited putative example (John doesn’t READ books) is not such a case. Still, other cases do not so readily succumb to such re-analysis, as with the following examples from Ladd’s (1980) and Selkirk’s (1984) discussions. ((i))

((ii)) ((iii)) (24) See Stevens (2014) who argues specifically against both the LEA analysis and the LF raising of arguments. (25) Due to reasons of space, I have not explored in this chapter either range of variation in prosodic marking of Givenness (for relevant discussion see Zubizarreta 1998, this volume; Cruttenden 2006; Zerbian 2012) or the syntactic expression of Givenness (see Grosz, this volume; Fanselow, this volume; López, this volume; Neeleman and van de Koot, this volume; Winkler, this volume).

Michael Rochemont

Michael Rochemont is Professor of Linguistics at the University of British Columbia. His current areas of interest are prosodic phonology, specifically the syntax phonology connection and relations between information structure categories and prosody; the syntactic expression of categories of information structure, both within and across languages; and givenness, particularly as expressed through deaccenting.

Page 26 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

Givenness

Page 27 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Arizona Library; date: 26 July 2018

(Contrastive) Topic

Oxford Handbooks Online (Contrastive) Topic   Daniel Büring The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara Print Publication Date: Jul 2016 Subject: Linguistics, Morphology and Syntax Online Publication Date: Mar 2015 DOI: 10.1093/oxfordhb/9780199642670.013.002

Abstract and Keywords This chapter discusses the semantics and pragmatics of contrastive topics vis-à-vis focus. A semi-formal characterization of its main properties is given, using the techniques of alternative semantics and questions under discussion. This treatment is compared to various analyses proposed in the literature for contrastive topics and arguably related constructions, such as the English rise–fall–rise contour. Finally a brief discussion of noncontrastive topics is provided. Keywords: topic, contrastive topic, alternative semantics, question under discussion, rise–fall–rise

4.1 Contrastive topics: the un-focus TO approach contrastive topics, we take focus as our point of departure (non-contrastive or thematic topics will not be discussed until Section 4.5 below).1 A focus-marked (Fmarked) constituent, roughly, is interpreted as ‘the new information in response to a question’:2 (1) Either because of the meaning of focus, or because of the pragmatics of question–answer pairs, focus is interpreted exhaustively (‘↝’ marks a pragmatic inference): (1)

Page 1 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic This sometimes carries over to sentences with two intonational prominences: (2)

It seems therefore plausible to analyse (2) as a double focus: (3) (p. 65)

In some instance (or some contexts), however, the two prosodically highlighted

elements are interpreted asymmetrically: (4)

The answer in (4) suggests a continuation along the lines of ‘… whereas someone else wants to kick so-and-so out’. In that case, she is—by assumption—marked as a contrastive topic (henceforth CT), rather than a focus: (5)

Like focus, a CT relates to alternatives (‘someone else’), but whereas with the double focus in (2) all combinations of alternatives are, pragmatically, excluded (‘no one else kicked anyone out’), with the CT+F in (5), the exclusion only concerns who she wants to kick out (see Rooth, this volume, for more on alternatives). It is in fact implied that others want to do some kicking out as well. The notion that CT marking results in a set of alternative propositions which are explicitly not used for exhaustification lies at the heart of the recent proposals in Hara and van Rooij (2007), van Rooij (2010), and Tomioka (2009, 2010). In fact, at least Tomioka (2009, 2010) does not explicitly state anything more about what is done with these alternatives. The idea is that from the mere existence of such non-excluded alternatives, a hearer can deduce that the speaker must find these alternatives potentially relevant, and at least possible (otherwise she would have explicitly excluded them, i.e. have used a focus instead). This may ultimately be the most elegant account of CT, but for the time being we will outline an account here which specifies the reasoning that leads from the presence of

Page 2 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic non-excluded alternatives to their actual pragmatic effects in more detail, as part of the linguistic rule system. In Büring (2003) I argued for the following view: Whereas F relates a declarative sentence to alternative propositions, CT+F relates it to alternative questions. In (4), the alternative propositions say things like ‘she wants to kick George out’, ‘she wants to kick Marcy out’, etc., whereas the alternative questions are ‘Who does Bob want to kick out?’, ‘Who does Kim want to kick out?’, etc. (I use single quotes to characterize meanings). The alternative propositions are alternative answers to the question, so by the pragmatics of questions, and following the line of analysis just quoted, they are excluded: she doesn’t want to kick George/Marcy…. out, only the speaker. The alternative questions, on the other hand, are not excluded in any way; rather at least one must be pertinent to the conversation. This is the case in (4): The original question was who they want to (p. 66) kick out. Given that the answer is limited to her intentions, there must be other relevant people—the other members of the group referred to by they—for whom it is pertinent to ask: Who does (s)he want to kick out? This basic idea can be implemented via an extension of alternative semantics for focus (von Stechow 1981; Rooth 1985, this volume), which derives from a sentence like (4), repeated with information structural marking in (6), the set of F-alternatives, (6a), as well as the set of CT-alternatives, (6b):3 (6)

We will write SCT+F for a sentence containing CT+F (analogously for F+F, F etc.), and ⟦SCT+F⟧O , ⟦SCT+F⟧f , and for its ordinary meaning, F-, and CT-alternatives, respectively.

Page 3 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic What to do with the CT-alternatives? That is, how are they interpreted? For the purpose of this paper, (7) will serve as our sole rule: (7)

The requirements on SCT+F introduced by the CIR should be understood as conventional implicature triggered by the presence of CT marking. Applied to (6), (7) requires that speaker and hearer can identify at least one question instantiating ‘Who does y want to kick out?’ which is pertinent and independent of (6) itself. We will elaborate on what exactly is meant by Pertinence, Independence, and Identifiability in the course of this chapter, but it should be clear already that in a context in which it was asked who X want to kick out, X some group, there are at least as many such pertinent questions as there are members of X. If one knows who the members of X are, one can identify the questions about them. And only one of those questions is resolved by—i.e. not independent of—the answer She wants to kick me out (namely ‘Who does she want to kick out?’); the other questions (‘Who does Jeanne want to kick out?’…) are independent of that answer, meeting (7). Note that (7) is not met in (2) (Did you kick her out?—SHE kicked ME out!). If she were a CT (and me a focus), (7) would require that there is at least one question like (p. 67) ‘Who did y kick out?’ which is pertinent to the conversation and logically independent of (i.e. not resolved by) the answer. But this is not the case: What we are interested in is whether he kicked her out, or she him. Answering that she kicked him out resolves this issue completely, in violation of the CIR in (7), in particular Independence, (7b). Crucially, failure to meet (7) does not mean that (2) is infelicitous, only that it cannot be a CT+F structure. As stated above, it does have a well-formed structure on which both she and me are F, to which (7) simply does not apply.

Page 4 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic A question not addressed so far is whether the F+F structure in (2) is prosodically different from the CT+F structure in (4). Or, put differently, whether prosody can disambiguate the declarative sentence in (8) between a double focus (answering (8a)) and a CT+F structure (answering (8b)): (8)

This question has not been systematically investigated, but it is usually assumed that the two intonational contours resulting from the different questions in (8) are, or at least can be, different. Impressionistically speaking, the F+F contour has a high pitch accent on the first F she, followed by a low stretch (presumably H* L– in Mainstream American English ToBI notation, cf. Beckman et al. 2005) and another high pitch accent on me, followed by a low boundary (H* L–L%).4 The CT+F contour, in contrast, has a rising pitch accent on she, after which the pitch remains rising/high (L*+H), followed again by a high pitch accent on me, and a low boundary (H* L–L%). In a very stylized form, the CT on she would be realized as a ‘rise– fall–rise’ (L*+H L–L%)—what Jackendoff (1972) calls a B-accent—whereas the F-accent would just be a high pitch accent, followed by a low tone (H* L–). In what follows I will assume that CT+F sentences are indeed prosodically different from F+F sentences. There are other ways to tease the two patterns apart. In a CT+F sentence, CT may be fronted across F, as in (9b):5

(p. 68)

(9)

Page 5 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic An F+F sentence does not allow for non-canonical ordering of the two F, cf. (10):

(10)

For pragmatic reasons, the answer in (10)—unlike that in (9)—must be an F+F sentence: If Muckensturm wrote the song in 1972, there cannot be other pertinent questions like ‘When did Batiston write it?’ (or ‘Who wrote it in 1982?’), given that songs are only written once. Accordingly, (10b), which involves preposing of one focus across the other, is out. We can thus take the possibility of non-canonical order of two prosodically prominent constituents to be indicative of a CT+F pattern. I will now review some of the uses that have been taken to be typical for CT+F patterns in the literature.

4.2 CT phenomenology 4.2.1 Partial topics Answers to multiple wh-questions, or a single wh-question containing plurals, typically allow CT+F answers: (11)

The CT values of the answers in (11) are ‘What did x bring?’ and ‘Where does your xsibling live?’, respectively. Given the more general questions about the guests/siblings, we can see how the CIR in (7) is met: The questions in

are obviously pertinent,

and independent of the answers in (11). Languages with topic-marking morphemes like Korean -nun or Japanese wa likewise use these markers, together with intonational prominence, in such discourses:6 (p. 69)

(12)

Page 6 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic The question need not be overt in order for CTs to occur. Simple pair-lists that answer an obvious implicit question like ‘Who ate how many mbeju?’ suffice, as the following example from Guaraní shows:7 (13)

It is not necessary that the speaker knows what the answer to the other question is. The following example from Hungarian is representative for contexts showing that:8 (14)

Example (14) is instructive in two respects: First, the speaker need not believe that someone else slept in a different place (see also Section 4.3 below); and second, (s)he does not even have to think that the other question does in fact have a well-defined answer, as long as the question itself (‘where did the others sleep’) is pertinent. This shows that it is accurate for the CIR to refer to alternative questions, rather than their answers.

4.2.2 Shifting topics (15)

The CT-questions here are ‘How is/was Bo on day x?’ and ‘Where did x buy this book?’. Which of them are pertinent? For (15b), the very questions overtly asked is pertinent and not resolved by the answers. Similarly, for (15a) the question how Bo is today is pertinent, since its answer will indicate whether he will come to school today. (p. 70)

Page 7 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic (Note that we do not aim to answer the question why the answers in (15) are relevant to the question in particular, but only why they can bear the CT+F pattern they do.)

Page 8 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic

4.2.3 Purely implicational topics (16)

In this case, the answer directly resolves the question that was asked. But the CT indicates additional questions: Where was the chauffeur? The cook?9 Are they pertinent? Quite plausibly ‘yes’ in a case like (16), where the questioner is easily construed as trying to find the murderer. This is not always the case, of course: (17)

The odd implication of the CT-marking on Mann is that someone else might have written the novel elsewhere, which defies word knowledge (similarly to (10b) above). Somewhat in-between are cases like (18): (18)

Using the CT-marking on the first person subject here adds the implication that someone else’s whereabouts (at the time of their learning about Chernobyl) are a pertinent question. Unlike in the murder case (17), this is not so easily accommodated here, but unlike in the Thomas Mann case, one can probably come up with something that the speaker considers pertinent, maybe that her dog was vacationing in the Ukraine. Even in (16), though, there is a feeling that the speaker stretches the use of CT, flouting the CIR, which govern its use. Why? Recall from (7) that the CIR requires that the pertinent questions referenced by CT should be identifiable—(7c). This amounts to knowing at least some value for x in ‘Where was x at the time of the murder?’ which makes for a pertinent question (see Section 4.4 below for more). While, for example, (p. 71) with partial topics (Section 4.2.1 above) the question itself names a group containing such x (so Identifiablity is met), additional common knowledge between the speaker and addressee about who the suspects are is needed in (16); and the more unclear the group of potential xs is, the more enigmatic the answer appears, as in (18) (where additionally, Pertinence is a problem). Examples (16) and (18) show that CT-marking does not necessarily just ‘echo’ something that is in the context already, but may itself contribute, possibly by way of accommodation, the notion that more questions are pertinent at this point. Conversely, the CT marking on for example, the gardener in (16) can be omitted without loss of

Page 9 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic coherence. But without it, there is no implication of other suspects (via other questions). So here the speaker is making a choice as to whether to indicate the presence of other questions prosodically or not.

4.2.4 Ineffability In some cases, CT marking appears impossible, regardless of context. One such case from German is (19) (Büring 2003: 534): (19)

Without CT-marking (but retaining F on the finite verb), (19) would be perfectly wellformed. But CT-marking on the determiner alle leads to unacceptability (even though alle can be CT-marked under other circumstances, as we will see momentarily). Why is that? Assuming that F on a finite verb signals polarity focus, the F-alternatives of (19) are that all politicians are corrupt, or that they are not. This means that the CT-alternatives are questions like ‘Are Q politicians corrupt?’, where Q is some determiner. Now note that none of these questions are independent of (19)’s assertion: If all politicians are corrupt, that logically entails the answers to ‘Are most/some/the… politicians corrupt?’, namely: ‘Yes’. In other words, independently of actual context, (19) cannot possibly meet the CT-Interpretation Rule (7), specifically Independence, (7b), and hence is ungrammatical (or ‘unpragmatical’, if you like). If we change the example slightly, CT+F is possible again: (20)

The CT-alternatives of (20a) are the same as in (19), but since the sentence itself neither entails nor precludes that all, or most, politicians are corrupt, these will be unresolved (and potentially pertinent) questions. In (20b), with the added degree expression, CT-alternatives are ‘How corrupt are Q Politicians?’, that is, questions that could be answered by Only a handful are totally corrupt or Some are not at all corrupt. And while the latter is again resolved by (20b)’s assertion, the former, for example, is not. There are thus unresolved (and potentially pertinent) questions among the CTalternatives, as required by CIR. (p. 72)

Page 10 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic

4.2.5 Scope inversion If a SCT+F is structurally ambiguous between a construal on which it violates the CIR— along the lines just discussed in Section 4.2.4—and one on which it does not, CT marking will effectively disambiguate the sentence towards the latter. A case in point is sentence (21), with CT+F marking as indicated, which can only mean that not all politicians are corrupt. Without CT marking, it can either mean that, or that all politicians are noncorrupt:10 (21)

The same can be observed for example in Hungarian (from Gyuris 2002: 80): (22) only has the ‘not… everybody’ reading, although quantificational elements in Hungarian usually have a preference for surface scope: (22)

Let us focus on the available construal first: The sentence asserts that not all politicians are corrupt, and its CT-alternatives will be ‘Is it false that Q politicians are corrupt?’, for example ‘Is it false that many politicians are corrupt?’. This question is not resolved by the assertion, and it is plausibly pertinent, meeting the CIR. On the other construal, the sentence says that all politicians are un-corrupt, with CTalternatives like ‘Are Q politicians corrupt?’. But since the sentence asserts that none of them are, any such question is resolved by the assertion alone. As was the case with (19), this construal of (21) cannot possibly meet the CT Interpretation Rule (7). But in (21), unlike in (19), there is a second construal which is (or can be) felicitous, so that the sentence is acceptable, though not ambiguous, with CT+F marking. A similar effect can be observed in (23): (p. 73)

(23)

Page 11 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic With the CT+F marking as indicated, the sentence invites the question ‘Then why DID you drink?’. For that to make sense, the sentence itself must be interpreted as ‘it is not the case that my being sad is the reason for my drinking’. If it meant ‘my sadness is the reason I don’t drink’ it would be incoherent to ask what, instead, is the reason the speaker drinks. Without this CT+F marking, on the other hand, that latter reading is easily possible. So here, just as in (22), the CIR disambiguates by way of rendering one logical construal of the sentence contradictory.

4.3 Single CT and F+CT English has a rise–fall–rise (RFR) pattern—L*+H L– H% in MaeToBI notation—which can occur sentence finally, that is, without a following F:11 (24)

For both the ‘sole RFR’ and the ‘F+RFR’ we should ask whether the RFR is the same as CT in CT+F. The parallels are striking, not just prosodically, but pragmatically. For example RFR on all in (25) disambiguates the otherwise scope ambiguous string towards the ‘not all’ reading, much like CT+F on alle… nicht (‘all…. not’) did in (21) above:12 (25) Likewise, Japanese stressed wa, and Korean stressed nun, the pragmatics of which appear very similar to English and German CT marking, can appear without an accompanying focus, yielding scope disambiguation:13 (26)

Furthermore, RFR may occur in partial answers and shifting topics, similar to the cases discussed in Sections 4.2.1 and 4.2.2 above:14 (p. 74)

(27)

Page 12 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic Constant (2006, 2012) assumes that the accent in RFR is a focus, and that RFR operates on the focus alternatives, conventionally implicating (28): (28)

15

Example (28a) states that there are alternatives in the focus value which are independent of the assertion of SRFR (as well as the Common Ground). It should be transparent that that derives the scope disambiguation effects, as well as cases of wholesale infelicity in case the assertion of SRFR entails or contradicts all its alternatives: (29)

16

It should be noted that (29b) (as well as (19)—all politicians ARE corrupt) has Falternatives which are entailed (‘some/most of my friends liked it’), as well as ones that are contradictory (‘none/fewer than half of my friends liked it’). This is captured by the word ‘informative’ in (28a), as well as the word ‘independent’ in the CIR, (7b). Weaker conditions which merely require alternatives that are compatible (but possibly redundant, such as Wagner 2012: 24, ex (46)), or non-redundant (but possibly known to be false, such as Oshima 2008: 7, ex (17)), or simply not equivalent to SRFR (e.g. Ludwig 2008: 391, ex (19)) will systematically fail to derive the desired result. Turning to the examples in (27), here, too, (28a) is crucial, in particular the ‘salient’ bit: In (27b), repeated below, as well as (30b), we are looking for alternatives like ‘I fed x’ which are informative; this holds for any alternative x ≠ ‘the cat’. Assuming that the speaker fed the cat only, none of these is ‘safely claimable’, either, so (28) appears to be met in both (30a) and (30b): (p. 75)

(30)

Page 13 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic Crucially, then, such alternatives must be salient in (30a), precisely because the question asks about them (namely whatever other animals are in ⟦the animals⟧Ο; the anomaly of (30b), on the other hand, must arise because the question only makes one proposition salient: ‘I fed the cat’. Since that proposition is entailed by the answer, (28a) is violated, not because there are no informative unclaimable F-alternatives, but because none of them qualifies as ‘salient’ in the context of (30b). Similarly in (27a) and (27c). The alert reader will notice that ‘salient’ in (28) plays much the same role as ‘pertinent’ in (7a) in the CIR. Unsurprisingly, then, cases like (25)–(27c) can be analysed using the CIR as well, if we replace the informative and non-claimable proposition required in (28) by the yes/no-question based on it, that is, ‘Did not n of my friends come?’, ‘Can x come to tea?’, ‘Did you feed x?’, and ‘Will you have an x?’. This was suggested in Büring (2003: 532), where it was assumed that the CT-value of a declarative with CT but without F is a set of yes/no-question meanings (see Constant 2012, sec. 5.3 for more discussion).17

Page 14 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic

4.4 Details, open questions, and alternatives 4.4.1 Last answer Crucial to the treatment of scope disambiguation in Section 2.5 was the fact that an SCT+F must not be a complete answer to the questions in its CT-value, that is, that the latter contain at least one independent question. It bears pointing out that, according to the CIR in (7), there need not be an actual open question (in ) after uttering (p. 76) SCT+F; there only needs to be a question that is not resolved by SCT+F alone. Example (31) helps to illuminate the difference:

(31)

Sentence (31a) is infelicitous as an answer, which is predicted: The CT-value of (31a) contains questions like ‘Where is x from?’, and for the pertinent ones, x must be her or him. Since (31a) resolves both of these questions, there is no question in that is independent of (31a) itself, in violation of Independence, (7b). Now, the two answers in (31b), taken together, also completely resolve the question, so why is CT+F possible here, in particular in (31b-ii), which completes the answer to (31)? To see why, it is important to realize that the Independence part of the CIR, (7b), prohibits CT+F in sentences which themselves completely answer the questions in , but not in sentences which together with the common ground (in particular together with previous partial answers) completely answer the question. That is to say, the crucial difference is that (31a) itself is a complete answer, whereas (31b-ii) is merely a ‘completing’ or last answer. The former, but not the latter, are excluded from containing CT+F.18 Interestingly, as noted, for example, in Constant (2012: 430), the single peak RFR pattern discussed in Section 4.3 above cannot occur in a last answer (Constant 2012: ex 76): (32)

While the CT+F pattern in (32b) is fine here, a sole RFR on Persephone as in (32a) is not, pointing to an essential difference between the two contours.

Page 15 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic

4.4.2 Pertinence In a double correction like (33), it seems that any two teams could be replaced for England or Spain, preserving felicity:

(33)

(p. 77)

This is different when we use a CT+F pattern instead:19

(34)

Sentence (34c) is certainly fine with F+F intonation, but with a proper rise on Spain, the likely reaction to (34c) is: ‘Spain?! Who is talking about Spain?’ One might be tempted to conclude that CTs in general have to be ‘anaphoric’ (in some sense). But note the contrast with (35):

(35)

20

The oddness of (34c) must therefore have to do with corrections in particular. I would like to suggest that it has to do with Pertinence, (7a), that is, with which question is currently under discussion. Presumably the basic question in (34)/(35) was which games are on tonight. We then tackle the subquestion which team Brazil is playing tonight, suggesting (34)/(35). If the answer to that is ‘yes’, as in (35), we can continue with the general question of which games are on, saying who Spain is playing, as in (35). If the answer to ‘Is Brazil playing England tonight?’ is ‘no’, on the other hand, we cannot go on to the next question—who Spain is playing—as (34c) is trying to. We first have to resolve the question of who Brazil is playing.21 If this is the right conclusion to draw from these examples, it shows that the notion of Question under Discussion—and hence the mentioning of Pertinence in the CIR in (7a)—is highly relevant for the modelling of CT. After all, the fact that Spain is playing Germany is equally informative and even relevant in (34c) and (35). The crucial difference is that,

Page 16 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic since it addresses a new subquestion (about Spain), it can only do so after the current subquestion (about Brazil and/or England) has been resolved. To complete the picture, consider (36): (p. 78)

(36)

The difference between (34c) and (36) is that we can understand the answer in (36) as part of the strategy to figure out if and against whom Brazil is playing. We are thus not introducing Spain as a topic in its own right, but as part of figuring out if and who Brazil is playing.

Page 17 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic

4.4.3 Distinctness of foci The CIR in (7) requires that there be an independent and relevant question among the CT alternatives. It does not require that the answer to that question have a different focus than SCT+F. Thus, on the face of it, the condition allows for texts like (37): (37)

But (37) is clearly odd, and that oddness vanishes if we replace either occurrence of Paris (but not both) with Marseille. So it appears to be infelicitous to pair two CTs with the same focus. How to rule this out? One possibility is to blame the oddness of (37) on the Gricean Maxims of Manner: Basically the speaker should have said Jacques and Colette are from Paris, whereas otherwise one wrongly infers, via scalar implicature, from (37i) that only Jacques is from Paris (this was the line taken in Büring 1997b). Oshima (2008: sec. 3.3), van Rooij (2010: 14f) and Constant (2012), among others, point out that the effect appears too strong to be a conversational implicature. Van Rooij (2010) instead assumes that CT is grammatically exhausted independently, so that, for example, (37i) implies ‘only Jacques lives in Paris’; consequently (37ii) creates a contradiction. This, however, would also, wrongly, rule out (38ii), since it contradicts the CT-exhausting of (38i) (‘only Jacques is from Southern California’):22 (38)

Oshima (2008), too, assumes that CT-alternatives are not questions, but propositions, just like F-alternatives, namely propositions in which both CT and F are replaced by alternatives. That means that (37i) would require that there be some proposition ‘x is from y’, where x ≠ Jacques and y ≠ Colette, which is not assumed yet (hence a fortiori independent of ‘Jacques is from Paris’); ‘Colette is from Paris’ in (37) would not qualify here, since it sets y = Paris (whereas ‘Colette is from Marseille’ would). Unlike van Rooij (2010), this does not wrongly rule out (38). What Oshima (2008) (as well as van Rooij 2010) has difficulty explaining is that (39) is notably better than (37):23 (p. 79)

(39)

Page 18 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic This suggests that CT+F should not rule out the proposition that Colette is from Paris as satisfying the CIR for (37i) and (39i) (as van Rooij and Oshima do), but the particular realization of that proposition in (37ii). Plausibly one can blame the infelicity of (37ii) on the repeated focusing of Paris: There simply is no alternative ‘live in x’ to license F on the given element Paris in (37ii) (cf. Rochemont, this volume). This is avoided by focusing the particle too in (39ii) (see Féry and Samek-Lodovici 2006:146 for basically this idea).

4.4.4 CT—F Asymmetries In introducing CT-values, we have added an additional layer of semantic computation to the compositional semantics. We should ask whether this is strictly necessary. For example, we saw in Section 4.3 that scope inversion effects can be achieved by requiring that a proposition (rather than a question) which replaces CT and F be independent and informative. Of course, if we ignore the difference between CT+F and F+F altogether, it is impossible to differentiate whether (40) answers (40a) or (40b) (repeated from (8) above):

(40)

In Section 4.1 above I suggested that the answer in (40) would be pronounced differently, depending on the question, reflecting the difference between F+F (answering (40a)) and CT+F (answering (40b)). If this is correct, we cannot treat CT+F just like F+F. Assuming then that the rule for interpreting CT+F is different from that for F+F, couldn’t the former nevertheless operate on the same formal object, a set of propositions, as proposed e.g. in Ludwig (2008) or van Rooij (2010)? I submit that the answer is ‘no’. It would predict that CT and F would generally be symmetrical, since they both introduce alternatives at the same ‘level’, the F-alternatives. But CT and F, while often exchangeable, sometimes are not: (p. 80)

(41)

Were we to ignore the CT/F difference, both (41a) and (41b) would have the alternative set ‘x’s birthday is on y’, so any distributional difference between them would be

Page 19 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic unexpected. This alone shows that whatever interpretation rule applies to CT+F structures, it must be able to distinguish F- from CT-alternatives. But what is responsible for the difference between (41a) and (41b)? Presumably it has to do with the fact that speaker and hearer in (41) can list all the people who should appear in an answer (‘you guys’), but not all the birthdays that will appear in an answer (of course they know all potential birthdays, but not which ones are one of you guys’). This is what the word ‘identifiable’ was included for in (7c) of the CIR: If you guys is known to refer to Marcy, Elisabeth, Anton and Kim, speaker and addressee can identify the pertinent questions ‘When is Marcy’s birthday?’, ‘When is Elisabeth’s birthday’ and so on. But they cannot identify the four corresponding ‘inverse’ questions ‘Whose birthday is it on 9/12?’, ‘Whose birthday is it on 3/26?’ etc. Note that both sets of questions yield the same answers (that Marcy’s birthday is on 9/12, Elisabeth’s on 3/26 etc.), so a condition that only refers to the set ‘x’s birthday is on y’ could not possibly encode this effect. One might note that the same preference is observed with overt subquestions: When are you guys’ birthdays?’ is naturally followed up by (42a), not (42b)

(42)

So should this really be a part of the CT-condition, rather than a theory of querying? Note that even if we assume that there are covert sub-questions, and querying principles that choose (42a) over (42b), there still needs to be a part of the CT-condition that tells us that (41a) answers (42a), and (41b) (42b), and not the other way around. And that requires keeping CT- and F-alternatives different. So the main point—that the CT-condition must treat CT and F asymmetrically—remains valid.

4.5 Thematic Topics In this last section, I will briefly turn to non-contrastive topics, also called thematic topics. Examples of these include e.g. clitic left dislocation in Romance, preposing in English, or wa-marking in Japanese:24

Page 20 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic (p. 81)

(43)

Presumably, some or all of these constructions can host contrastive topics as well, especially if the topic constituent receives some extra prosodic prominence, say, by pitch accenting. What we are interested in now are examples where this is not the case, that is, where the dislocated/wa-marked phrase is neither a CT, nor an F, but still not just an ordinary part of the background. But why assume that there are designated topics in the background? Usually this characterization comes about as follows: Some (syntactic, morphological, or intonational) marking does not have a regular truth conditional effect, but seems to shape the pragmatic meaning of the sentence it occurs in, that is, the kinds of contexts where it is felicitous (sometimes, certain restrictions on the kind of constituents that can be so marked—for example ‘referring expressions only’—go hand in hand with that). If these effects are clearly not the ones found with F (answer to a question, new information, locus of correction…. ), or CT (see Sections 4.1–4.4 above), the marking is likely to be called (thematic) topic-marking.25 It is therefore perhaps unsurprising that attempts to characterize the properties of topic marking pragmatically have yielded widely diverging results. That said, the described pragmatic effects of topic-marking cluster around notions such as ‘psychological subject’, ‘what the sentence is about’, etc. As seen in (43c), a common way of eliciting a constituent x as a topic in this sense is to prefix the sentence in question with ‘Tell me about x’ or ‘What about x?’ Similarly, an utterance of a sentence with a topic x should be reportable by saying ‘(s)he said about x that….’; this latter fact is probably the reason such topics are often referred to as aboutness topics. But, as most recently discussed in detail in Roberts (2011), none of these tests seem sufficient or necessary to identify thematic topics, nor is it clear that they actually test for the same thing; or as van Bergen and de Hoop (2009: 173) put it in their foreword to a special issue on topics: ‘there is very little consensus among linguists on any…. specific definition. Multiple properties contributing to topichood have been described, but none of these properties seems either necessary or sufficient to classify something as a topic.’ Jacobs (2001) for one concludes

Page 21 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic that this diversity of characterizations is fundamental to the notion of topic itself: ‘there is no common functional (p. 82) feature (nor a common set of functional features) that justifies this classification [as ‘topic’; DB]’. A common way of modelling aboutness topics is to break the meaning of sentences with topics into two parts: the referent of the topic-marked expression (assuming for the moment that it is referential), and the property expressed by the non-topic part of the sentence. This is what Reinhart (1982: 6.2) calls the Pragmatic Assertion of a sentence. The discourse context, too, is structured into individuals, and properties that are agreed to hold of them. The standard metaphor for this is a stack of indexed file cards (one for each discourse referent), on which properties of the individuals are written (Reinhart 1982; Vallduví 1990; Erteschik-Shir 1997). For concreteness, associate a context C with a partial function fC from individuals to sets of properties, that is, a set of pairs 〈x, Φ〉, where x is an individual (or a discourse referent) and ɸ is a set of properties. So for any individual i that is part of the context, fC (i) gives us the set of all properties that are agreed to hold of i. Adding a sentence about i, that is, a sentence with a Pragmatic Assertion of the form (i, ϕ) to the context, results in a new context C′ such that fC′ is like fC except that Finally, assume that a context C is also associated with a set aC of aboutees, where aC is a subset of the domain of fC. The purpose of this set is to distinguish between individuals that have generally been introduced to the context C (the domain of fC) and those that are currently ‘under discussion’, aC. For illustration, consider the following short text from Tzozil:26 (44)

According to Aissen (1992: 51), Tzeb san-antrex, ‘a girl from S. A.’, is focused in (44a). In our model, it will expand the domain of the contextual function fC so as to include a discourse referent for the girl in its domain, and map that referent to the property of ‘being found in the straw mat (that fell from the tree)’. In (44b), ti tzeb san-antrex, ‘the girl from S.A.’ is topic-marked (by the prefix a and suffix e). Such topic-marking occurs when a topic is ‘new or shifted’ (Aissen 1992: 51). In our model, the discourse marker for Page 22 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic the girl is now added to aC, the set of aboutees, possibly removing other elements from that set. In the subsequent six sentences of the story, there is neither a full DP nor a pronoun referring to the girl, although they are about her. In other words, the topic marking in Tzozil is only used to establish an aboutee, not to refer back to it. On the other hand, some markings are supposed to pick out an already established aboutee, but cannot newly establish one. For example, Reinhart (1982: 63) blames the degraded status of (45b) vis-á-vis (45a) on the fact that the first sentence in (45) established Felix as an aboutee (in our terms), and that the as for…. construction refers back to an established topic, rather than establish a new one, as it would have to in (45b):27 (p. 83)

(45)

A thorough look at descriptions of topic-markings in various languages reveals that differences like these abound, making it impossible to talk about properties of ‘topics’ in general. An investigation of a suspected topic-marking in a given language should therefore start, not by assuming that, since it is a topic, it has a particular cluster of properties, but by establishing their exact specific properties independently, among them… 1. what items can be so marked (DPs only, definites only…. ); 2. whether the marking can… (a) establish a new discourse referent as the aboutee for the following, or (b) establish an existing discourse referent as the new aboutee (as in Tzozil), or (c) refer to an established aboutee throughout its tenure as ‘what the passage is about’? (d) … or do something altogether different; 3. whether elements that meet that description have to be so marked, or merely may be; 4. whether the same marking can serve other pragmatic functions; 5. whether there are other tests (than occurrence with that marking) to establish the status marked by it. Once careful descriptions along these lines are available, we can start a systematic investigation of topic cross-linguistically. Above, a way of modelling aboutness topics formally was sketched. It is not claimed that this particular version in fact models the effects of any particular marking in any particular language, rather it served to illustrate a common way of thinking about topics. Moreover, as Krifka (2008: sec. 5.1) notes, such a model ‘presupposes that information in human communication and memory is organized in a certain way so that it can be said to Page 23 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic be “about” something. This does not follow from a general definition of information. For example, relational databases or sets of possible worlds, both models for information, do not presuppose any relation of (p. 84) aboutness’. Given that there is no reason to assume a priori that information should be structured into aboutees and their properties, models of topic that reduce their meaning to such a structuring of information essentially leave us with the question: What is the observable effect? That is: How is an information state in which, say, the information that Mary visited Sue is stored under the Mary-entry different from one in which it is stored under the Sue-entry? Few proposals seem to address this crucial question. As an example of what we are after, take Portner and Yabushita’s (1998) claim that the descriptive content of an anaphoric definite in Japanese preferably describes properties that are filed under the entry of the intended referent (Portner and Yabushita 1998 pp. 125f): (46)

In the first clause, John carries the topic marker wa, so by hypothesis the resulting context stores about John that he met a woman in a café. In the second clause, the woman carries wa which establishes about her that she is a pianist. In the continuation (46a), this information is used to form a definite description referring to the woman, which is fine. In (46b) on the other hand, the information in the definite description is taken from the information stored about John—that he met the woman in a café—rather than about the

Page 24 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic woman. This is less felicitous. So we have an independent way of establishing how otherwise equivalent information is stored differently in the context. This effect, however, proves to be rather weak and subject to various exceptions, some of them pointed out in Portner and Yabushita (1998), others elsewhere (e.g. Fry and Nakamura 2000, who provide further counter-examples) casting doubt on the validity of this diagnostic. But it is such discernible effects that we need to find, for without them, the claim that topics determine the way information is added to the context is lacking empirical (p. 85) content, as are any conclusions about the structure of the context required to model this (see Dekker and Hendriks (1996) for similar criticism regarding Vallduví’s version of this model). The above discussion centred around ‘aboutness-topics’. A less widely used concept is what we may call frame topics. Chafe (1976: 50f) (endorsed by Li and Thompson 1976: 464), for example, explicitly dismisses the ‘what the sentence is about’ characterization for topic-marking in, for example, Chinese, and describes its effect like this instead: What the topics appear to do is to limit the applicability of the main predication to a certain restricted domain.… Typically, it would seem, the topic sets a spatial, temporal, or individual framework within which the predication holds. In English we do something similar with certain temporal adverbs. (Chafe 1976: 50) Example (47) illustrates with two Lahu examples from Li and Thompson (1976: 462): (47)

Extrapolating from Chafe, these sentences are about rice and noses, not fields and elephants. In any case they lack any contrast associated with the ‘as for…. ’ paraphrase and therefore constitute—according to Chafe—a category of their own. Krifka (2008: sec. 6) gives the following English examples: (48)

Page 25 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic (49)

Examples like these also lend themselves to an analysis as CT, given that we did not restrict CTs to DPs or indiviual-denoting expressions. Indeed Krifka (2008) subsumes both frame setters and contrastive topics under his notion of DELIMITATION. Summing up, the notion of ‘topic’ (without ‘contrastive’) should be used with great caution. There is no agreed-upon way to identify topics across languages, and therefore inconsistent claims about their properties abound. There is a semi-formal rendering of the notion of aboutness topic, which, however, relies on a notion of context which cannot be reliably established independently. This section can be read as a plea to refrain from using the notion altogether, and characterize ‘topic’-markings independently along the lines suggested in the questions above.

Notes: (1) Thanks to Manuel Križ, the editors, and an anonymous reviewer for valuable comments and corrections. (2) Small caps indicate intonational prominence; sentences in parentheses are given as context and not annotated for intonation. (3) See Büring (2003: sec. 12.1) for how to do that, Büring (1997b: sec. 3.3) for details. (4) See Eady et al. (1986), experiment 2, for comparison of this accent pattern with wide and narrow foci—though not with what we call CT+F here. Mehlhorn (2001) and Braun (2005) compare CT+F in German to ‘neutral’ sentences, but not systematically to what we call F+F here. (5) Sub-examples with letters are to be read as alternative continuations, so (9b) is not a reply or sequel to (9a), but—like (9a)—to (9). (6) Lee (1999); for Japanese see Uechi (1998), Tomioka, this volume. (7) Krivoshein De Canese et al. (2005: 81), via Tonhauser (2012: 273); the marker katur is glossed ‘contrast’. It can occur in non-initial answers within a CT+F answer sequence. (8) Gyuris (2002: 38f), see also É. Kiss, this volume. (9) Though of course, the gardener did it! (10) From Jacobs (1984), analysis following Büring (1997a).

Page 26 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic (11) See, among others, Ward and Hirschberg (1985); Pierrehumbert and Steele (1987); Hirschberg and Ward (1991); examples from Bolinger (1982: 507) and Jackendoff (1972: 261). (12) From Ladd (1980: 146). I notate the final R at the end of sentence here, indicating that it is not part of the pitch accent on the prominent word—all in (50)—but a rise at the right edge of the intonational phrase. (13) Lee (1999); see Hara (2008), Oshima (2008) and Tomioka (2009, this volume) for Japanese. (14) O’Connor and Arnold (1973: 173), Ladd (1980: 153), examples (16) and (19). (15) This half of the rule is also assumed in Wagner (2012: 24), where it is assumed that any additional meaning of RFR follows by Gricean reasoning, instead of being grammatically encoded. (16) Constant (2012, ex (33b)). (17) Strikingly, parallel German cases appear to have a focus on the finite verb or negation, as in (21) above, which—by standard F semantics—yields the meaning of a yes/ no-question as the focus value. Generally, German does not allow for CT (or RFR) without a following F, so one can hypothesize that German here chooses an F-marking—on the finite verb—which yields the same result that a CT-only/RFR sentence would (see Büring 2003: 532). (18) The interpretation rules for topic in Büring (1997b) has rightly been faulted for creating the ‘Last Answer Problem’, since it ruled out CT+F not just in complete answers, but also, wrongly, in last answers (Krifka 1999: sec. 3.4; Hara and van Rooij 2007: slides 19f; Aloni and van Rooij 2002: 32f). This was corrected via the ‘highest attachment’ condition in Büring (2003), as well as, explicitly, in the Independence condition in (7b) here. (19) To get in the right mood for the intonation patterns, you may try the left-dislocated versions first, i.e. No, Brazil, they’re playing Spain, etc. The contrast with No, Spain, they’re playing Germany is rather stark. However, what I claim here is that the same contrast can be observed with intonation alone, and is not a function of syntactic dislocation. (20) I am inclined to think that in (35) we accommodate that, for example, Spain belongs to the same group as England and Brazil, and the relevant task was to find out who (out of the four) was playing who tonight. That is to say, Spain, too, seems to have to be identifiable or anaphoric in some sense. If this were not the case, Spain would just have to be F, as in (33b). The point is that no such accommodation can come to the rescue of (34c), it seems.

Page 27 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

(Contrastive) Topic (21) In fact it seems likely that both the question who Brazil plays and who England plays have to be resolved (or aborted by means of saying ‘Brazil isn’t playing’ or ‘We can’t find out right now who England is playing’) before we turn to other countries as topics: Is Brazil playing England tonight?— ((i)) ((ii)) (22) Thanks to an anonymous reviewer for pointing out this example and its significance. (23) Van Rooij (2010: 15) actually notices the felicity of (39) and proposes that the presence of too in (37ii) ‘cancels’ the otherwise obligatory exhaustification of JacquesCT lives in ParisF to ‘only Jacques lives in Paris’. Note, though, that this does not offer an explanation for the acceptability of (38). (24) Catalan from López (this volume, ex (1b)), Japanese from Vermeulen (2011: ex (10a)). (25) See e.g. Aissen (1992: sec. 3.1) for a nice illustration. (26) Laughlin (1977: 67), via Aissen (1992: 50f). CL stands for ‘second position clitic’, ENC for ‘intonational phrase enclitic’. (27) As Reinhart herself later notes, as for is bad, too, unless ‘used to change the current topic of the conversation’ (Reinhart 1982: 64). Accordingly, (i) does not improve on (45b), suggesting that the problem must lie elsewhere: ((i))

Daniel Büring

Daniel Büring, Department of Linguistics, UCLA

Page 28 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Edinburgh; date: 25 December 2018

Focus Projection Theories

Oxford Handbooks Online Focus Projection Theories   Karlos Arregi The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara Print Publication Date: Jul 2016 Subject: Linguistics, Discourse Analysis Online Publication Date: Jun 2016 DOI: 10.1093/oxfordhb/9780199642670.013.005

Abstract and Keywords This chapter examines the phenomenon of focus projection: a sentence with prosodic marking of focus on a word can lead to ambiguity, in that different constituents containing the word can be interpreted as focused. Two general approaches to focus projection are compared. In Default Prosody analyses, focus projection is epiphenomenal, a consequence of general principles requiring that focus be prosodically prominent, coupled with default principles of default prominence. In the F-projection approach, focus projection is encoded more directly, in terms of licensing rules for F-marking on constituents. Overall, the evidence in the literature favors the Default Prosody approach, and this chapter concentrates on comparing these two general approaches, as well as briefly summarizing other aspects of focus projection. The chapter ends with discussion of Basque data that are problematic for widely assumed generalizations about focus projection, and a possible way of understanding the data in the Default Prosody approach. Keywords: Basque, default prominence, F-marking, F-projection, focus, focus projection

10.1 Introduction IN languages that mark focus via prosodic prominence, a sentence with a prosodic peak on a given word is often compatible with more than one focus reading.1 For instance, the sentence Mary killed Bill with prosodic prominence on Bill over all other words in the sentence can be understood with focus on Bill, but this is also the prosodic profile corresponding to a sentence-wide focus reading. This, however, is not the case for Mary in this sentence: prosodic prominence on this constituent is only compatible with focus restricted to Mary. This phenomenon, and its constraints in different languages, is Page 1 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Focus Projection Theories typically referred to in the literature as ‘focus projection’, and has received quite a bit of attention. In this chapter, I compare two types of accounts of focus projection, which I term Default Prosody and F-projection approaches. Under the former approach, constraints on focus projection are the consequence of default prosodic prominence rules that are independent of focus or any other IS-related concept. The only principle directly relating focus and prosody is Focus Prominence, a very general constraint requiring the focus to be prominent. On the other hand, the F-projection approach assumes a more direct relation between prosody and (syntactic) focus-marking, and default prosody rules (whether they exist or not) play no role in accounting for focus projection facts. This chapter is organized as follows. Section 10.2 reviews the basic facts, and Sections 10.3 and 10.4 summarize specific implementations of the Default Prosody and Fprojection approaches, respectively. The two approaches are further compared in Section 10.5, which provides crucial evidence from the literature in favor of the Default Prosody approach. The review of the literature concludes in Section 10.6, which provides a brief summary of other aspects of theories of focus projection not covered in (p. 186) previous sections. Finally, Section 10.7 discusses potential counterevidence to Focus Prominence from focused unaccented words in Northern Biscayan Basque, and I propose a small modification of the principle in order to accommodate the facts. Throughout this chapter, I give only very basic illustration of focus projection and related facts, concentrating on English, and to a lesser extent, on other West Germanic languages and Basque. The interested reader should consult the works cited in the chapter to obtain a more accurate and comprehensive picture of the empirical domain, as well as of the different analyses that have been proposed in the literature.

10.2 Focus projection: The basic facts Focus projection2 is the expression used to refer to the fact that, in several languages, phonological marking of focus (prosodic prominence) on a word is compatible with semantic focus on different constituents containing that word.3 The phenomenon, which was described for English in Halliday (1967: 207–208) and was first brought to the attention of generative linguists in Chomsky (1971), can be illustrated by the following English sentence, where small capitals indicate prosodic prominence:4 (1) Prominence on the direct object Bill in this sentence is compatible with focus on the direct object itself, but also on the containing VP killed Bill5 as well as the whole sentence.6 This ambiguity can be verified in several ways, for instance, by checking the felicity of uttering (1) as an answer to the following questions:

Page 2 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Focus Projection Theories

(2)

The fact that (1) is a congruent answer to all three questions shows that it can be interpreted with focus on the direct object (as an answer to (2a)), the VP (2b), or the entire (p. 187) sentence (2c).7 We can thus say that focus can project from the minimal constituent containing prosodic prominence to other constituents dominating it. Although much of the literature concentrates on the facts of English and other Germanic languages, the basic phenomenon is not restricted to this family. For example, the following sentence in Basque8 can have different focus readings (see A. Elordieta 2001: 130–134 for relevant examples and discussion):

(3)

As in English, the claim that the sentence has several focus readings can be diagnosed by the fact that it can be an answer to the following questions:

(4)

Page 3 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Focus Projection Theories Thus, (3) can be understood with focus on the prosodically prominent direct object Jon, but also on other constituents containing it, namely the VP Jon hil and the whole sentence. Observations of this type have led to the following hypothesis (see Sections 10.3 and 10.4 for specific implementations): (5)

9

(p. 188)

This derives the facts above. In the case of (1), since Bill is dominated by (at

least) three different nodes (the direct object, the VP, and the sentence), prosodic prominence on this word is compatible with (at least) three different focus readings. Several different theories of focus projection have been proposed in the literature. The main general observation that these theories attempt to account for is the fact that focus projection is not as unrestricted as suggested by the data discussed above. For instance, although focus can project from the direct object to the VP or the entire sentence in (1), focus projection from other constituents is severely limited. For instance, consider the following variation on (1) in which prosodic prominence is on Mary instead of Bill: (6) Given Focus Prominence, this sentence is expected to not have an object-focus or VPfocus reading, since neither constituent contains Mary. Indeed, it cannot be an answer to (2a) or (2b). Similarly, (1) cannot be understood with focus on the subject (i.e. it cannot be an answer to the subject wh-question below), since this constituent does not contain Bill. Furthermore, the principle also correctly predicts the fact that (6) has a subjectfocus reading (since the subject contains Mary) and can thus be an answer to Who killed Bill? However, a theory of focus projection that has no principle other than (5) wrongly predicts that (6) can be understood with focus on the entire sentence in an out of the blue context, since, trivially, the sentence contains Mary. Two main approaches have been adopted in the literature with the objective of accounting for the observed facts of focus projection. Both types of analyses complement Focus Prominence with other principles and rules that further restrict the possible relation between prosodic peaks and focused constitutents. In the Default Prosody approach, constraints on focus projection are the result of default prosodic prominence rules that are independent of focus. Under this view, Focus Prominence is the only principle that directly relates focus and prosody. This contrasts with the F-projection approach, in which focus and prosodic prominence are related in a more direct way by a set of rules that can license focus on different constituents of a sentence given a specific distribution of prosodic peaks in the sentence (i.e. the rules license ‘projection’ of focus Page 4 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Focus Projection Theories from a word with a prosodic peak to constituents containing that word). Note that in this chapter I make a terminological distinction between ‘focus projection’, namely, the fact that a particular sentence is compatible with several focus readings, and ‘F-projection’, a particular type of analysis of this fact. This important terminological distinction is often not made in the literature reviewed below. These two approaches to focus projection are summarized and compared in the next three sections, where it is argued that the Default Prosody approach provides a more satisfactory account of the facts.

(p. 189)

10.3 The Default Prosody approach

In the Default Prosody approach, focus projection is accounted for in terms of Focus Prominence and default prominence rules. In this section, I discuss the analysis of focus projection in Jackendoff (1972: ch. 6), which provides the first explicit analysis of the facts under this approach. For reasons of space, I limit the main discussion to Jackendoff’s account as a representative of the Default Prosody approach, and only briefly summarize the main alternatives in Section 10.6. Jackendoff’s analysis is based on Focus Prominence, as defined in (7), and the Default Prominence hypothesis shown in (8):10 (7)

(8)

Under this approach, other than by (7), focus does not determine the prosodic shape of a sentence. Specifically, once the effects of Focus Prominence are taken into account, the relative prosodic prominence of any two constituents in a sentence is determined by principles of default prominence that are independent of focus. The analysis is also based on two further assumptions: (9)

Page 5 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Focus Projection Theories (10)

Consider again the sentence Mary killed Bill. Recall that the basic facts of focus projection are that, while prominence on the direct object Bill is compatible with focus on the direct object, the VP, or the entire sentence, prominence on the subject Mary is only compatible with focus on the subject. In Jackendoff’s analysis, the sentence can have all of the following syntactic representations, each of which corresponds to a different focus reading: (p. 190)

(11)

In each of these structures, a constituent is F-marked (i.e. it is assigned a feature F). Semantically, this means that the F-marked constituent is focused (Jackendoff 1972: 245– 247; see Rooth 1992 for a widely assumed formalization of the relation between F and the semantics of focus). The principles in (7)–(10) ensure that the correct word in the sentence is assigned prosodic prominence. If the subject is focused (i.e. F-marked as in (11a)), Focus Prominence (7) ensures that prominence (nuclear stress) is on Mary, whereas if the object is focused (11b), this principle ensures that prominence is on Bill. The role played by the principles determining default prominence (8)–(10) in these cases is trivial, since the F-marked constituents in these cases only contain one (overt) word. The adoption of (8)–(10) (especially Default Prominence) is crucial for accounting for what we can informally refer to as the projection examples, namely those in which the Fmarked constituent contains more than one word. In particular, sentence-wide focus (11c) results on prominence on the direct object Bill: Focus Prominence would trivially be satisfied by assigning prominence to any word, but Default Prominence and (10) determine that prominence is on the rightmost word, that is Bill. The role of Default Prominence can also be observed in the derivation of the VP-focus reading: Focus Prominence ensures that the VP is more prominent that the subject, but within the VP, the object is assigned prominence because it contains the last word in the VP. Focus Prominence alone cannot derive this, since no particular constituent within the VP is Fmarked (i.e. informationally more salient than any other constituent within VP). Therefore, Default Prosody (together with the other elements in the analysis) correctly predicts that prominence on the object, but not the subject (or the verb), is compatible with a focus on VP and the entire sentence.

Page 6 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Focus Projection Theories This analysis can easily be extended to account for focus projection facts in other languages, with the caveat that other languages might differ from English in their rules of default prominence. For instance, default prominence in sentences in Basque is not on the last word of the sentence, but on the constituent immediately preceding the main verb.11 Thus, F-marking on the direct object, the VP, or the entire sentence in (3) above will result in prominence on the direct object, since the latter immediately precedes the main verb in all the former constituents.12 Importantly, this theory accounts for focus projection facts without rules or principles that directly determine focus projection: the fact that (transitive) VP focus in English requires prominence on the object is not derived by ‘projecting’ the focus from the object to the VP, but is a consequence of default prominence rules (which in English assign prominence to the rightmost constituent, i.e. the object in a transitive VP). (p. 191)

Subsequent literature has challenged most of the components of Jackendoff’s analysis, and the following sections provide a critical overview of this literature. Focus Prominence remains essentially unchallenged (with important modifications due to cases where the domain of focus is not an entire sentence, as in Rooth 1992 and Truckenbrodt 1995: ch. 4), but all the other components of the theory have been argued against. In the next two sections, I concentrate on the F-projection approach, whose most interesting feature is that it provides an account of focus projection that is not based on Default Prominence. Section 10.6 provides a brief summary of alternatives to Jackendoff’s analysis that differ mostly on the principles that determine default prominence.

10.4 The F-projection approach As mentioned above, the defining feature of the F-projection approach is that it accounts for focus projection facts in terms of principles and/or rules that license focus on different constituents in a sentence given a specific distribution of prosodic peaks in the sentence (Selkirk 1984, 1995; von Stechow and Uhmann 1986; Rochemont 1986). Under this approach, default (i.e. IS-independent) rules of prominence play no role in accounting for focus projection. In this section, I provide an overview of Selkirk (1995) as representative of this type of approach. Part of the motivation for Selkirk’s analysis (and more generally, for the F-projection approach) comes from cases of so-called default accent, first discussed at length in Ladd (1980: ch. 4). Consider the following congruent question–answer pair, in which the focus in the answer is a VP containing an element that is co-referent with an element in the question (Bill): (12)

Page 7 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Focus Projection Theories As an answer to (12a), the prominence in (12b) must be on the verb, not the object. In Jackendoff’s theory, this is predicted to be possible only if the context requires a focus on the verb. However, the context in this case (i.e. the wh-question) determines that the focus in (12b) is the entire VP, not just its head. This contrasts with the following question–answer pair (assuming that no mention of Bill has been made in the conversation up to this point): (p. 192)

(13) In this case, the prominence in the answer is on the object, as expected for VP focus. The crucial difference is that, by mentioning Bill, the question in (12a) makes the object in (12b) given,13 which is not the case in (13). Selkirk uses examples of this type to justify a theory that has the following components. Relying on the fact that a prosodically prominent word has a pitch accent, Selkirk takes the latter (and not stress) to be the primary exponent of focus.14 Pitch accents are freely assigned to words in the syntactic structure, and the rest of the analysis ensures that placement of a pitch accent on a given word is paired with the correct semantic interpretation. More specifically, a set of F-assignment rules regulates which constituents can be F-marked, given a specific distribution of pitch accents. Unlike the Default Prosody approach, focus projection is directly encoded in the representation of the sentence in terms of the licensing of F-marking, and default prominence plays no role in the analysis. The F-assignment rules are the following (Selkirk 1995: 555): (14) (15)

Note that (15b) makes argument structure an important factor in determining focus projection. This represents another important departure from Jackendoff’s analysis (Section 10.3 above), in which focus projection (as encoded by rules determining (p. 193) default prominence) is not related to argument structure (see Section 10.6 for a brief discussion of this aspect of Selkirk’s and others’ views on the role of argument structure in focus projection). The implicit asymmetry in (15) between arguments and modifiers (which are not mentioned by the F-projection rules) is due to observations made in Gussenhoven (1983a; see Section 10.5 below for relevant discussion).

Page 8 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Focus Projection Theories A different set of principles relates F-marking to the semantics (and pragmatics) of focus and givenness (Selkirk 1995: 555–556): (16)

Statement (16a) is the equivalent of the Focus Prominence principle: it requires the focused constituent to be F-marked, and given the F-assignment rules, this consituent must contain a pitch accent. Note that the rules in (16) treat F-marking and non-Fmarking in a different way: while the latter always entails givenness, the former only entails newness in F-marked constituents dominated by other F-marked constituents. As discussed below, this asymmetry is crucial in accounting for sentences in which the focused constituent is given. Consider first (1), in which the direct object has an accent. Given the F-assignment rules, it can have all of the following representations:

(17)

In all these representations, accent on the direct object Bill licenses F-marking on this constituent (by the Basic F-rule). In the absence of any other F-mark, as in (17a), the interpretive rule (16a) determines that the focus of the sentence is the direct object, as desired. F-marking on the direct object can license F-marking on the verb (by Argument Projection), which in turn licenses F-marking on the VP (by Head Projection), resulting in (17b), in which the VP is the focus (by (16a)) and both the verb and the direct object are new. Head Projection can also license F-marking on the entire sentence,15 as in (17c), in which case the entire sentence is the focus. The F-marking on (17b) and (17c) is also compatible with an accented verb, which accounts for the optional presence of an accent on the verb in all-new F-marked transtivie (p. 194) VPs. This contrasts with the behaviour of the subject Mary, for which the absence of F-marking in these cases is correctly predicted to entail that this constituent is given (by (16c)). This contrasts with a sentence in which both the subject and direct object are accented: (18) In this case, the subject is new, since it has an embedded F-mark. The analysis thus correctly predicts that all-new sentences must have an accent on the subject (as well as the direct object). Page 9 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Focus Projection Theories As shown above, Argument Projection (15b) accounts for the fact that focus can be projected from the direct object to the dominating VP (via the V head). In Selkirk’s (1984) original formulation of this rule, F-marking on any argument of a head could license Fmarking of the head. The observation that Argument Projection must be restricted to internal arguments is due to von Stechow and Uhmann (1986: 308), and accounts, for instance, for the fact that prosodic prominence on a transitive subject, as in (6), does not seem to license sentence-wide focus (see Section 10.5 for relevant discussion on this fact). Importantly, default prominence plays no role in this theory. The fact that under both a VP-focus and a sentence-focus reading, prominence is on the object in a transitive sentence has nothing to do with default prominence rules in English, but with rules that constrain the distribution of F-marks in a sentence. One of the main advantages of this approach over Jackendoff’s Default Prosody approach is that it makes more nuanced (and correct) predictions about the relation between the prosodic shape of a sentence and the discourse status of its subconstituents. This is due to its reliance on the distribution of accents (as opposed to nuclear stress), and a more fine-grained theory of the informationtheoretic interpretation of F-marking (which takes givenness in addition to focus into account). A further advantage of the F-projection analysis is that it provides a straightforward account for sentences of the type discussed at the beginning of this section, in which givenness within the focused constituent results in ‘shift’ of the accent to the head of the constituent. Consider (12), repeated here: (19) The analysis licenses the following F-marking on the answer:16 (20) In this case, the accent on the verb directly licenses F-marking on VP (by Head Projection), which can thus be interpreted as the focus. This shows that no accent on the object is necessary to license focus on the VP. In fact, in this particular context in which (p. 195) Bill/him is given, accenting the object is not possible, since that would entail Fmarking on the object, which by (16b), would have to be interpreted as new. The analysis also accounts for sentences in which the focus is given, as in the following question–answer pair: (21)

Given the question, the focus of the answer must be the direct object, which limits Fmarking to just this constituent: Page 10 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Focus Projection Theories (22) Since (16b) only requires embedded F-marked constituents to be new, the focus (i.e. an unembedded F-marked constituent) can be given, as shown by this example. In summary, Selkirk’s analysis differs from Jackendoff’s in a number of ways, including the central role played by argument structure in determining focus projection, and the adoption of a more articulated theory of the relation between F-marking and semantic/ pragmantic interpretation. In addition, Selkirk’s analysis of focus projection does not rely on principles of default prosody. As discussed in the next section, this latter aspect of the theory leads to certain predictions about the interpretation and prosodic prominence of parts of the focused constituent of a sentence that have been shown to be wrong in the later literature. This work thus provides crucial evidence that default prominence plays a central role in accounting for focus projection.

10.5 Evidence for the Default Prosody approach The F-projection approach makes the following strong predictions about focus projection: (23)

Jacobs (1988, 1991), Schwarzschild (1999), and Büring (1996, 2006) argue that these predictions are wrong and conclude that focus projection is primarily determined by principles of default prominence.17 As discussed below, counterexamples to Limited Projection are relatively simple to produce, but it is not completely clear that No (p. 196)

All-Given Focus can be shown to be false. Nevertheless, the argument against Limited Projection is strong enough to justify abandonment of the F-projection approach. Limited Projection follows straightforwardly from the F-assignment rules in (14)–(15). Fmarking of a phrase is only licensed by Head or Argument Projection. Since the Basic Frule ultimately grounds F-marking on accent, it follows that either the head or an internal argument of the head must contain an accent. In other words, focus pojection cannot proceed from any subconstituent that is not a head or an internal argument. The following question–answer pair provides the first counterexample to Limited Projection (from Büring 2006: (13), attributed to Büring 1996):

Page 11 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Focus Projection Theories (24)

The adjective is the only accented word, which is therefore F-marked by the Basic F-rule. Since it is a modifier (not an argument) of the NP object it is contained in, no F-projection is allowed to any other constituent, including the dominating NP. The F-assignment rules can thus only license the following: (25) However, the question requires focus on the NP, which wrongly predicts the answer to be unacceptable. What seems to be the required F-marking on the answer given the question is the following: (26) This requires projection from a modifier, which is not allowed by the F-projection rules. The following is a another counterexample to Limited Projection (from Büring 2006: (17)):

(27)

The F-assignment rules do not allow projection from the transitive subject to the sentence in (27b), as shown in (27c). Since the question requires focus on the entire sentence, (p. 197) the answer is wrongly predicted to be infelicitous.18 See Schwarzschild (1999: 167–169) and Büring (2006: section 3) for related counterexamples. The prediction No All-Given Focus (23b) is a bit more involved. First, a focused element is an F-marked constituent not dominated by any other F-marked constituent, by (16a). Second, according to the F-assignment rules (14)–(15), this F-marked constituent must contain at least one F-marked subconstituent (unless the F-marked constituent is a terminal node, in which case F is licensed directly by the Basic F-rule (14)). Since this subconstituent is an embedded F-marked node (i.e. it is dominated by the focused node), it must be new, by (16b). Thus, a (nonterminal) focused node must contain at least one part that is new. Counterevidence to No All-Given Focus is discussed by both Schwarzschild (1999: 171– 173) and Büring (2006). The following is a relevant question–answer pair from Büring (2006: section 6): (28)

Page 12 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Focus Projection Theories The question requires focus on the VP in the answer, but both elements (the verb and the direct object) in this VP are given: (29) Although the direct object is accented, it cannot be F-marked: that would make it an embedded F-marked constituent, hence new. However, for reasons given in Büring (2006: section 6), it is not completely clear that the context in (28a) does not require F-marking on Bill in (28b) (despite its being given), due to the fact that called Bill’s sister in (28a) contrasts with called Bill in (28b). Hence, the argument against F-projection based on No All-Given Focus is not conclusive. To summarize so far, it looks like F-assignment to phrasal nodes is unconstrained: it can be licensed by any F-marked subconstituent, and, in fact, if conclusive arguments against No-All Given Focus can be found, it might even be possible to license it directly, without any embedded F-marking. Although the predictions are stated in terms of Selkirk’s (1995) specific implementation of F-projection, this criticism of the approach is more general, and independent, for instance, of the central role that argument structure plays in Selkirk’s analysis, since the counterexamples discussed above provide evidence that there need not be any special relation between F-marking in a phrase and a constituent in it, however that relation is defined. As a consequence, Schwarzschild and Büring argue for a Default Prosody approach, namely one that assumes both Focus Prominence, which requires the focus to be prominent, and Default Prosody, which assigns prosodic prominence in a phrase (p. 198) independently of focus. This is, of course, not to say that these and other authors adopt the same analysis as Jackendoff (summarized in Section 10.3). In particular, one of the main insights of F-projection approaches is that givenness is an important factor in determining the relation between prosody and semantic/pragmatic interpretation, and current Default Prosody approaches to focus projection take givenness into account in some way or another (i.a. Neeleman and Reinhart 1998; Zubizarreta 1998; Schwarzschild 1999; Samek-Lodovici 2005; Wagner 2005; Büring 2006; Féry and Samek-Lodovici 2006; Reinhart 2006). Another aspect in which versions of Default Prosody differ from Jackendoff’s (and from each other) is what exactly constrains default prominence (hence, indirectly, focus projection), which is the main topic of the next section.

Page 13 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Focus Projection Theories

10.6 Other aspects of the theory of focus projection As summarized in the previous section, the literature has reached a consensus with respect to what kinds of principles govern focus projection. Specifically, it seems that both Focus Prominence and Default Prominence are necessary components of any successful account. Where the authors disagree is on the specifics of Default Prominence, that is, on the question of what syntactic and/or semantic aspects of an expression determine the patterns of default prosodic prominence within that expression. This section summarizes and compares the different approaches to Default Prominence present in the literature. Since the questions addressed here are mostly not about focus projection per se, but about patterns of default prominence, the discussion is brief, and the interested reader is referred to the work cited in this section, as well as Zubizarreta (this volume). One of the core cases of focus projection and default prominence discussed throughout this chapter has to do with transitive sentences in which both the verb and the object are included in the focus, for instance, in VP-focus cases, as in the following question–answer pair:

(30)

The relevant generalization that needs to be captured is that the direct object is more prominent than the verb. As was shown in Section 10.3, Jackendoff’s Default Prosody analysis accounts for this in terms of Chomsky and Halle’s (1968) cyclic Nuclear Stress Rule, which is based on linear order: the object is more prominent than the verb because the former follows the latter. This word-order-based approach to default prominence was adopted in a number of works, including Liberman and Prince (1977), Ladd (1980), Culicover and Rochemont (1983), and Halle and Vergnaud (1987). Starting with Schmerling (1976), a number of authors have criticized this approach to default prominence, based on both English and other languages. For instance, it cannot (p. 199) account for unaccusative sentences, where the subject is typically more prominent than the verb: (31)

Moreover, in Germanic OV languages, the object is more prominent than the verb, even though the latter is rightmost in the VP:

Page 14 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Focus Projection Theories (32)

Based on this and related evidence, a number of accounts have been proposed in which principles of default prominence make direct reference to argument structure, including Schmerling (1976), Gussenhoven (1983a), Jacobs (1988, 1991), Zubizarreta (1998), Schwarzschild (1999), Büring and Gutiérrez-Bravo (2001), and Büring (2006, 2012). The basic idea in this approach is that the overarching generalization in these cases is that the internal argument of a head (e.g. the subject of an unaccusative verb, or the direct object of a transitive verb) is more prominent than the head.19 With a similar objective of accounting for cross-linguistic generalizations involving default prominence, Cinque (1993) proposes that the crucial notion involved is syntactic depth of embedding, not argument structure. More specifically, Cinque proposes a very minimal Nuclear Stress Rule that, by applying cyclically to successively larger constituents, derives prosodic prominence (stress) on the most deeply embedded constituent in the sentence. In the Germanic transitive VP examples discussed above, relying on the fact that direct objects (more generally, complements) are phrasal, some word in the direct object is more deeply embedded than the verb within the VP, and thus is assigned prominence over the verb. Other syntax-centred accounts based on similar ideas include Zubizarreta (1998, whose analysis also takes argument structure into account) and Arregi (2002). A somewhat different type of syntax-centred approach to default prominence, based on cyclic phase-by-phase computation, can be found in Legate (2003), Kahnemuyipour (2004, 2009), Adger (2007), and Kratzer and Selkirk (2007), who develop ideas originally due to Bresnan (1971). An important variant of the Default Prosody approach is defended in Neeleman and Reinhart (1998), Szendrői (2001, 2003), and Reinhart (2006: ch. 3).20 In this version of the (p. 200) approach, the main factor determining focus projection is Default Prominence, but the relation between prosodic prominence and focus is not mediated by syntactic Fmarking. Rather, the Focus Prominence principle assumed in this analysis directly relates semantic focus to prosodic prominence in the sentence. For instance, the principle in Neeleman and Reinhart (1998) states that the focus set of a sentence consists of the constituents containing the main stress in the sentence. This means that several constituents can be the focus in a given sentence, but focus is not itself directly encoded in the syntax. Rather, discourse conditions determine which is the actual focus in the context the sentence is uttered. Finally, another aspect in which analyses of default prominence can differ has to do with the type of structure over which prominence is determined. In most of the works cited above, in addition to Truckenbrodt (1995), Samek-Lodovici (2005), and Féry (2011), Page 15 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Focus Projection Theories prominence is determined on the basis of prosodic structure, a structural representation of sentences that is built on—but which is independent of—their syntactic structure. This follows a long-standing tradition in phonology according to which the relation between syntax and phonology is mediated by prosodic structure. On the other hand, the syntaxcentred accounts in Cinque (1993), Zubizarreta (1998), Arregi (2002), and Kahnemuyipour (2004, 2009) make direct reference to syntax, without the mediating role of prosodic structure.

10.7 Unaccentable elements and the status of the Focus Prominence principle Despite the variety of approaches to focus projection discussed above, the principle of Focus Prominence, which requires the focused constituent to be prosodically prominent, remains essentially unchallenged. Data from Northern Biscayan Basque (NBB)21 prove to be problematic for this principle. The basic facts are discussed for the variety of Lekeitio in Hualde et al. (1994) (see also A. Elordieta 2002, G. Elordieta 2003, 2007, as well as Arregi 2002, 2006 for the variety of Ondarru). As illustrated below, many words in NBB are unaccentable in certain syntactic positions, even when interpreted as focused. This provides prima facie counterevidence to Focus Prominence, as these words can be (semantically) focused even if they are not prosodically prominent. In this final part of this chapter, I give a brief description of the relevant NBB facts, and propose a modification of Focus Prominence that can account for the data. In NBB, some words are lexically accented and others are not. The former always surface with a pitch accent (H*+L) on some syllable,22 and the latter only surface with a pitch accent (also H*+L) assigned at the phrase level in a specific syntactic position described below. For instance, ergative plural definite lagúnak ‘friends’ is lexically accented and thus always carries an accent on its penultimate syllable. On the other hand, its singular counterpart lagunak is not lexically accented, but can contain a phrasal accent when it surfaces in a specific syntactic position, namely when it is contained in a phrase immediately preceding the verb, which, as discussed more generally for Basque in Sections 10.2 and 10.3 above, is the prosodically most prominent phrase in the sentence. Within this phrase, the accent surfaces on the last syllable (in some varieties the phrasal accent is on the penultimate syllable). A lexically unaccented word can thus have an accent and carry prosodic prominence in a very limited set of contexts: it has to be the word containing the last (penultimate in some varieties) syllable in a phrase immediately preceding the verb. This is illustrated in the following sentence: (p. 201)

Page 16 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Focus Projection Theories (33)

The immediately preverbal subject neure lagunák has a syntactically determined phrasal accent that falls on the last syllable of the phrase. Both neure ‘my’ and ergative singular lagunak ‘friend’ are lexically unaccented, hence the only accent that is possible in this phrase is the phrasal accent on lagunak, which happens to contain the last syllable in this phrase. Furthermore, the phrase containing that accent must also contain the focus of the sentence, which follows from Focus Prominence, under the natural assumption that the phrasal accent makes the accented phrase prosodically prominent (see the discussion of Basque in Sections 10.2 and 10.3). However, the process that assigns this accent to the preverbal phrase treats this phrase as unanalysable prosodically,23 and can only place the accent on the last syllable of the phrase, which leads to clear violations of Focus Prominence. For instance, (33) can be uttered felicitously in a context requiring a focus on neure ‘my’ (e.g. with the continuation es seuriak ‘not yours’; see Hualde et al. 1994: 62). Even in this case, the phrasal accent must be on the last syllable of the phrase immediately preceding the verb, and must thus be assigned to the last word in this phrase. Under no circumstance can the phrasal accent be shifted to the first word (*neuré lagunak). This, however, does not prevent the first word neure from being interpreted as focused (contexts forcing focus on the last word lagunák or the entire phrase are of course also (p. 202) possible). Thus, (33) has a reading in which neure is focused, even though this word does not contain an accent, in clear violation of Focus Prominence.24 Note, however, that some version of Focus Prominence must be in play in NBB. If a word is focused, it must be contained in the phrase that contains a phrasal accent, even if the word itself is not assigned the accent. In (33), a focus on neure is only possible if it is contained in the phrase immediately preceding the verb; for instance, placing an overt direct object between the subject neure lagunak and the verb in (33) makes a focus on neure (or any other element in the subject) impossible. Although Focus Prominence as standardly understood is violated, something like this principle ensures that the focused constituent is in the correct syntactic position. Let us assume, following the discussion above, that the preverbal phrase is unanalysable prosodically, so that any of its subconstituents are in effect prosodically invisible. Then, the following modification to Focus Prominence, adapted from Arregi (2002: 172–175), accounts for the NBB facts:

Page 17 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Focus Projection Theories (34)

By restricting prosodic peaks to prosodically visible constituents, this reformulation of Focus Prominence takes into account the fact that, even though focus can affect the prosody of a sentence, there are certain (language-particular) prosodic principles and constraints that cannot be violated, even under focus. In (33), neure lagunák ‘my friend’ is prosodically visible, but its subconstituents are not, even if they are focused. The revised Focus Prominence principle thus determines that neure lagunák, not neure, has a prosodic peak (i.e. the phrasal accent on the last syllable of lagunák), even if the focus is just neure.

10.8 Conclusion The current literature on focus projection is largely based on a consensus on two basic principles: Focus Prominence, which requires a focus to be phonologically prominent, and Default Prominence, which determines prosodic prominence independently of focus. In effect, focus projection is an epiphenomenon, derived from these principles, and no mechanisms of actual focus projection are needed. Although the NBB facts reviewed in the last section show that more research is needed on the relation between prosody and focus, especially in languages outside Germanic, it seems that they do not warrant straying too far from this consensus.

Notes: (1) I would like to thank Michael Wagner, Caroline Féry, Shinichiro Ishihara, and an anonymous reviewer for very helpful comments and suggestions. All errors are mine. (2) The use of this expression to refer to the phenomenon described below is due to Höhle (1982: 99). (3) For the purposes of this chapter, ‘focus’ is understood as denotation focus in Krifka’s (2008) sense. (4) For the moment, I intentionally remain vague as to what exactly ‘prosodic prominence’ means, since theories of focus projection differ on this point. See Sections 10.3, 10.4 and 10.6 for details. (5) The present discussion abstracts away from details of clause structure that seem ultimately not to be relevant to these descriptive remarks. For instance, the constituent referred to as ‘VP’ here may be larger, under the assumption that V moves out of VP.

Page 18 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Focus Projection Theories What is important is that there be a constituent whose only overt subconstituents are the verb and the direct object. (6) Note that this description of (1) is compatible with the existence of phonetic differences correlating with the different focus readings, and, indeed, both Halliday (1967) and Chomsky (1971) note that those differences exist. See Gussenhoven (1983b), Breen et al. (2010), and references cited there, for relevant experimental literature showing that these phonetic differences are real. (7) See Krifka (2008) and Rooth (this volume), for discussion of the role that focus plays in determining question–answer congruency under the notion of focus assumed here. (8) Unless otherwise noted, Basque examples are from Batua, the standard dialect. (9) The first formulation of a principle along these lines in the generative literature is in Chomsky (1971: 201). (10) Jackendoff (1972: 237) states these two hypotheses in terms of a single condition that also subsumes (9) below: ‘If a phrase P is chosen as the focus of a sentence S, the highest stress in S will be on the syllable of P that is assigned highest stress by the regular stress rules.’ For his formal implementation, see Jackendoff (1972: 241–242). (11) See Hualde et al. (1994), A. Elordieta (2001: 130–143, 2002), Arregi (2002: ch. 4, 2006), and G. Elordieta (2003) for more detailed description and analysis of the facts in different dialects. (12) This also correctly predicts that F-marking on the subject in (3) is not possible, since the subject is not immediately preverbal in this sentence (Basque is an SOV language), and unlike English and other Germanic languages, sentence prominence cannot be shifted to a different constituent. As a result, a focused argument or modifier must surface to the immediate left of the verb, which may result in deviations from the default SOV order (Etxepare and Ortiz de Urbina 2003). In addition, these alternative orders impose tighter restrictions on focus projection. See A. Elordieta (2001: 138–142) and Arregi (2002: 189–199) for specific analyses of these facts within the Default Prosody approach. (13) See Rochemont (this volume) for discussion of the notion of givenness. The definition of ‘given’ and ‘givenness’ assumed here is from Krifka (2008). (14) On the relation between pitch accent and stress (including default phrasal stress) in this analysis, see Selkirk (1995: section 2). (15) This can be derived under the assumption that the verb is the head of the sentence. With respect to this, Selkirk (1995: 556) states that F-marking of the VP licenses Fmarking of the entire sentence ‘via licensing of the various intervening inflectional heads’. What seems to be implicit here is a principle to the effect that F-marking of the

Page 19 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Focus Projection Theories complement of a functional head h licenses F-marking of h, which may be considered a subcase of Argument Projection. (16) Example (19b) is also compatible with F-marking on just the verb, which must therefore be the focus. Thus, (19b) can also be an answer to What did Bill’s mother do to Bill/him?. (17) Due to space limitations, I cannot offer an overview of all the arguments in the literature against the F-projection approach, and limit myself to those related to the predictions in (23). Other arguments can be found in Wagner (2005: 285–303), Büring (2006: section 4), and Breen et al. (2010: 1092). (18) The observation that focus can project from a transitive subject (in German) is due to Jacobs (1988: 132; see also Jacobs 1991: 20–21). (19) Selkirk (1984, 1995), von Stechow and Uhmann (1986), and Rochemont (1986) can also be considered to be accounts along these lines, but, as discussed in Section 10.4 above, these works adopt an F-projection approach in which argument structure is relevant in determining F-projection, rather than default prominence. (20) The approach to focus projection adopted in Chomsky (1971) also falls within this variant of the Default Prosody approach. (21) The label ‘Northern Biscayan Basque’ is used in the literature on Basque accentuation (e.g. Hualde 1999, 2003) to refer to the group of varieties of Basque that have pitch-accent systems. This accent-based grouping does not correspond to standard dialectal groupings based on other criteria (i.a. Zuazo 2013). (22) The specific placement of the accent within the word is subject to some variation. See references cited above. (23) The phrasal domain the accent is assigned to can be quite large, e.g. embedded adjunct clauses, or DPs containing relative clauses. See Arregi (2006) for an analysis of these facts. (24) As expected, lexically accented words contrast with lexically unaccented words precisely in this respect. When focused, lexically accented words are prosodically prominent, even when not immediately preverbal. See references cited above for relevant examples.

Karlos Arregi

Karlos Arregi is Associate Professor of Linguistics at the University of Chicago. His research is in syntax and morphology, and their interaction, as well as their interfaces with phonology and semantics, with a special focus on Basque and

Page 20 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Focus Projection Theories Romance. He is the author of numerous articles in journals including Linguistic Inquiry and Natural Language Semantics.

Page 21 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Constraint Conflict and Information Structure

Oxford Handbooks Online Constraint Conflict and Information Structure   Vieri Samek-Lodovici The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara Print Publication Date: Jul 2016 Subject: Linguistics, Morphology and Syntax Online Publication Date: Mar 2015 DOI: 10.1093/oxfordhb/9780199642670.013.27

Abstract and Keywords This article examines the insights brought about by a conflict-based approach to the study of information structure. It does so mainly, but not exclusively, through a chronological survey of particularly significant analyses that modelled the syntactic displacements induced by focalization as the effect of prosodic constraints governing the position of prosodic prominence. The historic and conceptual relations between these analyses are highlighted, together with the main theoretical issues they raise and address. While most analyses are based on optimality theory, the article does not assume any prior knowledge of this framework and is accessible to all scholars. Keywords: prosody, focalization, constraint conflict, optimality theory, syntactic displacement

Page 1 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure WHETHER the universal constraints of grammar conflict with each other or not is an empirical question. If they do, analyses that take this into account will provide better models of the underlying linguistic reality. The study of information structure (IS) is of particular interest in this respect, as constraint conflict approaches have proved particularly apt at modelling the conflicting constraints affecting constituents carrying discourse functions and showing how the associated empirical data reflect the different ways these conflicts are resolved. Complex patterns can then be explained in terms of simple and independently necessary constraints, while handling exceptions in a principled way and deriving cross-linguistic variation with minimal appeal to language specific provisos. This chapter provides a survey of conflict-based analyses of IS in the last fifteen years (for a different approach see Aboh, this volume). Most analyses concern the effects of prosody on the syntax of focus, but purely syntactic analyses are mentioned too. Purely phonological studies on IS and prosodic phrasing are instead absent except for aspects of Truckenbrodt (1995) which fed many subsequent syntactic and prosodic analyses (but see Truckenbrodt, Zubizarreta, and Myrberg and Riad, in this volume). Conflict-based analyses not involving IS are also ignored except for Harford and Demuth (1999) which first considered how syntactic variation could emerge from the conflict between prosodic and syntactic constraints under an optimality-theoretic perspective. For reasons of space, I concentrate on what I consider particularly significant claims, inevitably simplifying the original analyses and keeping the references there cited to a minimum. To convey the relations holding across different analyses, I also use invariant names for constraints with identical or similar definitions (similar in that the differences are irrelevant for the patterns discussed here), often departing from the original names. Readers interested in specific analyses are strongly invited to consult the original papers. All analyses are cast in terms of Optimality Theory (Prince and Smolensky 1993, 2004). OT provides a formal definition of grammaticality under constraint conflict and (p. 204) a rigorous framework for testing the empirical predictions of conflict-based analyses. I will keep technicalities to a minimum and keep the discussion accessible to all readers. Nevertheless, it is worth mentioning now that under OT, constraints are universal and individual languages are identified by particular rankings of these universal constraints (where ‘C1>>C2’ indicates that constraint C1 is ranked above constraint C2). A linguistic expression is grammatical when its underlying structure is the best possible under the constraint ranking of the corresponding language (i.e. the structure is optimal for that language). A structure is optimal when it outperforms any other competing structure. This is determined as follows: given two structures A and B, A is better than B if the highest ranked constraint on which A and B differ is one that A violates less than B. The consequences of these definitions will become clear as soon as we consider concrete examples. Those interested in a more thorough introduction to OT may consult Prince and Smolensky (1993, 2004), Kager (1999), and McCarthy (2001). For a general review of OTsyntax see Müller (2015).

Page 2 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure This chapter contains three main sections. Section 11.1 illustrates the potential insights offered by a conflict-based perspective using a recent focalization study from Italian (Samek-Lodovici 2015). Section 11.2 surveys several past analyses of the prosody–syntax interface in relation to IS, highlighting the historic and conceptual relations holding between them. Section 11.3 concludes the chapter considering possible future developments.

11.1 Constraint Conflict Affecting the Focalization of Italian Post-verbal Constituents Italian VP-internal constituents can be contrastively focalized in situ (Brunetti 2004; Samek-Lodovici 2015). See for example the focused subject in (1a) which occurs postverbally consistently with its specVP position. Like any VP-internal negative constituent, it also requires licensing by the negative marker ‘non’ located in T (Penka 2011). This is expected if the subject occurs in situ and also shows that it is not placed in a focus projection a la Rizzi (1997, 2004) because negative phrases in this position need no licensing.1

(1)

The post-verbal area of the clause hosts movement operations that are nontrivially conditioned by the discourse status of the involved constituents. Consider the pattern in (2) from Samek-Lodovici (2015). An unfocused object can raise above a subject focalized in situ, as in (2a), but the same movement is ungrammatical under any other assignment of discourse functions. The object cannot move when contrastively focused (2b), nor when object and subject are both unfocused (2c), both contrastively focused (2d), or both within a wider presentationally focused phrase (2e). The corresponding data are provided in (3)–(7); native speakers must always read the context sentence before assessing the data. The symbols ‘F’ and ‘NewF’ mark contrastive and presentational focalization and main stress is represented in capitals. In (6), the first of the two foci carries secondary stress, represented in small caps. In (7) the position of the subject is marked by the stranded quantifier, represented as ‘QS’. The pattern is fully general, also (p. 205)

Page 3 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure applying to pairs involving subjects and sentential complements, objects and sentential complements, or lower adverbs in Cinque’s (1999) hierarchy.2

(2)

(3)

(4)

Page 4 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure (p. 206)

(5)

(6)

(7)

How do we model the observed movement? An analysis in terms of the intrinsic features of the moving constituent would have to explain why the movement in (2a) does not generalize to (2b–e). Stipulating which the feature triggering movement is only available (p. 207) to unfocused constituents accounts for the ungrammaticality of (2b), (2d), and (2e), but it incorrectly predicts (2c) to be grammatical since raising here involves an unfocused object. The problem is that (2a) and (2c) are only distinguished by the focused or unfocused status of the subject, which is a property external to the object that cannot be captured in terms of the object’s features. The ungrammaticality of (2c) thus

Page 5 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure constitutes a serious obstacle to any analysis where movement may only be triggered by the intrinsic features of the moving phrase. The paradigm instead follows straightforwardly from a constraint conflict perspective. As argued in Zubizarreta (1998), Szendrői (2001), and Samek-Lodovici (2005, 2015), the association of main stress with focus places focused items under conflicting requirements. The constraints governing stress placement in Italian, here collectively identified as ‘RIGHTMOST STRESS’, require stress to occur clause-rightmost and therefore favour the raising of lower unfocused constituents to the left of focus, as this aligns focus and the associated stress with the clause right edge. But this operation is opposed by the constraint against free movement, here called ‘STAY’ as in Grimshaw (1997) and capturing Chomsky’s (1995) economy of movement. As described below, the focalization pattern in (2) above emerges straightforwardly whenever the conflict between RIGHTMOST STRESS and STAY is resolved in favour of RIGHTMOST STRESS. This is best expressed through an OT analysis where for every possible assignment of discourse functions to the subject and the object, the structure raising the object and the one leaving it in situ compete against each other and are assessed relative to the ranking RIGHTMOST STRESS>>STAY (i.e. RIGHTMOST STRESS dominates STAY), with the structure that satisfies this ranking best being chosen as grammatical. As explained below, this ranking selects raising the object as optimal only when the subject is focused and the object unfocused, as in (2a) above, whereas leaving the object in situ is optimal in all other cases. When the subject is focused and the object unfocused, raising the object as in (2a) is optimal because it leaves the focused subject and the associated stress clause right-most, thus satisfying RIGHTMOST STRESS at the cost of violating STAY. This is illustrated in tableau (8) (where the symbol ‘*’ represents individual constraint violations while the main stress is represented as ‘x’). The structure leaving the object in situ in (8b) violates RIGHTMOST STRESS because the object intervenes between main stress and the clause right edge, but it satisfies STAY because it involves no movement. The structure raising the object in (8a) instead satisfies RIGHTMOST STRESS but violates STAY. Since RIGHTMOST STRESS outranks STAY, the conflict is resolved in favour of RIGHTMOST STRESS, thus choosing movement in (8a) against in-situ realization in (8b). Following OT conventions, the optimal structure is signalled by the hand symbol ☞. (8)

Page 6 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure Any other discourse function assignment selects in-situ realization of the object as optimal, completing the derivation of paradigm (2). Consider first tableau (9), illustrating the case where focalization affects the object while the subject is left unfocused. Leaving the object in situ in (9a) satisfies both constraints because movement is absent and stress occurs rightmost, whereas raising the object in (9b) violates both constraints because movement is present and the subject intervenes between stress and the clause right edge. Consequently, in-situ realization in (9a) is optimal, accounting for the ungrammaticality of object raising in pattern (2b). (p. 208)

(9)

When the subject and the object follow a focused verb and are both unfocused, RIGHTMOST STRESS is violated by both structures because the main stress on the verb is always two constituents away from the clause right edge; see (10a) and (10b). Raising the object as in (10b) adds a violation of STAY, making this structure less optimal than (a). This accounts for pattern (2c) above. (10)

When subject and object are both focused, as in (11a) and (11b), RIGHTMOST STRESS is satisfied by both structures provided that stress falls on whichever of the two constituents occurs rightmost. Once again, raising the object adds a violation of STAY, making (11b) suboptimal relative to (11a). This accounts for pattern (2d).

Page 7 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure (11)

Similarly, when subject and object participate to a wider presentational focus, as in (12), RIGHTMOST STRESS is satisfied by both structures provided that stress falls on whichever of the two constituents occurs rightmost. As before, raising the object adds a violation of STAY and is suboptimal, accounting for pattern (2e). (p. 209)

(12)

A finer-grained analysis of the conflict between stress alignment and movement is available in Samek-Lodovici (2005, 2015). Some of the theoretical insights made available by a constraint conflict perspective, however, are already apparent in the analysis just described. Simple, independently needed constraints—Paradigm (2) is accounted for in terms of the interaction of simple constraints which are independently needed for Italian but also outside it, since rightmost stress assignment and movement economy are found in many languages. By exploiting their conflict, the analysis asserts that paradigm (2) is a direct consequence of the constraints ultimately governing stress placement and movement economy. This is significant, as it avoids saddling the grammar with constraints specific to the observed paradigm, such as ‘no scrambling above higher generated foci’. Paradigmspecific constraints tend to be excessively descriptive and unnecessarily add to the complexity of human grammar, making its acquisition appear more complex than it needs to be. Cross-linguistic variation—Paradigm (2) is a direct consequence of ranking RIGHT- MOST STRESS above STAY. Grammars with the opposite ranking would resolve the conflict in tableau (8) above in favour of STAY, preferring (8b) to (8a), thus disallowing object raising even in (2a). The corresponding language would differ from Italian in that all patterns in (2) would be ungrammatical, including (2a). These are also the only languages predicted to be possible: unless new constraints are introduced, languages where all Page 8 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure patterns in (2) are grammatical, or where everything is grammatical but (2a) (the mirror image of Italian) are predicted to be impossible. Cross-linguistic variation thus emerges from the different structures that are selected as optimal across different rankings of the same constraints (Prince and Smolensky 1993, 2004). Cross-linguistic variation is also rooted in UG constraints, being present only where they conflict and manifested in the forms dictated by the conflicting constraints. Interface modelling—Interface phenomena emerge from the conflicts arising between constraints in distinct linguistic modules (e.g. prosody and syntax). The conflict is the way distinct modules interface with each other. While this does not eliminate the need for interface constraints, it reduces their complexity by restricting the set of phenomena requiring them. For example, stress placement in Italian and other Romance languages (p. 210) is best analysed through purely prosodic constraints requiring right-aligned prominence across all major prosodic constituents in Selkirk’s prosodic hierarchy (Selkirk 1984, 1986, 1995; Truckenbrodt 1995). We may model stress in this principled way and still derive paradigm (2) in terms of the conflict between STAY and the prosodic constraints just mentioned (Samek-Lodovici 2005, 2015. See also Section 11.2). Once again, this leads to a model of human grammar with fewer, simpler, and independently required constraints. Systematic exceptions—Systematic exceptions to otherwise regular paradigms should receive principled explanations. Constraint conflict predicts the existence of such exceptions: they are bound to arise for any structures affected by higher-ranked constraints that happen to conflict with the constraints immediately responsible for the observed empirical paradigm. For example, paradigm (2) extends beyond subjects and objects, but focused finite verbs constitute a systematic exception because they raise above unfocused subjects in specVP. This challenges the analysis because focused items should not raise above higher unfocused ones as this worsens stress alignment. Under a constraint conflict perspective, however, such exceptional behaviour is the inevitable consequence of the constraint forcing verb raising in Romance. When ranked higher than RIGHTMOST STRESS, this constraint will raise focused verbs from V to T even if this adversely affects stress placement. The same constraint does not apply to any other VPinternal constituents, explaining why these items follow the distribution in (2). What is noteworthy here is the possibility to account for the exceptional behaviour of finite verbs through an independently established non-controversial property of theirs rather than by stipulation. The same solution is not as straightforwardly available outside constraint conflict, because the conditions being posited to prevent focused items from raising are likely to apply to finite verbs as well, unless verbs are excluded by stipulation. However worded, such a stipulation is the price paid for assuming constraint conflict to be impossible.

Page 9 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure

11.2 Constraint Conflict at the Syntax–prosody Interface The last fifteen years have seen a flourishing of analyses that accounted for focalization paradigms across a substantial set of languages by exploiting the conflict between syntactic constraints governing the position of specific constituents and prosodic constraints governing stress placement. Most of them benefit of the theoretical properties discussed in the preceding section. Here, I describe a few, chosen for the particular properties they illustrate.

11.2.1 Harford and Demuth (1999)—Prosody Forcing Verb Raising Harford and Demuth contributed one of the very first OT analyses showing how syntactic phenomena can emerge from conflicts between syntactic constraints and (p. 211) independently attested prosodic ones. While not directly related to IS, their study showed that the direct interaction between syntax and prosody could lead to a simpler model of syntax, one free of the additional conditions that would otherwise be needed to account for the same phenomena. Furthermore, if prosodic constraints affect the syntax of focus, as claimed by the other analyses examined in this section, we ought to see syntactic effects even outside focalization and Harford and Demuth’s study provides such an example. Specifically, Harford and Demuth (1999) observe how the order of subject and verb in the object relative clauses of the SVO Bantu languages Sesotho and Chishona depends on the prosodic weight of the complementizer introducing the relative clause. In Sesotho, where the complementizer tseo is disyllabic, the subject obligatorily precedes the verb. In Chishona, which involves a monosyllabic complementizer dza, the complementizer obligatorily cliticizes onto the verb once the verb has raised to C, resulting in subject verb inversion (except for a marginally acceptable marked construction where the subject precedes the complementizer). A similar alternation is found in Kiswahili, where the SV order is found after the multisyllabic complementizer amba-cho, formed by the lexical item amba plus the relative morpheme cho expressing agreement with the relative head noun, but not when the complementizer is reduced to the monosyllabic cho alone, which always forces verb raising to C and the ensuing subject verb inversion.

(13)

Page 10 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure (14)

Harford and Demuth show that the observed patterns emerge from the conflict between the independently attested constraint MINPW (minimal prosodic word) requiring words to be at least bimoraic (McCarthy and Prince 1991), which in the Bantu languages discussed here translates in the required presence of at least two syllables, and the constraint STAY against movement introduced in the previous section. Multisyllabic complementizers, such as Sesotho’s tseo, satisfy MINPW and therefore can occur as free-standing words. This, in turn, allows the verb to remain in T, determining the SV order shown in (15a) below. Raising the verb to C as in (15b) violates STAY and provides a suboptimal structure when compared to (15a). (15)

Monosyllabic complementizers, like Chishona’s dza, violate MINPW when occurring as free-standing words, see (16a). Raising the verb to C, as in (16b), violates STAY but provides a head the complementizer can cliticize to, forming a multisyllabic word that satisfies MINPW. Under the ranking MINPW>>STAY, which Harford and Demuth claim to occur across all three languages, (16b) outperforms (16a) because it satisfies the higher ranked MINPW and is selected as optimal, thus determining the observed subject inversion pattern. (p. 212)

(16)

The conflict between MINPW and STAY thus immediately accounts for the correlation between verb raising and the prosodic weight of complementizers in these Bantu languages. A prosody-free account would end up modelling the effects of MINPW through additional syntactic conditions that Harford and Demuth’s analysis shows to be unnecessary.3

Page 11 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure

11.2.2 Truckenbrodt (1995)—Focalization Affecting Prosodic Phrasing This survey cannot address the vast literature focusing on the prosodic effects of IS, but Truckenbrodt’s (1995) seminal research in this area needs to be mentioned because the prosodic constraints there proposed have frequently been used in many subsequent analyses of IS-effects in syntax. Here, I consider his analysis of the prosodic phrasing of the Bantu languages Chi Mwi:ni, Chicheŵa, and Kimatuumbi. As for the effects of focalization, I examine Chicheŵa alone (Truckenbrodt 1995: ch. 6). Truckenbrodt notes how the distinct mappings of ditransitive VPs into phonological phrases in these three languages follow from three simple constraints. The constraint NONREC against recursion disallows placing a phonological phrase (or ‘pp’) inside another pp. The constraint STRESSXP requires lexically headed maximal projections to be prosodically prominent in the pp containing them (i.e. it requires them to be stressed at pp-level). The constraint WRAPXP requires lexically headed maximal projections to be contained into a single pp. The different pp-phrasings found in Chicheŵa, Chi Mwi:ni, and Kimatuumbi when none of the items in the VP is narrowly focused are provided in the first column of (17), (p. 213)

with pp-boundaries shown as parentheses and pp-level stress represented as an ‘x’ mark over the stressed constituent. Each distinct phrasing violates one constraint. Chicheŵa violates STRESSXP because the first NP is not phrasally stressed.4 Chi Mwi:ni violates WRAPXP because the entire VP, including the verb and both NPs, is not contained in a single pp. Kimatuumbi violates NONREC due to pp-recursion. (17)

Each phrasing arises from a specific ranking of these three constraints. For example, when NONREC and WRAPXP both dominate STRESSXP, wrapping all items into a single pp as in Chicheŵa provides the best solution to the conflict involving these three constraints, as the other two phrasings violate either NONREC or WRAPXP. Similarly, Chi Mwi:ni’s phrasing emerges when NONREC and STRESSXP dominate WRAPXP and that of Kimatuumbi when WRAPXP and STRESSXP dominate NONREC.

Page 12 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure Crucially, the same independently established constraints also account for how focalization affects Chicheŵa’s prosodic phrasing through their interaction with the additional constraints FOCUS and ALIGN-Ø (defined in Truckenbrodt 1995: 178–82). The first constraint, here renamed ‘STRESSFOCUS’, requires focused constituents to receive the highest prosodic prominence in their focus domain. The second constraint, here renamed RIGHTSTRESS(pp), demands every pp to show rightmost stress, thus being similar to the RIGHTMOST STRESS constraint discussed in Section 11.1 but defining the pp as the relevant domain. In Chicheŵa, focalization forces a pp-boundary at the right edge of the focused constituent. For example, focalizing the first object of a ditransitive VP forces a ppboundary after that object, see (18a) below. This new phrasing diverges from the discussed VP-wide pp found when narrow focalization is absent. Under Truckenbrodt’s model, the prosodic rephrasing induced by focalization follows from the constraint ranking characterizing Chicheŵa, where STRESSFOCUS and RIGHTSTRESS(pp) dominate WRAPXP, and WRAPXP dominates STRESSXP (thus consistently with the earlier ranking where WRAPXP dominated STRESSXP). As tableau (18) shows, the attested phrasing in (18a) violates WRAPXP because the VP is no longer contained in a single pp, but it satisfies the (p. 214) higher ranked constraints STRESSFOCUS and RIGHTSTRESS(pp) by respectively ensuring that the focused NP is stressed and that each pp is rightmost stressed. Furthermore, the ranking ensures that the alternative phrasings in (18b) and (18c) are all suboptimal relative to (18a). The stress in structure (18b) is not pp-rightmost, as the second NP intervenes between stress and the right pp-edge. Structure (18b) fails STRESSFOCUS because the focused NP is unstressed. (The constraint NONREC has been omitted because it is satisfied by each listed competitor.) (18)

Amongst its many merits, Truckenbrodt’s analysis is significant because it shows how language-internal variation induced by focalization in Chicheŵa and cross-linguistic variation in pp-phrasing both follow from the same set of simple, general, universal constraints, none of which details how specific constituents are to be phrased nor how focus is to affect phrasing. The observed phrasing patterns all emerge as the best

Page 13 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure possible solutions, under specific constraint rankings, of the conflicts that arise between the proposed constraints.

11.2.3 Samek-Lodovici (2005)—Focus-induced Variation at the Syntax–Prosody Interface Truckenbrodt’s (1995) model of prosodic phrasing enabled linguists to apply Harford and Demuth’s hypothesis about the interaction of prosody and syntax to focalization patterns. Since then, many focalization paradigms across several languages have been shown to emerge from the conflicts arising between syntactic constraints and Truckenbrodt’s prosodic ones (on prosody affecting word order see also Neeleman and van de Koot, this volume). Most of these analyses concern patterns in a single language. Under OT, however, different rankings of the same constraints impose different solutions to existing constraint conflicts, potentially accounting for cross-linguistic variation in the syntax of focalization much like Truckenbrodt-derived variation in prosodic phrasing. For example, Samek-Lodovici (2005) derives aspects of the focalization paradigms of Italian, French, and English (amongst other examined languages). When the entire clause is focused, these languages share an ‘S V O IO’ order and also a (p. 215) similar prosodic structure where the entire clause is wrapped in a single intonational phrase ip with rightmost stress. As the table in (19) shows, cross-linguistic variation emerges when focalization is narrowed to single constituents. Focused subjects are necessarily preverbal in English and French but may occur post-verbally in Italian. Focused objects precede indirect objects in English but may follow indirect objects in Italian and in (some varieties of) French. Examples involving simple tenses show a similar distribution, but the presence of V-to-T movement in Italian and French and its absence in English slightly adds to the complexity of the analysis and is thus here ignored. (19)

Greatly simplifying the original analysis, variation is analysed as arising from the conflict between the syntactic constraints STAY and EPP—where STAY penalizes movement and EPP requires subjects to raise to specTP (Grimshaw and Samek-Lodovici 1995, 1998; Grimshaw 1997)—and Truckenbrodt’s prosodic constraints STRESSFOCUS and RIGHTSTRESS(ip), where STRESSFOCUS requires focus to carry stress while RIGHT Page 14 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure STRESS(ip) requires stress to occur rightmost in the ip encompassing the clause, thus providing a more accurate definition of the constraint RIGHTMOST STRESS used in Section 11.1. Leaving STRESSFOCUS aside, although I will return to it later, the three focalization patterns in (19) arise from the three rankings in (20).

(20)

When the entire clause is focused, subjects raise to specTP in all three languages. As (21) shows, subjects raise because in all these languages EPP dominates STAY. This ranking favours the raised subject in (21b), which satisfies EPP, rather than the subject in-situ of (21a), which satisfies STAY but violates the higher ranked EPP. The ranking of RIGHTSTRESS(ip) is irrelevant here since it is satisfied by both structures by stressing the verb. (For space reasons the listed structures abstract away from past-participle movement, thus showing the verb in-situ preceded by in-situ subjects even though pastparticiple movement forces the opposite order in Italian and French. The violations of STAY caused by the omitted movement affects all competing structures equally and can therefore be safely ignored.) (p. 216)

(21)

The position of focused subjects in each language follows from the ranking of RIGHTSTRESS(ip) relative to EPP. If RIGHTSTRESS(ip) dominates EPP, as is the case in Italian, focused subjects remain in situ and the verb raises above them as in (22a). Under the opposite ranking, found in English and French, focused subjects raise to specTP as in (22b).

Page 15 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure (22)

The position of focused objects, instead, follows from the ranking of RIGHTSTRESS(ip) relative to STAY. If RIGHTSTRESS(ip) dominates STAY, as in Italian and the relevant varieties of French, the indirect object raises above the object as in (23a). Under the opposite ranking, found in English, the indirect object remains in situ as in (23b). (23)

The different positions of focalized constituents across otherwise similar languages thus emerge from the possible resolutions of the conflicts arising between syntactic and prosodic constraints induced by narrow focalization. Constraint conflict makes it possible to derive the distinct patterns in terms of fully general, universal, constraints, none of which directly concerns the position of focalization. Descriptive parameters, such as Vallduví’s (1991) plasticity parameter, which separate languages where focalization affects prosody (e.g. English) from those where it affects syntax (e.g. Italian), can be dispensed with, since the properties they describe follow from the conflicting constraints themselves. Similarly, Zubizarreta’s (1994, 1998) insight about the central role played by prosody in Romance focalization is accounted for through the independently necessary prosodic model of Truckenbrodt (1995) rather than by positing syntax-based parametric stress rules as in Zubizarreta (1998, this volume).

11.2.4 Zerbian (2006), Downing (2006)—Extending the Crosslinguistic Typology to Unstressed Foci (p. 217)

Truckenbrodt (1995), Samek-Lodovici (2005), and most analyses of focalization exploiting the conflicting demands of prosody and syntax under focalization concern languages where the constraint STRESSFOCUS outranks any other conflicting constraint, thus ensuring its satisfaction. For example, in Samek-Lodovici (2005) STRESSFOCUS dominates EPP in Italian and RIGHTSTRESS(ip) in English. Were this not the case, Page 16 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure placing stress on unfocused verbs would incorrectly be selected as the optimal realization of focused subjects in intransitive clauses in both languages. This is shown in (24) where stressing the final unfocused verb in (24a) violates STRESSFOCUS and yet it beats the stressed post-verbal focused subject found in Italian and shown in (24b) because (24b) violates the higher ranked constraint EPP. Structure (24a) also beats the stressed preverbal focused subject of English in (24c) because (24c) violates the higher ranked constraint RIGHTSTRESS(ip). (24)

If cross-linguistic variation follows from constraint reranking, as predicted by OT, there ought to exist languages that rank STRESSFOCUS sufficiently low in the constraint hierarchy as to leave focused constituents unstressed. Zerbian (2006) shows that the Bantu language Northern Sotho instantiates one such language. The focused constituents of this language are never marked by pitch accents, prosodic phrasing, or tone sandhi. Yet, as Zerbian points out, its focalization pattern follows from the same conflicting constraints proposed in Samek-Lodovici (2005) and Truckenbrodt (1995), provided that RIGHTSTRESS(ip) and STAY both dominate STRESSFOCUS. A similar analysis is also available in Downing (2006) for unstressed foci in the Bantu language Chitumbuka.5 (For a more general survey of IS in Bantu languages see Downing and Hyman, this volume.) Amongst several other results, Zerbian (2006: 166) shows that when focalization affects the verb of a transitive sentence, the canonical SVO order of (p. 218) the language remains unaltered. Rightmost stress—which in this language is expressed via syllabic lengthening—falls on the unfocused final object, see (25a), rather than on the focused verb. Competing structures that stress the focused verb, whether raising the object as in (25b) or leaving it in situ as in (25c), respectively violate the higher ranked constraints STAY and RIGHTSTRESS(ip) and are therefore ungrammatical.

Page 17 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure (25)

Zerbian’s and Downing’s results are significant in many respects. First, they add to the growing cross-linguistic typology of focalization patterns determined by the conflict between syntactic and prosodic constraints. Second, they provide interesting evidence for an OT approach to constraint conflict by showing that even constraints long held as inviolable, such as the association of focalization with prosodic prominence (Jackendoff 1974), are indeed violable and violated, as expected if conflicting constraints are reranked across different languages. Finally, they provide a particularly clear case for the need to distinguish formal constraints from the empirical generalizations that support them (McCarthy and Prince 1994). As their analyses show, the observation that the empirical effects mandated by STRESSFOCUS are absent in some specific languages, does not entail that the constraint STRESSFOCUS is not present in the grammars of those languages. Rather, what is specific to these languages is the low rank of STRESSFOCUS relative to RIGHTSTRESS(ip) and STAY: their satisfaction forces the violations of the lower ranked STRESSFOCUS and prevents it from having visible effects. Under a constraint conflict perspective, the analytical worth of a constraint must thus be assessed from the overall success of the analyses it is part of and the contribution it brings to those analyses, rather than the existence of direct empirical correlates across all known languages.6

11.2.5 Dehé (2005)—Revealing the Syntactic Effects of Prosody Where They are Least Expected (p. 219)

Dehé examines the impact of focalization on the syntax of particle verbs in English and German. Her analysis is significant in two important respects. First, it shows that prosodic constraints affect word order in English as well, thus challenging the widespread view that English word order is rigid and entirely driven by syntactic constraints as proposed in Vallduví (1991), Zubizarreta (1998), and Szendrői (2001). Second, her analysis provides independent evidence for Samek-Lodovici’s (2005) claim that constraint conflict concerns individual constraints, not entire linguistic modules such as Syntax vs Prosody, thus allowing individual constraints across distinct modules to be

Page 18 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure intermingled as in the ranking for French focalization ‘EPP>>RIGHTSTRESS(ip)>>STAY’ described above where the prosodic constraint occurs sandwiched between two syntactic ones (see also the discussion of Szendrői 2001 in Section 11.2.6). With respect to English particle verbs, Dehé (2005) observes that when the object is focused, the verb and its particle are contiguous and precede the final stressed object, see (26). When focus narrows to the verbal complex alone, however, the verbal particle must follow the object and carry main stress, see (27). The two sentences are respectively assigned the structures in (28). In both, the object moves to SpecAgrOP for case reasons and the verb moves from V to v. The only difference concerns whether the latter movement involves the entire verb–particle complex, as in (28a), or just the verb, with the verbal particle stranded in situ as in (28b). (26) (27) (28)

According to Dehé, pied-piping verbal particles to v as in (28a) violates the syntactic constraint NPPP (NoParticlePiedPiping). The alternation between (26) and (27), as well as several other data examined in Dehé’s paper, can then be accounted for in terms of the conflict between RightStress(pp) and NPPP. When the object is focused, pied-piping the verbal particle as in (29a) below violates NPPP but places stress pp-finally as required by RIGHTSTRESS(pp). Stranding the verbal particle as in (29b) satisfies NPPP but violates RIGHTSTRESS(pp) because stress is no longer rightmost. The ungrammatical status of (29b) shows that RIGHTSTRESS(pp) outranks NPPP, thus letting (29a) emerge as optimal. (p. 220)

(29)

When focus affects the verb, pied-piping the verbal particle as in (30a) violates NPPP as well as RIGHTSTRESS(pp), because stress is no longer rightmost. Stranding the verbal particle in V as in (30b) is instead optimal, as it satisfies NPPP by avoiding pied-piping and also RIGHTSTRESS(pp) by placing stress rightmost. The different positions taken by Page 19 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure verbal particles under different focalization assignments are thus a direct consequence of the prosodic constraint RIGHTSTRESS(pp), which prevents particles from stranding when they are unfocused even though their stranding is favoured by the lower ranked NPPP. (30)

The prosodic constraint RIGHTSTRESS(pp) is itself dominated by syntactic constraints. As Dehé points out, this conclusion follows when considering the position of focused simple verbs. As (31) shows, focused simple verbs raise to v rather than being stranded in V as in (31b), even though the latter structure would place stress rightmost as required by RIGHTSTRESS(pp).

(31)

Dehé attributes this pattern to the effects of the syntactic constraint OBHD (OBLIGATORY HEADS) requiring syntactic heads to be non-empty (Grimshaw 1997; Vikner 2001). As shown in (32), the higher ranking of this constraint relative to RIGHTSTRESS(pp) ensures that simple verbs are raised to v even when focused, even though this places stress nonrightmost and violates RIGHTSTRESS(pp). (32)

(p. 221)

(33)

Dehé’s analysis thus supports a view of grammar where prosodic constraints affect word order even in languages with an otherwise fairly rigid syntax such as English. Furthermore, the ranking emerging from the above discussion, in (33), alternates

Page 20 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure syntactic and prosodic constraints, thus supporting the hypothesis that cross-modular conflict concerns individual constraints rather than linguistic modules in their entirety.

11.2.6 Szendrői (2001) and Other Analyses Szendrői’s constitutes one of the first conflict-based studies of the relevance of prosody for IS. It proposes that conflict affects linguistic modules in their entirety, not individual constraints from distinct modules (although constraints may still conflict within each module). Linguistic expressions are represented as pairs involving a syntactic and a prosodic representation. Prosody and Syntax act as meta-constraints that order these pairs according to the wellformedness of their prosodic or syntactic representation. A third module, named ‘Map’, orders the pairs according to constraints that govern the mapping between prosodic and syntactic representations. These three modules conflict whenever they select different pairs as best wellformed. The expression selected as grammatical depends on the ranking of Prosody, Syntax, and Map in each language. For example, English ranks Syntax and Map above Prosody and consequently shows an invariant syntax and a flexible stress system (Szendrői 2001: 173). Amongst several interesting results, Szendrői’s model derives the distribution of focalization in Hungarian from its stress system, thus dispensing with the need for a dedicated focus position. It also accounts for focalization patterns in English and Italian through distinct rankings of the linguistic modules. Szendrői’s research raises the issue of whether conflict may involve individual constraints across distinct modules. As mentioned, such conflicts are claimed to be necessary in Dehé (2005) and Samek-Lodovici (2005), but see Teeple (2007) for a defence of Szendrői’s approach and a critique of Dehé and Samek-Lodovici. The issue is whether the results in Dehé’s, Samek-Lodovici’s, and other similar analyses, can still be derived once crossmodular conflicts between individual constraints are removed. Further analyses of IS based on the prosody–syntax interface—The last fifteen years have seen many analyses examine the key role played by prosody in shaping IS, especially with respect to focalization. Several of them directly contributed to the research programme outlined by the studies described so far by examining the impact on the syntax of focalization of constraints strictly or loosely based on Truckenbrodt’s (1995) prosodic model. They are listed below ordered by the examined language. Arabic—Teeple (2011) on focus and accusative pronouns in Arabic. Dutch—Bouma and de Hoop (2008) on pronominal scrambling. French—Hamlaoui (2008) on the preference for cleft-sentences in specific focalization contexts. Hamlaoui (2011) on the influence of prosody and discourse factors on the syntax of wh-phrases.

Page 21 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure German—Büring (2001, 2002) on focalization and (in)definiteness in German double object constructions. Schmid and Vogel (2004) on dialectal variation in verbal clusters. Féry (2006) on object fronting under sentence-wide focus. English—Anttila et al (2010) on the dative shift alternation and its triggers. Mandarin Chinese—Li (2009) on information focus. Spanish—Hoot (2012) on rightmost focus. Western Chadic Languages—Zimmermann (2006) on a unified analysis of focalization in Hausa, Tangale, and Guruntum. Lovestrand (2009) on focalizaton in Hausa. Cross-linguistic studies—Büring and Gutiérrez-Bravo (2002) and Gutiérrez-Bravo (2002) on focus-induced word order variation in English, Spanish, and German. Dehé (2004) on word order variation in double object constructions in Icelandic and German. Vogel (2006) on object shift in Danish, Swedish, and English. Büring (2006) on the potential elimination of focus projection rules with data from English and German. A second set of analyses pursued a similar programme but under a substantial revision of Truckenbrodt’s prosodic model that involves the addition of phase-sensitive prosodic constraints following and adapting Kratzer and Selkirk (2007), where the boundaries of prosodic phrases are required to match those of syntactic phases.7 See amongst others Cheng and Downing (2009, 2012) on the focalization and dislocation patterns of Durban Zulu, Elfner (2011, 2012) on the different positions of weak and strong pronouns in Modern Irish and Scottish Gaelic, López (2009) on the conflict between linearization and prosodic phrasing and its relevance for the representation of left and right dislocation in Spanish, and Göbbel (2012) on how prosodic phrasing affects the availability of relative clause extraposition in English. (p. 222)

Non-prosodic analyses—Older analyses and a few more recent ones propose constraints that directly encode the preferred position of focused or topic constituents, rather than deriving their position from their relation to prosody. They include, amongst others, Grimshaw and Samek-Lodovici (1995, 1998) and Samek-Lodovici (1996) on the impact of IS on the syntax of Italian subjects; Bresnan (1997) on Chicheŵa’s pronominal system; Sells (1997) on Korean morphology; Samek-Lodovici (1998, 2001) on focalization in Kanakuru, a Chadic language, and its cross-linguistic implications; Costa (1998, 2001) on IS and word order variation in several languages; Choi (1999, 2001) on the import of IS in scrambling; Legendre et al. (2000) on IS factors in BasqueV2; Payne and Chisarik (2000) on focus and negation in Hungarian; Choi (1999), Müller (1999), Heck (p. 223) (2000), and Büring (1997) on Lenerz’ paradigm8 (Lenerz 1977); Keller and Alexopoulou (2001) on IS in Greek; Engdahl et al. (2003) on Swedish word order; Vikner and Engels (2006) on object shift and scrambling in Scandinavian and Continental West Germanic languages; Gabriel (2010) on the distribution of focalization in Argentinian Spanish; and El-touny (2011) on focalization in Cairene Arabic wh-phrases.

Page 22 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure

11.3 Conclusions Besides the issues identified in the above discussion, such as the potential restrictions on cross-modular conflicts, there are three main directions in which research in this area might develop. A first obvious one involves the study of IS in additional constructions and languages such as those examined in Aboh (this volume), adding to the empirical coverage of conflict-based analyses. A second desirable development would be the analysis of IS-functions other than focalization; a notable instance in this respect is Choi’s (1999) analysis of four different discourse functions in terms of the possible values of two underlying features [±discourse prominence (i.e. saliency)] and [±new]. Finally, the existing analyses could be integrated further, possibly identifying a single set of constraints that accounts—via constraint reranking—for the entire IS typology currently described. As this survey shows, this is already happening, as IS-paradigms across several different languages have been successfully analysed through similar constraints. But further progress appears possible. For example, could OBHD be replaced by the MINPW constraint used by Harford and Demuth (1999), with head movement occurring to provide the target position with sufficient prosodic content? If so, the pervasive V-to-T and T-to-C movement would itself be an instance of prosodically-induced syntactic movement. Similarly, while the NPPP constraint in Dehé (2005) cannot be directly reduced to STAY (c.f. Dehé 2005: 221, fn4), it could perhaps be reduced to specific rankings of the simpler constraints shown to derive the effects of STAY in Grimshaw (2001, 2002, 2006)9. The typological import of Grimshaw’s constraints in all other examined analyses is also worth pursuing. There is clearly plenty of interesting work ahead.

(p. 224)

Notes: (1) Placing the subject in Rizzi’s focus position also requires remnant movement of the TP non parlerà to a higher position. This disrupts the c-commanding relation between ‘non’ and ‘nessuno’ necessary for licensing (Cardinaletti 2001; Samek-Lodovici 2006, 2015). (2) The ungrammaticality of (2b) is not caused by the focused subject intervening between ‘non’ and the negative object. As the grammatical (5a) shows, a focused verb can intervene between non and nessuno without disrupting licensing. Similarly, focused subjects may intervene between non and an unfocused negative object; see (i) where the auxiliary inflection makes MILANESI a subject (Samek-Lodovici 2015).

Page 23 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure ((i))

(3) Harford and Demuth’s analysis also suggests the interesting possibility that the constraint OBLIGATORY HEADS, requiring heads to be syntactically realized (Grimshaw 1997) and often used to explain verbal movement (e.g. Dehé 2005), might at least in part be subsumed by MINPW. Empty syntactic heads would attract lower heads in order to become legitimate minimal words. I leave this suggestion open to further research. (4) The VP as a whole satisfies STRESSXP, because the phrasal stress on the second NP also counts as stress for the entire VP. The same holds for the Chi Mwi:ni’s and Kimatuumbi’s structures. (5) Like Zerbian (2006), Downing (2006) argues for a conflict-based analysis where STRESSFOCUS is dominated by prosodic constraints. See also Cheng and Downing (2009: fn13, 2012: fn7) involving lack of stress on focus in Durban Zulu, but proposing new constraints that might affect the analysis of Chitumbuka (Downing p.c.). (6) This does not imperil falsifiability, including falsifiability of specific constraints. See for example Samek-Lodovici (2005: 735ff) where the need to account for non-culmination in Chicheŵa forces a weaker version of STRESSFOCUS. See also Ishihara (2011) and Hoot (2012). (7) Preliminary comparisons between Truckenbrodt (1995) and Kratzer and Selkirk (2007) are available in Truckenbrodt (2012) and Elfner (2012). An overview of different theories of the syntax–prosody interface is available in Elordieta (2008). Some problematic aspects in Kahnemuyipour’s (2004) stress model, which inspired Kratzer and Selkirk (2007), are considered in Samek-Lodovici (2013). (8) Lenerz’ paradigm concerns the scrambling of German objects above preceding indirect objects, which is possible when the object is unfocused and the indirect object focused but not vice versa. As mentioned by an anonymous reviewer, it would be interesting to examine the potential relations between Lenerz’ paradigm and paradigm (2) at the beginning of this chapter. (9) Specifically, Grimshaw’s constraints ALIGNHEADLEFT and ALIGNCOMPLEFT would favour raising a simple verbal head V above raising a complex verbal head ‘V+P’, since the latter gives rise to additional left-alignment violations. It remains to be examined whether an appropriate ranking exists that is consistent with this intuition and Dehé’s analysis and data.

Vieri Samek-Lodovici

Page 24 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Constraint Conflict and Information Structure Vieri Samek-Lodovici is a Reader in theoretical linguistics at University College London. He has published several articles on the syntax of focalization and its interaction with prosody and discourse givenness. He has also worked on a wide range of other topics, including argument structure, agreement, optimality theory, and the relation between optimality theory and minimalism. At the beginning of his career, he worked for several years as a computational linguist.

Page 25 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Information Structure: A Cartographic Perspective

Oxford Handbooks Online Information Structure: A Cartographic Perspective   Enoch Aboh The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara Print Publication Date: Jul 2016 Subject: Linguistics, Discourse Analysis Online Publication Date: Jun 2016 DOI: 10.1093/oxfordhb/9780199642670.013.004

Abstract and Keywords This chapter discusses the cartographic approach to clause structure according to which information structure directly relates to syntactic heads that project within the clausal left periphery. This view is supported by data from languages in which informationstructure-sensitive notions (e.g. topic, focus) are encoded by means of discourse markers that trigger various constituent displacement rules. Such empirical facts are compatible with the cartographic view in which lexical choices condition information packaging and clause structure. Put together, the cross-linguistic data presented in this chapter indicate that [FOCUS], [TOPIC], and [INTERROGATIVE] represent formal features that are properties of lexical elements and may sometimes trigger generalized-piping and snowballing movement. Keywords: topic, focus, interrogative, cartography, generalized pied-piping, snowballing movement

8.1 Introduction A question that is central to modern comparative syntax is whether there could be direct relations between information structure packaging and structure building (e.g. Chomsky 1971; Jackendoff 1972; Gundel 1974; Valduví 1990). Within the Minimalism framework, for instance, Chomsky (1995: 220) briefly addressed this question suggesting that surface effects (e.g. commonly associated with Topic–Focus articulation) ‘involve some additional level or levels internal to the phonology component, postmorphology but prephonetic, accessed at the interface along with PF and LF’. This view restates Chomsky’s (1971) original conclusion that information-structure-sensitive word order variations derive from PF (Phonological Form) and LF (Logical Form) interface properties.

Page 1 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 31 October 2018

Information Structure: A Cartographic Perspective Section 8.1.2 briefly presents some recent proposals that adopt (some version of) this view. Zubizarreta (1998, 2010, this volume), for instance, argues that informationstructure-sensitive surface rearrangements derive from the interaction between checking operations involving purely formal features, and phonologically conditioned movement operations (i.e. P-movement). Under this view, the interaction between information structure and clause structure is mediated by p-features. Szendrői (2001) and Fanselow (2006, 2007) adopt a more radical view in which surface manifestations strictly derive form syntax–PF interface phenomena that relate to information structure only indirectly. Alternatively, the cartographic approach, discussed in Sections 8.1.3 and 8.1.4, postulates that information structure directly relates to syntactic heads that project within the clausal left periphery (Rizzi 1997). This view is supported by data from cross-linguistic data (e.g. Romance, Germanic, Kwa, Bantu, Chadic, Gur, Sign Languages). In (p. 148) some of these languages, information-structure-sensitive notions (e.g. topic, focus) are encoded by means of discourse markers that trigger various constituent displacement rules. Such empirical facts are compatible with the cartographic view in which lexical choices condition information packaging and clause structure. Section 8.1.5 concludes the chapter.

8.2 Information structure: a syntax–PF interaction Studies on stress assignment and focus meaning in English (e.g. Chomsky 1971, 1977; Jackendoff 1972; Selkirk 1984, 1995; Zubizarreta 1998, 2010, this volume) show that: (i) There is a correlation between stress assignment and focus interpretation. (ii) Stress assignment rules make reference to syntactic notions (e.g. c-command). (iii) Stress assignment rules correlate with prosody-driven movement (e.g. scrambling). (iv) Prosody-driven word order alternations affect meaning. In accounting for these four interrelated observations, Zubizarreta (1998) argues for a syntactic model that postulates two types of features: f(ormal)-features, involved in checking operations, and p(honological)-features, relevant for P-movement. f-features and p-features enter the computation at different points: the computation of p-features occurs after f-features have been ‘checked’ or ‘valued’. A converging sentence is therefore one for which both formal operations converge. Because Zubizarreta’s (1998, 2010) analysis in terms of P-movement involves the syntactic component, this view does not exclude the possibility that some languages (e.g. Bantu, Kwa) can display discourse marking functional items that may encode discourse-related f-features responsible for rearrangement rules. The question then remains how P-movement is parametrized.

Page 2 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 31 October 2018

Information Structure: A Cartographic Perspective Fanselow (2006, 2007), Szendrői (2001, 2003) and much related work argue for a ‘pure’ syntax and seek to exclude discourse-related features from the syntactic component altogether. Under such views, the relation between clause peripheral phenomena and information structure is indirect. Szendrői (2003: 44, 47), for instance, claims that movement of the focused constituent to sentence-initial position in Hungarian results from an interaction between Hungarian stress rule, a stress–focus correspondence principle, and a stress-driven movement rule. Assuming that topics are adjoined, Szendrői (2001, 2003) argues that the constituent a kalapját ‘her cap’ in (1a) raises to the position left-adjacent to the verb in the clausal left periphery in order to bear the main stress rather than to check the focus feature in a focus phrase (as originally proposed by Brody 1990, 1995). An adaptation of Szendrői’s (2001: 42) representation is given in (1b). (p. 149) (1)

In this representation, the topic a nő ‘the woman’ is adjoined to FP containing the focus element. FP is a mere label for the Functional Projection hosting the fronted constituent. Although representation (1b) derives the right word order, a number of questions arise with regard to the structural and licensing properties of FP/FPFOCUS. One also wonders whether the heads of these projections can be morphologically expressed.

8.3 Mapping the clausal left periphery Rizzi (1997) addresses these questions showing that a unitary approach to the complementizer system (CP) coupled with a free adjunction rule, as in (1b), cannot account for the following sentences.

Page 3 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 31 October 2018

Information Structure: A Cartographic Perspective (2)

Sentence (2a) indicates that there are structural positions between the complementizer and the subject that license the fronted negative constituent under no circumstances and the inverted modal should. Similar data led Rizzi (1997) to conclude (p. 150)

that the clausal left periphery (CP) involves a complex structure. Rizzi’s conclusion is compatible with the Italian examples (2b), which instantiates two complementizers che (Paoli 2001), and (2c) which indicate that the space between the two complementizers can host a rigid sequence of topics and focus represented in structure (3). (3) ForceP is the interface between the propositional content expressed by TP and the superordinate structure (i.e. the main clause or the discourse). It encodes clause-typing. TopP and FocP on the other hand license topicalized and focalized constituents, respectively. Rizzi (1997: 288) argues that: the topic–focus system is present in a structure only if needed, i.e., when a constituent bears topic or focus features to be sanctioned by a spec-head criterion. If the topic focus field is activated, it will inevitably be sandwiched in between force and finiteness, as the two specifications must terminate the C system upward and downward, in order to meet the different selectional requirements and properly insert the C system in the structure. FinP expresses tense and modal specifications that match with those of the TP domain. It represents the interface between the proposition and the complementizer system. Since Rizzi’s seminal paper, there has been a wealth of literature in support of the cartography approach to the clausal left periphery and its relation to information structure (cf. Haegeman 1995, 2003, 2012; Rizzi 1996, 1997, 2001; Belletti and Rizzi 1996; Belletti 1999, 2001, 2002, 2004, 2009; Poletto 2000; Frascarelli 2000; Benincà Page 4 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 31 October 2018

Information Structure: A Cartographic Perspective 2001; Aboh 2004a, 2004b; Aboh et al. 2007; Cinque and Rizzi 2008; Aboh and Pfau 2011; Benincà and Munaro 2011; Biloa 2013; among others). In terms of these studies, prosodic phenomena such as those discussed in Szendrői (2001, 2003), Fanselow (2006, 2007), are analysed as properties of the clausal left periphery. In the following paragraphs, I illustrate these views based on data from the Gbe languages, a subgroup of the Kwa family. Unlike the Romance and Germanic languages mostly discussed in the literature, these languages offer the empirical advantage that they express the functional heads in (3) by means of segmental and supra-segmental material. Indeed, studies on information structure that treat discourse-related constituent displacements as PF phenomena do not usually take into account the fact that what is seen as mere intonational pattern may sometimes be related to underlying morphemic specifications. This can be shown by the fact that some languages display segmental material in the contexts where other languages apparently make use of prosody.1 The Gungbe sentence in (4a) is comparable to the Hungarian example (1a). Yet, this example differs from the Hungarian (p. 151) one only to the extent that, aside from using the clause initial position as Hungarian does, Gungbe exhibits a topic and a focus marker where Hungarian relies on intonation. (4)

Likewise, fronted constituents in Hungarian and Gungbe tend to respect a hierarchy in which topics precede focus, as in (1b) and (4b). Given these representations, the only difference between the Hungarian structure in (1b) and the Gungbe one in (4b) is that the

Page 5 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 31 October 2018

Information Structure: A Cartographic Perspective latter involves FocP and TopP which host the focused and topicalized constituents attracted by the focus and topic heads. Because these functional heads are obligatorily expressed in Gungbe topic and focus constructions and exhibit a particular syntax cross-linguistically (cf. Aboh et al. 2007; Biloa 2013), one cannot argue that they are mere equivalents of phonological prominence in Hungarian (pace Eilam 2011). Instead, Gungbe-type languages suggest that crosslinguistic surface manifestations, sometimes analysed as mere PF phenomena, could be expressions of tonemes functioning as discourse markers. This requires a change of perspective: information structure is primarily expressed by morphemes that may be segmental or tonemic. Due to their syntactic properties, these morphemes may trigger rearrangement operations which in turn may affect intonation (Aboh 2010a). Although this view may look a bold statement while analysing Indo-European languages, it is compatible with the following pieces of data from Fongbe and French.

Page 6 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 31 October 2018

Information Structure: A Cartographic Perspective

8.3.1 Tonemes as functional heads Yes–no questions in Gungbe and Fongbe (two closely related tone languages) are marked by a sentence-final particle. The particle is a floating low tone (i.e. a supra-segmental element) in Gungbe, but a full segment in Fongbe. Sentence (5a) is a declarative in both Gungbe and Fongbe. Note that the verb wá ‘come/arrive’ bears a high tone. In the yes–no questions (5b vs 5c), this verb is realized with an additional low tone (i.e. wâ) in Gungbe (5b), unlike in Fongbe in which a yes–no question includes the final marker à (5c). (p. 152)

(5)

Aboh (2004a, 2010a) accounts for the contrast between (5b) and (5c) arguing that the clause-final floating low tone represents the Gungbe question marker. This toneme has a syntax comparable to that of the Fongbe segmental question marker à: it occurs clausefinally, and when it co-occurs with other discourse markers it occurs to the right edge from where it attaches to the proposition. In this regard, Aboh (2004a, 2004b, 2004c, 2010a) reports that even though the topic and focus markers realize the clausal left periphery in (4a) they can also mark the clause as a whole. In such contexts, these markers may cluster clause-finally together with the interrogative maker. This is instantiated by the Gungbe example in (6a) and its Fongbe equivalent in (6b). These sentences are felicitous in a context where a speaker being exasperated by the cries of the child asks whether she has not yet eaten as planned. In the Gungbe example, the floating tone attaches to the last particle in the sequence, yielding the form wɛ̏ with an extra low tone (6a). This contrasts with Fongbe in which the same focus marker retains its original low tone and precedes the question particle à (6b).2

Page 7 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 31 October 2018

Information Structure: A Cartographic Perspective (6)

In accounting for the distribution of these discourse markers in relation to the expression of negation in Gbe, Aboh (2010b) argues that the Gungbe example in (6a) results from snowballing movement of the proposition (i.e. represented here by FinP) into (p. 153) [spec FocP], forming the sequence ví lɔ́ má ɖù nú wὲ!’ THE CHILD HAS NOT EATEN’ in (7). This sentence would be felicitous in a context where the speaker is exasperated by a repetitive question of an interlocutor about why the child is crying.

(7)

In the context of a yes–no question like (6a), the sequence in (7) further pied-pipes to [spec InterP], whose head Inter is expressed by the floating low tone ‘ˋ’, thus generating the final wɛ̏ now represented in (8).

Page 8 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 31 October 2018

Information Structure: A Cartographic Perspective

(8)

This analysis extends to the Fongbe sentence (6b) in which the question marker à realizes Inter. Note that while negation is systematically expressed by the preverbal morpheme má in Gungbe, it may be encoded by the sentence-final particle ă in Fongbe (cf. Lefebvre and Brousseau 2002). In discussing this marker, Aboh (2010b) concluded that right-edge negative elements in Gbe are better analysed as modal elements belonging to the complementizer system where they encode (negative) evidentiality.3 Accordingly, the Fongbe negative particle is analysed on a par with left-peripheral markers (e.g. topic, focus, inter) which, as we saw already, occur to the right edge when they mark the proposition as a whole. In this analysis the complementizer system involves a negative phrase Neg°[C]‌ that projects below the focus phrase. Neg° attracts the proposition in its specifier thus forming the negative sequence in (9a). This sequence can be further marked for focus, as in (9b) which is equivalent to the Gungbe example in (7). (p. 154)

Page 9 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 31 October 2018

Information Structure: A Cartographic Perspective (9)

4

The sequence in (9b) may be subsequently questioned, for instance, in a context where a participant in the discourse doubts the utterance in (9b) and asks for confirmation as in the English sentence ‘Are you saying that it’s not the case that s/he has eaten?’ In Fongbe, such a sequence will be rendered as in our preceding example (6b). What is remarkable in this example is that the question particle à marks the sequence in (9b) which includes both the evidential negative marker and the focus marker, ǎ and wὲ, respectively. Following the rationale in (8), the Fongbe facts suggest that a sentence like (6b) derives from snowballing movement of the proposition to [spec NegP] within the left periphery, followed by pied-piping of NegP to [spec FocP], and pied-piping of FocP to [spec InterP] as illustrated in (10).

Page 10 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 31 October 2018

Information Structure: A Cartographic Perspective (10)

There is ample data from Niger-Congo and other languages which support this analysis. Nkemji (1995), Koopman (2000) and Biloa (2013) for instance, present a detailed discussion of various (Grassfield) Bantu languages, where generalized pied-piping of the type discussed here licenses interrogative mood, aspect, negation, and focus. Similarly, Aboh et al. (2005) and Aboh and Pfau (2011) argue that Indian Sign Language and Sign Language of the Netherlands (NGT), respectively display interrogative structures where the proposition must raise leftward to the specifier of the question particle that realizes Inter. Given these cross-linguistic facts, I conclude that (i) Discourse markers responsible for information structure packaging can take the form of (supra)-segmental material that encodes context independent meaning (e.g. topic, focus); and (ii) discourse markers expressing the features [TOPIC], [FOCUS] must be treated on a par with markers expressing the features [NEGATIVE], [INTERROGATIVE] together with which they realize the clausal left periphery. These markers trigger morphosyntactic operations that affect word order (see Section 8.1.4 for further discussion). The following section on French (p. 155)

Page 11 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 31 October 2018

Information Structure: A Cartographic Perspective yes–no questions is compatible with the proposed cartographic approach to grammatical tonemes.

8.3.2 A toneme in French yes–no questions French yes–no questions come in three forms that can be contrasted with the declarative sentence in (11a). Such a declarative sentence typically displays an SVO order and a final falling contour. Accordingly, example (11a) forms a minimal pair with the yes–no question in (11b) which is pronounced with a final rising contour (cf. Post 2000).5

(11)

The examples in (12a,b) illustrate two additional strategies in French yes–no questions: (12a) involves complex inversion (cf. Kayne 1972; Kayne and Pollock 2001), while (12b) includes the ‘question marker’ est-ce que in initial position (cf. Munaro and Pollock 2005). (p. 156)

(12)

Setting aside complex aspects of French intonation, a common denominator to the yes–no questions in (11b) and (12) is the use of a final rise (cf. Di Cristo 1998: 202). Even though example (12) illustrates questions which are marked by a syntactic or lexical device, these strategies do not exclude the final rise typical of French yes–no questions.

Page 12 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 31 October 2018

Information Structure: A Cartographic Perspective Looking at these examples from a Gbe perspective, a natural explanation for French yes– no questions would be to propose that examples (11b) and (12a,b) involve a yes–no question particle: a toneme that determines the final rising intonation. As in Gbe, this question morpheme expresses interrogative force. This would also mean that the syntactic strategies in (12) are not primarily motivated by the feature [INTERROGATIVE] even though they may be conditioned by its syntax. Keeping to the same rationale, I therefore propose that the French question particle realizes Inter and attracts the proposition as a whole into its specifier position [Spec InterP]. Example (11b) can therefore be described as in (13) in which Inter contains a null morpheme indicated by a high tone ‘ˊ‘ for expository reasons (irrelevant projections are ignored).

(13)

Thus, data from Fongbe, Gungbe, and French support the cartographic view of the clausal left periphery. While it is traditionally accepted that Inter involves a formal feature [INTERROGATIVE] associated with interrogative grammatical items, the status of focus markers as grammatical elements with the formal feature [FOCUS] is still much debated. Looking at the distribution of the feature [FOCUS] and its interaction with generalized pied-piping, (p. 157) the following section argues that [FOCUS] must be analysed on a par with ordinary formal features.

8.4 The feature [FOCUS] and generalized piedpiping in Gungbe

Page 13 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 31 October 2018

Information Structure: A Cartographic Perspective The preceding section showed that Gbe languages involve discourse markers such as illustrated by the question–answer pair in (14). In the question (14a) the sentence-final floating low tone affects the last word: the determiner lɔ̂. In the answer (14b), however, the focus marker wὲ attaches to the focused phrase àvún lɔ́ ‘the dog’, while the topic marker yà attaches to the topic phrase àsé lɔ́ ‘the cat’. Note that in this case, the determiner marking the cat displays a high tone contrary to the example (14a) in which it bears an additional low tone expressing INTERROGATIVE.6 (14)

The description of these examples as compared to similar markers in other languages led me to conclude that discourse markers (e.g. focus, topic) as well as speech modality markers (e.g. interrogative) are expressed cross-linguistically by (supra)-segmental material and therefore project in syntax where they trigger specific rearrangement rules (i.e. Internal Merge). Given this stance, it must be the case that the discourse features [TOPIC], [FOCUS] are visible to the computational system similarly to other commonly assumed formal features.7 This would mean that [TOPIC] and [FOCUS] are context independent features that enter probe-goal relations whenever they are present in the derivation (Rizzi 1997). This view predicts that in addition to entering probe–goal relations, [TOPIC] and [FOCUS] can trigger other phenomena contingent on chain formation. A case in point is generalized pied-piping as discussed below.

(p. 158)

8.4.1 FOCUS, snowballing movement, and generalized pied-

piping In terms of Aboh (2004a, 2004b, 2004c), snowballing movement, illustrated in (8), (9), (10), and (13) is an instance of generalized pied-piping in which the probe (e.g. Inter, Top, Foc) attracts the event head expressed by the verb. Consequently, FinP (or TP) as a whole, is fronted to the specifier of the relevant probe. Interestingly, however, this kind of pied-piping can also happen when the probe attracts a focused argument that cannot be extracted, as shown by the example in (15). Here, the wh-phrase ɛ́tɛ́ is attracted to the focus position, but the whole bracketed OV sequence containing the wh-phrase and the verb phrase is pied-piped to the left of the focus marker wὲ. Note that this example is interpreted as an object wh-question even though the wh-phrase is embedded in a larger constituent.

Page 14 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 31 October 2018

Information Structure: A Cartographic Perspective

(15)

In order to understand this example, we need to step back and consider wh-questions in Gungbe. In Gbe languages, the wh-phrase must occur to the left of a focus marker (16) (cf. Ameka 1992, 2010; Lefebvre and Brousseau 2002; Aboh 2004a, 2004b, 2004c, 2006a, 2006b, 2007a, 2010a; Fiedler and Schwarz 2007). These examples briefly illustrate subject (16a), object (16b) and time adjunct (16c) wh-questions in Gungbe. Example (16d) shows that in-situ questions are excluded in Gungbe. (16)

Aboh and Pfau’s (2011) cross-linguistic study demonstrates that in wh-questions, the wh-phrase does not raise to license a wh-feature as usually assumed in the literature, but in order to check the focus feature under Foc (expressed here by the marker wὲ), see also Bošković (2001, 2002) for discussion. This view is fully supported by example (16) as well as the following indirect questions. (p. 159)

Page 15 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 31 October 2018

Information Structure: A Cartographic Perspective (17)

In these examples, the verb kànbíɔ́ ‘to ask’ selects for an interrogative embedded clause, hence the clause-final additional low tone on the topic marker yà in (17a). The latter marks the embedded interrogative phrase and therefore occurs last (see Section 8.1.3 for discussion). In addition, the embedded proposition contains a focus phrase (FocP) that attracts focused elements (i.e. DPs and wh-phrases alike, 17b,c), as partially represented below.

Page 16 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 31 October 2018

Information Structure: A Cartographic Perspective

(18)

In this representation, Inter expressed by the interrogative floating low tone ‘ˋ‘ marks TopP headed by the marker yà. The latter marks the proposition containing FocP whose specifier is realized by the focused phrase Suru or the wh-phrase mɛ́nù itself marked by the focus marker wὲ under Foc. It appears from this description that the proposition embedding the topic phrase raises to [spec InterP] to license interrogative, while movement of the focused phrase to TopP licenses the topic feature under Top. Finally, movement of the focused constituent or the wh-phrase to [spec FocP] licenses the focus feature under Foc. That full clauses, DPs and wh-phrases are sensitive to movement to [spec FocP] indicates that the relevant feature responsible for this syntactic operation is the feature [FOCUS], rather than [WH]. With this description in mind, let us now go back and consider object wh-questions in Gungbe OV sequences as illustrated in example (15). (p. 160)

8.4.2 Generalized pied-piping in Gungbe OV sequences

Page 17 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 31 October 2018

Information Structure: A Cartographic Perspective Aboh (2004a, 2004b, 2009), and Aboh and Smith (2012, 2014) and references cited there discuss OV sequences in Gbe in great details. Due to space limitations, I will not repeat this discussion here, but I refer the interested reader to these studies and references therein. For the current discussion, it suffices to remark that Gungbe displays aspectually determined OV sequences of the type in (19). In this example, the construction encodes progressive aspect.

(19)

Two observations are in order: (i) OV sequences are typically introduced by an aspectual particle or verb, here tò. The aspectual morpheme tò also encodes location and has the property of changing its form into tè, when its complement is displaced as indicated by the contrast in (20).

(20)

(ii) OV sequences are typically delimited to the right by a final particle as indicated by the elements gbé and jí in (21). Example (21a) encodes purposive, while (21b) expresses inceptive. (p. 161)

(21)

Page 18 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 31 October 2018

Information Structure: A Cartographic Perspective With regard to example (19), the clause-final particle (glossed as PCL) is a floating low tone that attaches to the last word, hence the high–low tone on the last syllable of Alúkû in (19) unlike in (16) where this personal name ends with a high tone (i.e. Àlúkú).8 Aboh (2004a, 2004b, 2009) derives OV sequences from an underlying VO structure by suggesting that the sequence to the right of the aspectual verb tò in (19), involves a complex structure embedding both an edge FP, headed by the particle that occurs clausefinally and an inflectional domain (IP) headed by an aspect phrase. It is further proposed that the inflectional domain involves an EPP position which is licensed by fronting of the object to a preverb position. Subsequently, IP raises to [spec FP], as a consequence of which the particle typical of OV sequences occurs clause finally. This analysis is described in (22).

(22)

An important conclusion reached in Aboh (2009: 13) is that OV sequences form a constituent. In addition, the preverb object position qualifies as an EPP position, that is, a freezing position as defined in Rizzi and Shlonsky (2007). Be it so, object extraction to the clausal periphery for focus purposes cannot transit via this internal EPP position: the language must find a way to circumvent the freezing position. In this regard, Gungbe shows that when the object is extracted for focus or wh-question, the verb must reduplicate (23a) unless it is preceded by an aspect marker as in (23b). Example (23c) shows that failure to reduplicate the verb or to insert an aspect marker in the preverb slot leads to ungrammaticality.9 (p. 162) (23)

Under the assumption that reduplication is a form of INFL—comparable to the preverbal aspect—that allows the licensing of a null expletive in the preverbal EPP position, Aboh (2004a, 2007b, 2009) concludes that the focused phrase or the wh-phrase does not transit Page 19 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 31 October 2018

Information Structure: A Cartographic Perspective through the preverbal EPP position on its way to [spec FocP]. If this were the case, the wh-phrase or the focused constituent would be frozen in place leading to the ungrammatical sequence (24), where [spec FocP] remains empty.

(24)

Yet, one possibility that comes to mind is that in contexts where the wh-phrase gets frozen in the EPP position, the language may allow generalized pied-piping of the whole constituent containing the wh-phrase. This is indeed what happens in example (15), which we can now describe as being subsequent to (24) where the wh-phrase is frozen in the EPP position. As a consequence the constituent containing the wh-phrase and the VP raises to [spec FocP] to check the focus feature. The derivation goes as follows. We first begin with the sequence in (25a) with the whphrase in its base position. In (25b), the wh-phrase raises to the preverbal EPP position where it freezes. Then the focus marker merges (25c) and attracts the wh-phrase bearing the focus feature. The latter, however, cannot extract from the EPP position. Consequently, the constituent containing the wh-phrase and the VP is pied-piped to [spec FocP], (25d = 15). (25)

The change of the progressive aspect marker from tò into tè in a similar way to example (20b) provides us with a further piece of evidence that what fronts to [spec FocP] is the complement of tò, that is, FP. The focus feature on the wh-phrase is therefore responsible for the generalized pied-piping of FP to [spec FocP]. That this is indeed the case can further be shown by the fact that predicate focus in OV sequences also results in the generalized pied-piping of FP to [spec FocP]. (p. 163)

Page 20 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 31 October 2018

Information Structure: A Cartographic Perspective

(26)

As discussed in Aboh (2006a, 2010a) and Aboh and Dyakonova (2009), the Gbe languages display an asymmetry between predicate focus in VO structures versus OV structures in that the latter allow generalized pied-piping of the constituent containing VP to [spec FocP], while the former results in doubling structures in which a token of the verb realizes the focus position while a doublet occurs within IP as indicated in (27).10

(27)

The cited references provide very detailed accounts of these structures in Kwa languages and beyond showing that generalized pied-piping is triggered in (26), but not in (27), due to the fact that the verb cannot be extracted from OV structures because it is part of a constituent that is licensed in the specifier of FP as described in (22). This is not the case in VO constructions.

8.5 Conclusion Put together, the cross-linguistic data presented here indicate that [FOCUS], [TOPIC], and [INTERROGATIVE] represent formal features that are properties of lexical elements and may sometimes trigger generalized pied-piping and snowballing movement. While these phenomena are not unexpected in a probe–goal theory where specific (p. 164)

features can be probed over by higher heads and may thus lead to pied-piping of additional material, it seems to me hard to argue that these phenomena can alternatively be classified as mere surface PF manifestations. Note that pied-piping of FinP, FocP, and TopP as illustrated in this chapter is no more spectacular than commonly assumed phrasal movement. Indeed phrasal movement systematically results in generalized piedpiping. For instance, in the sequence, ‘which house did John buy?’ where the phrase [which house] fronts, the wh-feature being attracted to the left periphery is not a property of ‘house’ but ‘which’. Yet, what must front in this example is the whole sequence [which house]. Based on this, Chomsky (1995: ch. 4) argues that generalized pied-piping is a fundamental property of syntactic computation and it is contingent on the formation of chains. The latter, we know, only involves formal features. This chapter shows that discourse features [FOCUS] and [TOPIC] can participate in such generalized

Page 21 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 31 October 2018

Information Structure: A Cartographic Perspective pied-piping (i.e. chain formation), a piece of evidence that they belong to the set of formal features visible to the syntactic computation.

Notes: (1) See Ladd (2008) for discussion. (2) The Fongbe example is adapted from Lefebvre and Brousseau (2002: 135, 485). (3) Lefebvre and Brousseau (2002: 128) reports that ‘ă appears to express the speaker’s disagreement with the content of the proposition.’ (4) According to Lefebvre and Brousseau (2002) an English expression like ‘he did not eat’ will be rendered in Fongbe by means of a preverbal negative marker which also happens to be má in Fongbe.

((i))

(5) These examples are comparable to the Gungbe sentences in (5a,b). (6) Although I illustrate the focus marker in Gbe on the basis of Gungbe, I would like to stress that all Gbe languages display such a marker. The facts presented here are therefore not isolated. (7) See Chomsky (1995: 235ff) on formal features. (8) This is additional evidence that tonemes can indeed function as grammatical particles in Gbe and arguably in other languages as well. (9) See Aboh (2004a, 2007b, 2009), Aboh and Smith (2012) for discussion on OV sequences and reduplication. (10) See Aboh and Dyakonova (2009) for an analysis of these structures in terms of parallel chains.

Enoch Aboh

Enoch O. Aboh is Professor of Linguistics and Learnability at the University of Amsterdam. He explores issues of learnability of human languages with a special focus on theoretical syntax as related to the discourse-syntax interface, language creation and language change.

Page 22 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 31 October 2018

Information Structure: A Cartographic Perspective

Page 23 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 31 October 2018

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar

Oxford Handbooks Online Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar   Sophie Repp The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara Print Publication Date: Jul 2016 Subject: Linguistics, Discourse Analysis Online Publication Date: Jun 2016 DOI: 10.1093/oxfordhb/9780199642670.013.006

Abstract and Keywords This article critically evaluates the notion of contrast and discusses the role that contrast has been claimed to have in grammar. It argues that a precise understanding of grammatical effects of contrast can only be gained if both the contrastive constituents with the kind of alternative set they evoke as well as the discourse relations that connect the discourse segments containing the contrastive constituents are subjected to detailed analysis for their effects on grammar (prosody, morphosyntax). It presents three hypotheses specifying the details for the identification of (a) contrast-related alternative formation, (b) contrastive discourse relations, and (c) grammatical manifestations of contrast. It reviews previous research on contrast in relation to these hypotheses, examining the linguistic materials that have been used to elicit grammatical manifestations of contrast, and discussing specific findings for particular languages from the prosodic and the morphosyntactic literature on contrast. Keywords: contrast, constrastive constituent, alternative set, contrastive discourse relation, correction, contrastive accent, focus fronting

Page 1 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar

14.1 Introduction CONTRAST is a much-studied phenomenon in linguistics. For many languages, linguists have identified prosodic and/or morphosyntactic marking strategies that seem to be applied if a sentence is in a contrastive relation with another sentence. However, as we will see in this chapter, despite the apparent abundance of evidence for contrast playing a role in grammar, the question of whether contrast indeed plays a role in grammar—or, to approach the matter in a more cautious way—in the grammar of particular languages, is a rather tricky one and not easy to answer. The reason is an imprecise understanding of the notion of contrast, which more often than not is defined only intuitively. As a pretheoretical term, contrast is understood as referring to differences between similar things. For instance, in the Oxford English Dictionary (http://www.oxforddictionaries.com) contrast is described as ‘the state of being strikingly different from something else in juxtaposition or close association’, and in the Merriam-Webster online service (http:// www.merriam-webster.com) as ‘the difference or degree of difference between things having similar or comparable natures, [and] the comparison of similar objects to set off their dissimilar qualities’. Applying these pre-theoretical notions to a discourse of two sentences s1 and s2, we can say that s1 and s2 may be construed as being in a contrastive relation if s1 contains an element α that can be construed as an alternative to an element β in s2, where being construed as an alternative reflects the notions of juxtaposition and comparison in the dictionary definitions. A prototypical example for such a discourse is (1), a sequence of two sentences with two contrast pairs, which in addition to the contrastive elements contain identical material, i.e. display some parallelism, which we may assume helps ‘setting off the dissimilar qualities’. Example (2) is less prototypical because (p. 271) there is no parallelism. Still, the subjects of the two sentences are overt alternatives to each other and therefore may be viewed as contrasting with each other. (3), from Rooth (1992: 80), illustrates contrast between elements within one sentence.1 (1) (2) (3)

The alternativeness of elements—that is of constituents or their denotations—is only one angle from which the notion of contrast can be approached. Another is the role of discourse relations between two sentences—or more generally—between two discourse segments. If we compare (1) above and (4) below intuition tells us that (4), which contains but, feels ‘more’ contrastive than (1)—which illustrates the idea of ‘degrees of difference’ in the dictionary definitions. We could also say that (4) has some additional contrastive meaning component, which, if not indicated by some special prosody, is absent in (1).

Page 2 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar (4)

In this chapter, I argue that we need to look both at the way alternatives are construed, that is at the issue of contrastive constituents, and at the discourse relations that connect two discourse segments in order to gain a precise understanding of the notion of contrast (cf. Umbach 2004), and as a consequence, a good understanding of the grammatical effects contrast may or may not have in particular languages. In order to do this I first discuss views from the literature on these two aspects of contrast, and I formulate three hypotheses which are to serve as a proposal for critical evaluations of existent findings about grammatical reflexes of contrast, as well as for future investigations on this topic. Then I report empirical findings from the literature against the background of these hypotheses.

Page 3 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar

14.2 Elements of contrast 1: alternativeness of constituents The examples in (1)–(4) all contain overt pairs of alternatives, that is the alternatives are expressed linguistically, which is often considered to be necessary for applying the notion of contrast. Sometimes the contextual or situational salience or predictability of an unexpressed alternative is considered sufficient, for instance a strong (p. 272)

accent on the subject in PETE went to Rome would signal that Pete contrasts with some implied alternative, that is a person that did not go to Rome, even if this sentence were uttered out of the blue (e.g. Halliday 1967; Chafe 1976; Pierrehumbert and Hirschberg 1990). Still, overtness of alternatives is usually viewed as a reliable indicator for the presence of contrast, and it is one of several conditions that have been proposed to be a necessary condition for contrast marking in grammar. Quite similar to but subtly different from the overtness condition is É. Kiss’s (1998) proposal, according to which there must be a restricted set of alternatives in the context (also Bolinger 1961; Chafe 1976) which are clearly identifiable by the discourse participants. For instance in (5), John in the second sentence would be marked for contrast in languages that mark contrast, because the first sentence provides a restricted alternative set such that the alternative chosen in the second sentence (John) as well as its complement set ({Pete, Josie}) are clearly identifiable. (5) This view of contrast is not symmetrical in the sense that the first sentence contrasts with the second and vice versa. Rather the second sentence contains an element for which there is an alternative set in the context. This element is assumed to be contrastive and consequently might be marked by specific prosodic or morphosyntactic means in a given language. This reflects the observation that often only the second element in a pair of contrastive elements is marked (see below). É. Kiss also applies her notion of contrast to question–answer discourses, for example after the question in (6A) the focus in the answer (6B), John, would be contrastive, whereas after (6A′) it would not be contrastive because (6A) but not (6A′) provides identifiable alternatives: two people in the situational context versus an open set of people.

(6)

Another non-symmetric view on contrast is that the alternative set must not be provided by the context but is made available by the sentence containing the contrast-marked element (López 2009). For instance, a wh-question makes available an alternative set and therefore involves contrast, that is the wh-phrase is contrastive—even though there is no Page 4 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar overt alternative in the left context. The answer to a wh-question, however, does not involve contrast because the alternative set has already been made available by the question. The second element of the contrast pairs in (1)–(4) is contrastive on this view because it can be construed as being in an alternative set with an element in the context, that is the alternative set will be made available when the second sentence is uttered. A very widely applied definition of contrast is that there must be an alternative to the element at issue such that substituting the original with the alternative results in a false statement (e.g. Halliday 1967; Chafe 1976; Kenesei 2006; Neeleman and Vermeulen 2012; many of the empirical studies discussed below). This captures the intuition that corrections, see (7), are a contrastive discourse type (see Section 14.3), as well as the intuition that the sun in (8) cannot be contrastive because—ordinarily—only the sun can shine through the clouds (Kenesei 2006). The exclusion of alternatives view therefore brings exhaustivity (roughly the meaning contribution of only) into the picture. (p. 273)

(7) (8) As for (1)–(3) above, we find that they can be considered as not involving contrast on this view: (1) comes with the implicature that Pete did not go to London but this is only an implicature: the discourse can be continued with and Pete went to London too. Alternatively, one could argue that the contrastive reading of (1) is one with a silent Exhaust operator (Chierchia 2004), whereas in the non-contrastive reading this operator is missing, that is (1) is ambiguous. For (4), the reading without Exhaust is less easily available, that is the implicature is less easy to cancel. A weaker version of the exclusion view is that the speaker is committed to the chosen alternative but not to another (Pierrehumbert and Hirschberg 1990), which would fit better with (1) involving contrast. As adherents of the exclusion view do not necessarily require the excluded alternative(s) to be overt, this notion of contrast can be very broad but (2) and (3) above would still not involve contrast. Note that although I discussed the exclusion view in this section on alternatives it actually imposes conditions on propositions associated with discourse segments (see Section 14.3), rather than on alternative-denoting contrastive constituents. Another broad definition of contrast, according to which (1)–(4) would all be contrastive, is that alternatives always contrast with each other no matter what the alternative set looks like or what operators operate on it, simply because alternatives are different from each other (Vallduví and Vilkuna 1998; Selkirk 2008; Katz and Selkirk 2011). This notion of contrast is virtually identical to the notion of focus in Alternative Semantics (Rooth 1985, 1992, this volume; Krifka 2008). Indeed, for (3) above, which Rooth (1992) pretheoretically describes as involving contrast, Rooth argues that it is just one of several discourse types whose interpretation is best explained under the assumption that focus— as indicated by the presence of a pitch accent, which signals the presence of a syntactic F-feature on the focused constituent—introduces a set of alternatives from which the focused element is drawn. Other discourse types involving focus and thus alternatives are question–answer sequences, or sentences with a focus particle. Proponents of the Page 5 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar alternatives = contrast view assume that what is sometimes referred to as new information focus does not involve focus, at least not in the Alternative Semantics (p. 274) sense. In other words there is new information on the one hand, and ‘contrastive’ focus on the other (Katz and Selkirk 2011). Finally, contrast has been related to the interlocutors’ belief systems: the alternative selected by the speaker is unexpected, or in some other way remarkable (e.g. Halliday 1967; Frey 2006, 2010). Other researchers view unexpectedness as only loosely connected with, or independent from contrast (e.g. Zimmermann 2008; Brunetti 2009a). These issues also touch on the role of discourse relations (section 14.3). This overview shows that opinions on what contrast is differ dramatically even if one only looks at possible restrictions on the (set of) alternatives. This has consequences for the evaluation of observations concerning grammatical manifestations of contrast. Statements like In language x contrast is marked in way y might mean very different things depending on the definition of contrast applied. A priori, languages might differ in their grammatical sensitivity to particular characteristics of the (set of) alternatives. For instance, the alternatives = contrast view might make the right predictions for the application of particular marking strategies in language x whereas in language y similar marking strategies might require the presence of a clearly identifiable alternative set. It is therefore necessary to take into account particular characteristics of the alternative set, and thus of the constituents denoting the alternatives, when making claims about contrast marking.

Page 6 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar In hypothesis C-Const in (9) I define three semantic relations between constituents of two sentences that according to the literature potentially turn these constituents into contrastive constituents. I describe these relations for overt constituents and ignore the issue of contextually salient but non-explicit alternatives. (9)

2

(9a) defines the basic case with an explicit alternative in the context. It borrows from Rooth (1992), who proposed that an F-marked constituent β is to be construed as (p. 275)

contrasting with a constituent α, if the ordinary semantic value of α is an element of the focus-semantic value of β.3 In other words, the denotation of α must be in the set of focus alternatives of β. For instance, in a context where we are interested in what John ate, we might say John ate an apple. Then he ate [a banana]F. where α = an apple and β = a banana. Examples (1)–(4) in the introduction are all of the explicit alternative type. Rooth points out that the relation between α and β is normally symmetric: they are both Fmarked. This captures the intuition that contrast is a relation between two elements and not a feature of just one element. However, as already mentioned in the discussion of example (5) above and as we will see in more detail later, often only element β is marked grammatically for contrast. Therefore I consider symmetry as not being crucial here. In the second semantic relation, see (9b), the context contains several explicit elements that form a set of which β can be a member, for example John bought a banana and an apple. He ate [the banana]F. This is the explicit alternative set type. We encountered it in É. Kiss’s (1998) contrast definition. Finally, the focus-semantic value of β might correspond to the ordinary semantic value of α but not vice versa: John was choosing fruit. He picked [a banana]F, where α = fruit. This is the implicit alternative set type, see (9c). It includes cases where α is a wh-constituent, assuming that wh-constituents are indefinites. Hypothesis C-Const will be put to work in Section 14.4.

Page 7 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar

14.3 Elements of contrast 2: discourse relations In the discussion of (4) in Section 14.1, I mentioned that contrast might be a gradable phenomenon, which is a view that is defended in several works, for example Molnár (2006), Paoli (2009), Calhoun (2010) (also cf. Bolinger 1961; Lambrecht 1994; Asher and Lascarides 2003). The idea is perhaps most intuitive for different discourse relations (as opposed to contrastiveness in terms of alternatives). In Section 14.1 I said that discourse (4) with but is probably more contrastive than its cousin (1) without but. Corrections are intuitively even more contrastive. Now, if contrast comes in different degrees we may expect that these degrees correlate with the application of additional or different grammatical means. For instance the peak of a pitch accent may be raised higher and higher with an increasing degree of contrastiveness, or languages may differ as to how contrastive discourses must be before certain marking strategies are applied. Many studies (e.g. in the Romance prosody literature, for Spanish Sosa 1999; Nibert 2000; Face 2001, 2002; Hualde 2002; Willis 2003; Face and Prieto 2007; Simonet 2010, O’Rourke 2012; Vanrell (p. 276) et al. 2013) ignore this potentially important aspect. Consider (10), a typical stimulus, where participants utter the same sentence twice. The context for the two utterances differs. The critical element is the NP banana, which in the first utterance is a new non-contrastive constituent (there are no contextual alternatives), and the discourse relation with the previous discourse segment is non-contrastive. In the second utterance, banana is a given contrastive constituent (the context provides the explicit alternative apple), and the discourse relation is highly contrastive (a correction). (10)

This paradigm reliably produces prosodic differences between the two utterances but since the utterances differ with respect to several features, we cannot identify the feature that is responsible for the effects. Is it the presence vs absence of the overt alternative? Is it the non-contrastive vs highly contrastive discourse relation? Do we need both the alternative and the highly contrastive relation? Would less contrastive relations have the same effects? Looking at the inventory of discourse relations in various discourse theories, we find that all theories have a relation CONTRAST. How this relation is defined is—maybe unsurprisingly—different in each theory, which also has to do with the different overall number of discourse relations and the concomitant degree of specificity for the individual relations. For instance, CONTRAST in SDRT (Asher and Lascarides 2003) subsumes the Page 8 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar RST relations CONTRAST, CONCESSION, and ANTITHESIS (Mann and Thompson 1988; Mann and Taboada 2014). Wolf and Gibson (2005), following Hobbs (1985), distinguish CONTRAST from VIOLATION OF EXPECTATION (≈ CONCESSION) but not from ANTITHESIS. The basic ingredient to the CONTRAST relation in all theories is that there must be similarities as well as dissimilarities between two discourse segments, much like in the dictionary definitions discussed in Section 14.1. Additional meaning aspects like an incompatibility between the denoted states-of-affairs (ANTITHESIS, see (11) for a correction example), or a violation of expectation (see (12)) are either thought to be a possible additional aspect of the CONTRAST relation, adding to the degree of contrastiveness (Asher and Lascarides 2003), or they are assigned to a different discourse relation (ANTITHESIS, CONCESSION). (11) (12)

The above-mentioned theories also have a discourse relation of similarity, which is called PARALLEL (Hobbs 1985; Asher and Lascarides 2003), SIMILAR(ITY) (Wolf and Gibson 2005) or, less transparently, LIST (Mann and Taboada 2014). I will use the term SIMILAR. (p. 277) In a SIMILAR relation, similarities between corresponding sets of entities or events are established, for example (13) describes two similar—in fact, identical—actions carried out by different agents. The SIMILAR-typical marker too occurs. Examples like (14) are also often classified as SIMILAR: various people are engaged in similar activities —they are all gardening. However, examples like (14) are also often classified as a CONTRAST relation: there are similarities and dissimilarities. (13) (14)

The problem with the similarity–dissimilarity rhetoric is that it is vague. I propose to classify (14) as SIMILAR because meaning components like an incompatibility of state-ofaffairs or violation of expectation are absent. Whether or not SIMILAR discourses involve contrast beyond the level of contrastive constituents (C-Const), however, is doubtful. Example (15) is a minimal variant of (14) with a causal discourse relation. Intuitively, (15) does not involve more or less contrast than (14). I suggest therefore not to view cases like (14) as contrastive discourse relations in a substantial sense, that is independently of constituent alternatives. (15)

In (16) below I define the SIMILAR relation as one where the discourse segments make the same contribution to the current question under discussion. This captures the

Page 9 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar intuition that SIMILAR discourses are smooth discourses without real or perceived incompatibilities. Contrastive meaning components like incompatibility and violation of expectation are often conveyed by discourse markers like but, although, still, etc. In the literature, most expository examples for the CONTRAST relation—independently of the particular theory —contain the conjunction but,4 which we already encountered in Section 14.1. But is traditionally assumed to signal that the first conjunct serves as an argument for some background assumption whereas the second conjunct serves as an argument against it (e.g. Anscombre and Ducrot 1977). For instance, the insertion of but between the two discourse segments in (14) may be taken to indicate that the background assumption was that John and Pete would mow the lawn together. Example (14) with but implies that this expectation is violated. More generally we may say that but signals that the two conjuncts make opposing contributions to the current question under discussion (e.g. Lang 1991; for recent refinements see Sæbø 2003; Umbach 2005). I use the label OPPOSE for this kind of discourse relation to avoid the term contrast, and to make it more general than violation of expectation. (p. 278) The definition is given in (16) below. Intuitively, OPPOSE is more contrastive than SIMILAR, that is we are dealing with a truly contrastive discourse relation. On the other side, OPPOSE is intuitively less contrastive than the ANTITHESIS relation, recall (11) above, because there is no correction involved. I use the label CORR for correction rather than ANTITHESIS to cover both monologic ANTITHESIS and dialogic rejections, as the latter are often used in investigations of contrast.5 The definition of CORR is also given in (16). The discourse relations I have examined so far tend to involve discourse segments that are associated with declaratives. In Section 14.1 we saw that sequences of interrogatives and declaratives, interrogative discourses for short, have also figured prominently in the contrast debate. If an interrogative discourse consists of a question and a congruent answer (= Q–A in (16) below) the discourse relation involved intuitively is non-contrastive. If, however, the declarative sentence in the discourse is used to reject the question as in A: When did Pete complain? B: JOHN complained!, where B rejects a presupposition of the question, namely that Pete complained, we have a CORR relation.

Page 10 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar Hypothesis C-DRel in (16) summarizes the discussion of discourse relations and their relation to contrast and potential degrees of contrast. It covers the non-contrastive discourse relations Q–A and SIMILAR because they have been used in empirical investigations of contrast unlike other non-contrastive relations. Note that the issue of degrees of contrast has also been discussed in relation to focus size, which I ignore in my hypotheses but discuss in Section 14.5. (16)

6

Page 11 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar

(p. 279)

14.4 Contrast in grammar

Taking together our observations about potentially contrastive constituents and about potentially contrastive discourse relations, I propose the following hypothesis for the role of contrast in the grammar of a particular language: (17)

C-Gram does not cover cases where a subset of the contrastive constituent types is marked in a subset of the discourse relations. There are many combinatorial possibilities (not all of which are equally plausible from a conceptual point of view). Languages might choose specific marking strategies for various combinations.7 Depending on the empirical situation, specific theoretical notions have to be defined to capture such licensing conditions. They cannot be captured by a general notion of (p. 280) contrast, and using a more specific terminology will help to highlight the licensing conditions of individual cases. In my assumptions I deviate from much of the earlier literature, which has tried to come up with a notion of contrast that holds across languages.

Page 12 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar

14.5 Test paradigms In this section I describe discourse types that have been used in tests for prosodic and/or morphosyntactic manifestations of contrast—either as contrastive or as non-contrastive discourses, depending on the notion of contrast applied. OPPOSE(i) discourses have not been tested systematically so I will not discuss them.8 Interrogative discourses with wh-questions usually are Q–A(n) discourses. Whquestions can license wide (= broad) focus in the answer, or different kinds of narrow focus (cf. Arregi this volume): (18)

Wide focus in Q–A(n) discourses unequivocally is considered to be non-contrastive. Narrow focus in Q–A(n) is considered to be non-contrastive by some authors (many of those cited in Section 14.6), and contrastive by others (e.g. Calhoun 2006, 2010; Lee and Xu 2010). Calhoun (2010) proposes that the size of the focus domain correlates inversely with the degree of contrast. If we implement ‘size’ in terms of the containment relation between constituents, the object focus in (18c) is narrower, and thus would be more contrastive than the VP focus in (18c): an object is contained in the VP. The semantic relation between the contrastive constituents typically is the implicit alternative set ImplAltSet but the alternatives can also be explicit (ExplAltSet), for example Who of Mary and Ann did John call? Section 14.6.1 reports on prosodic reflexes of a narrowing focus domain in Q–A(n) discourses. However, whether or not these reflexes should be related to contrast is unclear: the focused constituent always has the same relation with its antecedent in terms of alternative type. Looking at other discourse relations, we find that a wide focus can easily be contrastive, for example in CORR(ii), see (27) further below. Interrogative discourses with polar questions are mainly used as CORR(ii) discourses. In (19), B does not answer A’s question with a simple yes or no. Rather B seems to answer the implicit wh-question Who sang last night?. The subject in the question, John, is replaced by a different subject in the answer, Pete (= ExplAlt). With the appropriate prosody, for example with a high tone on Pete followed by post-focal deaccentuation (p. 281) and a low boundary tone at the end of the utterance, the answer can be interpreted as a congruent answer: Pete is interpreted as an exhaustive focus and the answer implies that John did not sing last night: it is a pragmatic no-answer. (19)

Page 13 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar B’s utterance in (19) is a rejection of the question as pragmatically inappropriate: the possibility, entertained by A, that John might have sung last night, is rejected as a nonpossibility, hence the classification of (19) as a CORR(ii). See Section 14.6.1 for studies testing such discourses. The rejection reading is not the only reading available in (19). There is a rejection only if element α in the question, John, is unaccented. If α is accented—and thus focused—the possibility that someone other than John sang is already part of the question meaning. Glossing over some details, this corresponds most closely to Q–A(n). Yet another interpretative option is that B’s utterance is a refusal to answer, because B does not know the answer, or does not wish to answer. B’s utterance would have a different prosody then, possibly the rise–fall–rise contour of contrastive topics (cf. Büring, this volume). Intuitively, replies signalling ignorance are less contrastive than corrections but this is an unexplored issue. Interrogative discourses with polar questions and implied answers may also contain ImplAltSet constituents, see (20A,B), which is modelled on Bartels and Kingston (1994: 5). The indefinite subject in A’s utterance delivers the implicit alternative set. With the exhaustive focus prosody (falling accent on β, son, in B) the implied answer is yes, rather than no as in (19), that is the discourse relation is not CORR(ii) but resembles the variant of (19) with focused α (≈ Q–A(n)). (20)

Interrogative discourses with alternative questions are Q–A(n). They involve ExplAltSet constituents: the answerer chooses one of at least two alternatives provided in the question, see (21A,B). (21)

Declarative SIMILAR(n) discourses have been used in various test paradigms. One is a coordination with parallel conjuncts like (1) in the introduction, that is with ExplAlt constituents. These constituents have been proposed to be contrastive topics with a focus (e.g. Braun 2005; Winkler 2005) but they do not necessarily come with the rise–fall–rise contour or with the implicational pragmatics usually associated with the term contrastive topic (cf. Büring, this volume). The word topic is here usually applied in the aboutness sense (Reinhart 1981). For clarity, I call such topics non-implicational (p. 282) contrastive topics. A (slightly modified) example from a read-aloud production study by Braun (2006: 461) is (22). The context introduces a set of explicit alternatives, and two target

Page 14 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar utterances each contain an element from this set, that is the contrastive constituents have an ExplAltSet relation with the context, and an ExplAlt relation with each other. (22)

Another example with two kinds of alternative relations where the—this time interrogative—context provides an ImplAltSet, is (23) from Horvath (2010: 1357): (23)

Another SIMILAR(n) paradigm is (24), modelled on Krahmer and Swerts (2001: 394). ExplAlt constituents occur as constituents of DPs in fragmentary utterances. In studies using this paradigm, participants describe scenes to each other that are visible to themselves but not to their interlocutor (Dutch: Krahmer and Swerts 2001; Swerts 2007; Swerts et al. 2002; English: Speer and Ito 2011; Italian: Swerts et al. 2002).

(24)

There are paradigms of this sort with full clause utterances (Romanian: Swerts 2007), which have also been used in psycholinguistic comprehension studies (English: Dahan et al. 2002; Chen et al. 2007; Ito and Speer 2008; Watson et al. 2008; German: Weber et al. 2006; Japanese: Ito et al. 2012). A fourth SIMILAR(n) paradigm is (25) from Katz and Selkirk (2011: 774). They investigate prosodic effects of the presence versus absence of F-marking. The context introduces two elements, both of which can be construed as alternatives with an element in the final target sentence (= ExplAlt). The target sentence contains the focus particle only and hence is assumed to involve F-marking. There is also an ImplAltSet relation with a constituent earlier in the context: the paintings he buys.

Page 15 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar (p. 283)

(25)

Declarative CORR(ii) discourses can be dialogues ((26), (27)), and monologues ((28), (29)). (26)

(27) (28) (29)

Examples (26)–(28) have ExplAlt constituents. In (27) the contrastive constituent is a full sentence, which illustrates that contrast may involve wide focus.9 In (29), which is modelled on Sudhoff (2010: 1462), the context introduces a set of students whose members are left implicit (= ImplAltSet). Something is claimed about this set, which then is corrected: the claim is reduced to a claim about one student. For this discourse to be felicitous, the particle only is required, that is prosodic marking alone cannot signal the exhaustiveness. Declarative CORR(ii) discourses have been tested for example in German (Sudhoff 2010), Mandarin Chinese (Chen 2006); Korean (Jun and Lee 1998), and Yucatec Maya (Kügler and Skopeteas 2007).

14.6 Empirical findings In this section I discuss a selection of findings from the prosodic and syntactic literature on contrast. Because of space limitations I ignore interactions of prosody and syntax.

Page 16 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar (p. 284)

14.6.1 Prosodic reflexes of contrast

There are three kinds of prosodic reflexes that contrast has been claimed to have. The first is a special contrastive accent on the contrastive constituent, that is contrast is claimed to correlate with categorial differences and thus a special phonology. The second kind is gradual changes in acoustic features, that is a special phonetics. Relevant features are maximum and mean pitch, pitch excursion (= difference between minimum and maximum pitch of the accentual tones that are associated with the accented syllable), pitch register (range between minimum and maximum pitch used by a speaker in an utterance), pitch compression (reduction of pitch register), pitch peak position (earlier or later), duration, and intensity. Many of these features show higher measurements (e.g. a longer duration) for the more contrastive discourse in particular comparisons. Sometimes the differences are subtle in the sense that there are no reliable differences between a contrastive constituent in one utterance and its counterpart non-contrastive constituent in a minimally different utterance, that is there are no absolute differences, but there are differences between differences, that is relative differences. For instance, the difference in intensity between a contrastive constituent and other constituents in the same utterance might be larger than the difference in intensity between the counterpart noncontrastive constituent and other constituents in the non-contrastive utterance. The third kind of reflex is rephrasing: phrase boundaries are removed or added, weakened or strengthened, which is reflected for example in segmental processes, duration, or boundary tones. The following overview of findings in English, German, Spanish, and Portuguese illustrates some central issues in the investigation of prosodic effects of contrast. For English, Pierrehumbert and Hirschberg (1990) proposed a special contrastive accent, a rising L+H*, as opposed to H*, which marks new information. The contrast notion underlying this claim is very wide with L+H* apparently indicating a salient, but not necessarily linguistically explicit, set of alternatives. The proposal has been extremely influential: the existence of a contrastive accent is often taken for granted, not just for English, but also for other languages. However, none of the production and contextual appropriateness rating studies for English (e.g. Bartels and Kingston 1994; Welby 2003; Breen et al. 2010; Katz and Selkirk 2011) could support the existence of a special accent, see Table 14.1 for paradigms that were applied to test for it. Some psycholinguistic comprehension studies report effects of accent type (L+H* vs H*) in SIMILAR(n) discourses with vs without ExplAlt constituents, for example quicker identification of referents (Ito and Speer 2008) or enhanced recall (Fraundorf et al. 2010) but in the materials these studies used, accent type always correlated with acoustic measures like pitch height, duration and/or intensity (all higher for L+H*). Consequently, these results cannot serve as evidence for a special contrastive accent. Acoustic measures in English do differ in various test paradigms. For instance, narrow focus contrastive constituents have different (p. 285) (p. 286) phonetic characteristics in CORR(ii) discourses than in Q– A(n) discourses (Breen et al. 2010). It is tempting to interpret this as an effect of discourse relation (C-DRel) but note that the discourse relation (CORR(ii) vs Q–A(n)) is conflated with Page 17 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar the alternative type (ExplAlt vs ImpAltSet). Breen et al. also found effects of the presence vs absence of ImplAltSet constituents in Q–A(n): there were absolute and relative differences between narrow focus contrastive constituents and their counterpart noncontrastive constituents in wide focus discourses. This finding could be an effect of Fmarking, which needs to be investigated in future research on different alternative types. Katz and Selkirk (2011) found effects of F-marking in SIMILAR(n) discourses: if there are ExplAlt constituents these discourses have different characteristics than if there are no contrastive constituents. Table 14.1 Test paradigms used in some of the studies discussed in this section. The example numbers refer to the illustrations in Section 14.5. Study

Task

Bartels & Kingston (1994)

English contextual rating, two comparisons

Breen et al. (2010)

English quasifree production, dialogue

Test paradigms compared

• Q-A(n) broad focus, (18a) • Q-A(n) narrow focus, ImplAltset, (18c) • polar interrogative CORR(ii), ExplAlt, (20)

Katz & Selkirk (2011)

Baumann et al. (2006) and Baumann et al. (2007)

English readaloud production, monologue

• SIMILAR(n), with new non-contrastive constituents

German readaloud production, dialogue

• Q-A(n) broad focus, (18a)

• SIMILAR(n), ExplAlt & ImplAltSet,a (25) with new contrastive and new noncontrastive constituent

• Q-A(n) ‘semi-narrow’ focus, ImplAltSet, (18b) • Q-A(n) narrow focus, ImplAltSet, (18c) • CORR(ii) with polar question, ExplAlt, (20)

Braun (2005, 2006)

German readaloud

• SIMILAR(n) without ExplAltSet or ExplAlt

Page 18 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar

Sudhoff (2010)

production, monologue

• SIMILAR(n) with ExplAltSet & ExplAlt, (22)

German readaloud production, monologue

• SIMILAR(n), ExplAlt • SIMILAR(n), ExplAltSet, (25) • declarative CORR(ii), ExplAlt, (28) • declarative CORR(ii), ImplAltSet, (29)

Chung (2012)

Frota (2002)

Spanish readaloud production, dialogue

• Q-A(n) narrow focus, ImplAltSet, (18c)

Portuguese readaloud production, dialogue

• Q-A(n) broad focus, (18a)

• declarative CORR(ii), ExplAlt, (26)

• Q-A(n) narrow focus, ImplAltSet, (18c) • SIMILAR(n), ExplAltSet, (22) • polar interrogative CORR(ii), ExplAlt, (20)

(a) The use of ImplAltSet is inconsistent in the materials. The study focused on ExplAlt. In German, there is evidence for different accents in some of the potentially contrastive discourse types. The results are somewhat preliminary though because in several studies the findings were not statistically reliable. Still, in CORR(ii) discourses contrastive constituents seem to be realized with rising accents (L*+H or L+H*) rather than with H* more often than in SIMILAR(n) discourses (Sudhoff 2010). In Q–A(n) discourses a narrowing focus domain seems to be accompanied with a categorial distinction between downstepped versus unchanged H*, and in CORR(ii) discourses upstepped H* occurs more often than in Q–A(n) with narrow focus (Baumann et al. 2006; Baumann et al. 2007). Whether or not different alternative sets (C-Const) play a role for the choice of accent is unclear. For SIMILAR(n) discourses with non-implicational topics, which are typically realized by rising accents, Braun (2005, 2006) found phonetic differences between discourses with vs without ExplAltSet/ExplAlt constituents but no special accent. Frascarelli and Hinterhölzl (2007), who analysed a corpus of radio conversations, identified a special accent in SIMILAR(n) with ExplAlt (L*+H vs L+H* for new aboutness topics). As for the acoustic correlates of contrast, the German findings resemble those for English. Turning to Spanish, recall from Section 14.3 that many studies are not so informative due to the use of a coarse test paradigm (ex. (10)), but there are also some more detailed findings. In Castilian Spanish, narrow focus contrastive constituents have a higher maximum pitch (García-Lecumberri 1995; Chung 2012), a longer absolute duration, an Page 19 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar earlier pitch peak, and a greater pitch range (Chung 2012) when they are in CORR(ii) discourses compared to Q–A(n) discourses. The earlier pitch peak corresponds to a phonological difference between L+H* (CORR(ii)) versus L*+H (Q-A(n)) (Chung 2012). Thus, there seem to be prosodic effects of the discourse relation but, as in Breen et al. (2010) for English, the discourse relation correlates with alternative type (ImplAltSet/ ExplAlt). For Portuguese, Frota (2002, 2012) identified a special accent (earlier peak) on a narrowly focused contrastive constituent in Q–A(n) and in CORR(ii) discourses (H*+L) compared to non-contrastive constituents in non-contrastive discourses (H+L*). In SIMILAR(n) discourses with ExplAlt non-implicational topics, there are phrasing effects (Frota 2002): topical subjects are mapped exhaustively onto an intonational constituent, (p. 287) non-contrastive new broad-focus subjects and q–a (n)/CORR(ii) narrow focus subjects are not. To summarize the prosodic findings, in some languages (English, German, Spanish) contrastive constituents are marked differently in discourses with contrastive discourse relations than in discourses with non-contrastive discourse relations (but there is some conflation of discourse relation and alternative type). In non-contrastive discourse relations, the presence or absence of ExplAlt/ExplAltSet matters for prosodic marking in English and German, which points to an influence of F-marking whose demarcation from contrast is an open issue. Many effects are of a gradient phonetic nature rather than of a categorial phonological one (esp. English), but in some languages (Spanish, Portuguese), there is a special accent.

14.6.2 Morphosyntactic reflexes of contrast Morphosyntactic reflexes that are typically associated with contrast, are the movement of a contrastive constituent to a left-peripheral position in the clause, the use of specific constructions, for example clefts, and the use of specific morphological markers (e.g. Japanese wa, see Hara 2006; Vermeulen 2013; Tomioka, this volume). Studies investigating syntactic reflexes of contrast also usually address the issue of how precisely an information-structural category like contrast can have an effect on a syntactic derivation (for discussion see e.g. Aboh; Neeleman and van de Koot; Samek-Lodovici, all in this volume). Among those who assume contrast-coding features in the syntax are Molnár (2006) and López (2009). Neeleman and Vermeulen (2012) assign the syntaxcontrast association to mapping procedures at the interfaces. In what follows I discuss a small selection of contrast-related findings about leftperipheral movement in Hungarian and Italian. Both languages have figured prominently in the syntactic contrast debate, and the discussion illustrates some of the intricacies of the issue. I start with Hungarian and É. Kiss’s (1998) definition of contrast, where a constituent is contrastive if every member in its alternative set is clearly identifiable (= ExplAltSet or ExplAlt; recall Section 14.2). É. Kiss argues that contrast is a feature of one Page 20 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar of two focus types, identificational focus (the other type being information focus). Identificational focus is specified for [±contrastive] and [±exhaustive], the particular specification, [+]‌ or [−], being language-dependent. For Hungarian, É. Kiss argues that identificational focus is [+exhaustive] but [−contrastive]. She observes that both in Q–A(n) discourses with ExplAltSet constituents (Who of you two broke the vase?) and in Q–A(n) discourses with ImplAltSet constituents (Who broke the vase?), the focused XP in the answer occurs in a designated left-peripheral position for identificational focus (cf. Brody 1990) if and only if it is interpreted as exhaustive (cf. É. Kiss, this volume). Thus the availability of implicit vs explicit alternative sets has no syntactic effects, only exhaustiveness does. Horvath (2010) shows that in SIMILAR(n) discourses with ExplAlt constituents and ImpAltSet in the context (see (23), Section 14.5) none of the contrastive constituents occurs in the focus position. (p. 288) However, in declarative CORR(ii) discourses the contrastive constituent does (e.g. Szabolcsi 1981). The latter fact can be ascribed to the exhaustive meaning contribution of corrections. Thus, Hungarian leftperipheral movement is not influenced by alternative type (C-Const) but the discourse relation is relevant because CORR(ii) typically implies exhaustiveness. Comparing this situation to Italian, where so-called focus fronting (FF) has been argued to involve contrast (e.g. Rizzi 1997),10 we find that in Q–A(n) discourses the potentially contrastive constituent in the answer is not fronted with ImplAltSet (e.g. Rizzi 1997; Zubizarreta 1998; Belletti 2004) but with ExplAltSet it can be (É. Kiss 1998; Brunetti 2004). Thus Italian differs from Hungarian in its sensitivity to alternative type. In Italian SIMILAR(ii) discourses there is no FF (Bianchi and Bocci 2012). In CORR(ii) discourses with ExplAlt constituents, as in Q–A(n) with ExplAltSet, FF has been claimed to be optional (Frascarelli 2000; Brunetti 2004; Samek-Lodovici 2006), although López (2009) maintains that the non-movement variant in CORR(ii) is only felicitous in the presence of the answer particle no, see the interrogative CORR(ii) discourse in (30): (30)

López (2009) suggests that no expresses the contrast and that the in-situ focus is noncontrastive information focus (also Brunetti 2009b). However, Bianchi and Bocci (2012) present quantitative experimental evidence from declarative CORR(ii) discourses, where speakers choose FF in only 25 per cent of corrective sentences without no, so there is optionality, and even a preference for the non-movement variant. Thus, a contrastive

Page 21 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar constituent may, but need not, front with ExplAlt/ExplAltSet, and it does not front with ImplAltSet. There are various proposals to account for the optionality of FF in CORR(ii). Brunetti (2009a) argues that for a fronted focus, the alternative set can be identified more easily than for an in-situ focus because focus projection is impossible. Easier identifiability is advantageous when the relevant set of alternatives is not clearly marked in the context, for example in dialogic corrections or when the speaker’s utterance conflicts with implicit beliefs of the interlocutor, or is otherwise ‘unexpected’. The proposal cannot account for the effects of alternative type but the relevance of (un)expectedness for (p. 289) syntactic movement is also emphasized in Zimmermann (2008), Frey (2010), and Paoli (2009). A different account comes from Bianchi and Bocci (2012), according to whom the contrastive constituent in CORR(ii) discourses always moves, the movement being licensed by an incompatibility presupposition (recall that C-DRel identifies incompatibility as essential for CORR(ii)). Crucially, at the syntax–phonology interface the lower rather than the higher copy in the movement chain may be realized, if that produces a less marked prosodic structure. Since FF produces a marked prosodic structure the prosodic constraint favours the realization of the lower copy in CORR(ii). The constraint is violable so sometimes we observe FF. This brief discussion has shown that two left-peripheral movement types in Hungarian and Italian are sensitive to different ‘ingredients’ of contrast. Whereas Italian FF is sensitive to alternative types, Hungarian focus movement is not. In both languages, the discourse relation matters: CORR(ii) requires leftward movement in Hungarian, and allows it in Italian. The particular motivation for CORR(ii) having these effects might be different, with Hungarian being sensitive to exhaustiveness, and Italian possibly being sensitive to the particular discourse relation.

14.7 Conclusion I have argued that contrast is a multi-faceted phenomenon, and that it is important to subject these facets to detailed investigation in individual languages. Languages show similarities, for example in the probability with which the CORR(ii) discourse relation has prosodic or syntactic reflexes, but languages also differ in their sensitivity to discourse relations and/or alternative types. Fine-grained comparisons of alternative types and/or discourse relations often yield rather subtle phonetic differences, and contrast-related effects in the syntax may be optional, which provides challenges in the modelling of the effects. Overall, I conclude that grammars of individual languages are sensitive to aspects of contrast, and that which aspects these are requires careful specification.

Page 22 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar

Notes: (1) In English, contrast is usually marked by pitch accents at least on the second contrastive element β. I do not indicate pitch accents in this chapter. (2) (a) is a special case of (b) but since the presence vs absence of ⟦β⟧ in the context might affect contrast marking (a) and (b) are listed separately. (3) Rooth (1992: 81, 85ff.) eventually dispenses with this definition and treats contrast as a subcase of focus. (4) Spenader and Lobanova (2009) show that the different contrastive relations defined in RST tend to come with different markers. (5) Evidence in Bianchi and Bocci (2012) suggests that the monologique vs dialogique nature of CORR has syntactic reflexes in Italian. Also cf. Steube (2001) for different correction types. (6) The abbreviations I use contain subscripts as a mnemonic for the potential degree of contrastiveness (nihil to ii). (7) Even the two ‘contrast options’ in C-Gram leave room for flexibility. For instance, languages may differ in the specific class of contrastive constituent they mark by grammatical means. (8) But see Umbach et al. (2004). (9) To my knowledge, wide focus CORR(ii) discourses have not been tested experimentally. (10) In Italian, as in other Romance languages, the association of left-peripheral movement with contrast concerns FF as well as clitic left dislocation, see López (this volume) for the latter.

Sophie Repp

Sophie Repp is Assistant Professor of Linguistics in the Department of English and American Studies at the Humboldt-Universität zu Berlin. Her research interests include the syntax, semantics, pragmatics and prosody of ellipsis, negation, sentence topics, non-assertive speech acts and question-answer dialogues, as well as processing aspects of some of these phenomena.

Page 23 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Contrast: Dissecting an Elusive Information-structural Notion and its Role in Grammar

Page 24 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 22 January 2019

Focus Sensitive Operators

Oxford Handbooks Online Focus Sensitive Operators   Sigrid Beck The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara Print Publication Date: Jul 2016 Subject: Linguistics, Semantics Online Publication Date: Dec 2014 DOI: 10.1093/oxfordhb/9780199642670.013.007

Abstract and Keywords This chapter investigates operators that evaluate alternatives. An indirect analysis of association with focus (Rooth (1992) is assumed, according to which there is just one focus evaluating operator. In addition to focus, questions and polarity items are considered. Both also involve an operator that evaluates alternatives. The alternative semantic tier is thus used in several types of construction, and the interaction of the respective operators has to be investigated. A compositional semantic analysis of alternatives is argued for that is based on distinguished variables. Keywords: focus, alternatives, squiggle operator, questions, polarity items, intervention effects

12.1 Introduction THIS

chapter examines operators that evaluate focus. I adopt an alternative semantics for

focus as in Rooth (1985, 1992); see also Rooth (this volume). Focus sensitive operators are thus expressions that operate on alternatives.1 Two questions central to this chapter are raised in Section 12.2. The first is which alternative sensitive operators evaluate the alternatives introduced by focus. An obvious approach would be to take the expression whose interpretation is sensitive to focus (quantifiers like only, even, also, always, etc.) to be the alternative evaluating operator. However, Rooth (1992) offers a very general theory according to which focus is uniformly evaluated by the abstract operator ~. Beaver and Clark (2008) in turn argue against a uniform theory and in favour of an analysis that has the focus evaluated by a set of expressions. Section 12.2.1 contrasts Rooth’s (1992) theory with Beaver and Clark’s

Page 1 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators (2008) analysis and proposes a way of incorporating their insights into Rooth’s theory. Following Rooth, the ~ operator is the alternative evaluating operator, not quantifiers like only, even, always, etc. (see Krifka, this volume, for a discussion of these expressions as focus sensitive). The second question is where else in the grammar alternatives are used, and hence what other alternative evaluating operators there are besides the one(s) that evaluate(s) focus. Alternatives have been argued to be at work in the semantics of questions (Hamblin 1973; von Stechow 1991), polarity items (Krifka 1995; Lahiri 1998; Chierchia 2006), Japanese -mo/ka constructions (Shimoyama 2001, 2006) and others. Expressions that operate on alternatives thus also include the operators involved in question formation, negative polarity item licensing, and so on. Section 12.2.2 introduces the (to my mind) clearest other cases of alternative semantic constructions. The types of data covered by the discussion are illustrated in (1) (alternative triggers are highlighted in boldface and indicated by ‘Alt-Trigger’; alternative evaluating operators are also in boldface and indicated by ‘OP’). (p. 228)

(1)

A first important point made in this chapter is that alternative semantics should be developed under a general perspective, where questions, focus, etc. all use the same concept and formal semantic object. Having identified a set of operators whose semantics is based on alternatives, the relation between the expression that introduces alternatives and the evaluating operator is examined in Section 12.3. In Section 12.3.1, we consider how fixed the relation between the kind of alternative trigger and the evaluating operator is. Some triggers come with a formal requirement identifying particular evaluating operators that have to be in the structure. In Section 12.3.2, we look at configurations with several triggers and several

Page 2 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators operators, in particular multiple focus configurations (e.g. Krifka 1991; Rooth 1996; Wold 1996). Example (2) illustrates what types of data are relevant to the discussion. (2)

Our understanding of such data is crucial for the semantics of alternative sensitive operators including in particular focus evaluation. The data indicate that some alternative sensitive operators are selective while others are unselective. A second important outcome of this chapter is thus that the option for selective alternative evaluation has to be there. This requires an alternative semantics in the style of Kratzer (1991) and Wold (1996) rather than Rooth (1985, 1992), which I specify in detail in the appendix. Section 12.4 concludes the chapter. (p. 229)

12.2 Collecting triggers and operators The goal of this section is to identify a set of alternative triggers and alternative sensitive operators. I first discuss focus and then turn to further constructions with an alternative semantics. The section summary lays out my view of which triggers and operators we should minimally consider when we try to figure out the grammar of alternatives.

12.2.1 Focus I begin with a Roothian (1985, 1992) implementation of the notion of alternatives (see also Rooth (this volume) for discussion and Appendix A for the formal details of Roothian composition of alternatives). Under Rooth’s theory, (3a) has the ordinary semantic value in (3b) and the alternative semantic value in (3c).

(3)

Page 3 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators Combining (3) with the adverb only results in a focus sensitive interpretation. Truth conditions change according to where the focus falls.

(4)

This observation suggests an analysis of (4a) according to which the operator only takes as its arguments both the ordinary and the alternative semantic value of its sister; that is, only could be a focus sensitive operator, as worked out in (5) (see e.g. Rooth 1985). ([[α]]o stands for the ordinary semantic value of α and [[α]]Alt for the alternative semantic (p. 230)

value. In (5), [the AUdiobook]F is the alternative trigger and only the alternative

evaluating operator.) (5)

The seminal paper of Rooth (1992) instead pursues a much more general analysis of focus. The focus sensitive interpretation of quantifiers like only is distinguished from the interpretive impact of focus itself. Rooth points out that the various interpretive effects of focus—question/answer congruence, contrast, and constraining domains of quantification, among others—all have one thing in common: focus restricts the value of a free variable. The free variable is anaphoric and can play various roles in the semantics/ pragmatics (see Velleman and Beaver, this volume, for discussion of the pragmatic effects of focus). Rooth captures this generalization by introducing the operator ~ defined in (6) (slightly simplified version; additionally I will leave out the second and third conjuncts in the definedness condition in what follows). This operator always evaluates focus alternatives.

Page 4 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators (6)

The free variable C in (6), the focus anaphor, can be put to various uses. In particular, it can be the domain of quantification of an adverb like only. The interpretation of our example (4a) is now derived as in (7). The adverb only itself in (7a) would be focus sensitive, in the sense that the interpretation of a sentence in which it occurs depends on where the focus falls; but only is not itself a focus evaluating operator. (7)

According to this theory, there is just one focus evaluating operator, the ~. What we call focus sensitive expressions interact with this operator via the focus anaphor, and this gives rise to focus affected interpretations. (p. 231)

Rooth’s theory has recently been challenged in particular by Beaver and Clark (2003, 2008). Beaver and Clark observe that Rooth’s analysis leads us to expect that all focus sensitive operators should be alike. They point out important differences in behaviour between, in particular, the operators only and always. They conclude basically that a Roothian theory captures the properties of association with always perfectly, but does not account for the different properties of association with only. Therefore they propose a different analysis for only in which it is directly focus evaluating. This is incompatible with Rooth’s uniform theory of focus evaluation. Since this issue concerns precisely the set of focus evaluating operators, their challenge is discussed here. Let us first consider a pair of examples from Beaver and Clark that illustrates how only and always are, at first glance, very similar.

Page 5 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators

(8)

Next, (9) and (11) provide two types of example that make only and always come apart. The first data type concerns expressions that cannot bear focus, so-called ‘leaners’ like unstressed pronouns (’im in (9)). The example pair shows that while always permits an interpretation in which it appears to associate with the leaner, this is impossible for only. (9)

Beaver and Clark interpret this as evidence that always associates with a contextually provided domain of quantification, as indicated in (10). It does not directly work with focus. This is exactly what we would expect under a Roothian analysis. But then the same cannot be true for only (contra the Roothian analysis in (7a)). (10)

The second type of example from Beaver and Clark that I will mention here is illustrated in (11). They observe that the restrictor for always can be found by means other than focus, for example in (11a) a reading is prominent in which the presupposition of manage contributes the relevant restriction, and the focus on exams is ignored for the purpose of the quantification effected by always. The parallel reading is unavailable for only, illustrating that only invariably associates with focus. (p. 232)

Page 6 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators (11)

Beaver and Clark provide more arguments for distinguishing only from always than I can report here—see in particular Beaver and Clark (2008). But (9) and (11) are sufficient to demonstrate that some aspect of only is missing in the Roothian analysis. What do Beaver and Clark suggest? My version of their semantics for only is given in (12) (their actual semantics differs in various ways; (12) is adapted to the present discussion but preserves their main points). Example (12) differs in two respects from the Roothian analysis in (7). First, it directly accesses the focus semantic value of only’s sister. Secondly, it adds the presupposition that the focus semantic value of only’s sister be identical to the current question under discussion QUD (Roberts 1996).

(12)

The definedness condition is important for our understanding of only. Beaver and Clark suggest that the lexical contribution of only is to tell us about how the discourse is being managed. Only is not merely a universal quantifier. The interpretive impact of only +S can be described as ‘S is an exhaustive answer to the current question under discussion’. Beaver and Clark argue that its core meaning is pragmatic, in the sense that its lexical semantics is about discourse management. Since focus is equally about discourse management, only’s relation to focus is different from the relation of always to focus. Always being a regular universal quantifier over situations, it interacts with focus because it interacts with context. A Roothian analysis captures that. The differences that Beaver and Clark observe between only and always are intriguing. At the same time, giving up Rooth’s strong theory that focus is always evaluated by the ~ loses an important generalization. According to Rooth, focus can never do anything but affect the value of a free variable (compare in particular Rooth’s 1996 discussion of the non-existing verb tolf). I am not aware of counterevidence to this generalization. (p. 233) Note that the truth conditions in (12) are the usual ones and fall firmly within the range of phenomena Rooth’s theory is designed to capture: quantifiers must have a restriction, which may be covert, and covert variables can be constrained by focus. Furthermore, I Page 7 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators suggest that the definedness condition derives much of the empirical impact that Beaver and Clark aim for. It is there that the discourse management function of only is located. Thus I propose the analysis in (13) for only which combines a Beaver and Clark meaning of the adverb with Rooth’s ~. (13)

Comparing (12) with (13), there is a missing step in that (12) but not (13) ensures that the domain of quantification of only is the alternative semantic value of only’s sister: C=[[S]‌]Alt. This is strictly speaking optional in (13). But question/answer congruence demands that S contain a constituent whose alternative semantics is the same as the current question. If [[S]]Alt = QUD and C = QUD, then we have in the end the same interpretive requirements as (12) imposes. The case of always would be different because there is no requirement that the domain of quantification be identical to the QUD. Future research will have to show to what extent such an analysis can capture all the differences that Beaver and Clark observe between only and always. For now, I adopt Rooth’s (1992) theory, according to which the focus evaluating operator is uniformly the ~.

12.2.2 More alternatives As mentioned in the introduction, the notion of alternative has played a role in the analysis of further linguistic phenomena besides focus. For questions, this was proposed as early as Hamblin (1973) and von Stechow (1991). To illustrate, the WH-question in (14a) evokes the alternatives in (14b). A listener confronted with a question is asked to indicate which proposition(s) from the set is/are true. Similarly in the alternative question in (15).

(14) (15)

I assume that the same notion of alternative is at work in questions and in focus, and that the same semantic objects are involved in their analysis. A compositional analysis of (14) in terms of alternative semantic values is sketched in (16). The WH-phrase is the alternative trigger. Alternatives are compositionally inherited up to the point where they are evaluated. The evaluating operator is the Q operator, which elevates the alternative (p. 234) semantic value of its sister to the level of ordinary meaning (i.e. the alternatives Page 8 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators evoked are the meaning of the question). The case of the alternative question is parallel except that the disjunction is the alternative trigger. More detailed formal definitions can be found in Appendix A. (16)

Another well-established analysis in terms of alternatives is the licensing of negative polarity items. I concentrate on Krifka’s (1995) analysis of the licensing of weak NPIs. The familiar pattern is repeated in (17): NPIs like anything are licensed in downward monotonic contexts (Ladusaw 1979), that is, in contexts that permit inferences from supersets to subsets. Negation is an operator that reverses the inferential properties of a sentence. It turns ‘normal’, upward monotonic (18a) into downward monotonic (18b). NPI anything is allowed in (17b) but not (17a).

(17) (18)

Krifka’s goal is to explain how NPIs come to impose such a requirement on their sentence context. The first key ingredient to his analysis concerns the semantics of the NPI. The NPI has an ordinary semantic value, which is very weak—the weakest element on an inferential scale. It also triggers the introduction of alternatives into the calculation, all of which are semantically stronger than the NPI itself. The alternatives project to sentence level in the usual manner.

(19)

(20)

Page 9 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators The second key ingredient to Krifka’s analysis is the assumption of an operator generating scalar implicatures, ScalAssert. I sketch its semantics in (21) (Krifka 1995; more modern versions are available in Chierchia 2006; Fox 2007; Chierchia et al. 2011; and (p. 235) subsequent work). It operates on the ordinary and the alternative semantic value of its sister; it says that the proposition expressed by the ordinary semantic value is true, and that all alternatives that are not entailed by that proposition are false. That is, all semantically stronger alternatives to the ordinary meaning asserted are false. (21)

The operator is part of the structure of sentences with NPIs. In cases like (17a), it is predicted that assertion and implicatures together are systematically contradictory, cf. (22). This is Krifka’s idea for the reason for the unacceptability of the example. (22)

Something has to distinguish the acceptable example (17b) from the unacceptable example (17a). The inference reversing operator NOT now comes into play. Alternatives of the structure with negation are as in (23b). The alternatives are now less informative than the assertion. The ScalAssert operator is harmless in this case: there are no stronger alternatives, and so nothing is added to the assertion. (23)

Page 10 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators (24)

Thus we have an analysis of why weak NPIs require special contexts. The reason lies in the interaction of the somewhat pathological alternatives triggered by the NPI with the ScalAssert operator. See also Lahiri (1998) and Chierchia (2006) for further relevant discussion including strong NPIs and other polarity sensitive items; for present purposes I will leave it at the sketch above. (Although it would seem an obvious way to go, I will also not explore the possibility of decomposing ScalAssert into an operator only and an operator (like the) ~.) A final case of alternative semantics I want to mention is Japanese -mo/ka constructions, because they open an angle to cross-linguistic variation in the domain of alternative semantics. It is well-known that an indeterminate phrase in Japanese (nani ‘what’ in (25a)) can associate with the question particle -ka (see e.g discussion and references in Shimoyama 2006), and this looks similar enough to the English question in (16), with -ka appearing to instantiate the question operator Q. More surprising from an anglocentric perspective is the construction with -mo, which expresses universal quantification. The example in (25b) is an instance of a Japanese mo- construction taken from Shimoyama (2001). There is a DP ‘which student’s mother’, attached to which is the morpheme -mo. The whole DP semantically contributes ‘every student’s mother’. (p. 236)

(25)

It seems that the morpheme -mo contributes universal quantification over the indeterminate phrase. Sentence (26) is another example; here we have a DP corresponding to ‘the teacher that which student invited’, to which -mo attaches.

Page 11 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators Sentence (27) illustrates that in the absence of an indeterminate phrase, -mo is semantically like also/even. (26)

(27)

Shimoyama argues that quantification is not actually over the indeterminate phrase directly, but over alternatives, as indicated in (28)–(30). The indeterminate phrase introduces alternatives, which are passed on compositionally and become relevant as we encounter the operator -mo. -Mo expresses universal quantification over the alternatives provided by its sister, cf. (29). The example is interpreted in (30), resulting in the desired semantics. According to this analysis, we must add Japanese -mo to the set of alternative evaluating operators. (28)

(p. 237)

(29)

(30)

This survey of constructions involving an alternative semantics is not intended to be exhaustive. It is very likely that there are more alternative triggers, and more potential alternative evaluating operators, than the ones mentioned so far. In addition to negative polarity items, free choice items are plausible candidates for alternative triggers (e.g. Chierchia 2006; Kratzer and Shimoyama 2002). These works also mention morphological/ typological considerations (see references cited there) that should make us look for a common core meaning of such items—alternative triggering. Alternative questions brought disjunctions to our attention, and hence we should wonder about elements like either which seem to evaluate the alternatives introduced by the disjunction Page 12 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators (Zimmermann 2000; Aloni 2003). Cross-linguistically we might find more elements like mo/ka with an alternative semantics (some possibilities from Mandarin Chinese are considered in this light in Beck 2007). I will not consider any of these possibilities in more detail in what follows. Let us instead take stock of what we have.

12.2.3 Summary We have found a set of expressions that have been analysed in the semantics literature as introducing alternatives into the semantic calculation, including the ones in (31). The assumption that all of them operate with the same semantic object is made explicit in (32). (31)

(32)

This position entails that there are several alternative sensitive operators, according to our present knowledge including at least the following: (33)

These are operators whose meaning makes reference to [[.]‌]Alt. We need a system of alternative semantic values and a grammar of alternative evaluation that works for all of them. The next section takes some steps in this direction. (p. 238)

12.3 Grammar of alternative evaluating operators This section refines the analysis of the grammar of alternatives that we have so far. The first subsection contains a syntactic addition. The second subsection expands on the

Page 13 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators grammar of alternative sensitive operators and will lead to a revision concerning the compositional semantics of the alternative semantic tier.

12.3.1 Relation operator—trigger Now that we have a set of alternative triggers and a set of alternative evaluating operators, the question arises which operator evaluates which alternatives. With the Roothian composition of alternatives discussed so far (with the specific assumptions laid out in Appendix A), we might expect free association: any operator can evaluate alternatives triggered by any type of trigger. This does not seem to be right; (34a) cannot be understood as a multiple question (34d), for example, which we would expect if the alternatives triggered by focus were evaluated by the Q operator. Thus some form of constraint has to be added to the analysis we have at the moment. (Note also that the alternatives our grammar generates are (34c), and this is unsuitable for evaluating the alternatives triggered by focus and the WH-phrase in two different places. This semantic issue will be addressed in the next subsection.)

(34)

Let us examine the opposite position then: is the connection between operator and trigger completely fixed? For example, there is a WH-phrase in the structure iff there is an evaluating Q, a focus iff there is an evaluating ~, and so on? This position may be too strong. Notice that the Q operator can evaluate the alternatives triggered by a WHphrase or by a disjunction in the case of an alternative question. Similarly, Krifka (1995) motivates the ScalAssert operator more generally with structures that contain a focus and are informationally ordered—so ScalAssert would be able to evaluate weak NPIs and focus. Conversely, however, we need to be sure that there is a ScalAssert in the (p. 239) structure whenever a weak NPI is in the structure, otherwise the licensing conditions on NPIs would no longer follow. It seems that an operator would in principle be willing to work with alternatives no matter what triggered them. The triggers, however, are picky. We need to be able to ensure that a particular operator is present in the structure when a certain type of trigger is. Example (35) summarizes the constraints that I think are plausible.

Page 14 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators (35)

Chierchia’s (2006) analysis contains the ingredient missing in our discussion so far: The alternatives introduced by an NPI have to be evaluated by a matching operator. This is enforced by giving the NPI an uninterpretable feature which he calls [+σ]. The feature has to be checked against an operator bearing the same feature [+σ], but interpretable— this is our ScalAssert operator. The structure with the feature that enforces this formal connection is given below. (36) A parallel constraint has to be operative for the other alternative triggers that are picky about their evaluating operators (e.g. [+WH], a feature familiar from the WH-criterion, see for example Haegeman 1991 for discussion and references). Thus a syntactic/formal component has to be added to the semantic analysis.

12.3.2 More than one operator Next we turn to structures that contain more than one alternative evaluating operator. A strictly Roothian alternative semantics makes wrong predictions about them. This subsection will thus result in a revision towards an alternative semantics in the style of Kratzer (1991) and Wold (1996).

Page 15 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators 12.3.2.1 Interaction of ~ and Q We begin by considering the two alternative evaluating operators ~ and Q. We are interested in nested configurations. There are four possible combinations:

(37)

The issue is in each case whether the higher operator can evaluate an appropriate Alt-Trigger, which skips the intervening lower operator. If this is possible, then alternative evaluation by the lower operator is selective: it affects some, but not all alternative triggers in its scope. Note that our current system is unselective: both the ~ and the Q operator simply take the alternative semantic value of their sister, and that includes all the alternatives triggered in the scope of the operator. Our prediction is thus that the higher operator cannot evaluate appropriate alternatives in any of the four configurations. What are the facts? (p. 240)

It is fairly clear that an intervening Q operator can be skipped. Thus Q is a selective operator. Combination (37a) is instantiated by Baker ambiguities (Baker 1970). The example in (38) permits a reading according to which it expresses a multiple matrix question. I indicate interpretations informally by coindexing alternative trigger and evaluating operator. Example (39) is even clearer. Focus on embedded John can be evaluated in the matrix clause, skipping a Q operator. Parallel considerations let Wold (1996) draw the conclusion that the indices in (38), (39) on the alternative triggers are a special kind of variable, and that the indices on the alternative evaluating operators act as variable binders, selectively binding some alternative variables and passing on others (Kratzer 1991 was lead to a similar conclusion on different grounds). See Appendix B for an implementation of this version of alternative semantics. (38)

(39)

Page 16 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators The other two configurations in (37) have the ~ as the lower operator. Combination (37c) amounts to an intervention effect exemplified in (40) (see e.g. Beck 1996, 2006; Pesetsky 2000), which has been argued to arise because the ~ accidentally, unselectively, accesses the alternatives triggered by the WH-phrase. We concentrate on (37b), however. A relevant example is given in (41) (Rooth 1996; also Krifka 1991; Wold 1996). (40)

(41)

The interpretation of (41) that is of interest here is one in which each focus sensitive operator associates with one of the foci. This becomes more plausible in a context like (42). Only associates with Marilyn (which is realized as a second occurence focus; see Baumann (this volume on second occurence focus) and also associates with Bob Kennedy. (p. 241)

(42)

Such data have been discussed controversially in the literature. Rooth (1996), Wold (1996), and Krifka (1991, 2006) take them to be acceptable under the relevant interpretation (modulo, perhaps, constraints imposed by syntactic structure). But von Fintel (1994) already observes that at least some such structures are impossible (cf. (43B2)). (43)

Beck and Vasishth (2009) conducted an experimental study that shows them to tend towards the unacceptable in general, and to be significantly worse than data with the same interpretation that avoid the problematic nested configuration. One of their test items is given below.

Page 17 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators (44)

I will side with Beck and Vasishth and consider (42) to be deviant under the relevant interpretation. This means that also cannot access the focus on Bob Kennedy across the intervening only and its ~. This, in turn, I interpret as the ~ being an unselective operator, evaluating all alternatives in its scope. Those alternatives are then lost for the purposes of all further alternative generation upstairs. (p. 242)

The upshot is, then, that we can stick with Rooth’s original definition of the ~

operator, but have to revise the definition of the Q operator from (16c) to a selective operator. This is done in Appendix B. There I specify a Woldian alternative semantics, which in contrast to a Roothian is able to do just that.

12.3.2.2 Further configurations The foregoing discussion has not included two other alternative evaluating operators mentioned earlier, ScalAssert and -mo/ka. There is some interesting discussion in the literature concerning structures with several alternative evaluating operators including these two. Linebarger (1987), Chierchia (2006), Guerzoni (2006), and Beck (2007) discuss data instantiating the configuration in (45). They are judged deviant. An example is given in (46). (45)

Page 18 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators (46)

Similarly, Shimoyama’s (2001, 2006) main point is to show that indeterminate phrases always have to associate with the closest potential evaluating operator. A relevant example is (47). (47)

(48)

This indicates that we need to decide for each operator whether or not it is selective. Concentrating on the three operators we have considered for English, this means that we should systematically investigate the configurations in Table 12.1. The first column of Table 12.1 stands for the various intervention effects (called the general minimality effect in Beck 2007). The ~ operator that evaluates focus is unselective and causes higher operators to lose access to alternatives triggered in its scope, hence (p. 243) these structures are ungrammatical. The second column stands for the data that show that Q, on the other hand, is transparent, hence selective. The missing data point ‘NPI out of questions’ means NPI licensing across a Q operator and is illustrated by (49) ((49c) only has a biased interpretation, unlike (49a), so the NPI is not licensed by the question in (49a) and (49b) is the relevant configuration). These structures thus seem acceptable.

Page 19 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators (49)

Table 12.1 Operator combinations high OP \ low OP

~

Q

ScalAssert

~

* (multiple focus)

√ (Focus out of Q)

_ (Focus out of NPI)

Q

* (intervention effect)

√ (Baker Ambiguity)

_ (NPI inside Q)

ScalAssert

* (intervention

√ (NPI out of Q)

_ (nested NPIs)

NPIs) The third column concerns transparency of the ScalAssert operator as the lower operator. This issue has not, to my knowledge, been investigated. The sentence in (50) is an example of focus being evaluated across an NPI licensing configuration (‘Focus out of NPI’ in Table 12.1). In (50), even associates with the red cross, which is contained in the argument of ScalAssert. The focus alternatives triggered by focus on the red cross thus need to skip the intervening ScalAssert operator in order to associate with even. The sentence seems fine, which would indicate (perhaps surprisingly, in view of its semantic similarity to only) that ScalAssert is selective, like Q and unlike the ~. But of course this has to be looked at more carefully before we draw such conclusions (hence the missing acceptability judgement for such structures indicated by the _ in Table 12.1). (50)

12.3.3 Alternative evaluation—upshot The data considered in this section show that some of the operators that evaluate alternatives are selective, while others are unselective. To capture this, I propose a grammar (p. 244) of alternatives based on variable binding, with details provided in Page 20 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators Appendix B. Looking at several different alternative constructions simultaneously and at their interaction has led us to discover an important feature of the grammar of alternatives and a revision of what is standardly used. In addition to the semantics, syntax also has some work to do regarding the association of triggers with their evaluating operators: we have to be able to establish a formal/structural connection between operator and trigger.

12.4 Conclusions and open questions 12.4.1 Summary This chapter has offered a fairly general and abstract discussion of focus sensitive operators. Two considerations have guided me in this. First I have followed Rooth (1992) in distinguishing between the interpretation of focus itself, and the interpretation of expressions that yield focus affected readings. I find Rooth’s general perspective on the contribution of focus still the most insightful one. This means that the set of focus sensitive operators in the narrow sense boils down to just the ~, and it is the ~’s behaviour in the grammar we must investigate. The analysis of focus affected interpretations involves, in addition to the ~, the interaction of the grammar with context, for example in the shape of covert domain restrictions. Secondly, I have argued that an insightful theory of focus alternatives links them with alternatives as used in other constructions like questions, polarity items, and so on. Thus the question operator and the ScalAssert operator also count among the alternative sensitive operators. This is not, per se, a new or controversial position; but I have followed up some theoretical consequences of making this connection and ended up arguing for a version of alternative semantics with distinguished variables instead of alternative sets.

12.4.2 Open questions A general question is how we identify the set of alternative evaluating operators. From what I have said above, it looks like a small set. An important choice point here is the distinction between abstract operators vs visible morphemes (e.g. ~ vs only). It is difficult to decide in individual cases what the actual alternative sensitive operator is. We have seen that there is some cross-linguistc variation w.r.t. the set of alternative evaluating operators available. Is there also a stable core—for example the ~, Q, and ScalAssert? If so, do these three have the same semantics cross-linguistically?

Page 21 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators Perhaps the most important question to my mind is this. If the way I sketch things in this chapter is on the right track, then there are two separate ways to effect quantification in natural language: quantification over ordinary variables and quantification over (p. 245) distinguished variables. This is particularly obvious in the Kratzer/Wold system where the operators are variable binders. Why should the grammar double mechanisms in this way? And when does a grammar choose which quantificational system?

Appendix Compositional semantics for alternative semantic values For a theory of compositional interpretation for ordinary semantic values, I adopt Heim and Kratzer (1998). Our task, then, is to enrich this theory so as to include the compositional calculation of alternative semantic values. (While the theory as offered is limited to an extensional semantics, I hypothesize that the compositional theory itself is basically sufficient as is. An intensional semantics is added in this manner, for example in Hohaus and Beck (2010), and Beck and von Stechow (2015). But see Heim and Kratzer (1998 ch. 12) and von Fintel and Heim (2008) for different extensions to intensionality). Below, I first offer an implementation of Rooth’s alternative semantics in this theory (which runs into a known problem with lambda abstraction) and then an implementation of the Kratzer/Wold version (in which the problem goes away).

Appendix A: Roothian alternatives added to Heim and Kratzer In (1) I specify the rules that allow us to interpret the most basic examples. The first clause of these rules is always from Heim and Kratzer; the second clause for alternative semantic values supplements it. Examples (2) and (3) provide an illustration.

Page 22 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators (1)

(p. 246)

(2)

(3)

In (4) I repeat the (simplified) definition of the ~ and in (5) I give an example of its use (question/answer congruence). In (4), the first clause of the definition ensures that the alternatives generated in the sister constituent of the ~ are ‘stored’ in the focus anaphor C (to be made use of by anaphoric processes like question/answer congruence). The second clause says that the ~ adds nothing to the ordinary semantics. The third clause says that the alternatives within the scope of the ~ do not contribute to the generation of alternative sets above the ~. That is, those alternatives are hereby evaluated and are not passed on compositionally.

Page 23 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators

(4)

(5)

What is missing from the Heim and Kratzer system of compositional interpretation are the rules Predicate Modification and Predicate Abstraction. The first can be unproblematically provided as in (6a) and would be applicable in examples like (6b), cf. (6c–6e). (6)

The rule of Predicate Abstraction, however, is a problem. We need it for instance to cover examples like (7). It is tempting to write (8), but note that the metavariable x in (8), (p. 247) focus semantic value definition, after the ‘|’, is not bound. Rewriting the formula as in (9) reveals that it doesn’t give us the desired result (thanks to Ede Zimmermann p.c. for making this clear to me): (7)

Page 24 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators (8)

(9) ‘λx. γ′=1’ is a constant function, which is not intended. This is the motivation behind Rooth’s (1985) official proposal (and also von Stechow’s 1991 version of it), which uses a framework in which meanings are functions from variable assignments to denotations of the usual types (so if g is the type of variable assignments, then referential expressions would now be and propositions would be and so on). In such a framework the problem can be avoided. See Poesio (1996), Shan (2004) and especially Novel and Romero (2010) for discussion. We have independent reasons for redesigning the system specified so far, however. The rule in (10) adds wh-phrases and the Q operator, and (11) illustrates that Q is necessarily unselective—an undesired feature of alternative semantics, cf. the discussion of (38) and (39) above. Appendix B introduces my official proposal, which allows selective alternative evaluating operators.

(10) (11)

Page 25 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators

Appendix B: A version of Wold’s (1996) alternative semantics Wold (1996) perceives the need for selective evaluation of alternatives and proposes a version of alternative semantics in which alternative triggers introduce variables. An alternative sensitive operator can then bind (selected) variables. Wold implements this in a very clever way using only the ordinary assignment function g, (p. 248)

combined with definedness conditions that ensure that we generate alternatives in just the right places. I concede that this is possible and elegant, but I invariably get confused. The version specified below uses a proposal by Kratzer (1991) according to which alternative triggers introduce distinguished variables, to be interpreted by a second variable assignment function h for the distinguished variables. I begin once more with a set of composition rules that gets us off the ground: (12)

(13)

At the level of focus semantic values, the focus is replaced by a distinguished variable. The alternative semantic value is of the same type as the ordinary semantic value. Alternative sensitive operators bind the distinguished variables. I assume that we always

Page 26 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators start with the empty assignment for h. Rule (14) is the ~ operator in such a Wold/Kratzer system (the requirement that h be total is to say that the ~ unselectively binds all distinguished variables in its scope). (14)

This is Rooth’s ~, an unselective operator which doesn’t allow any alternatives to be passed on beyond its scope, as argued for above. (A Woldian selective ~ is defined in (15) (and this is presumably what we would use if we didn’t side with Beck and Vasishth (2009) on the multiple focus issue). But my proposal is (14).) An example of (14)’s (p. 249) application is given in (16); (16) is the same as what we calculated above with Roothian alternatives. It shows that the familiar alternative sets can be generated. The result can be used in a discourse like (5) above.

(15)

(16)

I specify the rule for Predicate Abstraction in (17). The problem with lambda abstraction doesn’t arise in this version of alternative semantics. Under a Woldian theory, there is then no need to make all meanings functions from assignments to normal denotations. My motivation for the Woldian system comes from questions, which I add to the theory in (18)–(23).

Page 27 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators (17)

(18)

(19)

(20)

(21)

(22)

The question operator saves structures with WH-expressions from uninterpretability: (p. 250)

(23)

I have simplified things, leaving out restrictions on WH-phrases and multiple WHquestions, but this is essentially how I propose questions fit into a two-tier semantics with ordinary and alternative semantic values (see Beck 2006 for more details and discussion).

Notes: (1) I am very grateful to Thomas Ede Zimmermann for discussion of points important to this paper. I would also like to thank the participants of my class on focus semantics in the summer semester 2012 at the Universität Tübingen: Nadine Bade, Polina Page 28 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Focus Sensitive Operators Berezovskaya, Lucas Champollion, Remus Gergel, Jutta Hartmann, Stefan Hofstetter, Vera Hohaus, Anna Howell, Saskia Ottschofski, Konstantin Sachs, Sonja Tiemann, and Sara Wallsworth. This paper builds on Beck (2006) and Beck (2007) and I am still grateful to everyone mentioned there. I will also use this opportunity to thank Shravan Vasishth for working with me on Beck and Vasishth (2009), which plays an important role for the perspective taken here. Finally, thank you to three anonymous reviewers for OUP and to the editors.

Sigrid Beck

Sigrid Beck is Professor of Descriptive and Theoretical Linguistics at the Universität Tübingen. Her specialization is in semantics, with an emphasis on new data sources for semantics, such as crosslinguistic variation, language acquisition, and language change. She is the author of recent articles in Language Acquisition, Journal of Semantics, and Natural Language Semantics.

Page 29 of 29

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Quantification and Information Structure

Oxford Handbooks Online Quantification and Information Structure   Manfred Krifka The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara Print Publication Date: Jul 2016 Subject: Linguistics, Discourse Analysis Online Publication Date: Jun 2016 DOI: 10.1093/oxfordhb/9780199642670.013.35

Abstract and Keywords The chapter provides an overview of the interaction between quantification and information-structural properties, especially focus, givenness, and topic. While quantification affects truth conditions, information-structuring devices can have an effect on the interpretation of quantificational expressions. After a short introduction to the nature of quantification, the chapter covers the main areas where such effects have been identified, in particular in adverbial quantification, in generic clauses, in conditional sentences, and in sentences with nominal (or determiner) quantification, including intersective determiners and comparative determiners like most. It reviews different theoretical proposals how this sensitivity to information-structural categories arises, in particular whether they are related to focus or givenness. It also discusses cases in which the quantifier itself is topical, given, or focused. Keywords: quantification, determiners, focus, topic, givenness, scope, presupposition

13.1 What is quantification? And why is it relevant for information structure? IN linguistic semantics, the term quantification refers to evaluating the relation of two sets, prototypically expressed by quantificational determiners like every, some, no, or most. This is illustrated in the following examples; imagine a set of dice of different colours, some of which might be loaded.

Page 1 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 01 November 2018

Quantification and Information Structure (1)

If α is an expression, ⟦α⟧ stands for the truth-conditional meaning of α; ⟦black die⟧ is the set of black dice, and ⟦loaded⟧ is the set of loaded objects. Quantificational determiners express relations between the two sets, for example every, the subset relation. The NP black die is called restrictor, and the VP loaded, (nuclear) scope, or matrix. Quantificational statements like (1) are further restricted to the entities being talked about (like the dice under investigation); this is the domain of quantification, which can be specified, for example as far as these dice are concerned, but it is often left implicit. (2)

Quantifiers have been studied in Generalized Quantifier theory (cf. Barwise and Cooper 1981; for recent overviews Szabolcsi 2010; Keenan 2011; for typological (p. 252) treatments Matthewson 2008, Keenan and Paperno 2012). They increase the expressive power of human languages compared to languages that only have referential DPs, like John or the thief. Human language also exceeds the expressive power of first-order predicate logic, which has two (interdefinable) quantifiers, existential ∃ and universal ∀, which would not be able to express (1c). The examples in (1) differ in their truth conditions. So, why have a chapter on quantification in this handbook, as information structure is concerned with the way how information is presented, or ‘packaged,’ in the sense of Chafe (1976)? Because the marking devices of information-structural features like focus, givenness, and topic influence the interpretation of quantificational clauses (cf. Partee 1991, 1999). This can be seen best with a type of quantification not expressed by determiners (called Dquantification), but by adverbial quantifiers like always, generic sentences, or modal operators like must (called A-quantification; cf. Bach et al. 1995). In the following example, the stress pattern, normally expressing focus, leads to truth-conditional differences.

(3)

The effect of stress in quantificational statements has been observed by various authors. Halliday (1970) noticed it with modal statements. The warning sign Dogs must be carried gets the unintended reading that the escalator is only for dog-holders with stress on dogs. Sgall et al. (1987) pointed it out for clauses with frequency adverbials, as in Londoners most often go to Brighton, which gets different truth conditions with stress on LONDONERS (‘all Brighton visitors are Londoners’) or BRIGHTON (‘all visits by Londoners are to Brighton’). Krifka et al. 1995 attended to ambiguities in generic Page 2 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 01 November 2018

Quantification and Information Structure sentences as in Typhoons arise in the Pacific, which are a generic statement about typhoons with stress on PACIFIC, and about the Pacific with stress on TYPHOONS (cf. also Cohen and Erteschik-Shir 2002). There is one important finding of generalized quantifier theory that will play an important role, conservativity. According to Keenan and Stavi (1986), it is sufficient for judging the truth of a D-quantificational statement to consider the entities in the restrictor, which is given by the NP argument. For (1), it is sufficient to check the black dice; other entities such as the white dice or the roulette wheels are irrelevant. The restrictor indicates the entities that the sentence is about (here, the black dice). This corresponds to the notion of topic, whereas the nuclear scope corresponds to the notion of comment. With D-quantifiers, the syntactic structure [[DP D NP] VP] indicates how the clause should be evaluated: NP is the restrictor, VP the nuclear scope, and D expresses the relation between restrictor and nuclear scope. With A-quantifiers as in (3), the identification of restrictor and nuclear scope is supported by prosodic means or syntactic movement, like scrambling, see below. Configurations of quantifier, restrictor, and nuclear scope have been called tripartite structures by Partee (1991) and Heim (1982). There are certain phenomena relating to quantification that are treated in other articles of this handbook. In particular, Chapter 12 treats particles like only, also, and even (p. 253) that have quantificational force. And Chapter 3 treats phenomena relating to definite and indefinite DPs that have quantificational effects on the level of discourse.

13.2 Information structure in A-quantification 13.2.1 Focus effects in a situation-based theory Adverbial quantifiers, as observed in (3), have been analysed as quantifying over times, cases, or situations (cf. e.g. Lewis 1975). Their restrictor can be provided by a whenclause. Assuming situations (cf. von Fintel 1994), we can illustrate this as follows: (4)

Rooth (1985) explains the influence of information structure on quantification as an effect of focus, understood as indicating alternative meanings that play a role in semantic interpretation (cf. Chapter 2). In addition to the regular meaning ⟦α⟧ there is a focus

Page 3 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 01 November 2018

Quantification and Information Structure meaning ⟦α⟧f, a set of meanings of the same type, which is relevant for the interpretation of certain operators, including adverbial quantification: (5)

(6)

The regular and the focus meaning of (5) without always are as follows: (7)

The restrictor is provided by the union of the focus meaning, ⋃⟦α⟧f. With always we get the following interpretation for (5): (8)

(p. 254)

For (6) this would result in the following interpretation:

(9)

This represents the intended reading, but at the price of assuming different ways in which adverbial quantifiers are interpreted: In (4) the restrictor is provided by a whenclause, in (5)/(6) by the focus meaning.

13.2.2 Focus effects mediated via context Starting with Rooth (1992), focus effects were seen as mediated through context, thus leading to an overall more restrictive theory of how focus influences interpretation; von Fintel (1994) carried this out for A-quantification. Previous work by Kasper (1987), Schubert and Pelletier (1987), and Berman (1989) had suggested that the presuppositions of a clause form the restrictor of an A-quantifier. This is illustrated by (10), where hit the target presupposes that the subject aims and shoots at a target.

Page 4 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 01 November 2018

Quantification and Information Structure (10)

Presupposition is a form of givenness, an information-structural notion (Chapter 7). The focus structure gives a linguistic clue about the kinds of situation talked about, which in turn influences quantificational statements. Von Fintel takes up a device introduced by Rooth (1992) for context-sensitivity: The alternatives at the propositional level that are introduced by focus restrict a contextual parameter C, which is related to the clause by a squiggle operator ~: (11) C is interpreted as a subset of the focus alternatives ⟦ϕ⟧f (von Fintel also requires that ⟦ϕ⟧ and at least one other element must be in ⟦C⟧). C is further constrained by the presuppositions of the clause, by a when-clause, or by the discourse topic of the sentence. With adverbial quantification, C determines the restrictor. Consequently, focus is just one of the factors restricting the domain of quantification. Consider (12): (12)

This states that in all situations s in ⋃⟦C⟧, Mary takes John to the movies. As C is not explicitly restricted further, we should take it to be the maximal set satisfying (p. 255) the provision, which results in the intended reading: whenever Mary feels lonely and takes someone to the movies, she takes John. C can be restricted further by the context, for example by the leading question (13), in which case the interpretation would be appropriately restricted to situations in which Mary takes John, Bill, or Sam to the movies. (13)

Beaver and Clark (2008) distinguish three ways of how focus can affect meaning: quasi association, where a non-veridical propositional operator like negation applies to a proposition, which in turn has to be an answer to a question under discussion; free association, where an operator depends on the setting of a contextual variable; and conventionalized association, as with additive and scalar particles like also, only, and even, where association with focus arguably is hard-wired into the grammar. Quantification belongs to the second kind: Focus helps to restrict a domain variable, which then is intersected with the restrictor of a quantifier (cf. de Hoop and Solà 1996 for a related proposal). Other constructions whose interpretation depends in similar ways on Page 5 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 01 November 2018

Quantification and Information Structure focus are superlatives like MARYF gave John the biggest box ( = Mary gave a bigger box to John than anyone else gave to him) and emotive factives like The students were glad that BRADYF taught semantics. ( = The students preferred Brady over another person to teach semantics). But for quantifiers, Beaver and Clark (2008) make the point that their focus dependency cannot be explained by mere sensitivity to presupposition, as has been argued in the light of examples like (10) by Rooth (1999) and Cohen (1999). Consider (14). (14)

Realize is factive, hence (14b) presupposes that Sandy rides a Harley Davidson to town and attracts a lot of attention. If this presupposition would restrict the domain of always, then (14b) should mean ‘Whenever Sandy is riding a Harley Davidson and it attracts a lot of attention, then Sandy realizes that it attracts a lot of attention’. Example (14b) does not have this meaning. Beaver and Clark argue that (14a) introduces a set of situations of Sandy going to town, one for each Friday, that cannot be reduced further by the presupposition of the following sentence. Rather, this presupposition is accommodated: we understand (14) as saying that every Friday situation that Sandy goes to town is a situation in which her Harley Davidson attracts a lot of attention. The resulting picture then is as follows: quantification depends on a given set of situations, and the presuppositions within the quantified sentence corresponds to that set of given situations. Focus in turn helps to determine the presuppositions of the sentence, (p. 256) by the union over the alternatives of the focus. We do not have to stipulate a direct relation between quantification and focus.

13.2.3 The requantification problem Von Fintel (1994) notices a problem with quantification over situations which arise with indefinite DPs. Consider (15), after Krifka (1995).

(15)

We understand this as: when a cowboy chews something, he chews tobacco. But the representation given actually states: for any situation s, if there is a cowboy that chews something in s, then there is a cowboy (potentially another one!) that chews tobacco in s. For this, a situation s would constitute a positive instance in which one cowboy chews tobacco, and nine others chew bubble gum. We cannot achieve binding between the x in the restrictor and the x in the scope. This is the requantification problem.

Page 6 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 01 November 2018

Quantification and Information Structure Von Fintel (1994) suggests that we quantify over minimal situations in the restrictor and in the scope (cf. also Berman 1987; Heim 1990; Elbourne 2005): each minimal situation s in which a cowboy chews something is a minimal situation in which a cowboy chews tobacco. This strategy gives us the right truth conditions. But it forces an assumption of identity between the two cats in (16): (16) This interpretation might be possible (von Fintel discusses examples like Show me a man who plays hard and I show you a man who deserves a beer). But clearly, (16) invites the reading ‘If a cat x is hungry, then there is a distinct cat y such that y cries’. The problem does not go away in a dynamic framework that is designed to capture the introduction and uptake of discourse referents, such as Heim (1982) or Kamp (1981). If we adapt the interpretation rules for focus and A-quantifiers to such a framework, as in Rooth (1995), then a cowboy in the scope in (15) would require the introduction of a novel discourse referent, leading to the wrong interpretation. Rooth’s solution is to relax the novelty condition and allow for the index to be in the input meaning. However, this creates problems in cases like (16), where we actually want the indefinite DP in the scope to come with the novelty condition. Hinterwimmer and Schueler (2012) propose that sentences like (15) are not about situations, but events. As the participants have unique roles in events, we do not need any reference to minimal situations: For any event e of chewing, there is a unique agent, and hence the agent of the event in the restrictor and the agent of the event in the scope must be the same. Example (16) is different: There are two clauses, which allow for the agent of the first (p. 257) event (being hungry) and the agent of the second (crying) to be different. Hinterwimmer and Schueler (2012) also discuss (17a) in which the indefinite DP in the scope is interpreted as co-referent with the indefinite DP in the restrictor, in contrast to (17b). (17)

Hinterwimmer and Schueler argue that in case the predicate in the restrictor and the scope is the same, as praise in (17a), there is a tendency that they pick out the same event; this is not the case if the predicates are different, as in (17b). In addition, as the descriptive part of the indefinite in the scope is partly novel in (17a), the speaker could not use a pronoun or a definite description, and so there is no pragmatic inference that the two entities are distinct.

13.2.4 Givenness in a representation based on discourse referents

Page 7 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 01 November 2018

Quantification and Information Structure Chierchia (1995) and Krifka (2001a) propose that it is givenness, not focus, which affects adverbial quantification. They analyse the accentual pattern of (15) as the result of deaccentuation of a cowboy. (18) The contextual effects of adverbial quantification discussed in 2.2 can be explained in terms of givenness, as it is inversely related to focus (cf. Schwarzschild 1999). Krifka argues that this solves the requantification problem if we assume that adverbial quantification involves discourse referents anchored to entities. Givenness implies that the entity referred to is already present in the context (cf. Chapter 3). Now, the discourse referent of a given indefinite as in (18) is obviously not present, as then a pronoun or definite DP would have been used. Krifka calls such indefinites non-novel, and assumes that they trigger accommodation of a context in which they are given. The quantifier always then states that each way in which the context can be accommodated makes the sentence true. (19)

In support of the givenness analysis, Krifka points out that non-novel indefinites can be marked by the word given, and that they carry other givenness markings, for example they are marked as topics in Japanese. (p. 258)

(20)

(21)

Furthermore, scrambling in languages like German can be seen as evidence for givenness of indefinite DPs that yields a restrictor for A-quantifiers (cf. Chapter 19). This has been argued for by Diesing (1992) with data like the following:

Page 8 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 01 November 2018

Quantification and Information Structure (22)

Finally, accommodated indefinites support cataphora, as in If a VEGETARIAN owns it, he usually takes good CARE of [a donkey]G. This is to be expected if their antecedent is given (cf. Reinhart 1986). One argument for the analysis based on discourse referents over situation- or event-based theories relates to stative predicates, as in (23). This is a statement about square numbers, not about situations containing such numbers, whatever this should be. (23) Krifka also considers non-novel indefinites in the restrictor of biclausal quantificational sentences, arguing that they trigger accommodation of their discourse referent as well. (24)

Sentence (24) asymmetrically quantifies over donkeys (cf. Kadmon 1987), which is indicated by deaccenting a donkey. This reading can be achieved if we assume that its discourse referent is accommodated, and usually quantifies over accommodations: most ways to accommodate the context such that x is a donkey support the following: if a vegetarian owns x, he takes good care of x.

13.3 Information structure in Dquantification (p. 259)

13.3.1 D-quantification over situations While information-structure effects have been initially discovered with A-quantification, they also affect sentences with D-quantification. Krifka (1990) discusses D-quantifiers

Page 9 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 01 November 2018

Quantification and Information Structure that involve quantification over events as in (25), and Geilfuß-Wolfgang (1995) and Eckardt (1999) treat examples like (26) in detail.

(25)

(26)

Geilfuß-Wolfgang 1995 develops an analysis in which the focus in the scope of the quantifier affects the nominal complement of the quantificational determiner. Assume that MOST expresses the relation between two sets A, B that holds if #(A⋂B) > ½ #(A), cf. (1c), then we have the following interpretation for (26). Just as with A-quantification, the restrictor is determined by a combination of factors, here the NP argument of the determiner and the focus of the nuclear scope. (27)

In (26) quantification appears to be over entities (tickets). However, as each ticket is sold exactly once, it may be that quantification is over events. Eckardt (1999) takes up this issue with examples like (28).

(28)

This cannot simply be understood as a quantification over cars that Ludwig washed: if Ludwig has ten cars and washes six cars once with X-Polish, and washes two others fifty times with another product, (28) is not an appropriate description. It is also not just the washing events that count: if Ludwig washes one of his cars fifty times with X-polish, (p. 260) and six of his cars once with some other product, (28) does not seem to be true either. Eckardt suggests that such sentences suggest a one-to-one mapping between entities and events; in our case, that each car was washed exactly once. Hence focus sensitivity appears to be a property of event or situation quantification, even with Dquantifiers.

13.3.2 Information structure effects in the restrictor: many and few

Page 10 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 01 November 2018

Quantification and Information Structure Focus does not only influence D-quantification in the scope of the quantifier, but also in the nuclear scope. This was first pointed out by Herburger (1993, 1997), based on an observation in Westerståhl (1985), with examples like the following: (29)

(30)

The usual restrictor–scope relation appears to be reversed; now the nominal complement determines the scope, and verbal predicate the restrictor. Hence, such uses of quantifiers do not satisfy conservativity. Herburger also showed that in cases when part of the restrictor is focused, the non-focused parts remains in the restrictor, resulting in a compositionality problem.

(31)

The phenomenon occurs with few and many; it does not show up with every or most:

(32) (33)

Herburger (1997) argues that the determiners for which it obtains must be weak in the sense of Milsark (1974): they must form quantificational sentences that can occur as subjects of there-sentences (cf. There were many cooks among the applicants vs *There were most cooks among the applicants). Such determiners are intersective, that is, the truth condition of [[D NP] VP] can be stated as a condition on the intersection ⟦NP⟧ ⋂ (p. 261) ⟦VP⟧ (cf. Keenan 2003). Intersective quantifiers are symmetric, that is, the truth conditions of [[D α] β] are the same as for [[D β] α], hence the distinction between restrictor and scope is irrelevant. Consequently, the syntactic position of the determiner D is no longer required to identify the restrictor. Herburger suggests that there are two construals, one in which the determiner phrase DP takes scope, with the nominal complement as restrictor, and one in which the determiner D raises and takes scope over the nominal complement and the restrictor, interpreted conjunctively.

Page 11 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 01 November 2018

Quantification and Information Structure

(34)

Example (34b) expresses the unary reading of the quantifier that also occurs in theresentences, as in (35). In this, the determiner is a cardinality predicate on a set—here, that the set {x | x is a child ∧ x is playing} contains many / three elements.

(35)

In (34b) there is no syntactically marked restrictor, hence focus will affect the interpretation just as with A-quantification. When we analyse a case with focus on incompetent following Rooth (1985), we get the reading in (36), where we assume that the alternative to incompetent is competent. (36)

This corresponds to the paraphrase given in (31). The approach extends naturally to cases in which the focus is located within the verbal constituent: (37)

We can also deal with cases of multiple focus spread over the nominal complement of the determiner and the verbal predicate. The resulting meaning is that among the cooks that applied, the number of competent late-appliers was low. (38)

(p. 262)

Herburger’s analysis has been criticized by Büring (1996), de Hoop and Solà

(1996), Cohen (2001), and Beaver and Clark (2008). In particular, the syntactic structure (36) does not correspond to the fact that syntactically, few is a nominal determiner, resulting in a D-Quantifier, not an adverbial or A-Quantifier. As an alternative, Büring

Page 12 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 01 November 2018

Quantification and Information Structure (1996) suggests the following general scheme for the interpretation of an expression with focus: (39)

Notice that according to (39), focus is not bound by the quantifier; rather, it is a general interpretation scheme for focus effects. (40)

The interpretation in (40) holds if many incompetent cooks applied is judged true, whereas many competent cooks applied is not judged true, or true to a lesser degree. This asks us to compare the following two statements: (41)

Interpretation (40) says that many INCOMPETENTF cooks applied is true iff (41a) holds, but (41b) does not hold to the same degree. This results in the correct interpretation. The sentence asserts that the number of incompetent ones among the cooks that applied was higher than expected, not that the number of competent cooks that applied was higher than expected. Herburger (1997) wrongly predicts similar effects for other intersective quantifiers. Focus does not seem to have an influence on truth conditions in (42). (42) Büring (1996) has the same problem, as the interpretation scheme (39) is compatible with quantifiers like three as well. What is special for many and few? As argued by Lappin (1988), these quantifiers are comparative. Solt (2009) analyses them as dimensional adjectives like big and small that relate individuals to a standard value of a scale that is dependent on some comparison class. The comparison class required for the interpretation of many and few is delivered by the restrictor of the quantifier, which depends on focus alternatives. This kind of sensitivity to the alternatives provided by focus shows up in sentences involving the as compared to construction: (43)

Page 13 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 01 November 2018

Quantification and Information Structure The analysis Solt proposes for few COOKSF applied amounts to (44) when referring to focus alternatives, where ⋃⟦COOKSF⟧f is the union of the alternatives, here ⟦cook⟧ ⋃ ⟦somelier⟧ ⋃ ⟦pastry chef⟧. (p. 263)

(44)

This states that the number of cooks that applied is considerably smaller than to be expected, given the proportion of cooks among the set of alternatives. For example, assume that among the alternative set ⋃⟦COOKF⟧f, there are 80 per cent cooks, 10 per cent someliers, and 10 per cent pastry chefs, and that there were ten applicants, of which five were cooks, four were someliers, and one, a pastry chef. Under these circumstances, Solt’s analysis predicts that Few COOKSF applied is true, as 5 > 0.1 × 10. This is correct, even though there were more cooks that applied than there were someliers. Notice, also, that (44) represents not a proportional reading, but a cardinal reading of few / many, a point that is argued to be correct by Solt as well.

13.3.3 Information structure effects with proportional determiners Sauerland (2014) also discusses cases in which focus affects the interpretation of Dquantifiers; interestingly, this involves proportional quantifiers that are not intersective like many and few. The phenomenon was first recognized for Korean by Park (2007); related effects are observed in French and (for non-subject quantifiers) in English as in (45), a newspaper headline. Sauerland discusses German cases like (46). (45)

(46)

Sauerland argues with case and agreement facts that the syntactic structures of the DPs are different: in (46a) it is [DP 60 [D′ [D Prozent] [DP (der) Frauen]]], whereas in (46b), [DP [DP 60 Prozent] [NP Frauen]]. He assumes that (46b) allows for an interpretation like Herburger’s analysis in (34b), yielding the logical form (47): (47)

Page 14 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 01 November 2018

Quantification and Information Structure The availability of (47) corresponds to the fact that in (46b), the DP 60 Prozent and the NP Frauen seems to be more loosely connected, and rather ambiguous as to which constituent forms the head. This structure is made clear with the alternative (48), in which the quantifier is expressed by an adverbial. (p. 264)

(48)

In (47) the union of the focus meaning can determine the restrictor of the quantificational DP, 60 Prozent, similar to few in (36): ‘60 percent of the men or women that voted were women’. In spite of these parallels, it is not clear whether the focus sensitivity of many/ few and proportional quantifiers is the same phenomenon, given the problems of Herburger’s original analysis, and its crucial assumption that the determiners it applies to are intersective, not proportional.

13.3.4 Focus-sensitive determiners So far we have investigated focus effects in the scope and the restrictor of nominal quantificational determiners. There are also determiners that appear to be sensitive to the focus within them. Krifka (1999) discusses cases with comparative quantifiers such as (49).

(49)

Classical General Quantifier analysis did not distinguish between the determiners and more than three and four, representing them as λP′λP[#(P⋂P′) > 3] and λP′λP[#(P⋂P′) ≥ 4], respectively. These are truth-functionally equivalent, hence semantically indistinguishable. The reason for the representation of numerals like four as expressing ≥ 4 is to account for the fact that Four boys left is true if more than four boys left; it is only by scalar implicature that this sentence gets the interpretation that exactly four boys left. Sentence (49a) lacks the implicature that excludes more than four boys being left, a fact not expected in General Quantifier theory. Krifka proposes that the determiner head, more, binds the set of alternatives introduced by focus, making them unavailable for scalar implicature. This is implemented in a theory in which the numeral three does not form part of the determiner, but is a numeral adjective that applies to sum individuals consisting of three atomic individuals. This provides the correct results for sentences with more than one quantifier, as in More than four girls kissed more than seven boys (in the cumulative reading), and it is extended to quantifiers headed by less or fewer as in (49b). Krifka (1999) applies a similar analysis for quantifiers headed by at least and at most. Newer work has pointed out that these superlatives are different in important respect.

Page 15 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 01 November 2018

Quantification and Information Structure Geurts and Nouwen (2007) have argued that they are of a modal nature, and Cohen and Krifka (2011, 2014) proposed that they represent quantifications over speech acts. In any case, they also are sensitive to focus, just like comparative quantifiers. (p. 265)

The focus of more than can be broader than the numeral, as in (50). (50) Another case of a determiner that appears to be inherently focusing is German lauter, which is derived from the adjective meaning ‘pure’ (cf. Eckardt 2002).

(51)

In contrast to only (or German nur), lauter is a proper determiner that attaches to nominal expressions (mass and count), but not to names or to other constituents.

13.4 Information structure on determiners and quantifiers The cases discussed so far involved the interaction of determiners and quantifiers and expressions with a special information status, especially focused and given items. We now consider determiners and quantifiers that are themselves focused and/or given.

13.4.1 Focus on determiners In (52), the determiner is in focus, and the quantifier is a contrastive topic: (52) In this case, some carries a rising accent, which is often considered to be a special marking for contrastive topics (cf. Chapter 3); here it is understood as indicating alternative topics. As usual, focus indicates alternatives, the alternatives being other quantifiers. The alternatives to some, naturally, are other quantifiers such as all, most, or no. These alternatives are used for the computation of scalar implicatures: If there is an alternative α that leads to a sentence that entails the one that is uttered, but not vice versa, then the sentence with this alternative is implicated to be false. In (52), the sentence All cooks were hired implicates Some cooks were hired, but not the other way

Page 16 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 01 November 2018

Quantification and Information Structure round (things are less clear with most). Hence, (52) implicates that not all cooks were hired. The relevant alternatives in the case of (52) are sentences in which the quantified DP is a topic—for example, [all cooks]Top were hired, [most cooks]Top were hired, etc. This suggests a context question like What happened to the cooks? Were they hired or fired? This context question requires that there is a given set of cooks that the discourse is about. As a consequence, this forces a partitive reading of some cooks, a phenomenon observed and explained in this way in Büring (1996). A partitive (or strong) reading of a quantificational DP is one in which the nominal part is given in the context; it can generally be expressed with a partitive construction such as some of the cooks. However, the partitive reading does not hinge on the topichood of some cooks. Example (53) also has a partitive reading without a topical interpretation of some cooks, as a context question like the one indicated is enforced. (p. 266)

(53) Again, a salient alternative is All cooks were hired. Hence, a partitive reading of a quantifier is enforced if there is a focused-induced alternative that is itself partitive. If the focus-induced alternatives are not easily construed as partitive, as with number words, then focus does not lead to a partitive reading: (54) While the numeral three can be used in partitive constructions—as in three of the cooks were fired—focus on three as in 54 does not invoke other partitive quantifiers as alternatives. This also shows up in the following pattern of implicature cancellation: (55)

Hence focus data show that there are two classes of quantification determiners, those that are naturally interpreted as partitive, and those that are not so interpreted.

13.4.2 Topical and focused quantifiers A quantified DP can be in focus, as in the following case: (56)

Page 17 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 01 November 2018

Quantification and Information Structure As a consequence, the relation between quantifiers and topicality is certainly less strict than suggested in Section 13.1. It is conceivable that quantifiers start out with restrictors that are topics, but then, as DPs, are subjected to general syntactic rules, and as a result occur in other DP positions as well. We would perhaps predict that quantifiers (p. 267) occur first in topic positions in language acquisition, and that quantifiers occur more frequently in topic positions—but this is still unknown. The information structure status of quantifiers influences their scope. Sæbø (1997) observes that topical quantifiers have wide scope over focused quantifiers. (57)

Krifka (2001b) argues that topical quantifiers can even scope over speech acts.

(58)

While every guest in (58a) can scope over the wh-constituent, leading to a quantification into a question (‘For every guest x, what did x bring to the party?’), this is not available in (58b) (‘For which y does it hold that every guest brought y?’). The scopal behaviour can be captured structurally within a framework that assumes that topics occupy syntactic positions that c-command foci, as in the ‘cartographic’ theory of Rizzi (1997). As for quantification, such phenomena have been treated by Szabolcsi (1997), who identifies three pre-verbal positions in Hungarian that differ in their information-structural potential, and can be occupied by different quantificational expressions. The outermost position is reserved for referential DPs, but can also be accessed by quantifiers headed by a legtöbb ‘most’. The next position is taken by distributive quantifiers, for example DPs headed by minden ‘every’, and the immediate preverbal position is accessible to ‘counting’ quantifiers headed by for example kevés ‘few’ or hatnál több ‘more than six’. Sometimes, one and the same DP can occur in distinct positions, leading to distinct interpretations (e.g. ‘Mary saw two movies, namely X and Y’ vs ‘The number of movies Mary saw was two’).

Page 18 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 01 November 2018

Quantification and Information Structure Ebert (2009) performs similar tests with German, using an observation by Frey (2004) that only topics can occur within the German middle field in front of certain sentence adverbials. Consider the following example: (59)

(p. 268)

In the position of D, determiners like ein ‘a’, zwei ‘two’, einige ‘several’, viele

‘many’, die meisten ‘most’, and perhaps also jeder ‘every’ are fine, whereas determiners like kein ‘no’, wenige ‘few’, weniger als drei ‘less than three’ or höchstens drei ‘at most three’ are bad. Quantifiers like mehr als drei ‘more than three’, genau drei ‘exactly three’, fast jeder ‘nearly every’ or mehr als die Hälfte der ‘more than half of the’ are degraded in the D position. The first group of quantifiers can be understood as topical, in contrast to the second. The generalization, following Szabolcsi (1997), is that monotone decreasing quantifiers are definitely excluded from the topic position. These are quantifiers like no student, few students, and less than three students for which it holds that if D(α)(β) holds, and β′ applies to a subset of β, then D(α)(β′) holds as well—for example, if no student slept is true, then no student slept and snored is true. Why should this be so? Szabolcsi (1997) invokes the notion of a minimal witness set. A witness set for a quantifier Q is a set W such that Q(β) holds if and only if Q(β⋂W) holds. For example, for every student the minimal witness set is the set of all students, and for three students a minimal witness set is any set that consists of three students. Now, a statement with a topical quantifier can be rephrased as being about such a minimal witness set of the quantifier: [[D students] left] is interpreted as: there is a minimal witness set W of [D students], and W left. Szabolcsi notes that for monotone decreasing quantifiers like fewer than three students the minimal witness set is empty, and hence this paraphrase does not work. We can modify Szabolcsi’s criterion by talking about witness sets in general. Notice that the truth conditions of sentences of the form [[D NP] VP] with quantifiers [D NP] that are not upward monotone cannot be rephrased as statements about their witness sets. Ebert (2009) suggests that topical quantifiers have the additional property that rephrasing them as statements about their witness sets does not change their anaphoric potential. This applies to quantifiers like three students, and distinguishes them from quantifiers like every student. The latter quantifier introduces discourse referents with a

Page 19 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 01 November 2018

Quantification and Information Structure life span limited to the sentence in which it occurs, whereas its reformulation in terms of statements on minimal witness sets involves discourse referents with unlimited life span. Topicality has also been invoked to explain the discourse effects of indefinites. Cresti (1995), as well as Ebert (2009), have suggested that the wide-scope effects of specific indefinites come about as a result of their topicality.

13.5 Conclusion It was the goal of this chapter to present the many ways in which quantification interacts with the information-structural features like focus, givenness, and topic. I have argued that there is a fundamental connection, insofar as the restrictor of a quantifier can be understood as given. This fact relates to natural-language quantifiers being conservative. There are different ways by which the restrictor can be determined. With adverbial quantifiers, or A-quantification, it is particularly obvious that information-structural (p. 269) features play a major role, be it focus, givenness, or topicality expressed by prosody, syntactic movement, or morphological markings. With nominal quantification, or D-quantification, the restrictor is more rigidly defined by syntactic structure, as the NP complement of the determiner. However, we have seen a variety of exceptions to this rule with the determiners many / few, with proportional determiners, and with focusing determiners, where D-quantification is subject to similar rules as A-quantification. We furthermore have observed that the topichood or focusation of quantifiers themselves can affect their scope, and other aspects of their interpretation.

Manfred Krifka

Manfred Krifka is professor of General Linguistics at Humboldt Universität zu Berlin and Director of ZAS, the Centre for General Linguistics, Berlin. He specializes on semantics and pragmatics on a broader variety of topics, and on descriptive linguistics (Austronesian languages of Vanuatu).

Page 20 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 01 November 2018

Predicate Focus

Oxford Handbooks Online Predicate Focus   Malte Zimmermann The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara Print Publication Date: Jul 2016 Subject: Linguistics, Morphology and Syntax Online Publication Date: Mar 2015 DOI: 10.1093/oxfordhb/9780199642670.013.26

Abstract and Keywords This chapter discusses a grammatically defined sub-class of focus: that on verbal predicates and on functional elements in the extended verbal projection. The phenomena falling under the label of predicate focus are introduced, and it is shown that predicate focus is interpretable on a par with argument or term focus on DPs and PPs. A unified structured-meaning approach that treats focus as the psychological predicate of the clause allows for singling out DP-terms and transitive verbs as categories in need of explicit marking when focused. A cross-linguistic overview of the grammatical strategies for marking predicate focus is provided, focusing on asymmetries in the realization of predicate as opposed to in terms of obligatory marking, grammatical strategy, and complexity. The information-structural and grammatical factors behind such focus asymmetries are discussed with some tentative universals concerning the explicit marking of information-structural categories on verbal predicates. Keywords: focus asymmetries, structured propositions, alternative semantics, term focus, predicate focus, common ground management, focus operators, TAM-focus, focus realization

THIS chapter discusses a grammatically defined sub-class of focus, namely the focus on verbal predicates and on functional elements in the extended verbal projection. Section 16.1 introduces the phenomena falling under the label of predicate focus, and Section 16.2 shows that predicate focus is, in principle, interpretable on a par with argument or term focus on DPs and PPs. Still, a unified structured-meaning approach that treats focus as the psychological predicate of the clause allows for singling out DP-terms and transitive verbs as categories in need of explicit marking when focused. Section 16.3 provides a cross-linguistic overview of the grammatical strategies available for marking predicate focus, focusing on asymmetries in the realization of predicate as opposed to term focus in terms of obligatory marking, grammatical strategy, and complexity. Section 16.4 discusses the information-structural and grammatical factors behind such focus

Page 1 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus asymmetries, and it concludes with some tentative universals concerning the explicit marking of information-structural categories on verbal predicates.

16.1 Predicate focus The term predicate focus refers to all instances of focus on lexical verbal predicates, such as V and VP, cf. (1a,b), and on functional elements in the extended verbal projection, like tense, aspect, and mood (TAM-focus), cf. (2a–c). Throughout, we control for focus by placing a wh-question and/or an assertion in need of correction in the preceding context (C); cf. for example Rooth (1992), Krifka (2008), and chapters 2 and 14 (this volume). Focus is marked by pitch accent on the focused constituent in intonation languages such as English; see chapter 9 (this volume). (1)

(p. 315)

(2)

Page 2 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus This characterization of predicate focus excludes focus on nominal arguments or adjuncts (term focus), as well as verum focus on the truth value of a given proposition. By excluding instances of verum focus (Höhle 1992 and chapter 15 this volume), the notion of predicate focus is more constrained than Güldemann’s (1996, 2003) predication focus and Hyman and Watters’ (1984) auxiliary focus, but, at the same time, it is less constrained by including instances of lexical focus on V and VP. By including instances of TAM-focus, the present notion of predicate focus is also less constrained than that of Lambrecht (2000) and van Valin (2005), which is restricted to the VP-comment and its subparts in sentences with topical subjects. Arguably, instances of thetic all-new focus fall under the notion of predicate focus as well. Following Erteschik-Shir (1997), one may analyse the entire vP in (3), including the indefinite subject (Diesing 1992), as predicating over an overt or covert situation argument: (3)

16.2 Semantic interpretation of predicate focus This section shows that predicate focus on verbal or TAM-categories can be semantically interpreted in full parallel to term focus on nominal categories, both in alternative semantics (Rooth 1992 and Chapter 2, this volume) and in the structured meaning approach (e.g. Jacobs 1983; von Stechow 1990), the two major frameworks for representing the semantic contribution of focus. In addition, it is shown that the structured-meaning approach, together with the conception of focus as the psychological predicate of the clause, singles out DP-terms and (di)transitive verbs as categories in need of (special) grammatical focus-marking.

Page 3 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus

16.2.1 Alternative semantics In alternative semantics, the semantic focus value of the focused constituent contains comparable alternatives of the same semantic type. Projecting to the sentence level, the resulting alternative propositions will differ only in the position of the focused constituent, while (p. 316) sharing the background part of the information. Examples (4a– c) illustrate the parallel interpretation of object term focus (4a) and predicate focus on V (4b) and T (4c), respectively: (4)

16.2.2 Structured meaning In the structured-meaning approach, the propositional meaning p of a clause is factored into a pair of focus (F) and background (BG) meaning, such that functional application of BG to F, or vice versa, gives back the original meaning p. The background meaning is arrived at by λ-abstracting over the focused part of the clause. The structured propositions in (5a–c), in which BG and F are represented as the pair , show again that this procedure applies to instances of term focus and predicate focus alike. (5)

Example (5c) shows that there is no principled problem with factoring out the meaning of functional operators, such as tense, which is here analysed in terms of existential quantification over times (Prior 1967). In (5c), the background part is analysed as a higher-order predicate over the propositional tense operator in keeping with the function–argument structure of term and V-focus in (5ab). Alternatively, the function– argument structure can be reversed by simplifying the background to the proposition λt. Peter pet- the cat at t, which then serves as argument for the tense operator in (5c). This

Page 4 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus alternative representation transparently reflects the semantic nature of tense as predicating over propositions.

16.2.3 Thetic sentences and focus on the progressive aspect Thetic all-new sentences are interpreted in full parallel. The background part of such sentences typically contains a temporally located situation (6a), but sometimes the tense component can also make up part of the focus. In such cases, the background will only contain (p. 317) an untensed situation (6b), which is plausibly construed as contextually bound: the background situation s1 in (6b) functions as the argument for the focused proposition. (6)

Finally, observe that—from a focus-semantic perspective—there is nothing special about the (focus) semantics of the progressive or imperfective aspect, which locates the reference time tR inside the running time of the situation described by the verbal predicate (Klein 1994). Just like tense operators (5c), aspectual operators can be focused, in which case they take a proposition as their background argument. Sentence (7a) could be a reply to ‘Has Peter petted the cat already?’ (7)

This observation is relevant in connection with Güldemann’s claim (1996, 2003) that progressive/imperfective aspect and predication focus are inherently connected. This is meant to account for the fact that focus-marking on terms is often impossible in progressive sentences in African languages (see also Hyman 1999 and Chapter 39, this volume). Whatever the reason behind this gap in the focus paradigm, it cannot be focussemantic in nature, as there is nothing in the meaning of the progressive in (7a) that would make it inherently predicative. A more promising possibility—to be investigated in future research—would be that the difficulties with realizing term focus in progressive sentences follow from conceptual problems with the backgrounding of temporally

Page 5 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus unbounded situations, which form a characteristic part of the progressive meaning, cf. (7b).

16.2.4 Association with focus The observed parallelism in semantic interpretation between term and predicate focus carries over to instances of association with focus-sensitive operators (Chapter 12, this volume), as illustrated in (8a,b) for object and VP-focus, respectively. (8)

On the earlier alternative semantics account of Rooth (1985) and on the structured-meaning account, the meaning of the focus particle only quantifies over focusalternatives and over different entities with the background property, respectively. In Rooth’s (1992) generalized treatment of focus semantics and association with focus, the quantificational domain of only is contextually specified, but is still constrained by the focus alternatives; see Chapter 12 (this volume) for details. The structured-meaning analysis for (8a,b) is shown in (9a,b), for simplicity leaving out the intensional meaning layer required by instances of VP-focus; see Krifka (2006). (p. 318)

(9)

16.2.5 Unified analysis and conclusion From a formal focus-semantic perspective, there are no principled differences between predicate focus and term focus. This semantic parallelism has led some researchers (see e.g. Löbner 1990; Wedgwood 2006; É. Kiss 2010) to treat all kinds of focus, including term foci, as instances of predicate focus semantically; see von Stechow (1991) for explicit discussion. The unified predicative analysis of focus implicitly takes up Paul’s (1880) characterization of focus as the psychological predicate of the clause. It tries to capture the intuition that the focused part of an utterance constitutes the new or relevant information that is predicated of an already established discourse entity—the psychological subject. Notice that the conception of focus as inherently predicative necessitates a reanalysis of focused DP-arguments as predicates. A Löbner-style (1990) reanalysis of object term focus in (5a) is shown in (10). Page 6 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus (10) The meaning in (10) is stronger than the one in (5a), though, in that it presupposes the existence of a unique individual that Peter petted in the context. For this reason, it cannot be an appropriate semantic analysis of accent focus in intonation languages (Rooth 1996). Still, (10a), or rather a situation-based variant of it, may well be the correct representation of focus in languages that mark focus by structural means, such as for example. clefting, movement, or morphological background-marking (Hole 2011). The situation-based representation in (10b) from Wedgwood (2006) merely requires there to exist some contextually given situation of Peter petting something, as marked by the epsilon-operator εs, with focus specifying the thematic argument of this background situation.

Next to focus on DP-terms, focus on (di)transitive verbs constitutes another marked case on the unified predicative analysis of focus. If the focus constitutes the main predicate of the clause, the background information will by necessity consist of several independent units, changing the representation in (5b) above to (5b′), in which the focus predicate takes three background arguments: (p. 319)

(5) In Section 16.3, it is shown that some languages, such as for example Nłeʔkepmxcin (Thompson River Salish), reflect the predicative nature of focus directly in their grammar, such that a focused constituent must be realized as (a projection of) the main predicate of the clause. It is also shown that the marked status of (di)transitive verb focus in the unified predicative analysis is reflected by the existence of special grammatical strategies for marking verb focus in many languages of the world. Section 16.4 elaborates further on the correlation between the information-structural concept of focus and the grammatical concept of predicate. For now, we conclude that any differences in the grammatical realization of predicate focus and term focus in a given language are not due to differences in their focus-semantic interpretation. Hence, they must follow for other reasons, such as grammatical constraints on focus-marking and the default interpretation of grammatical predicates as constituting the focus of the utterance.

16.3 Grammatical realization of predicate focus: asymmetries This section gives an empirical overview of the grammatical realization of predicate focus from a cross-linguistic perspective. Without aiming at a typologically balanced picture, the discussion concentrates on the following issues: (i) grammatical strategies employed for realizing predicate focus across languages (Section 16.3.1); (ii) (a)symmetries in the Page 7 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus marking of predicate focus and other constituents within individual languages (Section 16.3.2); (iii) mismatches, assimilation strategies, and complexities in the formal marking of predicate focus, with special attention to verb copying under focus (Section 16.3.3); (iv) the association with focus operators (Section 16.3.4). The survey shows that, across languages, the grammatical expression of predicate focus is marked when compared to term focus. In some languages, it requires only optional or no explicit focus marking at all. In others, it requires a special grammatical form. Moreover, in some languages the explicit grammatical realization of predicate focus can only be had at the cost of incurring mismatches between grammatical form and focus interpretation. Despite the observable peculiarities and complexities, Section 16.3.5 argues that there is no reason for assuming that predicate focus constitutes a distinct information-structural category of its own. The discussion is largely based on the grammatical (p. 320) realization of focus on lexical predicates (V, VP), with only occasional reference to focus on functional TAM-heads.

16.3.1 Grammatical strategies for realizing predicate focus Cross-linguistically, predicate focus is realized by a range of grammatical strategies, including prosodic marking, morphological markers, morphosyntactic devices, and syntactic reordering, the latter sometimes triggered by prosodic needs. Looking at prosodic marking first, intonation languages such as English and German with flexible accent position realize predicate focus by placing the nuclear pitch accent on the focused constituent. This was illustrated for English in (1) and (2) above. Next to these languages, there are intonation languages, such as French, in which the nuclear accent must be realized in a fixed phrase-final position (Lambrecht 1994). Consequently, the expression of verb focus requires that the verb be realized in clause- or phrase-final position, which, in transitive clauses, can be achieved by right-dislocation or fronting of the direct object, as in (11).

(11)

Page 8 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus Prosodic prominence is also employed for expressing focus on lexical verbs or other parts of the verbal complex in Japanese, such as, for instance, the passive morpheme rare in (12) (Shinichiro Ishihara, p.c.): (12)

Ishihara (p.c.) observes ‘that assigning a focal prominence on a part of the verbal complex does not necessarily mean that a phonological phrase boundary is inserted at the beginning of the morpheme’, as is the case with focus on DP- and PP-terms. This would suggest that Japanese is not a fully consistent phrasing language. Some Bantu languages such as Chichêwa realize predicate focus by means of prosodic phrasing; see for example Kanerva (1990), Downing (2003) and Chapter 39, this volume. Focus on the lexical verb in (13b) is marked by the insertion of a prosodic phrase boundary following the focused constituent. The phrase boundary is indicated (p. 321) by penultimate lengthening and tone lowering on the last syllable of the focused verb anaményá. (13)

Page 9 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus Looking at languages with morphological marking strategies, verb focus in the Gurlanguages Buli and Konni (Niger-Congo) in (14) is indicated by the sentence-final markers kámā and mìŋ, respectively (Schwarz 2010). The same markers are employed for marking verum or TAM-focus in (15), but NOT for marking focus on DP- and PP-terms; cf. (26) below.

(14)

(15)

Other languages employing morphological markers for expressing verb focus are Gùrùntùm (Chadic) and Medumba (Grassfield Bantu). These are discussed in Section 16.3.3 because they involve additional complexities.

Page 10 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus Two well-known languages with special morphological focus markers for predicate focus, Wolof and Aghem, may be more aptly characterized as employing a (p. 322) morphosyntactic strategy. This is because, apart from the morphological marker, the focused constituent must be realized in a special syntactic position. According to Torrence (2013), predicate focus in Wolof is realized by means of a predicate cleft construction built around the dummy verb daf, possibly derived from the light verb def ‘do’. In addition, the emphatic dummy verb shows number and person agreement with the subject, resulting in a complex formal paradigm of predicate focus markers discussed in detail in Robert (2000). Judging from the paraphrases given by the authors for (16a) and (16b), predicate focus clefts appear to be able to express at least V-focus (16b), VPfocus (16a.ii), and focus on verum or tense (16a.i). Glosses and paraphrases in (16b) are adapted from the original example in French (FMV = verbal focus marker, PR.O = pronominal object). (16)

Page 11 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus Predicate focus can also be marked syntactically by placing the focused constituent in a designated focus position; see for example É.Kiss (1998) and Chapters 8 and 33, this volume. In some languages, such reordering has been argued to be conditioned by prosodic factors, such as accent placement in Hungarian (Szendrői 2003) and prosodic phrasing in Nłeʔkepmxcin (Koch 2008). In the predicate-initial (Aux)VSO-language Nłeʔkepmxcin (Thompson River Salish), predicate focus is expressed by leaving the verbal predicate in its (default) initial position (Koch 2008; Koch and Zimmermann 2010). The sentences in (17ab) realize thetic all-new focus and VP-focus, respectively, (17)

Finally, while verb focus in Hungarian is typically expressed prosodically by assigning the main accent to the verb in situ, focus on the verbal head of verb–particle (p. 323)

combinations in Hungarian can be expressed by moving the head to the default focus position preceding the aspectual particle, where it is assigned main accent (Szendrői 2003): (18)

Having shown that predicate focus can be expressed by a range of grammatical strategies cross-linguistically, we next consider the question to what extent individual languages make use of the same strategies for marking predicate focus and focus on other grammatical categories, and to what extent they employ different strategies.

Page 12 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus

16.3.2 Symmetric and asymmetric focus-marking languages Languages differ as to whether they consistently employ one and the same strategy for realizing focus on all kinds of grammatical categories, or whether they use different strategies for marking predicate focus and term focus. Let us call the first class of languages symmetrically marking languages, and the second asymmetrically marking languages. To the former class belong languages that mark focus prosodically, but also languages that mark focus syntactically, possibly triggered by prosodic needs. Unrestricted intonation languages, such as English and German, constitute prototypical instances of a symmetrically marking language. As illustrated in (1) and (2), the device of free accent placement is a flexible enough tool to mark focus on maximal projections (DP, PP, VP), lexical verbs (V), and functional TAM-heads alike. Restricted intonation languages such as French (11), phrasing intonation languages such as Japanese (12), and pure phrasing languages such as Chichêwa (13) also belong to this group, at least with respect to the realization of focus on verbs and (in Japanese) other functional heads within the verbal complex. Examples (19a–c) illustrate the prosodic realization of object term focus on nominal categories in the three languages: (19)

To some extent, the syntactic focus-marking languages Hungarian and Nłeʔkepmxcin can be considered as symmetrically marking languages as well. In Hungarian, focused DP- or PP-terms are moved into the (accented) preverbal focus position, with the finite verb preceding the aspectual particle. The same movement to (p. 324)

Page 13 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus pre-particle position was observed with focused predicates in connection with (18); see Chapter 33, this volume, for further discussion of Hungarian. (20)

Of particular interest is Nłeʔkepmxcin which employs a purely predicative strategy for marking focus: the focused constituent must be realized as (part of the) sentence-initial main predicate in this VSO-language. If the focused constituent is predicative to begin with, as in the examples in (17ab) above, the focused constituent simply occurs in its default sentence-initial position. Bare NPs are predicative in this language and must be realized in sentence-initial position if focused (21a). Something special must happen with focused DP-arguments, however, which are referential, and must be turned into predicates by means of a clefting strategy (21b). See Koch (2008) and Koch and Zimmermann (2010) for arguments showing that the initial cleft-constituent is the grammatical predicate of the clause. (21)

Recalling the discussion of focus-semantics from Section 162, one can thus conceive of Nłeʔkepmxcin as a language that wears the predicative nature of focus on its sleeve. Cross-linguistically, it constitutes the rare case in which DP-term focus requires a more complex formal marking than predicate focus. (p. 325)

The fact that Nłeʔkepmxcin employs the same overall strategy for marking term and predicate focus (syntax), while exhibiting differences in the specific realization of predicate focus (no cleft copula) nonetheless, makes it a mixed language regarding the formal expression of focus. The same holds for Wolof, which marks focus by means of

Page 14 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus syntactic clefting throughout (Torrence 2013), but which employs different cleft copulas with predicate and term focus, respectively. Compare (16a,b) above with the object focus in (22): (22)

Aghem (Grassfields Bantu) is a mixed language as well, which differs from Wolof in that it features the same morphological marker in the marking of term focus and predicate focus on verbs, while the syntax of the two focus types differs: The morphological marker mô marks focus on verbal predicates and terms alike (23a,b) (Hyman and Watters 1984, Chapter 39, this volume), while the expression of so-called auxiliary focus on verum or TAM requires another morphological marker máà (23c). Crucially, though, the focused term in (23b) must occur immediately after the verb in the so-called IAV-position (Hyman and Polinsky 2010, Chapter 39, this volume), an option that is not available for the focused verb in (23a). (23)

It is possible that the element nó in (23a), which is glossed as a focus marker, has the grammatical function of filling the IAV-position as a dummy element, but this would still single out verb focus as involving a more complex structural representation. More generally, the differences in the realization of focus in Nłeʔkepmxcin, Wolof, and Aghem point to (p. 326) the existence of systematic asymmetries in the formal coding of predicate focus and term focus, which are regularly observed in languages with morphosyntactic focus-marking.

Page 15 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus

16.3.3 Asymmetries in the realization of term and predicate focus Languages with morphological and syntactic focus marking in the form of bi-partitioning (clefting, movement) frequently exhibit asymmetries in the realization of predicate focus as opposed to term focus. There appear to be three kinds of asymmetries, which concern (Section 16.3.3.1) the presence or absence of explicit focus marking; (Section 16.3.3.2) differences in grammatical strategy; and (Section 16.3.3.3) different degrees of structural complexity, that is simple as opposed to complex focus- marking.

Page 16 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus 16.3.3.1 Presence vs absence of focus-marking Some languages must realize predicate focus on TAM, V, and VP, by means of the canonical sentence form, without any explicit (i.e. non-canonical) marking for focus. This was shown for the focus/predicate-initial VSO-language Nłeʔkepmxcin above, but the same can be observed for the SVO-language Bura (Central Chadic, Afro-Asiatic), which is focus/predicate-final. Focus on V, VP, and TAM in Bura cannot be expressed by the only structural means of marking focus in that language, which is a cleft construction involving the focus copula an (24b) (Jacob et al. 2008; Hartmann and Zimmermann 2012). Instead, the canonical sentence structure in (24a) must be used.

(24)

In Hausa (West Chadic, Afro-Asiatic, SVO), predicate focus on V or VP is only optionally expressed by focus movement or the focus marker ne (Newman 2000; Hartmann and Zimmermann 2007). Predicate focus on any of the TAM-categories cannot be explicitly marked at all.

(25)

This is in striking contrast to the expression of term focus on subjects, which are obligatorily focus-fronted (accompanied by the focus form of the person–aspect-complex), and on non-subjects, which are regularly focus-fronted (Newman 2000). Presumably, the explicit marking of TAM-focus is blocked for structural reasons: TAM-categories are typically realized by functional heads or affixes, which cannot be targeted by (morpho)syntactic strategies of focus-marking, such as focus movement, clefting, or morphological focus markers. (p. 327)

16.3.3.2 Different strategies Many languages use different formal devices for the expression of focus on predicates and terms, respectively. Example (26) from Buli (Gur, Niger-Congo; Schwarz 2010) shows that term focus on the object is indicated by the pre-focal particle kà, and not by the

Page 17 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus clause-final particle kámā, which is reserved for the expression of predicate focus on V, VP, or TAM and verum focus; cf. (14), (15) above. (26)

In such languages, the grammatical strategy for marking term focus does not extend to predicates, mostly for structural reasons (see also the discussion of mixed languages in Section 16.3.2). As a result, a special grammatical strategy is required for expressing predicate focus.

16.3.3.3 Complex focus marking In many languages, predicate focus (on V and VP) is realized by the default strategy for marking term focus plus some additional complexities, which may come in the form of (i) assimilation to the term strategy; (ii) verb doubling; (iii) mismatches between grammatical structure and information structure. Beginning with assimilation, in many focus-fronting languages, explicit focus-marking of V(P)-predicates requires nominalization of the V(P), often accompanied by do-support. The main reasons seem to be that fronting is restricted to ([–verbal]) maximal projections (XPs), and that the sentence requires a finite verb to carry tense and agreement marking (Landau 2006). This is illustrated in (27) from Hausa, in which the fronted nominalized VP-predicate is replaced by the dummy verb yi ‘do’. Nominalization of the verb is indicated by low tone on the second syllable and the adnominal linker -n. The paraphrase shows that clefted predicates in English are non-finite as well: (27)

Page 18 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus Another way of assimilating the expression of predicate focus to that of term focus consists in the fronting of a nominalized predicate, which is doubled by a finite copy in the default position of the verb. This verb doubling strategy is found in languages as diverse as Vata (Kru, Koopman 1984), Hdi (Central Chadic, Frajzyngier and Shay 2002), Hebrew (Landau 2006), and Yoruba (Kwa, Niger-Congo, Aboh and Dyakonova 2009). Such nominalized (fronted) copies are also referred to as cognate objects on some accounts (Frajzyngier and Shay 2002). Example (28) from Manfredi (1993) illustrates for Yoruba. (p. 328)

(28)

In verb doubling languages, the lower copy appears to be necessary for carrying the finiteness features in the absence of do-support. Alternatively, it has been proposed that the lower copy is a resumptive form that licenses an otherwise illicit application of verb movement (Koopman and Sportiche 1986). A third instantiation of the assimilation strategy is found in the morphological focusmarking language Medumba (Grassfield Bantu): just like terms, focused verbs are marked by the prefocal marker á (29a), but with verb focus this á-marker must occur on an infinitival copy of the verb, which is right-adjoined to the VP-edge (29b) (Zimmermann and Kouankem in preparation); see Collins and Essizewa (2007) for parallel facts in Kabiye (Gur). The finite copy of the verb occurs in its default position and carries finite features: (29)

Page 19 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus Gungbe (Kwa, Niger-Congo) exhibits a different verb doubling pattern in that the focusfronted verb form is a finite copy of the verb (Aboh and Dyakonova 2009). This is not an instance of assimilation to the nominal term focus strategy. Rather, the additional complexity arises from the existence of two identical copies of the focused constituent. (30)

The third kind of additional complexity is constituted by mismatches between information structure and grammatical form. This happens when the default strategy for marking term focus is used for predicates, but the focused marker cannot surface on the focused constituent itself. Such a mismatch is illustrated in (31a,b) from Gùrùntùm (West Chadic, Hartmann and Zimmermann 2009), which marks V- and VP-focus by means of the regular term focus marker a. However, unlike (object) term focus in (31b), the focus marker does not precede the focused verb in (31a), but surfaces on the following object instead. (p. 329)

(31)

Hartmann and Zimmermann (2009) analyse the unexpected position of the focus marker in (31a) as emerging from the interaction of conflicting constraints: the informationstructural need for unambiguous focus-marking on the focused constituent is overridden by the structural constraint that the focus marker a must be realized on the nominal constituent that is structurally closest. The resulting structural ambiguity between narrow V-focus and object focus in Gùrùntùm is surprising from the perspective of European intonation languages at first sight, as it is not predicted by standard accounts of focus projection (Selkirk 1984; Chapter 10, this volume), which were mainly developed on the basis of such languages. On closer inspection, though, Gùrùntùm-style mismatches can also be found in familiar intonation languages. Consider, for instance, the realization of tense-focus in German (32), in which the focus accent is obligatorily realized on the stressed verbal stem sag-, instead of the focused suffix -te (Musan 2002). As noted by an

Page 20 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus anonymous reviewer, parallel facts obtain for English, even when the focused suffix is in a separate syllable, as in BUS-ted vs *bus-TED. (32)

In these cases, too, the mismatch between IS and grammatical realization appears to follow from a structural constraint that blocks the default focus strategy from applying to the focused tense morpheme. As in Gùrùntùm, shifting of the focus-marking device is subject to locality considerations in that it is the foot containing the focus that must (p. 330) be accented (Caroline Féry, p.c.). Notice that in those cases where accenting of the suffix is licit, as in (32′), it can only be interpreted as meta-linguistic correction, and not as a bona fide case of focus on the content of the accented morpheme. (32′)

16.3.4 Complex patterns in the association with predicate focus Just like morphological focus markers, focus particles (only, also, even) are in many languages restricted to combining with nominal (term) categories. In such languages, semantic association with predicate focus triggers assimilation processes and ISgrammar mismatches similar to those discussed in Section 16.3.3. For instance, the exclusive particle núm in Tangale (West Chadic) must syntactically precede the direct object even if it associates semantically with the verb (33i). The same order also gives rise to the (expected) association with the direct object in (33ii), without a change in prosody (Hartmann and Zimmermann 2007). (33)

Page 21 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus The reverse pattern is found with the additive particle məs ‘also’ in leftward-branching Ishkashim (Iranian). In (34), məs cliticizes on the DP to its left, which is the required structural configuration for association with nominal constituents, even though it semantically associates with the finite verb to its right (Karvovskaya 2013): (34)

One way to account for the mismatches in (33) and (34) would be to adopt Hole’s (2008) analysis of additives in Vietnamese, and to assume that the exclusive/additive particles in (33) and (34) are adverbial elements attaching to VP. The finite verb would then move into a higher INFL-position to the left (Tangale) and right (Ishkashim) of the particle, respectively: (35) This analysis would account for the word order facts, but it remains to be worked out to what extent an adverbial analysis of the particles is empirically tenable. (p. 331)

16.3.5 Arguments against a distinct IS-category of predicate focus The discussion in the preceding sections has shown that the grammatical realization of predicate focus tends to differ from focus-marking on DP and PP-terms across languages: First, many languages exhibit at best optional marking of predicate focus, or no explicit marking of predicate focus at all, whereas they show consistent marking of term focus— at least on subjects. This issue will be taken up again in Section 16.4.3. Secondly, languages may exhibit different grammatical strategies for marking predicate focus on verbs, as opposed to term focus. Third, many languages show a more complex marking of predicate focus on verbs in that the default strategy for marking (term) focus must be supplemented by additional grammatical mechanisms. However, the observable differences in grammatical marking should not be taken as evidence in favour of a distinct linguistic category of predicate focus. The arguments against such a category are threefold: first, from a structural perspective, not all languages exhibit asymmetric focus-marking, but there was shown to be cross-linguistic variation concerning whether or not predication focus and term focus are differently marked. For instance, the symmetrically marking intonation languages exhibit no formal differences in the marking of predicate and term focus. Secondly, from a focus-semantic perspective, it was shown that the semantic contribution of predicate focus and term focus is derived in parallel ways. Third, from the perspective of information-structure, the instances of predicate focus and term focus discussed here have the same status. The different formal markings on the two grammatical categories in asymmetrically marking Page 22 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus languages are found systematically in the same discourse environments, and they can be reliably elicited by means of the same empirical diagnostics, that is, answers to whquestions, corrections, and association with exclusive focus particles. To conclude, there is only one underlying IS-constituent of focus, which is realized differently on terms and predicates in many languages. In addition, the grammatical strategies for marking predicate focus are often more complex, and based on the default strategy for marking term focus. Next we turn to the question of which factors ARE responsible for the special grammatical status of predicate focus in so many languages.

16.4 Reasons for asymmetric marking and some tentative universals There are two independent factors that conspire to give rise to the cross-linguistically attested special status of predicate focus, resulting in a different, and often more (p. 332) complex grammatical realization in many languages. First, predicate focus is singled out as special by a general, possibly universal, cognitive constraint on Common Ground (CG) management in the sense of Krifka (2008) (Section 16.4.1). Secondly, grammatical strategies for marking focus are subject to structural constraints, which frequently block the default (term-based) strategy of grammatical focus-marking from applying to predicates (Section 16.4.2). The structural constraints are in part language-specific, but the cross-linguistic bias for term focus strategies may ultimately be induced by the more general cognitive constraint.

Page 23 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus

16.4.1 Cognitive constraint on the IS-status of terms and predicates The cognitive constraint is rooted in the way in which the information exchange in a discourse is organized. The information stored in the discourse models of the participants and in the CG is typically anchored to specific discourse referents. The privileged status of such discourse referents in the CG is reflected in numerous formal representations, such as the layered boxes of Discourse Representation Theory (Kamp and Reyle 1993), the file-card systems of Reinhart (1981) and Heim (1982), going back all the way to Russellian propositions of the form for Elisabeth II is wise (Russell 1910, 1912). Secondly, prototypical discourse-referents are concrete individuals, things, places, etc., which are typically denoted by DP- or PP-terms. Third, from the perspective of CG-management, the established discourse referents are fixed (≈ unfocused, without alternatives) at the current CG-state at which information on them is requested or provided. For instance, the question in (36) introduces the discourse referents Mary and the cat as fixed and requests a choice from a number of alternative predications (36) Fourth, and crucially, the information requested or provided in the form of a predication about the discourse referent must be in focus since the predication would be uninformative in the absence of alternatives: the informational needs of the discourse participants would already be satisfied. In sum, the cognitive constraint governing the anchoring of new information to pre-established discourse referents creates an inherent relation between focus and predication, which follows from basic cognitive constraints on inter-human communication via CG-management. This line of reasoning has immediate consequences for the formal realization of predicate and term focus, respectively. Verbal predicates (Vs, VPs) denote properties and activities of discourse referents, or situations containing them. Therefore, they constitute the default focus of an utterance, for which reason they need not be marked as such in order to be properly identified as focus. The optional or obligatory absence of formal focusmarking on focused predicates follows directly. Conversely, DP- and PP-terms (p. 333) denote discourse referents and do NOT constitute the default focus of an utterance, for which reason they tend to require marking under focus. This accounts for the observable cross-linguistic bias for explicit focus-marking on [nominal] terms as the less expected instance of focus. It also gives rise to the plausible speculation that focus-marking on nominal terms shows a wider cross-linguistic distribution, as well as an earlier emergence in language acquisition, and in the diachronic development of languages. Moreover, repercussions of the bias for focusmarking on terms are sometimes found even in flexible intonation languages such as

Page 24 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus English, in which the focus accent is located on the nominal object (37A1), and not on the verbal predicate (37A2), in the case of predicate focus on the VP (Krifka 2008).

(37)

On one standard account (Selkirk 1984), the VP-focus interpretation of (37) is the result of focus projection from the accented focus exponent to the VP; see also Chapter 11, this volume. The assumption of focus projection can be conceived of as mediating between focus on the VP-predicate, on the one hand, and its structural realization on a nominal argument, on the other. In the OT-accounts of Schwarzschild (1999) and Büring (2001), this bias for realizing VP-focus on nominal arguments is captured in the form of the OTconstraints HEADARG and ARGUMENT_OVER_PREDICATE, respectively.

Page 25 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus

16.4.2 Structural constraints on focus-marking The language-specific structural constraints follow from the cognitive bias for explicit marking of term focus, and from the fact that DP-/PP-terms, such as the/a man, at night, in the city, are non-verbal XP-categories. The morphological and (morpho)syntactic devices for marking term focus are therefore subject to structural and categorial restrictions that are typical of adnominal markers. First, an often observed structural constraint restricts focus movement and cleft copulas to apply to ([+nominal]) XPs. Second, a categorial constraint often restricts morphological focus markers to select for [+nominal] expressions. On a diachronic aside, Stassen (1997) has shown that in many languages such morphological focus markers derive diachronically from cleft copulas, from which they seem to inherit the relevant categorial and structural restrictions. In languages with structural restrictions on the formal marking of term focus, the relevant grammatical devices cannot be directly applied to the marking of focus on verbal predicates or predicative TAM-heads. This accounts for the asymmetries in the marking of predicate focus observed in Section 16.3.3. In some languages, such as for example Bura (Central Chadic), Duwai, and Ngizim (both West Chadic; Zimmermann 2011, the structural restrictions block the formal marking of predicate focus on V, VP, (p. 334) and TAM-heads, resulting in an absence of explicit focus-marking (A1) in these languages. Other languages must employ specific formal devices for marking predicate focus (A2) as the term focus markers are blocked from applying. In a third class of languages, the grammatical markers of term focus can be applied to the marking of predicate focus in principle, but only after certain formal adjustments have taken place, which result in additional complexity (A3): in assimilating languages, focused verbs must be turned into nominal XPs as a precondition for focus-fronting (Hausa, Yoruba) or for morphological focus-marking (Medumba). In genuine verb doubling languages Gungbe), a verbal copy must remain in the base position to carry finite features. Finally, there are mismatching languages (Gùrùntùm) that shift the focus realization from the focused (V-) predicate to an adjacent constituent. In sum, the special grammatical status of focused predicates follows from the combination of a general cognitive constraint, singling out the verbal predicate as the default locus of focus, and more language-specific, structural constraints on the grammatical focus-marking device.

Page 26 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus

16.4.3 Some tentative universals on focus/background marking in natural language In place of a conclusion, we propose a few tentative universals concerning the explicit marking of focus and background in natural language. In Section 16.4.1, it was established that the cognitive constraint governing the information exchange by means of CG-update results in the following default mappings from IS-status to grammatical category:

(38)

Default mappings between IS and grammar are not in need of explicit formal marking, from which it follows that predicate focus is often realized in the form of a canonical sentence without explicit focus-marking. Conversely, the assumption that non-default mappings between IS-status and grammatical category are marked, and hence in need of grammatical marking, gives rise to the following default patterns of explicit grammatical marking: First, focused terms (DPs, PPs) tend to be marked cross-linguistically (nominal bias). This holds in particular for focused subjects, which require obligatory marking across languages presumably because of their proto-typical IS-function as the sentence topic (Fiedler et al. 2010, and Chapter 4, this volume). Second, background or given predicates tend to be marked across languages. Indeed, natural languages exhibit numerous formal devices marking a V(P) as given or in background, ranging from the universally found strategy of ellipsis, as found for example in term answers, over deaccenting, to morphological background markers functioning as definite event determiners; for discussion see for example Baker and Travis (1997), Larson (2003), as well as Hole (p. 335) (2011) on the Mandarin particle de in shi…. de-clefts, and Grubic and Zimmermann (2011) on the verbal background marker -i in Ngamo (West Chadic). Based on these default marking patterns, we conclude this survey of predicate focus with the formulation of two tentative implicative and two, even more tentative, absolute universals. (39)

Page 27 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Predicate Focus The form of these universals are inspired by Skopeteas and Fanselow’s (2010) implicative universals concerning the formal realization of subject vs non-subject focus and new information vs contrastive focus, respectively. (40)

These absolute universals should be understood as expressing the strongest possible null hypothesis according to which certain aspects of the IS-grammar mapping are crosslinguistically invariant (von Fintel and Matthewson 2008). Future research will have to put this hypothesis to the systematic empirical test on the basis of a much larger language sample.

Malte Zimmermann

Malte Zimmermann is professor of semantics and grammar theory at Potsdam University, and currently director of the DFG-funded collaborative research centre SFB 632 „Information Sturcture“. His research interests range from laboratorybased experimental semantics to semantically informed field research on nonIndoeuropean languages, with a particular emphasis on West African languages.

Page 28 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Information Structure and Discourse Particles

Oxford Handbooks Online Information Structure and Discourse Particles   Patrick G. Grosz The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara Print Publication Date: Jul 2016 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jul 2015 DOI: 10.1093/oxfordhb/9780199642670.013.36

Abstract and Keywords Focusing on German as its object language, this chapter explores the interactions between discourse particles (e.g. ja, doch) and information structure. The first half of the chapter explores the traditional idea that discourse particles form a watershed between thematic (old) and rhematic (new) information. Concerning the ‘theme’, it is argued that unstressed elements to the left of discourse particles (in the ‘middle field’ between C0 and the verbal complex) must be aboutness topics. Concerning the ‘rheme’, it is shown that the focus exponent generally has to follow discourse particles. The second half of the chapter discusses current issues relating to the interactions of discourse particles and focus. Specifically, it investigates ‘relational’ discourse particles (e.g. doch and schon), which operate on a contextually salient alternative proposition q, giving rise to a discussion of the potential role that focus may play in the interpretation of such particles. Keywords: aboutness topic, discourse particle, focus, German, rheme, theme, watershed

17.1 Introduction: discourse particles THIS chapter is about discourse particles in the sense in which the term is used by authors such as Abraham (1991), Kratzer (1999), and Zimmermann (2011), corresponding to German Modalpartikeln ‘modal particles’.1 Since particles of this type are heterogeneous, both in their syntax and in their semantics, the class of discourse particles is difficult to define (cf. Zimmermann 2011). As a working definition, we can define discourse particles as a closed class of functional elements that contribute to Common Ground management in the spirit of Krifka (2008); that is, they encode specific instructions on how the Common Ground (Stalnaker 1974) should or should not be

Page 1 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 16 December 2018

Information Structure and Discourse Particles modified in the subsequent discourse. For a concrete analysis in this spirit, see Repp (2013), who argues that discourse particles mark their host proposition (in its entirety) as new vs given, expected vs unexpected, etc. with respect to the current Common Ground. To give a specific example, one meaning component that is typically associated with discourse particles can be paraphrased by ‘as you know’. This ‘as you know’ component has been attributed to German ja (e.g. Kratzer 1999), Dutch immers (Zimmermann 2011), Hungarian ugye (Gyuris 2009), Manado Malay kan (Stoel 2005), and St’át’imcets qa7 (Kratzer and Matthewson 2009); it plausibly exists in many other languages that utilize discourse particles. An approximation of German ja, representing the current standard view (e.g. Kratzer 1999; Zimmermann 2011), is given in (1).2 The lexical entry in (p. 337) (1) models ja as a truth-conditionally vacuous presupposition trigger, which makes reference to a contextually given speaker (speaker0), hearer (addressee0) and utterance situation (w0) (cf. Sauerland 2007; Schlenker 2007). In a matrix clause or an adverbial clause (cf. Thurmair 1989; Coniglio 2011), speaker and addressee will usually be identified with the actual speaker and addressee in the utterance situation (possibly via an additional context parameter c, cf. Kaplan 1989). By contrast, in complement clauses that contain discourse particles, the referents of g(speaker0), g(addressee0), and g(w0) can shift to a ‘reported’ speaker (e.g. the subject of a matrix verb of saying), hearer, and situation.3 (1)

In words, ja(p) triggers a presupposition that the contextually given speaker believes that the modified proposition p is true; it furthermore presupposes a belief concerning the contextually given addressee, namely that she either knows that p is true, or that the truth of p is evident in the utterance context (e.g. if p describes an eventuality in the utterance context, cf. Lindner 1991; Kratzer 1999). A concrete illustration of the meaning of ja is given in (2b) for the sentence in (2a). (2)

In terms of Common Ground management, ja conveys that uttering (2a) does not have the main goal of updating the Common Ground, since p is already in the Common Ground or

Page 2 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 16 December 2018

Information Structure and Discourse Particles evident in the utterance context. Such discourse-managing effects of discourse particles are highlighted by Thurmair (1989), Karagjosova (2004), and Repp (2013). Since the actual meaning of discourse particles is highly controversial, the presuppositional analysis in (1) should be viewed as one possible way of modelling ja. Alternative approaches entertain the view that discourse particles convey use-conditional/ expressive meaning (e.g. Kratzer 1999; Gutzmann 2012), or that they serve as modifiers of a speech act or an illocutionary operator (e.g. Thurmair 1989; Jacobs 1991; Lindner 1991; Waltereit 2001; Karagjosova 2004; Bayer and Obenauer 2011; Coniglio 2011). Due to the extensive literature on German discourse particles, which covers the topic of information structure to a significant extent, this chapter focuses on German. Other languages are briefly discussed in Section 17.4. German discourse particles include elements such as denn, doch,4 eben, halt, ja, schon, and wohl, to name a few (Thurmair 1989: 21). Due to the heterogeneity of this set, it is strategically useful to single out individual particles as case studies (such as ja in (1)–(2) above) and illustrate core properties for these samples; this strategy is pursued throughout this chapter. (p. 338)

This chapter focuses on two core issues pertaining to German discourse particles. Section 17.2 discusses the idea of Krivonosov (1977) that discourse particles play a central role in partitioning a clause into rhematic vs thematic information; the aim is to revisit this idea from a contemporary perspective on information structure. Section 17.3 focuses on current research that targets potential interactions between discourse particles and focus, raising the question of whether there are focus-sensitive discourse particles. Section 17.4 gives a cross-linguistic outlook and Section 17.5 concludes.

17.2 The watershed function Concerning a possible information-structural role of discourse particles, the central observation goes back to Krivonosov (1977: 202). He proposes that German discourse particles have a watershed function, which is roughly defined as follows: discourse particles are located to the left of rhematic (new) information and to the right of thematic (old) information. This insight is of particular concern with respect to the German middle field (cf. Thurmair 1989: 29–35), which corresponds to the area between the finite verb (in matrix clauses)/complementizer (in embedded clauses) and the verbal complex (cf. Wöllstein 2010: 41–50). Krivonosov’s template is illustrated in (3) for a matrix clause and in (4) for an embedded clause. In this illustration, Riko as a proper name qualifies as ‘old’ (thematic) information, whereas eine Frau ‘a woman’ as an indefinite qualifies as ‘new’ (rhematic) information.

Page 3 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 16 December 2018

Information Structure and Discourse Particles (3)

(4)

The most clear-cut contrast surfaces in examples like (5). The arbitrary pronoun man ‘one’ obligatorily precedes discourse particles, whereas the indefinite pronoun wer ‘someone’ must follow discourse particles (Haider 1993: 178); from Krivonosov’s perspective, this would suggest that man ‘one’ is always thematic, and wer ‘someone’ is always rhematic. (p. 339)

(5)

While Krivonosov’s view has largely been accepted (with further qualifications) in the literature on discourse particles (e.g. Hentschel 1986; Thurmair 1989), a core problem consists in the fact that the traditional notions of theme and rheme remain underspecified and thus difficult to operationalize (e.g. Krifka 2008: 265). Using Krifka’s (2008) terminology as a point of reference, it becomes clear that the traditional notion of theme is not equivalent to that of given information, but rather conflates the notions of topic and given information. Similarly, it is unclear whether rheme corresponds to the notion of focus (as is often assumed, e.g. in Thurmair 1989). Moreover, the traditional view that givenness and focus are complementary has been rejected in recent literature (Krifka 2008: 263). One core goal of this chapter is to tease apart the notions that are relevant for capturing the watershed function of discourse particles.

Page 4 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 16 December 2018

Information Structure and Discourse Particles There are three possible ways of operationalizing Krivonosov’s notion of a watershed function, (6a–c). We could identify: (i) theme with given and rheme with non-given information; (ii) theme with topic and rheme with comment; or (iii) rheme with focus and theme with non-focus.5

(6)

As we will see, the empirical facts are, indeed, somewhat more complex than anticipated by Krivonosov (1977). In Section 17.2.1, we start with a brief discussion of givenness as a possible equivalent to the concept of theme. Section 17.2.2 discusses the potential role of discourse particles as topic-marking devices. Section 17.2.3 addresses their potential role as focus-marking devices. Note that throughout Sections 17.2.1–17.2.3, I focus on the structurally ‘high’ discourse particles ja and doch, which are the prototypical watershedding elements. As shown in (p. 340) Coniglio (2005) and Grosz (2005) (based on Thurmair 1989), there are good reasons to believe that the heterogeneity amongst discourse particles correlates with different base positions. Lower discourse particles (e.g. ruhig and stressed JA in imperatives) may exhibit a very different distribution from the higher ones.

17.2.1 Water shedding and givenness Let us start by investigating the view that discourse particles separate given information from non-given information (Rochemont, this volume), (6a). Such a view appears to be corroborated by the example in (7) (based on Umbach 2004: 302; see Krifka 2008: 263). The sentences in (7b) and (7c) are disambiguated by the respective order of den Schuppen ‘the shed’ and the discourse particle ja. In (7b), den Schuppen ‘the shed’ must anaphorically refer to the entire farm, whereas in (7c), den Schuppen ‘the shed’ acts as a novel definite, introducing a new discourse referent for the (literal) shed associated with the farm.

Page 5 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 16 December 2018

Information Structure and Discourse Particles (7)

We can infer that (7b–c) have the information structure in (8a–b), thus corroborating Krivonosov’s view.

(8)

However, beyond clear-cut examples like (7), the preference for discourse particles to occur between given and non-given information appears to be a preference rather than a rigid constraint. Consider example (9), based on Krifka (2008: 264). Here, we observe that, while there certainly is a subtle preference for (9b), both (9b) and (9c) are possible (cf. Thurmair 1989: 29 for parallel examples). If discourse particles were to obligatorily follow all given information, (9c) should be deviant. This challenges the idea that given information must precede discourse particles, in favour of the weaker idea that there is a preference for given information to precede discourse particles.

Page 6 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 16 December 2018

Information Structure and Discourse Particles (p. 341)

(9)

Conversely, we may wonder if there is a more rigid requirement for non-given information to follow discourse particles. We observe that there are examples in which non-given information precedes discourse particles, as in (10) (based on Krifka 2008: 265), which can be uttered ‘out of the blue’. Crucially, Krifka (2008) claims that ‘a good friend of mine’ in (10) is actually a topic constituent that contains new (i.e. non-given) information, dissociating topicality from givenness. (10)

This indicates that the requirement for information that precedes discourse particles to be given may be a requirement for such information to be a topic. By default, as in (7b) and (9b), the two notions coincide, but, when they are teased apart, topicality matters, and not givenness, as suggested by (10). We thus proceed by discussing the possibility that discourse particles are topic-marking devices (Section 17.2.2).

17.2.2 Discourse particles as topic markers? To investigate potential interactions of discourse particles and topicality, I adopt Reinhart’s (1981) definition of aboutness topics. To quote Frey (2004: 163), whose approach I focus on, aboutness topics are ‘expression[s]‌ whose referent the sentence is about’. Frey investigates the syntactic distribution of (aboutness) topics, and posits the

Page 7 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 16 December 2018

Information Structure and Discourse Particles restriction in (11). However, he does not address discourse particles, which raises the question of how discourse particles fit into his view. (11)

In follow-up research, Frey (2006) proposes that German has three designated subject positions, illustrated in (12): the highest subject position is the position described in (11), and corresponds to the (aboutness) topic position, SpecTopP; the (non-topical) middle position corresponds to SpecTP; the lowest subject position corresponds to SpecvP. Crucially, Frey assumes distinct attachment sites for sentence adverbials as opposed to discourse particles. Sentence adverbials adjoin between SpecTopP and SpecTP, as shown for wahrscheinlich ‘probably’ in (12), whereas discourse particles adjoin between SpecTP and SpecvP, as shown for ja and doch in (12). (p. 342)

(12)

If Frey’s analysis is on the right track, we derive two predictions. Prediction 1: not everything that precedes a discourse particle is an aboutness topic (since it may be in a non-topical middle position). Prediction 2: aboutness topics must precede discourse particles (i.e. expressions that follow discourse particles cannot be aboutness topics). Below, I test these predictions and find that neither prediction can be confirmed. First of all, while (12) has its appeal, further qualification is in place, since the position of discourse particles is controversial. Meibauer (1994: 99) observes that the ‘traditional’ view (Brandt et al. 1992: 73; Brauße 1991; Doherty 1987; Jacobs 1991) holds that (unstressed) discourse particles precede sentence adverbials, as in (13a), which is the opposite of Frey’s (2006) position, reflected by (13b). These opposing intuitions are puzzling as long as one assumes the relative order of discourse particles and sentence adverbials to be rigid.

Page 8 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 16 December 2018

Information Structure and Discourse Particles (13)

6

Most plausibly, the assumption of a rigid order of discourse particles (PRT) and sentence adverbials (SADV) must be rejected. Meibauer (1994), Coniglio (2005), and Grosz (2005) observe flexibility in the ordering of discourse particles and sentence adverbials, that is, across different examples, native speakers tend to judge both the PRT verb order (1a). Focus surfaces in a different, though also fronted, position: unlike the Topic, it cannot be separated from the verb by other phrases, and in the case of a particle verb, the verb and the particle appear in an inverted verb > particle order (1b). (1)

A language in which some phrase structure configuration is mapped to Topic status can be referred to as topic-configurational. In a focus-configurational language, by analogy, some phrase structure configuration is associated with Focus status.4 Some languages, Hungarian included, are doubly discourse-configurational in that they are both topicconfigurational and focus-configurational. Cross-linguistically the two properties are independent of each other. Although up to the 1980s it was mostly viewed as peripheral, in recent decades the complex mapping between syntax and IS has been a central research topic in generative linguistic theory. The special interest of discourse-configurationality to the Principles and Parameters (P&P) approach to grammar is due to some of the key assumptions of this research paradigm. The first of these is the attribution of a fundamental role to hierarchical phrase structure in the explanation of syntactic phenomena. The second one concerns syntactic variation across languages, which P&P takes to be highly restricted, and parametric in nature. The third assumption is the Autonomy of Syntax hypothesis. Section 21.2 considers discourse-configurationality from the perspective of the first of these assumptions. It situates the notion in a wider context by clarifying its relation to (non-)configurationality, and by distinguishing it from the more inclusive concept of discourse-prominence. Section 21.3 provides an overview of major parameters of variation in discourse-configurationality. Section 21.4 turns to the relevance of the Autonomy of Syntax hypothesis in modeling discourse-configurationality. It outlines several prominent theoretical alternatives, concentrating on issues of grammatical

Page 2 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Discourse-configurationality architecture. Finally, Section 21.5 concludes with a summary of the pressing major issues at the forefront of current research.

21.2 The notion of discourseconfigurationality (p. 424)

21.2.1 Discourse-configurationality and (non-)configurationality The notion of discourse-configurationality is modelled upon the notion of configurationality. Although the two properties are conceptually related, each one is subject to independent parametric variation. Configurationality is ordinarily regarded as a property of languages in which grammatical functions like subject and object are correlated with particular phrase structural positions. Having dedicated canonical positions both for the subject and for the object, English unambiguously qualifies as a configurational language. A crucial phrase structural distinction between subject and object is that the object forms a constituent together with the verb (=VP), which excludes the subject. Using pre-functional projections terminology, subject can be defined as an NP immediately contained in S, and sister to VP; while object can be equated with an NP immediately contained in VP, and sister to V (Chomsky 1965). (2) On the assumption that subject and object occupy different, hierarchically asymmetrical positions, a number of subject/object asymmetries can be made to follow, without treating subject and object as theoretical primitives. These include differences in the distribution of anaphors and personal pronouns, asymmetries in word order, in quantifier scope, in negative polarity licensing, and in the availability of various displacements, to name a few examples (e.g. Chomsky 1981, 1986).5 In a wide variety of languages, which have come to be referred to as non-configurational, many, or possibly even most, of these linear and structural subject/object asymmetries appear to be absent. Significant from our present perspective is the fact that nonconfigurational languages tend to have free word order as far as grammatical functions are concerned, since the phrases that overtly express the verb’s arguments are not linked to a particular phrase structure position in the sentence (see Baker 1996, 2001).6 The surface phrase structure of sentences can, instead, encode the (p. 425) Information Structural status of sentence elements, among them their Topic and/or Focus role. Accordingly, a number of non-configurational languages have been shown to be discourse-configurational. For instance, East Cree has been described as syntactically

Page 3 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Discourse-configurationality marking (contrastive) Focus (Junker 2004), while the syntax of Warlpiri, a classic example of non-configurational languages, has been shown to encode both Topic and Focus (Legate 2002). In fact, the configurationality of grammatical roles and the configurationality of Topic or Focus discourse functions have even been approached as two alternative values along the same typological dimension, or in other words, as ‘two different types of configurationality’ (É. Kiss 1987: 250; Vilkuna 1989: 17–18; cf. also Sasse 1995: 1072– 1074). Yet, (non-)configurationality and discourse-configurationality are mutually independent properties. On the one hand, sentence structure, and hence constituent order, in non-configurational languages is not necessarily affected by Topic or Focus status, but it can be governed by other pragmatic factors instead. These include the Person/Animacy Scale (as in Fijian, see Aranovich 2013), evidentiality (as in Quechua, see Muysken 1995), relative degrees of newsworthiness (as in Coos, Cayuga, and Ngandi, see Mithun 1987), or givenness (see Siewierska 1993), among others. Thus, nonconfigurationality does not implicate discourse-configurationality. On the other hand, discourse-configurationality does not entail non-configurationality either (pace Abraham et al. 1986: 4–5; see É. Kiss 1995a): systematically encoding Topic or Focus status in its sentence structure does not prevent a language from also distinguishing phrases with different grammatical functions phrase structurally. This is the case, for instance, in languages like Japanese or Korean (Li and Thompson 1976; Saito 1985). Having said that, in a language in which configurationality and discourseconfigurationality co-occur, configurationality is at least partly masked in word order. This is because the phrase structure configurations associated with grammatical roles, such as those in (2), are routinely superseded in the syntactic derivation by other phrase structure configurations arising from the syntactic displacements of Topic and/or Focus constituents, such as those illustrated in (1).

21.2.2 Discourse-configurationality and discourse-prominence While the nature of the phrase structure configurations involved in discourseconfigurationality has been subject to debate (for which see Section 21.4), by definition the configurations themselves must be identified in terms of syntactic phrase structure (see note 4). In this respect discourse-configurationality should be differentiated from the notion of discourse-prominence. A language is discourse-prominent (or discourseoriented) if its sentential word order favours a grammatical description that attributes a key role to Information Structural categories, including Topic and Focus (for (p. 426) topic-prominence, see Li and Thompson 1976: 459; for focus-prominence, see Abraham and de Meij 1986). Although the two terms are oftentimes used interchangeably, discourse-prominence is a less restrictive notion than discourse-configurationality: it requires the relevant IS categories to be systematically associated neither to particular

Page 4 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Discourse-configurationality phrase structure configurations, nor even to any phrase structure configurations to begin with. I illustrate each of the latter two options in turns. Aghem is a Focus-prominent Bantu language, having a designated post-verbal Focus position (the so-called ‘immediately after verb’ or IAV position; see Watters 1979; Horváth 1995; Downing and Hyman, this volume). Polinsky and Hyman (2010), who provide a purely syntactic account of the Aghem IAV position, argue that what on the surface appears to be a designated immediately post-verbal locus for Focus, in fact corresponds to a set of syntactically restricted, but structurally different (base-generated) positions in the verb phrase. If there is no particular phrase structure configuration in which Focus systematically appears, then although Aghem can still be categorized as focus-prominent, it is not focus-configurational. In many languages the sentence-final Focus position has also been described as being structurally non-uniform. The Focus position at issue is sentence-final because what linearly ends up at the right edge of the sentence is mapped in these languages (by default) to the highest sentence-level prominence at the level of prosodic representation, viz. the Nuclear Stress (NS). Zubizarreta (1998) describes the Spanish and Italian sentence-final Focus position in these general terms (see also Vallduví 1992; SamekLodovici 2005). According to her account, if in a transitive sentence the subject is the Focus, then the object will undergo displacement in order to allow the subject Focus to end up in a linearly sentence-final position, where it will receive the NS in a default prosodic representation. I illustrate this with the VSO/VOS alternation in (3), analysing the VOS order for ease of exposition as derived by object-scrambling to the left of the subject (as argued by Ordóñez 1998). (3)

In non-contrastive contexts, (3a) can be interpreted either as a broad focus sentence or with focus on the object, while (3b) only has a subject-focus interpretation. Both the object-focus in (3a) and the subject-focus in (3b) are sentence-final, receiving the default nuclear prominence of the sentence, nevertheless they are in two different phrase structural positions: the Subject is in a specifier of VP, while the Object is in a complement position. More generally, although this is to be determined on a case-by-case basis, sentence-final Focus positions in languages with sentence-final NS are susceptible to an (p. 427) account in terms of prosodic requirements of Focus. The Focus needs to

Page 5 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Discourse-configurationality appear in a particular position at the level of prosodic representation, rather than at the level of phrase structure.7 In the same way, the favoured position for Focus in verb-final languages, namely the immediately pre-verbal position (Gunned 1988), is arguably also non-uniform structurally, and for the same reasons: the fact that it is linearly pre-verbal has been described as resulting from a general NS-Focus correspondence requirement (e.g. Neeleman and Reinhart 1998; Costa 1998 for Dutch; Ishihara 2001 for Japanese).8 To sum up the foregoing discussion, in some languages the phrase structural configuration in which Focus appears is variable; it is uniform only at the level of prosodic representation. From a syntactic perspective, in such languages the uniform sentencefinal or verb-adjacent linear position of Focus is merely an epiphenomenon, falling outside the scope of discourse-configurationality. At the same time, as these positions are grammatically defined by the notion of Focus, they will come under the more general rubric of discourse-prominence.

21.3 Variation in discourse-configurationality 21.3.1 The nature of the variation Discourse-configurationality, similarly to (non-)configurationality, has been viewed variously as a macro-parametric feature (Methuen 1987; É. Kiss 1995a; Ritter and Rosen 2005) or as a property of languages regulated by a small number of interacting parameters (see Miyagawa 2010). Alongside discrete parametric options, it has also been suggested that the extent to which different languages grammaticalize different discourse roles varies along a continuum (Sasse 1995; Öhl 2010). In this section, we survey the main dimensions of variation in discourse-configurationality, both within and across its two main manifestations: namely topic-configurationality and focusconfigurationality. This includes variation in terms of the syntactic positions (Section 21.3.2) and the IS categories involved (Section 21.3.3), and in terms of the uniqueness vs non-uniqueness in the mapping between IS categories and syntactic positions (Section 21.3.4). For a more (p. 428) comprehensive discussion of word order effects of Topic and Focus, see Neeleman and van de Koot (this volume); and for an extensive review of data drawn from different languages, see the studies in Part IV of this handbook.

21.3.2 Syntactic configurations Discourse-configurational languages, at least on the surface, differ with regard to the phrase structure position with which they associate a particular IS category of Topic or Focus. Beginning with topic-configurationality, a Topic typically occupies a left-peripheral Page 6 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Discourse-configurationality position of the sentence (Halliday 1967–68; Li and Thompson 1976: 465; Gundel 1988). The term left periphery refers to the region outside and to the left of the nucleus of the clause containing all the arguments in their canonical positions (see Rizzi 1997). A basic distinction is usually made between Topics that are syntactically part of the clause itself, and those that are syntactically (and prosodically) detached from it. Topics in Hanging Topic Left Dislocation (HTLD) are of the latter kind: they are attached externally to a syntactically complete sentence.9 Hanging Topics usually show no connectivity into the host clause, such as case-matching with a connected element inside it; they appear in a default case form (e.g. van Riemsdijk 1997; Frey 2004a), see (4). In this respect HTLD differs from Contrastive Topic Left Dislocation (CLD), in which the Topic is connected to a resumptive pronominal element inside the sentence. In CLD the Topic exhibits (some) connectivity effects, such as case-connectivity, in relation to the clause-internal element, see (5). This has been taken as evidence that the Topic in CLD is syntactically part of the clause containing the resumptive (for an overview, see Alexiadou 2006; for an account that challenges this view, see Ott 2014). In Clitic Left Dislocation (CLLD) (6), the resumptive correlate takes the form of a clitic pronoun (as in Greek, or Romance languages like Italian or Catalan); see Cinque (1990). (4)

(5)

(6)

As opposed to Left Dislocation constructions, plain Topicalization is a simple instance of syntactic movement in which the topicalized element is related to an unfilled gap (trace) inside the clause, as in (1a) above. (p. 429)

While designated Topic positions tend to occur at the left periphery of sentences, Topic status may also be associated with other positions. In one major type of Right Dislocation (RD), exemplified by She’s very nice, your Mum, the right-dislocated element is also apparently topical. Although it is discourse-anaphoric (Given), it is not clear that it has the status of an aboutness Topic (López 2009).10 Sentence-medial (or ‘low’) Topic positions have also been identified. For instance, Chinese has a unique Topic position below the subject (Paul 2002). Frey (2004b) argues that German is topic-configurational

Page 7 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Discourse-configurationality with regard to a sentence-medial aboutness Topic position that is located above the base position of sentence adverbials. Moving on to focus-configurationality, cross-linguistically the syntactic strategies of focusmarking characteristically belong to one of two main types. The position that Focus occupies may be sentence-peripheral or adjacent to the verb. The former may be in the sentence left periphery, as in Greek (Tsimpli 1995; see Skopeteas, this volume), Finnish (Vilkuna 1995), and Somali (Lecarme 1999), or in the right-periphery, as in American Sign Language (Wilbur 1994; see Kimmelman and Pfau, this volume). Some languages make use of both peripheries (such as Somali, where Focus is normally left-peripheral, but it is right-peripheral in the presence of a left-peripheral focus expletive, Lecarme 1999), although the interpretations in the two positions may differ (often in terms of contrastiveness, as in Russian, see Neeleman and Titov 2009). The focus position aligned with the verb may be a right-adjacent position, as in Aghem (Bantu, Watters 1979; see Downing and Hyman, this volume) or Western Bade (Chadic, Tuller 1992), or leftadjacent, as in Basque (Ortiz de Urbina 1989; see Arregi, this volume), Hungarian (FinnoUgric, Horváth 1986; see É. Kiss, this volume), or Malayalam (Jayeselaan 2001). A Focus position may be both sentence-peripheral and verb-adjacent at the same time, as is the case in languages like Basque, Hungarian, Spanish, and Romanian, in which both the Focus and the right-adjacent verb are raised to the left periphery. The two types of positions may also co-exist within a language, as in Western Bade (Schuh 1982; Tuller 1992), and in Ngizim and Tangale (where right-adjacency to the verb is slightly less strict, Tuller 1992). Although they constitute the predominant cross-linguistic options, sentence-peripheral and verb-adjacent Focus positions are not the only two possibilities. For instance, dedicated non-verb-adjacent sentence-medial focus positions have been argued for German by Frey (2004b) and Grewendorf (2005), and for Dutch by Neeleman and van de Koot (2008), in their analysis of scrambling in the middle field. In a different type of case, a normally right-aligned Focus position may end up non-sentence-final if it is followed by a right-dislocated element. In such a scenario an intonational (p. 430) break following it may nonetheless allow the Focus to occupy a right-edge position with regard to the intonation phrase in which it is contained (for Italian, see Samek-Lodovici 2006; cf. also the discussion of sentence-final Focus in Section 21.2.2). This is of course only a glimpse at the range of typological variation in the syntactic marking of Topic and Focus (from which we have excluded the issue of morphological marking).11 In the face of this apparent variation, it has been a central objective in the study of discourse-configurationality to uncover universals in the syntactic configurations that may underlie the positioning of Topic and Focus (under the proviso that some of the positions involved may turn out to be defined not syntactically but prosodically, see Section 21.2.2). We return to the nature of these phrase structure configurations as part

Page 8 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Discourse-configurationality of a discussion of the grammatical modelling of discourse-configurationality in Section 21.4.

21.3.3 IS categories As it is apparent from the foregoing discussion, discourse-configurationality manifests itself in heterogeneous ways across languages. Accordingly, several prominent types of discourse-configurationality are worth differentiating from a descriptive perspective, both according to the IS categories involved, which is the topic of this subsection, and according to the (non-)uniqueness in the mapping between IS categories and syntactic positions, to which we turn in Section 21.3.4. As defined in the Introduction, the property of discourse-configurationality holds of languages in which there is at least one phrase structure position such that all elements in that position are exclusively mapped to a unique information structural category that falls under the notions of Topic and Focus.12 Two apsects of this definition are to be emphasized. First, it does not admit languages with a syntactic position that is loosely associated with (some type of) Topic or/and Focus status, albeit in a non-exclusive way. For instance, in German a wide range of elements are associated with a contrastive Focus interpretation if they appear in the left-peripheral pre-V2 position. That, however, does not render German discourse-configurational, because the same pre-V2 position can also host phrases that are not interpreted as a contrastive Focus, including subjects, sentence adverbials, contrastive Topics, and non-contrastive Topics originating in the local clause (see Fanselow 2002).13 A second aspect of the above notion of discourse-configurationality that is worth highlighting is that it does not necessarily require a language to map the designated (p. 431) position at issue to an unqualified, broad notion of Topic or Focus, but also allows their information structural sub-categories to serve as values of the mapping function. Accordingly, Spanish, for instance, is deemed discourse-configurational on account of the fact that the elements in its left-peripheral ‘CLLD position’ are obligatorily interpreted as a contrastive Topic (Arregi 2003).14 As for Focus status, it is in fact doubtful whether any language demonstrates discourse-configurationality with respect to the general notion of Focus in the Roothian sense, which is based broadly on the relevance of alternatives and includes many different uses (including highlighting parallels in interpretation, as in socalled farmer-sentences like An AMErican farmer met a CaNAdian farmer, see Rooth 1992; Krifka 2008; see also Rooth, this volume). Instead, focus-configurational languages typically encode specific information structural sub-categories of Focus in their phrase structure, such as contrastive Focus, identificational Focus, and information Focus.15 Discourse-configurationality can be restricted in a language to a particular information structural sub-category of Focus or Topic in a way that cuts across these two IS notions. For instance, in Finnish a contrastive Topic must be fronted to a left-peripheral position, as in (7a). The same position, however, may alternatively be occupied by a contrastive Page 9 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Discourse-configurationality Focus. Thus, (7a) is felicitous both in the context of (7b) and in the context of (7c). This left-peripheral position is unavailable to non-contrastive Topics or non-contrastive Foci (see Vilkuna 1995; Vallduví and Vilkuna 1998). From this perspective, then, Finnish is discourse-configurational with respect to the IS category of Contrast.16 (7)

In brief, a second aspect of variation in discourse-configurationality concerns which IS category it involves.

(p. 432)

21.3.4 Uniqueness

Typically, discourse-configurational languages are characterized by a lax form of discourse-configurationality, which can be termed weak discourse-configurationality. In addition to having at least one phrase structural position that is mapped to a unique IS category, a weakly discourse-configurational language also has at least one other position that hosts elements that may (but need not) be associated with the same IS category. In other words, in weakly discourse-configurational languages the unambiguous syntactic marking of a particular IS category is optional. For instance, English and Hungarian are weakly discourse-configurational with respect to Contrast—although they employ two different syntactic constructions to mark it. In English, a fronted, left-peripheral position, unaccompanied by subject–auxiliary inversion or resumption, is mapped to either contrastive Topic or contrastive Focus status (Culicover 1991). In Hungarian, it is the CLD construction that maps the left-peripheral phrase exclusively to contrastive Topic or contrastive Focus status; see (6) above for an example.17 Crucially, in both of these languages, a contrastive (Topic or Focus) interpretation is available in at least one other position. Of particular relevance to the notion of weak discourse-configurationality is the syntactic behaviour of contrastive Focus in a remarkably wide range of languages. It is a crosslinguistically recurrent scenario that preposing a Focus to a left-peripheral position is associated with some special pragmatic import, not infrequently: Contrast (cf. Drubig 2003). Characteristically, such languages make available the same pragmatically marked interpretation outside the fronted position as well, rendering them only weakly configurational with respect to contrastive Focus (e.g. Italian, Rizzi 1997; Belletti 2004; Page 10 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Discourse-configurationality Bianchi and Bocci 2012; Spanish, Zubizarreta 1998; Korean, Vermeulen 2009; Finnish, Molnár and Järventausta 2003; Russian, Neeleman and Titov 2009; for a similar distribution of corrective Focus in Basque, see Ortiz de Urbina 2002; for Georgian, see Skopeteas and Fanselow 2010). Optionality in focus-fronting to a dedicated Focus position, hence weak discourse-configurationality, is not necessarily limited to contrastive Foci. For instance, in Sicilian (Cruschina 2012) and Greek (Gryllia 2008, Skopetas, this volume), both contrastive and information Foci are fronted only optionally. Allowing Foci to appear in more than one position may create the illusion in a language that syntactic focus-marking is optional, hence the language at issue is only weakly discourse-configurational. For instance, in addition to applying Focus fronting as in (1b), Hungarian also permits a Focus to remain in situ. However, the ex situ and the in situ Foci systematically differ in their interpretation, therefore there is no genuine optionality in fronting. All and only identificational Foci must be fronted to a dedicated left-peripheral pre-verbal position, and all and only plain information Foci must remain in situ (É. Kiss 1998, this volume; Horváth 2007, 2010). In other words, identificational Focus status is (p. 433) mandatory to be unambiguously marked in the syntax of Hungarian.18 Languages in which the mapping between a particular syntactic position and an information structural category of Topic or/and Focus is bi-unique can be termed strongly discourseconfigurational. With regard to Focus, strong discourse-configurationality is in fact rather rare to find cross-linguistically (if it exists at all, see note 18). The same pattern is less exceptional for contrastive Topic status, for which both strong and weak discourse-configurationality is attested.19 For instance, contrastive Topics are obligatorily fronted to a dedicated leftperipheral position in Japanese (Vermeulen 2013) and Hungarian (see Lipták 2011), while they also can appear outside their dedicated left-peripheral position in Korean (Vermeulen 2009; for Dutch, see Neeleman and van de Koot 2008). Strong and weak discourse-configurationality are complementary notions that divide the space of variation into two main types. Yet, further types may also be serviceable as descriptive categories. One special subtype that appears to be attested crosslinguistically differs only minimally from strong discourse-configurationality, as defined above. It is comprised by languages which must unambiguously mark an IS category in one of several distinct phrase structure positions, each of which is dedicated to that particular IS status. For instance, Aboh (2007) argues that Gungbe and Zulu are characterized by obligatory Focus-movement to either one of two Focus positions, associated with the same IS category of Focus. While the higher one of these two Focus positions is in the clausal left periphery, the lower one is above the verb phrase.20 It is reasonable to subsume such scenarios under a broader notion of strong discourseconfigurationality: in this broad sense strong discourse-configurationality is characterized by a bi-unique mapping between phrase structure and IS, albeit not necessarily with regard to a single position, but—disjunctively—with regard to a set of positions.

Page 11 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Discourse-configurationality

21.4 Modelling discourse-configurationality 21.4.1 Autonomy of Syntax and syntactic IS-features In this last part of the chapter, we turn to some outstanding theoretical issues in the modelling of discourse-configurationality, outlining various current approaches. The (p. 434) section considers the various roles that narrow syntactic IS-features play in recent models couched in transformational grammar (Section 21.4.1), discusses two challenges that the syntactic marking of Topic and Focus pose with particular regard to the Minimalist framework (Section 21.4.2), and outlines a cartographic and a noncartographic syncategorematic approach to Topic and Focus interpretation (Section 21.4.3). As is the case more generally, differences between alternative accounts are in no small part a function of fundamental assumptions regarding the architecture of grammar. In the case of discourse-configurationality, these differences are mostly related to the assumption of the Autonomy of Syntax and its implementation. The term Autonomy of Syntax covers a variety of distinct concepts (see Newmeyer 1998). For the present purposes, the autonomy thesis imposes the restriction that syntactic rules and principles can make no reference to notions of meaning, including Information Structure, and any mapping relating syntax and IS goes unidirectionally from syntax to IS (semantics/pragmatics plays an ‘interpretive’ role). No such IS-to-syntax mapping rules as ‘Topic must be clause-initial’ and ‘Focus must be clause-final’ are available. The autonomy hypothesis is a restrictive meta-principle of grammar that has by and large been adhered to in Chomskyan models, which constitute the theoretical purview of the present chapter.21 Competing Chomskyan approaches to discourse-configurationality differ significantly in the way they implement the autonomy hypothesis. Consider the grammatical architecture of the (inverted) T-model (aka Y-model), which has been the standard grammatical architecture for most of the history of transformational grammar. In this model both IS (a part of sentence ‘meaning’) and the phonological component are interpretive of syntax, and there is no direct interface between IS and phonological form. In order to account for the systematic relations between distinct information structural interpretations and their corresponding prosodic expression, the conditions triggering different IS representations must be syntactically represented. This has been a key role of discourse-related formal (syntactic) features like the [focus] feature, postulated by Jackendoff (1972). A [focus] feature establishes the needed link between focus interpretation and focus prosody.22 Importantly, the postulation of an IS-feature in syntax is not in itself a compromise to the Autonomy thesis, as long as no syntactic rules make reference to that feature. For instance, in Rooth’s (1992) influential theory of focus interpretation the [focus] feature enters the computation of meaning and prosodic form, but is otherwise syntactically inert Page 12 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Discourse-configurationality (the same holds for Jackendoff 1972).23 Paradoxically, a model in which syntax (p. 435) itself does not make any reference to a particular IS-feature lacks any syntax-internal motivation to postulate it as a formal syntactic feature to begin with, which renders that IS-feature altogether different from other narrow syntactic features. The alternative is to assume formal IS-features to be involved in conditioning syntactic rules, among them, movement operations. This is the basis of many narrow syntactic accounts of the word order effects of Topic and Focus across languages, reviewed in Section 21.3. A prevailing strategy to derive the positional restrictions involving Topic and Focus has been to posit corresponding IS-features on syntactic heads (distinct from Topic and Focus consitutents), with which a Topic or Focus is required to establish a local syntactic relation, by movement or base-generation. Horváth (1981, 1986) proposed that the [focus] feature is assigned by a syntactic head under government and adjacency in ways analogous to the assignment of structural Case. She applied this account to the obligatory displacement of the Focus as well as its adjacency to the verb in various languages; see Section 21.3.2. Brody (1990), on the other hand, assimilated the behaviour of Focus in Hungarian to that of interrogative wh-phrases by postulating a Focus Criterion, as a counterpart to Rizzi’s (1990) Wh-Criterion. Concurrently, Brody also proposed a dedicated functional projection, commonly labelled FocP, to house a fronted Focus. A criterion-based account requires a phrase (here: the Focus) bearing some feature [F]‌ (here: the [focus] feature) to be in a specifier-head relation with a functional head that bears the same feature [F] (Rizzi 1990).24 (8) is a partial analysis of (1b), with the fronted object NP appearing in the specifier of FocP (and the raised verb incorporated into the Foc head). (8)

The syntactic licensing of Case- and agreement-features as well as the Wh-/Neg-/Focuscriterion are unified and subsumed under the notion of feature checking in Minimalism (Chomsky 1993, 1995). Feature checking is the licensing of an uninterpretable feature by some matching interpretable feature, which takes place provided that the two enter a sufficiently local syntactic configuration. In (8) the [focus] feature is taken to be uninterpretable on the Foc functional head, and is in need of licensing (checking). As the [focus] feature is interpreted on the NP John, it can license the [focus] feature on Foc. Ultimately, this is what triggers the movement of the Focus NP to the specifier of FocP (Brody 1995).25 Models that posit IS-features as properties of heads of functional projections are able to account both for the distributional properties of Topic and Focus in a language and for (p. 436) the relevant intra- and cross-linguistic variation in a uniform way, by postulating (quasi-)lexical differences in the syntactic properties of the relevant functional heads. Among others, these differences include their location in the structural hierarchy (e.g. a high and a low FocP, Jayeselaan 2001; Belletti 2004) and their ability to recur or to host multiple specifiers (e.g. Serbo-Croatian allows multiple Focus fronting, Stjepanović 2003; Page 13 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Discourse-configurationality Hungarian does not, Surányi 2004). Given that analogous syntactic parameters govern the distribution of syntactic elements more generally (see Chomsky 1995), no dedicated grammatical mechanism is invoked to account for the syntax of Topic and Focus. Reference to IS-features in syntactic processes makes the syntactic reflections of IS modellable within narrow syntax without making direct reference to IS notions. There is no denying the fact, however, that such a use of IS-features voids the Autonomy of Syntax hypothesis of much of its empirical content: through formal IS-features syntax is empowered to refer to notions of meaning, contravening the spirit, although not the letter, of autonomy.26

21.4.2 Discourse-configurationality and Minimalism In relation to the syntactic marking of Topic and Focus, several issues are outstanding in the specific context of the minimalist research programme. One of these is related to Chomsky’s (1995: 228) restrictive principle of Inclusiveness. This principle states that a complex syntactic object can only come to have a syntactic property if it is a property of some lexical item that it contains. Assuming Inclusiveness, the Focus constituent can come to bear the [focus] feature only if [focus] is a feature of one of the lexical items contained in it. The problem is that features like [focus], [topic], or [contrast] are, plainly, not features of lexical items; thus Inclusiveness is violated (Zubizarreta 1998: 30; see also Kidwai 1999; Szendrői 2004; see Neeleman and Szendrői 2004; Horváth 2010 for further discussion).27 A second minimalist principle of key relevance to the syntactic marking of Topic and Focus is the economy of movement, according to which no movement takes place unless it serves the licensing of an uninterpretable feature (Chomsky 1993, 1995, 2001, 2004). As unlicensed uninterpretable features result in an ungrammatical derivation, movements are generally obligatory. As discussed in Section 21.3.4, however, a range of IS-related movements appear to be genuinely optional. This is an apparent problem for the principle of economy, which predicts that optional movements do not exist. Not all cases of positional optionality involve optionality in movement, however. Recall from Section 21.3.4 the optional fronting of certain types of Foci in Italian, (p. 437) Basque, Sicilian, and Greek. In these languages non-fronted foci must appear sentencefinally. This particular type of optionality may be analysed as involving obligatory movement to either one of two distinct possible landing sites, rather than a free choice between movement and no movement. Alternatively, the Focus constituent may turn out to target the very same structurally left-peripheral Focus position in both cases, with the surface word order difference deriving from the IS status of the remaining part of the sentence. In particular, sentences with a final Focus may differ from those with a leftperipheral Focus only in that in the former case the non-focus (background) part of the sentence has Topic status, and as such it has moved to a Topic position to the left of the left-peripheral Focus (9) (for empirical arguments in favour of this account of Basque, see Page 14 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Discourse-configurationality Ortiz de Urbina 2002; for the same derivation of sentence-final contrastive Focus in Italian, see Belletti 2004: 29; cf. also Samek-Lodovici 2006). As both movements are obligatory on the associated IS interpretation, no problem arises for the economy of movement. (9)

Among purported cases of optional IS-related movement a great many can be convincingly analysed as being only apparent. Those that defy such a characterization pose a genuine challenge to the minimalist principle of economy of syntactic derivation. Feature-checking based models of IS-related movements also face an issue of a different nature. Namely, the checking of discourse-features is unlike feature-checking as known from A-movements. In A-movements triggered by feature-checking needs, features receive one of several different values (e.g. values for number and person) and checking between the moved element (Goal) and the functional head (Probe) is mutual (e.g. the subject NP checks person and number features on finite T, while the latter checks the Case feature on the former; Chomsky 2000, 2001, Pesetsky and Torrego 2004). Partly for these reasons and partly because of the apparent threat posed to the autonomy thesis, Chomsky (2000 et seq) rejects the idea that the checking of dedicated features is involved in the domain of discourse-related movements. These movements are still triggered to perform feature-checking, nevertheless: namely, the checking of Edge-features (a generalized type of EPP-features) on the relevant Probe functional heads. Whether or not an Edge-feature is assigned to a functional head before it is Merged in the derivation is regulated by a general economy condition on optional rules (Chomsky 2000, 2001; see Reinhart 2006): (10)

In this implementation too, the movement operation itself does not look to semantic/pragmatic interpretation for its licensing. Nevertheless, syntax must consult the interpretive component to assess whether the assignment of the Edge-feature satisfies the economy condition in (10). Thus the fact that the syntactic operation of Edge-feature assignment is contingent on interpretation, specifically, on a comparison of two interpretations, remains a compromise to the autonomy thesis. (p. 438)

Page 15 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Discourse-configurationality

21.4.3 From cartography to syncategorematic rules of interpretation In the general approach to the syntax of Topic and Focus in terms of functional projections, the Focus or Topic constituent must ordinarily appear in a specifier position of a functional head that bears a corresponding formal feature, as in (8) above. Particular accounts may differ with regard to whether the functional heads hosting the [focus] or [topic] features are heads with an independent existence, such as T(ense) (as in Aissen 1992; Horváth 1995; Miyagawa 2010; Surányi 2012), or whether they are dedicated to hosting Focus or Topic, such as Foc and Top (as in Brody 1990, 1995; Rizzi 1997). This latter type of account is characteristic of the so-called cartographic approach to syntax. The cartographic approach draws up a detailed functional map of sentence structure in terms of an ordered set of functional projections, many of which are dedicated to specific information structural roles, especially—though not exclusively—in the left periphery (Rizzi 1997, 2001; Belletti 2004; Benincà and Poletto 2004). To illustrate, drawing on Rizzi’s (2001) hierarchy, Frascarelli and Hinterhölzl (2007) propose the cartography of this clausal periphery in (11) below. The specifier of ShiftP is the position of shifted (‘new’) aboutness Topics, which is above the location of contrastive Topics. These are higher than the position of left-peripheral Focus, which is in turn followed in the hierarchy by (possibly recursive) familiar aboutness Topics. (11) The cartographic approach has contributed a great deal of detailed, comparative knowledge regarding the distribution of constituents with different discourse functions in a variety of languages (for an overview of alternative cartographic accounts, see Aboh, this volume). The cartographic approach has been both criticized and defended from both theoretical and empirical perspectives. It is beyond the scope of this chapter to review and assess the arguments here (see van Craenenbroeck 2009; Cinque and Rizzi 2010). I restrict myself to merely pointing out one conceptual problem for variants of the model that assume both a [focus]/[topic] feature on the Focus/Topic constituent and a dedicated FocP/TopP projection. In such models there is a degree of unappealing redundancy with regard to the mapping from syntax to Focus/Topic interpretation: the Focus/Topic (p. 439) status of a constituent is encoded both by the IS-feature it bears and by the IS-related categorial feature of the dedicated functional projection that houses it in a specifier position (FocP/ TopP). The redundancy is only aggravated if the dedicated functional heads themselves are also associated with [focus]/[topic] features, in addition to their corresponding categorial feature. This argument from redundancy favours a variant of the cartographic model that encodes IS-features only in the syntactic category of the dedicated functional projection and does not concurrently postulate IS-features on Focus and Topic constituents. A constituent that is associated (at some stage of the syntactic derivation) with a FocP or TopP position will Page 16 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Discourse-configurationality receive a Focus or Topic interpretation as a function of the syntactic category of its host functional projection FocP/TopP. In other words, in this type of model, Focus/Topic interpretation is the result of syncategorematic mapping rules; it is a configurational, rather than a featural, property of the syntactic constituents occupying the relevant positions (cf. also note 5). Syncategorematic rules of interpretation do not necessarily rely on categorial [focus]/ [topic] features in the local context of Focus/Topic constituents, that is, on dedicated functional projections. Neeleman and van de Koot (2008) propose discourse templates, which are syncategorematic rules of interpretation that make no reference in their input to IS-features in any form, not even in the guise of dedicated projections. The particular mapping rules they postulate are (12b,c), which apply to the configuration in (12a). (12)

Neeleman and van de Koot (2008) demonstrate that these mapping rules can explain both the basic flexibility in the range of (in situ and derived) positions in which contrastive Topic and Focus interpretations are available in Dutch, and the complex set of restrictions that apply to syntactic scenarios in which more than one phrase is interpreted as a contrastive Topic or Focus. Both of these are cumbersome to derive for the cartographic approach, either in its IS-feature-based or in its syncategorematic implementation (for an extension of this account to Korean, see Vermeulen 2009). There is a price to pay, however, for the lack of IS-features in syntax. In particular, in order to capture prosody–meaning correspondences, the grammatical architecture needs to be enriched with a direct interface between IS and prosody (cf. note 7). With such an interface added to the T-model, one key task is to minimize the redundancy in empirical coverage between the prosody–IS mapping and the syntax–IS mapping, which requires a systematic investigation of the division of labour between the two interface subsystems. The issue is central to the current research agenda (for relevant discussion, see Zubizarreta, this volume, Samek-Lodovici, this volume, and Truckenbrodt, this volume); most of the work still lies ahead.

(p. 440)

21.5 Conclusion

The addition of the notion of discourse-configurationality to the configurational/nonconfigurational typology was motivated by the recognition that the InformationStructural notions of Topic and Focus may be just as systematically associated with particular phrase structure positions in some languages as the grammatical roles of argument phrases are in overtly configurational languages. This recognition has been Page 17 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Discourse-configurationality driving an ever-expanding body of research into the formal syntactic underpinnings of Information Structure, and into parameters of variation in this domain. Key questions at the forefront of this research concern the division of labour between prosody and syntax in capturing word order regularities in the expression of IS, the status of formal ISfeatures in narrow syntax (if any), and the nature and role of the syntactic configurations involved in the mapping between syntax and IS.

Notes: (1) I thank the anonymous reviewers, and especially, the volume editors for their helpful comments and valuable suggestions. The support of a grant of the Hungarian Scientific Research Fund (#NF-84217) and the Momentum grant of the Hungarian Academy of Sciences (#2011-013) is gratefully acknowledged. (2) For the notions of Topic and Focus, see Krifka (2008). For reasons of space, here I will not treat syntactic effects of other notions of Information Structure, like Givenness; these could possibly fall under a broader concept of discourse-configurationality. For Givenness, see Neeleman and van de Koot (2008, this volume), Kučerová (2012), Rochemont (this volume). (3) É. Kiss (1995a: 6) limits the notion of discourse-configurationality with regard to Focus to identificational Focus, which is defined as a Focus associated with a quantificational operator expressing exhaustive identification (É. Kiss 1998, this volume; Horváth 2007, 2010). (4) The term ‘phrase structure configuration’ in principle denotes any kind of syntactic position that is defined in terms of phrase structure. See Section 21.4 for relevant discussion. (5) See also Hale and Keyser’s (1993, 2002) configurational approach to thematic relations, according to which theta roles are emergent (syncategorematic) properties of constituents that appear in particular syntactic configurations in the phrase structure of ‘lexical’ phrases. (6) Baker (1996, 2001) adopts Jelinek’s (1984) Pronominal Argument Hyptothesis, according to which in non-configurational languages the true arguments are pronominal elements. For Baker, these pronominals are null pro pronouns, generated in a fully configurational structure. (7) For different prosodic accounts of Focus-related marked word orders, some of which assume the T-model coupled with a syntactic focus-feature (cf. section 21.4.1), while others posit a direct interface between prosody and meaning, see Truckenbrodt (this volume), Zubizarreta (this volume), and Samek-Lodovici (this volume).

Page 18 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Discourse-configurationality (8) The IAV Focus position in Bantu has also been analysed in prosodic, rather than syntactic, terms; see Costa and Kula (2008) for Bemba. For a prosodic account of the preverbal Focus position in Hungarian, see Szendrői (2003). (9) Topics in HTLD constructions are also referred to as external Topics, following Aissen’s (1992) analysis of Mayan languages. (10) For an overview of both left and right dislocation constructions, see Cecchetto (1999) and López (this volume). (11) This chapter concentrates on discourse-configurationality at the clausal level. For discourse-configurationality in the nominal phrase, see Giusti (2006), Bernstein (2001), Haegeman (2004), and Aboh (2010, this volume). (12) Wh-phrases occupying a dedicated Focus position are put aside, noting that they are commonly assumed to function as a Focus (Rochemont 1986; Horváth 1986). (13) This is so if the conservative assumption is adopted that all elements immediately preceding the Verb Second position occupy the same phrase structural position (for an alternative, see Frey 2006). (14) The ‘CLLD position’ in Spanish is a Topic position, unaccompanied by verb inversion. A phrase in this position, depending on its syntactic and semantic nature, may or may not be associated with a resumptive clitic pronoun (Arregi 2003; cf. also López, this volume). (15) The latter IS notions overlap with one another; languages may also display discourseconfigurationality with regard to their various intersections (e.g. É. Kiss 1998). Information Focus is taken to be a Focus that corresponds to the information gap in the current, possibly implicit information question being addressed (Roberts 1998). This IS category is orthogonal to ‘new information’ (Selkirk 2008; Rochemont 2013, this volume). (16) Contrast is assumed here to be a dependent property for which Topic or Focus status is a prerequisite (see Neeleman and Vermeulen 2012). For different notions of Contrast, see Repp (this volume). (17) For Italian CLLD, see Benincà and Poletto (2004). In many languages, CLD is narrowly limited to contrastive Topics (see Neeleman and van de Koot, this volume). While there is cross-linguistic variation with regard to C(L)LD, HTLD is universally associated with Topic status. (18) Notwithstanding this generalization (to which certain systematic exceptions exist, Surányi 2004), it is questionable whether identificational Focus is in fact an IS category in its own right (see Horváth 2007, 2010). (19) The issue of optionality in fronting to a Topic position is more involved in the case of non-contrastive Topics, mainly due to difficulties in reliably identifying them (Portner and Yabushita 1998, Roberts 2011; see also Büring, this volume). Page 19 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Discourse-configurationality (20) Compare Tuller (1992), Jayeselaan (2001), and Belletti (2004). In phase theory (Chomsky 2000, 2001), the Focus positions at the periphery of the clause and the verb phrase may be subsumed under a single notion of phase-peripheral Focus position. (21) Approaches grounded in the functionalist tradition, which reject the autonomy thesis, are beyond the scope of this chapter (e.g. van Valin 2005; Hengeveld and Mackenzie 2008; see also Primus 1993). (22) From the perspective of compositionality of meaning, narrow syntactic IS-features have another function: they furnish an appropriate formal object as input to the semantic derivation of the correlated IS interpretations (see Rooth 1992, this volume; Büring, this volume). (23) Although it does not enter transformational rules, Selkirk’s (1984, 1995) [focus] feature (F-marking) is not fully inert in narrow syntax, as it partakes in a syntactic mechanism of [focus] feature projection that is crucially sensitive to syntactic structure. For a critique of Selkrik’s (1984, 1995) theory of focus projection, see Büring (2006) and Arregi (this volume). (24) Brody (1990) conceptualized the [focus] feature on the Foc head as a categorial feature; see Rizzi (1997) for a generalized Focus Criterion. For early syntactic analyses of Topic-fronting relying on a formal feature [topic], see for instance Kim (1988) and Lasnik and Saito (1992). See Mallen (1992) and Rizzi (1997) for an analogous Topic Criterion. (25) A great deal of minimalist work assumes that [focus] and [topic] features enter feature checking, though Chomsky himself has never explicitly adopted such a view (Chomsky 1993 et seq; see also Section 4.2). (26) For further arguments against narrow syntactic IS-features, see Fanselow (2006, 2008, 2012), Zimmermann (2008), Horváth (2010), and Fanselow and Lenertová (2011). (27) One possible response to this quandary is to limit IS-related features to categorial features of dedicated functional heads, see Section 4.3. Another alternative is to treat all Foci/Topics as elements combining with a possibly null focus-/topic-particle that can bear the IS-feature as a lexical property. Such particles would be relatively unselective with regard to the syntactic category of their host (Bayer 1996; Horváth 2007; Cable 2010; cf. also Rooth 1985: ch. 3).

Balázs Surányi

Balázs Surányi is a research professor at the Research Institute of Linguistics of the Hungarian Academy of Sciences, and Professor of Linguistics in the Department of Theoretical Linguistics of Pázmány Péter Catholic University, Budapest. His principal research interests include the interface of syntax and Information Structure,

Page 20 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Discourse-configurationality minimalist syntax, and more recently, the interaction between Information Structure and intonation.

Page 21 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Tulane University; date: 01 January 2019

Dislocations and Information Structure

Oxford Handbooks Online Dislocations and Information Structure   Luis López The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara Print Publication Date: Jul 2016 Subject: Linguistics, Semantics Online Publication Date: Dec 2014 DOI: 10.1093/oxfordhb/9780199642670.013.003

Abstract and Keywords Dislocations are constituents in the periphery of the clause—or, depending on the analysis, outside the structure of the clause proper. In the canonical cases, they are doubled by a functional bundle and they are separated from the core clause by an intonational phrase boundary. In many languages we find that dislocations come in two classes: a class of dislocations that are syntactically linked to a position in the core structure (D-type) and a second class of dislocations that are connected only in the process of interpretation (H-type). This grammatical distinction maps onto a difference in the information structure properties: D-type dislocations are given, H-type dislocations signal topic promotion. Some languages seem to have only H-type dislocations, which take over the functions of both D-type and H-type dislocations. Keywords: resumptive, clitic, weak pronoun, hanging topic left dislocation, left dislocation, contrastive left dislocation, clitic left dislocation, clitic right dislocation, topicalization, islands, reconstruction, orphans, afterthought, intonational phrase, topic

20.1 Introduction THE Catalan example (1a) is a regular canonical sentence, with the Direct Object to the right of the verb as a constituent of the VP.1 Example (1b) exemplifies dislocation: the Direct Object appears displaced to the left periphery. Sentence (1b) also features another ingredient absent from (1a): a clitic pronoun that reproduces (a subset of) the grammatical features of the dislocated constituent. This clitic pronoun is traditionally referred to as a resumptive pronoun—although recently the more theory-neutral term correlate is also used. The indicates the place where one would normally expect to find the displaced constituent if it were not dislocated.

Page 1 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Dislocations and Information Structure

(1)

Dislocations are present in many languages of the world and we have thorough analyses of at least some of them. A short list includes the following references: Romance (Cinque (p. 403) 1977, 1983/1997, 1990; Rivero 1980; Dobrovie-Sorin 1994; Lambrecht 1994; Rizzi 1997; Frascarelli 2000; Villalba 2000, 2009; Delais-Roussaire et al. 2004; - SamekLodovici 2005, 2006; De Cat 2007a, 2007b), Greek (Anagnostopoulou 1994, 1997), Lebanese Arabic (Aoun and Benmamoun 1998), Germanic (Vat 1981/1997, Zaenen 1997; Van Riemsdjik and Zwarts 1997; Grohmann, 2003; Frey 2005; Ott to appear), Bantu (Bresnan and Mchombo 1987; Baker 2003), Mohawk (Baker 1996), and Czech (Sturgeon 2006). As is usually the case in the study of comparative syntax, we find broad similarities and fine-grained contrasts between languages, most of which are still lacking an analysis. Likewise, we sometimes find that two constructions that look alike in two different languages turn out to have different information structure (henceforth IS) functions. We should start by drawing a typological classification: polysynthetic vs non-polysynthetic languages. As argued by Baker (1996), in polysynthetic languages all the apparent arguments of the predicate are obligatorily adjoined to the clausal periphery—in effect, all arguments are always pro and the constituents that appear to be arguments are actually dislocated. Since dislocation is obligatory, there is no IS function connected to dislocation in a polysynthetic language. In non-polysynthetic languages arguments may be dislocated but they do not have to be, as in the Catalan example in (1a,b). This optionality is what gives rise to the possibility of utilizing dislocations for IS functions. In the following, I focus exclusively on non-polysynthetic languages.2 Other related topics that unfortunately I cannot discuss in these pages are clitic-doubled wh-movement and clitic doubling without dislocation. For classic references on these topics, see DobrovieSorin (1994) and Suñer (1988). In the remainder of this chapter I discuss the grammatical properties of dislocations (Section 20.2) and their information structure (Section 20.3).

20.2 Grammar of dislocations Sections 20.2.1 and 20.2.2 discuss the morphosyntax of dislocations. I show that some dislocations are clausal constituents while others seem to be quite independent (Section

Page 2 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Dislocations and Information Structure 20.2.2) and then I discuss the similarities and differences between right- and leftward dislocations (Section 20.2.3). Section 20.2.4 discusses the phonology of dislocations.

20.2.1 H-type dislocations and D-type dislocations Structurally, dislocations can be classified in two types depending on the sort of connection that the dislocated constituent holds with the core clause. Some dislocated constituents seem to have a very weak connection, of the kind that is only resolved during (p. 404) sentence interpretation without involving a syntactic dependency. They are sometimes referred to as Hanging Topic Left Dislocations or simply as Left Dislocations (although, in fact, they can also be oriented rightward, at least in some languages). Let’s call them H-type dislocations. Other dislocated constituents form a syntactic dependency with a head in the core clause structure. Common terms are Clitic Left/Right Dislocation or Contrastive Left Dislocation. I will refer to this latter group as D-type dislocations (the D/H-type distinction ultimately originates in Ross 1967 and Cinque 1983/1997 and there is an extensive body of work that builds on their insights). In the generative tradition, the D/H-type distinction is analysed as in (2). Example (2a) represents an H-type dislocation structure, which includes the dislocated constituent and a resumptive element. I include an empty pro-form in the position of in parenthesis because it is not certain that we need to assume an empty category in H-type constructions. Example (2b) represents a D-type structure, with a full copy of the dislocated constituent. This full copy is shown in brackets:

(2)

There are two types of empirical motivation for the distinction between D-type and H-type dislocations. (i) reconstruction effects, such as quantifier-variable relations and Binding Theory violations; and (ii) island effects. D-type dislocations show reconstruction and island effects.3 For instance, consider the following Spanish example: (3)

Page 3 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Dislocations and Information Structure In sentence (3), the dislocated constituent al árbitro is doubled by the clitic pronoun lo. The interesting point about this example is that ‘the referee’ cannot be co-referential with the epithet el muy tonto ‘that idiot’. This can be accounted for as a violation of Condition C of Binding Theory if the dislocated constituent has a copy in position , where it is ccommanded by the epithet. The type of evidence shown in (3) leads to the conclusion that D-type dislocations form a syntactic dependency with a position in the main clause—a dependency that can be formalized in terms of copy, among other possibilities. H-type dislocations show no reconstruction or island effects. Consider the following French example: (p. 405)

(4)

As De Cat points out, Léon and the pronoun il can be co-referent. This is not expected if there is a copy of the dislocated constituent in the position because it would give rise to a Condition-C violation. Rather, one has to conclude that we have a structure like that in (2a). H-type and D-type dislocations can also be teased apart by their morphological connectedness with the resumptive. H-type dislocations generally agree in gender, number, and person with the resumptive element, but not in case. In fact, H-type dislocations appear in a default case. D-type dislocates additionally exhibit the case form that corresponds to the grammatical function of . Consider now the following Spanish example: (5)

Page 4 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Dislocations and Information Structure The only visible difference between (3) and (5) is that the dislocated constituent does not bear a mark of accusative case in (5)—although is the direct object of vio and the corresponding resumptive, the clitic lo, appears in accusative case. This absence of case marking tells us that we are facing an H-type dislocation. As an H-type dislocation, el árbitro does not have a copy in . If that is the case, el árbitro and el muy tonto should be able to co-refer without triggering a Condition-C violation. In fact, that is the case. Thus, H-type dislocations seem disconnected from the core clause: they do not receive a θ-role from a predicate or establish any other form of syntactic dependency. This has led Shaer (2009) to propose that they are orphans (see Haegeman 1991/2009 for the first formulation of this concept). Orphans are phrases that do not integrate into a sentence but instead form discourse constituents directly. Alternatively, cartographic approaches locate H-type dislocates in a very high position within the CP-field (see Benincà and Poletto 2004).

(p. 406)

20.2.2 Resumption

Any sort of constituent can be a double for an H-type dislocation: agreement morphemes, clitics, weak pronouns, strong pronouns, even epithets. Since D-type dislocations include a copy within the core clause structure, it is traditionally understood that the list of potential resumptives is short: it has to consist of feature bundles that do not occupy an argument position or we would violate the θ-criterion: agreement morphemes, clitics (or at least, certain types of clitics), or weak pronouns (or at least certain types of weak pronouns). The southern Romance languages (see Bocci and Poletto this volume), Greek (see Skopeteas this volume) and Lebanese Arabic (see Aoun and Benmamoun 1998) are some of the languages in which D-type dislocates have a pronominal clitic as a resumptive. I have discussed some examples or Romance dislocations above. The following is a Greek example:

(6)

Notice that in (6a) the determiner is in nominative case while in (6b) it is in accusative case, matching the accusative case of the resumptive. As Anagnostopoulou argues, this morphological difference correlates with a difference in interpretation: in (6a) the Page 5 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Dislocations and Information Structure variable tu cannot be bound by the quantifier kathenas while in (6b) this binding is possible. This parallels what we see in Romance and leads to the conclusion that (6a) exemplifies H-type dislocation while (6b) exemplifies D-type dislocation. Additionally, these languages also allow for subject dislocation without a clitic. It is generally assumed that subject agreement is the resumptive in these cases. Consider the Spanish example (7). Notice that the lack of agreement between estos chicos ‘these guys’ and parece ‘seems’ indicates that there is no Raising to Subject: (7)

Different languages have different clitic repertoires: for instance, Catalan and Italian have clitics for locative arguments and non-specific indefinite objects while Spanish and Portuguese do not. This gives rise to almost minimal pairs such as the following: (p. 407)

(8)

Example (8a) is a bona fide Clitic Left Dislocation, with a locative clitic. What about (8b)? We could take the overt presence of the clitic to be crucial to define a D-type dislocation, and therefore (8b) would be analysed as a third type of construction, distinct from D-type or H-type dislocation. Alternatively, we could take the overt presence of the clitic as coincidental and regard the syntactic behaviour as fundamental. Since (8b) behaves exactly like (8a) with respect to island sensitivity and reconstruction effects, I take it that the label D-type dislocation can be used fruitfully to embrace dislocations without an overt resumptive such as (8b). Along these lines, post-verbal objects in otherwise SOV languages without clitic pronouns such as Japanese and Turkish could also be regarded as D-type dislocates (see Lambrecht’s 2001 discussion). In the light of this discussion, consider the following English examples:

(9)

Left Dislocation in English exemplified in (9a), is a form of H-type dislocation. Topicalization, exemplified in (9b), is the closest thing that we have in this language to a D-type dislocation (see Ross 1967; Rodman 1997; Birner and Ward 1998; Prince 1998; Page 6 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Dislocations and Information Structure Shaer 2009, among many others). English Topicalization is sensitive to islands and exhibits reconstruction effects (see Rochemont 1989), while Left Dislocation is not and does not. If we agree that the presence or absence of an overt resumptive is not crucial for the definition of a D-type dislocation, then we can place English Topicalization in the same group as Romance, Lebanese, and Greek CLLD. Finally, we can use German as an example of a language in which the resumptive of the D-type dislocated constituent is a weak pronoun (see Vat 1981/1997; Anagnostopoulou 1997; Van Riemsdijk and Zwarts 1997; Zaenen 1997 ; Grohmann 2003; Frey 2005, among many others). These languages have a set of pronouns, referred to as d-pronouns, that can be weak and a set of pronouns that can only be strong. Dislocated constituents doubled by a strong pronoun are H-type (Hanging Topics in many articles) while those doubled by a weak pronoun are D-type (referred to in the German tradition as Contrastive Left Dislocation (CLD) or Left Dislocation). The distinction is revealed by the regular reconstruction and island tests (see for instance Grohmann 2003; Frey 2005). Consider the German examples in (10). Sentence (10a), with the weak pronoun den and the dislocated constituent exhibiting accusative case on the determiner exemplifies CLD. Example (10b), with the strong pronoun ihn and a dislocate in default nominative exemplifies a (p. 408) Hanging Topic. Additionally, the dislocated constituent agrees in case with a weak pronoun but not with a strong pronoun.

(10)

Germanic d-pronouns are very close to Romance and Greek clitics but they are not identical in one crucial respect. Romance and Greek clitics are heads attached to another head. Germanic d-pronouns occupy what seem to be spec positions—for instance, the Vorfeld right before the V2 position, in (10a). If the d-pronoun is indeed an argument of the verb, the D-type dislocated constituent would seem to violate the θ-criterion. This has led to an analysis of German (and Romance) dislocations in which the pronoun and the dislocate start out as co-constituents of a big DP, which would account for the connectivity properties as well as the sharing of the semantic role (see Cecchetto 2000; Grewendorf 2008; see also Ott 2014 for a new perspective on dislocations in German). All the above languages have both H-type and D-type dislocations. However, some languages seem to have only H-type dislocations. This is the case of Tzotzil and Jakaltek, as argued by Aissen (1992) using prosodic structure as empirical evidence. Likewise, De

Page 7 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Dislocations and Information Structure Cat (2007a, 2007b) argues that French has only H-type dislocations. To get a taste of her argumentation, consider the following example: (11)

De Cat reports that the quantifier chaque cannot bind the variable ses. If French dislocations could avail themselves of a D-type strategy, the dislocated constituent would reconstruct to the position and the quantifier would c-command, and hence bind, the variable.4 There are many other languages and language groups that I had to leave out of this discussion of the syntax of dislocations. See Lambrecht (2001) for a detailed typology of dislocation constructions. (p. 409)

20.2.3 Left- and rightward dislocations Dislocations can also be classified according to their surface position: they can be oriented toward the left or toward the right. This yields a four-way typology: Right Htype, Left H-type, Right D-type, Left D-type. The following Spanish example instantiates a D-type rightward dislocation: (12)

D-type right dislocations in Italian, Catalan, and Spanish have received some attention lately. Most researchers have reached the conclusion that syntactically D-type right dislocation is very different from D-type left dislocation (see Cecchetto 1999; Villalba 2000, 2009; Cardinaletti 2001, 2002; Belletti 2004; López 2009a, 2009b; Feldhausen 2010; for dissenting opinions see Frascarelli 2000, 2004; Samek-Lodovici 2006 and for an entirely different perspective with basis in Germanic see Ott and de Vries 2015). Left dislocation is outside the main clause—the Tense Phrase (TP)—c-commanding all the constituents located in canonical positions: (13)

Page 8 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Dislocations and Information Structure Rightward D-type dislocated objects are in a mid-position. They c-command the constituents within the basic predicate structure but not a pre-verbal subject: (14) (Subjects can also be D-type rightward dislocates, but their configuration is hard to pinpoint, as explained in López 2009a). Given the configuration in (14), one should rightfully wonder how XP(D) ends up in the right periphery of the clause. Most analyses have opted for remnant movement of the predicate phrase (Cecchetto 1999; Villalba 2000, 2009; Belletti 2004). Alternatively, López (2009b) argues that the right-peripheral position of D-type right dislocations does not arise in syntax. Rather, he argues it is a specific linearization option made obligatory by prosodic requirements. Rightward H-type dislocations are understudied, but De Cat (2007a, 2007b) provides a discussion and a number of examples in French. The following French example includes two dislocated constituents, moi and sa fille. Sa fille is separated from its clause by an adjunct island. As mentioned, this independence from islands reveals that we are dealing with a Right H-type dislocation: (p. 410)

(15)

I also leave out what we could refer to as ‘medial dislocations’, dislocated constituents that do not show up on the clausal periphery. There is very little work on this area. I refer the reader to Delais-Roussarie et al. (2004) for some discussion.

20.2.4 The prosody of dislocations In recent years we have seen a growth of interest in the prosody of D-type dislocations. Within the Romance family, we have detailed analyses for Spanish (Zubizarreta 1998; Feldhausen 2012), Catalan (Astruc 2005; Feldhausen 2010), Italian (Frascarelli 2000) and French (Delais-Roussarie et al. 2004).

Page 9 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Dislocations and Information Structure For concreteness, let us assume a theory of prosodic structure in which prosodic words are dominated by phonological phrases and the latter by intonation phrases (Selkirk 1984; Nespor and Vogel 1986 are the classic references). These prosodic units are built by aligning their boundaries with syntactic boundaries, although they are not necessarily coextensive (see Truckenbrodt, this volume, for details). Consider the following example: (16)

The lowest relevant prosodic level is the prosodic word (w). It is structured around a lexical item that bears primary stress. Notice that a syntactic word like ‘the’ does not bear stress, and so it is cliticized to a word with stress to form a prosodic word. The next level up is the prosodic phrase (or an intermediate phrase, the difference is not important for our purposes). A prosodic phrase (PP) groups one or more prosodic words into a bigger constituent with a pitch accent. Notice that in example (16) the PP ‘kicked the ball’ includes two syntactic phrases, the VP and the DP. Finally, an intonation phrase (IP) embraces one or several PPs and can be circumscribed by pauses on each edge. The IP coincides with a (simple) clause but a non-restrictive relative clause or a parenthetical can also be intonation phrases. The careful argumentations in Aissen (1992), Frascarelli (2000), Astruc (2005), and Feldhausen (2010) show that in a variety of languages dislocations are separated from (p. 411) the rest of the clause by an IP boundary. The following exemplifies left dislocation in Catalan: (17)

Page 10 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Dislocations and Information Structure The following exemplifies right dislocation in Spanish:

(18)

When a sentence includes several dislocated constituents, each of them constitutes a separate intonational phrase. This is shown in the following Spanish example: (19)

Additionally, Frascarelli (2000) shows that left dislocations in Italian (at least D-type left dislocations) that consist of only one prosodic phrase may adjoin to the following intonation phrase:5 (20)

Page 11 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Dislocations and Information Structure Feldhausen (2010) extends his database to left dislocations in subordinate clauses in Catalan. He shows that left dislocations erect an intonation phrase boundary to their right while they phrase with the matrix clause to the left. The following is an example: (p. 412)

(21)

As for the internal structure, right dislocations are deaccented and their melody tends to mimic the contour of the matrix clause: if the matrix clause is a declarative clause that ends in a low tone, the right dislocation follows the same pattern but if the matrix clause is a question that ends in a high tone, the right dislocated constituent also ends in a high tone (Astruc 2005; Feldhausen 2010). Left dislocations can end in a monotonal rise or low tone, but also in a bitonal low-rise— as revealed especially in Frascarelli and Hinterhölzl’s (2006) corpus study of Italian dislocations. Frascarelli and Hinterhölzl additionally argue that the different tonal structures map onto different IS values. It has independently been argued (see Büring, this volume, Bocci and Poletto, this volume) that the notion Topic is a cover term for several types of entitities: aboutness topic, contrastive topic, and familiarity topic, each of which leads to dislocation. If we assume this state of affairs to hold, it would be desirable to investigate whether the types of edge tones and the types of topics that dislocations can express match one-to-one. Frascarelli and Hinterhölzl’s (2006) develop an analysis of this possibility with a database consisting of Italian dislocations.

20.3 Information structure properties Recall that in polysynthetic languages dislocated arguments are not connected to any particular IS function. That being the case, it follows that we should not expect that a particular pragmatic function p would be uniformly connected to a grammatical construction g across every language that we look at (as pointed out by Cinque 1983/1997). In fact, fine-grained comparative analyses yield subtle differences (see Frey 2005 for such an analysis of leftward constituents in English and German). It may even the case that, as Barbosa (2009) argues, the information structure function of a dislocated Page 12 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Dislocations and Information Structure subject is not the same as that of a dislocated object in the same language. Finally, recall that some languages have both H-type and D-type dislocations while some other languages only have H-type dislocations. It would not be surprising if H-type dislocations in the (p. 413) latter group of languages played a wider set of roles than in the former. Despite these caveats, it seems to me that we do find some cross-linguistic generalizations, features that we find often and features that we do not seem to find at all. In particular, it seems to me that languages that have both H-type and D-type dislocations divide the pragmatic cake similarly. In Sections 20.3.1 and 20.3.2 I discuss the IS roles of H-type and D-type dislocations in languages that have both. For reasons of space, I focus on English and Romance. In Section 20.3.3 I discuss French, a language that only has H-type dislocations.

20.3.1 H-type dislocations There are detailed analyses of the IS properties of left H-type dislocations in Italian (Cinque 1983/1997), Catalan (Villalba 2000, 2009), English (Prince 1998; Gregory and Michaelis 2001; Frey 2005), and German (Frey 2005), among others. A concept that reappears in these analyses is Topic Promotion. The idea is that H-type left dislocations are used to introduce a new referent to the discourse and make it salient enough that a following stretch of discourse—a sentence or several sentences—will be about this referent. (22)

In this example, ‘you and I’ is an H-type dislocated constituent. As such, it sets up the topic for the following sentences, all of which are about ‘you and I’. H-type dislocations do indeed seem to reject anaphoric relations and prefer to engage a new topic:

(23)

Arguably, the infelicity of (23) comes about because ‘that’ is an H-type dislocate (resumed by ‘it’) that is forced to be anaphoric to the antecedent ‘run for a school board position’.

Page 13 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Dislocations and Information Structure We may take the aboutness relation in a strict sense: the topic is relevant for the proposition in the main clause, the proposition in the main clause must expand our knowledge of the H-type dislocate (De Cat 2007a: 70). As De Cat argues, the following example (which originally was brought forward by Tanya Reinhart) suggests that this interpretation of the aboutness relation is the correct one: (24)

This dislocation is infelicitous because my losing a book does not expand what we know about Marilyn Monroe. (p. 414)

H-type right dislocations have not received much scholarly attention. In fact, judging from the literature available to me, only spoken French seems to use them productively (see in particular De Cat 2007a: 217–42). For the other languages, it seems to me that socalled Afterthoughts (Grosz and Ziv 1998; Villalba 2000, 2009: 116–22) qualify as an instance of rightward H-type dislocations. The following is an example (adapted from Villalba 2009: 120):

(25)

In an afterthought, the dislocated constituent is used to clarify the referent of an earlier pronoun. In this case, ‘Chris’ clarifies who ‘he’ refers to. Thus, afterthoughts can also be regarded as topic promotion devices. Unlike H-type left dislocations, H-type right dislocations cannot be entirely new to the discourse. In (26), the dislocated constituent clarifies the referent of the pronoun ‘it’, but since there is no mention of ‘stories’ in the previous discourse, the resulting discourse is infelicitous.

(26)

However, it is uncertain whether afterthoughts are indeed H-type dislocations. On the one hand, notice that the resumptive pronoun in (25) is within a relative clause, which precludes a movement analysis and suggests that regarding them as H-type dislocations is correct. However, they—afterthoughts, I mean—can appear as parentheticals in the middle of a sentence and they obligatorily agree in case with the resumptive (Villalba 2000, 2009). Both of these properties are unexpected of H-type dislocates.

20.3.2 D-type dislocations D-type dislocations, both left and right, are always Given (see Rochemont, this volume, for a discussion of this notion).6 A dislocated constituent refers back to an antecedent in

Page 14 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Dislocations and Information Structure the common ground. We need to discuss what an ‘antecedent’ is and what the expression ‘the common ground’ means. (p. 415)

I start by defining antecedents for dislocations. As pointed out by Villalba (2000, 2009), rightward and leftward dislocations bear a different relation to their antecedents. Leftward dislocations can refer to a subset of a referent in the common ground, or can be a member of a set or even a superset. In the following Spanish example the dislocated constituents are subsets of the antecedent (‘knives’ and ‘forks’ are subsets of the set ‘silverware’). (27)

In the following example, the dislocated constituent is a superset of the antecedent: (28)

Thus, D-linked dislocated constituents evoke a set of alternatives. The property of ‘evoking a set of alternatives’ is used here following the insights of Rooth’s (1985) study of focus as well as Büring’s (1999) analysis of CLD in German and Vallduví and Vilkuna’s (1998) generalization of the notion contrast (see also Büring, this volume and Repp, this volume). Let’s call this property contrast and claim that D-type dislocations are contrastive. Since dislocations are also given, the set of alternatives evoked by the dislocate must be found in the common ground rather than constituting a new referent.

Page 15 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Dislocations and Information Structure It is possible for a D-type left dislocation to be identical to the antecedent: (29)

This does not contradict that idea that D-type left dislocations are contrastive. The choice of a left dislocation (rather than a right dislocation) does open a set or sets of alternatives (i.e. dislocating ‘forks’ evokes sets such as ‘silverware’, ‘objects that should be (p. 416) laid on the table’, ‘objects that should be stored in this cupboard’ … ). Example (27) can thus be continued as in (30): (30)

D-type right dislocations must refer directly to a referent in the common ground. Examples (31) and (32) are Catalan. Example (31) is infelicitous because ‘the knives’, which is right dislocated, cannot take ‘silverware’ as antecedent. Sentence (31) contrasts with (32): (31)

(32)

Page 16 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Dislocations and Information Structure In (33) I provide a mid-way summary of what we have learned about the IS functions of D/ H-type dislocations so far. The reader should keep in mind at least two caveats: (i) what I say here about H-type right dislocations is provisional, since there is little precious literature on the topic, (ii) this discussion only holds of those languages where the H/Dtype distinction holds (for a different type of language, see Section 20.3.3). (33)

Let’s turn now to the notion of ‘common ground’ (see Krifka 2008 for a useful definition as well as Rochemont, this volume). Common ground could include three (p. 417) different types of objects, which I list here from (i) to (iii), from more to less circumscribed: (i) entities and events explicitly mentioned in the previous discourse or deictically obvious; (ii) entities and events made salient by the current communicative exchange even if not explicitly mentioned or deictically obvious; (iii) the knowledge base shared by the speakers in the communicative exchange (see Prince 1981 for a similar partition). We find that (iii) is not sufficient to license dislocation, but (ii) is. The Catalan exchange in (34) exemplifies (iii): (34)

Dislocation of a Barcelona is infelicitous, although it is to be expected that two Catalan speakers are very much familiar with Barcelona—it is part of their shared knowledge base. Only a regular presentational sentence will work because no mention of the city—or some other referent that includes or is included by Barcelona, such as ‘big cities’, ‘Gràcia

Page 17 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Dislocations and Information Structure neighborhood’ etc.—has been made or is somehow salient. Crucially, having mentioned other Catalan towns does not suffice. The reader can compare (34) with (35), (36), and (37), which work perfectly. Examples (35) and (36) exemplify (i). In (35), ‘Barcelona’ has been mentioned in the previous discourse, and therefore dislocating it is no problem: (35)

In (36), Speaker B mentions ‘a big city’, a set of entities that includes a big town like Barcelona. It is another example of (i). Consequently, dislocation is again possible: (p. 418)

(36)

Page 18 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Dislocations and Information Structure Let’s now explore an example of (ii). In the following Spanish example, Speaker A receives two alternative responses from Speaker B: (37)

Take Speaker B(1) first. Having dislocated a Juan in that context, a Juan is a member of the Gómez family, and Speaker B(1)’s dislocation is an example of (i). But now consider Speaker B(2). The response ‘me neither’ precludes Juan from being a member of the Gómez family. But we can still dislocate a Juan. Its dislocated status signals that it is linked to an antecedent that stands in the common ground of Speaker A and Speaker B(2). Given the sentence uttered by Speaker A, we can take the antecedent of the dislocate to consist of the set of people s that are likely to be encountered by speaker A and speaker B as they stroll around their village. This set s has become salient because of Speaker A’s utterance, although it is not explicitly mentioned or pointed at. The Gómez family is a subset of s and Juan is a member of s. Thus, the dislocation in Speaker B(2)’s exemplifies (ii).

Page 19 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Dislocations and Information Structure Let us finish this section with what D-type dislocations are not. The information structure label that is most commonly applied to dislocations is that of aboutness topic (see Alexiadou 2006, among many others). However, it is clear that D-type dislocations cannot be defined as aboutness topics cross-linguistically. This can be seen in the following Catalan examples:

(p. 419)

(38)

(39)

In (38), the dislocated constituent is a predicate adjective, which is given because the common ground includes the referent ‘Chris’ talents’. It can hardly be said that the sentence is about a predicate such as ‘intelligent’—rather, the sentence is about the referent of the null subject, the individual named ‘Chris’. Example (39) is an existential sentence. The dislocation of cadires is licensed because the common ground includes reference to ‘furniture’. But existential sentences are supposed to be the model examples of thetic propositions, which are defined by the property of not having an aboutness topic (or, alternatively, their aboutness topic is some space/time coordinates, see Erteschik-Shir 2006; Krifka 2008). Examples similar to (38) and (39) can be constructed in other languages, although not many of them have an indefinite non-specific clitic (like Catalan en) or a clitic for predicate adjective phrases (like Catalan ho). In those languages, a D-type dislocation without a clitic can be constructed. The following is in Greek:

(40)

The reader who does not approve of dumping may want to argue that the absence of a clitic makes (40) substantively different from the Catalan examples.

Page 20 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Dislocations and Information Structure Note that I do not argue that we cannot construct D-type dislocation examples in which the dislocated constituent can be claimed to be an aboutness topic. You plainly can. The question is whether this aboutness feature defines D-type dislocations as a distinct class of constructions and it seems the answer is no.

20.3.3 French: H-type dislocations take over Colloquial French uses H-type dislocations very frequently, while D-type dislocations are extremely rare (see footnote 4). The literature on French dislocations agrees that H-type dislocations are aboutness topics (Lambrecht 1981; Barnes 1985; De Cat 2007a). De (p. 420) Cat (2007a) goes one step further and argues that if a constituent is a topic, it has to be dislocated. Consider the following contrast:

(41)

(42)

Thus, with certain types of predicates, dislocation in colloquial French is obligatory, as shown in (41b). Not by chance, the predicates that require dislocation are individual level (IL) predicates like difficult, while stage-level (SL) predicates like be there do not require dislocation (the distinction between stage-level and individual-level predicates was first established in Carlson 1977). It has been argued that IL predicates form categorical propositions, with a subject (or topic) and a predicate (or comment) (see Kuroda 1972; Ladusaw 1994). SL predicates, on the other hand, form categorical or thetic judgements. De Cat’s proposal is that the reason why (41b) is infelicitous is because an IL requires a topic and a topic, in French, must be dislocated.

Page 21 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Dislocations and Information Structure A survey of the literature leads me to the conclusion that French dislocates play IS roles that in other languages are split between D-type and H-type dislocations. French dislocates routinely introduce a new topic: (43)

Nancy, the dislocated constituent, has not been mentioned before and cannot be related to an antecedent. Its pragmatic function seems to be Topic Promotion, just as the regular H-type dislocates of English and Southern Romance that were discussed in Section 20.3.1. However, French dislocations can also be given and contrastive. In the following, the kitchen is an element of the set of rooms of the house, thus playing the role that in other languages would be played by a D-type dislocation: (p. 421)

(44)

Recall that, in languages where the D/H-type distinction is robust, examples of Given Htype dislocations sound infelicitous (compare (23) and (44)).

Page 22 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Dislocations and Information Structure The following example confirms that, unlike typical H-type dislocations, French dislocations can be given. Ça ‘this’ refers to something in the magazine the speaker is looking at: (45)

Thus, I conclude that French dislocations are grammatically of the H-type (as argued by De Cat 2007a, 2007b). French H-type dislocations exhibit a range of IS functions—Topic Promotion, Given, Contrastive—that in other languages are distributed among D-type and H-type dislocations.

20.4 Conclusions Dislocations seem to be pervasive in the languages of the world. Following numerous precedents, I have shown that it is descriptively productive to tease apart H-type and Dtype dislocations as well as right and left dislocations. The differences between the different types of dislocations concern both their syntactic and their IS properties. I have also presented a brief description of their prosodic properties.

Notes: (1) I would like to thank Ingo Feldhausen, Xavier Villalba, and two anonymous reviewers for Oxford University Press for detailed comments on an earlier version of this chapter. I would also like to thank the volume editors for useful advice in terms of content and organization. Finally I would like to thank the language consultants, whether or not the data they provided me made it to the final cut: Anastasia Giannakidou, Kay GonzálezVilbazo, Dalina Kalluli, Inga Kolhof, Evalina Leivada, Lena Papadopoulous, Natalia Pavlou, Athina Sioupi, Volker Struckmeier, Greg Ward. None of them should be held responsible for any inaccuracies of the data presented here. (2) There exist mixed systems. Baker (2003) argues that in Kinande all subjects are obligatorily dislocated (and therefore dislocated subjects bear no particular IS function) while objects are optionally dislocated.

Page 23 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Dislocations and Information Structure (3) For reasons of space, I only discuss reconstruction effects in this chapter. For a discussion of islands in dislocations, see López (2009a: 212–39). (4) I follow De Cat (2007a, 2007b) in my discussion, who reports on a survey carried out on native speakers of various nationalities. Delais-Roussaire et al. (2004) argue that French has CLLD, although they find no interpretive or phonetic difference between HTLD and CLLD. The clearest cases of CLLD in French involve dislocated PPs in a subordinate clause. In corpora such examples are very rare, which suggests a diachronic process in which CLLD is yielding all territory to HTLD. (5) Frascarelli (2000) uses a reading task to elucidate the prosodic structure of dislocations in Italian. The adjunction effect is obtained when subjects are requested to read at an accelerated, but still natural, speed. (6) An anonymous reviewer argues that German CLD is not necessarily given. She provides the following discourse, in which the CLDed constituent and the resumptive together carry contrastive focus:

((i))

This sort of example gives us more reason to think that Germanic CLD and Romance/ Greek CLLD are more different than conventionally thought.

Luis López

Luis López is Professor of Spanish Linguistics at the School of Literatures, Cultural Studies and Linguistics and Co-director of the Bilingualism Research Laboratory at the University of Illinois at Chicago. His main field of interest is syntactic theory and the interfaces of syntax with information structure, semantics and prosodic phonology.

Page 24 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 16 January 2019

Second Occurrence Focus

Oxford Handbooks Online Second Occurrence Focus   Stefan Baumann The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara Print Publication Date: Jul 2016 Subject: Linguistics, Applied Linguistics Online Publication Date: Nov 2014 DOI: 10.1093/oxfordhb/9780199642670.013.38

Abstract and Keywords A ‘Second Occurrence Focus’ (SOF) is the semantic focus of a focus sensitive operator (like only) which is contextually given. SOF has been claimed to be phonologically unmarked, which poses a problem for association with focus theories assuming a direct relation between focus and pitch accent. This chapter discusses the main semanticpragmatic accounts of the SOF challenge but also empirical investigations which found that SOF actually is marked by secondary (i.e. non-nuclear) prosodic prominence, providing evidence in favour of association with focus theories. A similar prosodic pattern could be found in semantically and prosodically comparable structures such as cases of implicational bridging. Finally, an outlook on a possible unified approach of the phonological representation of second occurrence expressions is presented which is based on metrical stress. Keywords: Second Occurrence Focus, pitch accent, secondary prominence, association with focus, focus sensitive operator, givenness, prosody, metrical stress, bridging

24.1 Introduction A whole range of slightly different but related structures have been discussed under the notion of ‘Second Occurrence Focus’ (SOF). Taking the traditional view as a point of departure, we may conceive SOF as a specific type of focus, which is indicated morphosyntactically by a focus-sensitive operator (such as only or even), and which is at the same time contextually given—in contrast to ‘First Occurrence Focus’ (FOF), which is contextually new.1 An often cited example by Partee (1999: 215) is shown in (1), where

Page 1 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Second Occurrence Focus vegetables occurs as an FOF element in (a) and as an overtly repeated SOF element in (b).2 (1)

This combination of ‘focusedness’ on the one hand and ‘givenness’ on the other causes a problem in terms of the expected prosodic marking of SOF elements, which can be described as follows: According to (direct) ‘association with focus’ theories (e.g. Jackendoff 1972; Rooth 1985; von Stechow 1990; Krifka 1992), a focus-sensitive operator like only has to be associated with a focus in its syntactic constituent which is indicated by prosodic prominence. Note that ‘prosodic prominence’ has often been equated with the concept of ‘pitch accent’, in particular ‘nuclear pitch accent’, which is defined as the last and structurally strongest prominence in an intonation unit (cf. also Zubizarreta, this volume). We follow the view proposed in autosegmental-metrical approaches that the relevant prosodic domain for a nuclear accent is the intermediate phrase (ip; cf. Beckman and Pierrehumbert 1986; Ladd 2008).3 In the present chapter, we will use prominence as a cover term for both (postlexical, i.e. actually realized) stress (also referred to as metrical stress later in the text), indicated by non-tonal phonetic features such as duration and intensity, and (pitch) accent, which additionally involves pitch movement. It has been claimed (e.g. by Partee 1991, 1999; Krifka 2004) that SOF—in contrast to FOF—is not marked by prosodic prominence (note the apparent deaccentuation of vegetables in (1b) above). The existence of such an ‘inaudible focus’ would violate association with focus theories, since it would imply either that focus (at least SOF) is not necessarily marked by prosodic prominence OR that operators like only do not always associate with focus. (p. 484)

Many papers have discussed this problem from a semantics and pragmatics point of view, but more recently there has also been a growing interest in phonetic and phonological aspects of SOF, including the evidence found in empirical investigations. This evidence suggests that SOF is in fact prosodically marked but only by a secondary prominence without pitch movement. Since this specific prosodic marking is often regarded as the distinctive feature of SOF and other ‘second occurrence expressions’ (Krifka 2004), it is not an easy task to establish a conclusive definition of the phenomenon. The traditional definition given above is narrow in that it demands the occurrence of a focus particle and the literal repetition of a focused expression. In the following, we may also refer to this restrictive view as ‘proper SOF’ (Krifka 2004). However, in most of the more recent literature, a broader view has been advocated, suggesting that these conditions are neither sufficient nor necessary for a comprehensive account of SOF—including related second occurrence structures. Although it is agreed that SOF can not be distinguished from an ordinary focus by givenness alone (since an ordinary focus can be given as well; see next section), Page 2 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Second Occurrence Focus several studies insist that givenness in fact is a necessary precondition for SOF (which is the view of the so-called splitters, separating the effects of focus and givenness) while others consider only a division in focus and background necessary (view of lumpers, not separating the effects of focus and givenness). A final aspect of an ex negativo definition of SOF is that it is not just a second focus in the linear order of a clause or sentence (as may be suggested by (1))—rather, the focused phrase has to be derivable from the previous context (which implies that an SOF can also occur sentence-initially). The rest of the chapter is structured as follows: After a presentation of the main (semantic-pragmatic) accounts of SOF and their general models of focus and givenness, as well as empirical evidence for a prosodic manifestation of focus in SOF structures (Section 24.2), semantically and prosodically comparable configurations will be (p. 485) discussed (Section 24.3) leading to an outlook on a possible unified approach of the phonological representation of second occurence expressions (Section 24.4).

24.2 Solutions to the SOF problem As a first step towards a solution to the SOF challenge it has to be noted that the lack of prominence on SOF items cannot simply be explained by their givenness in the discourse, since there are many instances in West Germanic languages in which contextually given constituents do receive focus prominence, as in examples (2)–(4), all adapted from Büring (2013: 2).4

(2)

(3)

In (2b), John is focused or F-marked (cf. Rooth, this volume) since it is the answer to the preceding question. The referential expression is necessarily marked by a (nuclear) pitch accent, despite its contextual givenness (John has been mentioned in (2a)). Similarly, in cases of overt contrast or correction as in (3b), focus prosody (here: nuclear accent due to the correction to John) overrides the prosody which would be expected on the basis of the information status of the discourse referents alone (here: deaccentuation due to the givenness of John). The example in (4) contains foci which are associated with focus-sensitive operators. It reveals that associated foci can be given as well and, notably, this does even apply to ‘First Occurrence Foci’, making the term somewhat problematic here.5

Page 3 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Second Occurrence Focus

(4)

Actually, the referent in question, John, does not surface in the same syntactic position in (4a) and (4b), a fact that prevents deaccentuation according to findings by Terken and Hirschberg (1994). The authors could show that persistence of the same syntactic surface position facilitates deaccentuation, in particular if accompanied by a persistence (p. 486) of grammatical function. In fact, this is the case with only drank juice in (4), which is a ‘proper SOF expression’ defined by Krifka (2004) as a verbatim repetition of a phrase. Let us describe the SOF challenge more closely from a semantic point of view. The traditional assumption in direct association with focus models is that focus-sensitive operators determine a domain which contains a semantic focus. This semantic focus is obligatorily marked by a syntactic F-feature which in turn contains at least one element which is marked by a phonological focus, usually realized as a pitch accent (as in [eats VEgetables] in (1a)). The phenomenon accounting for the observation that not every element in a syntactic phrase corresponding to the semantic focus is prosodically prominent but generally only one element (the focus exponent) whose prominence ‘stands for’ the whole phrase, is known as focus projection (cf. Gussenhoven 1983; Selkirk 1995; Truckenbrodt 1995; Arregi, this volume). Along these lines, the challenge with SOF structures as introduced by Partee (1999) is that we have a semantic focus (vegetables in (1b)) which cannot be F-marked since it is not made prominent by a pitch accent. In other words, SOF structures such as those in (1b) demand a dissociation between semantic focus and phonological focus which is a contradiction in terms for traditional accounts of focus sensitivity (for a more detailed line of argumentation see Beaver et al. 2007). Two general solutions to the SOF challenge (given the restricted view on SOF we have used so far) have been put forward: Pragmatic theories of focus (Section 24.2.1) and modified association with focus accounts (Section 24.2.2).

24.2.1 Pragmatic accounts of focus The first solution is provided by pragmatic theories of focus starting with Rooth (1992) who allow that a semantic focus is not necessarily F-marked and thus not necessarily marked by pitch prominence. Approaches like these (e.g. von Fintel 2002; Roberts 1996; Schwarzschild 1999; Krifka 2004), called ‘strong theories of focus sensitivity’ by Rooth (1992), propose that a discourse entity becomes associated with a focus-sensitive operator (or, rather, with its free variable) simply for contextual reasons. That is, as opposed to direct association with focus theories, also termed ‘weak theories of focus sensitivity’ by Rooth, the interaction between focus and meaning is not mediated by rules of grammar (in particular syntax) but has to be explained by general principles of pragmatics (cf. Velleman and Beaver, this volume), as for example in the anaphora account of focus by von Fintel (2002), also argued for by Krifka (2004).

Page 4 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Second Occurrence Focus As an effect, there is no obligatory one-to-one relation between a focus—interpreted as a semantic focus of a quantificational or focus-sensitive operator—and its marking by a pitch accent any more. Instead, this relation is less direct and ‘reduced’ to an optional rule based on plausibility in the given context. In other words, some operators are claimed to be context-sensitive rather than focus-sensitive, in the same way as quantificational adverbials such as always in (5), whose domain can be restricted by context alone. (p. 487)

(5)

Generally, an SOF phrase may thus be fully deaccented due to contextual givenness, that is, it does not require a phonological reflex of an F feature. An example with the operator only is (6b) (derived from Partee 1991) with the repeated phrase the poor students. (6)

However, even proponents of pragmatic accounts concede that the realization of certain pronouns may challenge this approach. Von Fintel, for example, suggests that (6c) is only acceptable with a secondary prominence (indicated by small capitals) on the (noncontrastively used) pronoun those. One possible explanation for the alleged difference between (6b) and (6c) is based on the assumption that there are strong and weak varieties of pronouns, which are distinguished by a phonological feature [±stress]. The strong variety is used if the pronoun is in focus and thus surfaces with secondary prominence, that is [+stress] (see Krifka 2004: 205). Another explanation is that the operator only is directly sensitive to prosodic prominence after all, which would again be in line with an interpretation of only as being focus-sensitive (an assumption the association with focus accounts are based on; see the next section). There is evidence, though, that not all focus-sensitive operators have to be interpreted in the same way and that for some of them, for example always, a pragmatic interpretation may be more appropriate (see Beaver and Clark 2008: 154ff.).

24.2.2 Modified association with focus accounts The second solution is embodied in a number of models we may call ‘modified association with focus’ accounts. What they have in common is that they keep up the claim that SOF is in fact focused (or F-marked) and realized with prosodic prominence, but that this prominence is only secondary (i.e. non-nuclear). Among the advocates of these accounts there is some debate as to whether it is possible to find a uniform explanation for focusing and its prosodic realization or whether it is necessary to distinguish between Page 5 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Second Occurrence Focus two sources—namely contrast to alternatives and newness/givenness (with varying terminology)—that determine the assignment of prosodic prominence. Beaver and Velleman (2011) call the first group lumpers, advertising a one-factor account, and the second one splitters, favouring a two-factor account.

24.2.2.1 Lumpers The main proponents of the lumpers camp are Büring (2013) and Rooth (2010). They base their accounts of SOF to some extent on Schwarzschild (1999) who integrates the effects of focus into his definition of givenness (cf. Rochemont, this volume). Both (p. 488) Büring and Rooth claim that the difference between FOF and SOF can be accounted for in terms of domain size: While the domain of an FOF is maximal, that is, a whole sentence, and is thus marked by a nuclear accent, the domain of an SOF is smaller and can at most receive a secondary prominence (to be defined in greater detail below). Büring (2013) formulates this basic principle under the heading of the Domain Theory of Primacy, in combination with the FOCUSPROMINENCE principle (adapted from Truckenbrodt 1995): (7)

(8)

Büring claims that all foci have to be ‘interpreted’ by an operator. An operator may either be focus-sensitive (like the particles only or even) leading to an ‘associated’ focus, or not (like CONTEXTCONNECT (CC), being attached to the whole sentence) leaving the focus ‘free’. Primary foci (F1) are always free foci, occurring, for example, in question–answer pairs or in overtly contrastive focus structures. Thus, in a combination of a free focus and an associated focus as in (9), the first occurrence of faculty is primary and carries the nuclear accent, whereas the associated focus is embedded and thus secondary—both in terms of semantic interpretation and phonological marking, two levels which are closely related in this account. (9)

6

The combination of two associated foci, as in example (4) above, repeated here as (10), is a less clear case. Here, Büring claims that one of the associated foci additionally serves as a free, sentential focus which thus receives the nuclear accent.

Page 6 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Second Occurrence Focus (10)

The argument for choosing even John as the free focus is not its novelty in the discourse (since both candidates for the primary accent, John and juice, are contextually given) but the fact that even John fills the structural gap in the open proposition ‘x only drank juice at John’s party’, which can be interpreted as presupposed due to the indefinite expression (p. 489) many people. In other words, even John can be regarded as the answer to the contextually given question ‘Who only drank juice at John’s party?’ (cf. Roberts’ (1996) Question Under Discussion, or QUD) and is thus a legitimate free focus. Following the Domain Theory of Primacy, in combination with the FOCUSPROMINENCE principle, the free focus domain with John as focus exponent is maximal and contains the associated focus domain with juice as a secondarily prominent focus exponent. That is, juice only receives a postnuclear prominence. The structure in (10′) illustrates the relation of embeddedness more clearly than (10). (10′) Rooth (2010) summarizes this configuration of embeddedness of SOF in the scope of a primary focus and its phonological marking as follows:

(11)

7

In the schema, SOF phonology is supposed to be indicated by secondary, that is postnuclear, prominences. These are not equivalent to pitch accents but to metrical stresses which lack tonal movement. The main syllables of SOF items are made prominent in particular by increased duration and intensity in comparison with fully unaccented syllables. In contrast, ‘ordinary F phonology’ (which also holds for FOF as introduced above) is marked by fully-fledged (generally nuclear) pitch accents. We will come back to the phonetic realization of these prosodic prominences in detail in Section 24.2.3 below and in subsequent sections.

24.2.2.2 Splitters Being widely in accordance with the phonological analysis of SOF structures proposed by Büring and Rooth, advocates of the splitters camp (in particular Selkirk 2008 and Beaver and Velleman 2011) suggest a somewhat diverging semantic-syntactic-pragmatic model. As mentioned above, splitters argue for an independent role of newness/givenness from (contrastive) focus in the description of an element’s prosodic marking.

Page 7 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Second Occurrence Focus Selkirk (2008) explains the phonology of SOF items by a combination of F(ocus)-marking and G(ivenness)-marking, the latter inhibiting the assignment of a fully-fledged pitch accent. She proposes that F-marking is always contrastive, following Rooth’s (1992) interpretation of focus in his theory of alternative semantics. At the heart of Selkirk’s account is the FOCUSPROMINENCE rule introduced in (8) (which Selkirk calls Contrastive Focus Prominence Rule) and the constraint DESTRESS GIVEN adopted from Féry and Samek-Lodovici (2006) which guarantees that a given constituent cannot be (p. 490) marked by a primary (i.e. nuclear) accent. Selkirk (2008: 342) proposes four different possibilities, shown in (12) (with the predicted phonological markings added). (12)

Let us have a look at our SOF example (10). At first sight, it seems as if no primary accent could be assigned, since all relevant elements are discourse-given. However, Selkirk (2008: 341) adds a G-marking condition (13) by which she differentiates between two levels of potential givenness, the ordinary semantic value and the focus semantic value (following Rooth 1992). (13)

An element is given with respect to its ordinary semantic value simply if it has been mentioned before in the discourse. If an element is F-marked, however, it only counts as given if it has an antecedent for its focus semantic value. In the case of (10), this focus semantic value for juice is the alternatives set {drank beer, drank wine, drank water,…. }, which has been introduced in the context by the phrase only drank juice. Thus, juice is both F-marked (due to its association with the focus-sensitive operator only) and Gmarked. In contrast, John is only F-marked since an alternatives set such as {even Bill, even Mary,…. } has not been introduced before. In Selkirk’s terms, John only has an ordinary semantic value which becomes irrelevant if the constituent is F-marked at the same time. A Selkirk-like analysis of (10) is given in (14).

Page 8 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Second Occurrence Focus

(14)

A similar but more flexible approach has been proposed by Beaver and Velleman (2011) who replace Selkirk’s concept of focus by (pragmatic) importance and newness/givenness by (un-)predictability. Importance is marked by the F-feature as in Rooth’s and Selkirk’s approaches, and unpredictability is indicated by an N-feature (for ‘New’), which is derived from Schwarzschild’s (1999) interpretation of F-marking. The authors (p. 491) show that predictability is different from givenness, and that it is (un-)predictability which is relevant for an element’s prosodic realisation.

(15)

In (15b), John is contextually given but is unpredictable as an answer to the question. Thus, it is indexed with N. At the same time, it is pragmatically important and thus Fmarked. As a consequence, it receives a pitch accent, despite its givenness at word level. Note that Beaver and Velleman (2011: 1674) generally interpret association with focus in terms of answering a QUD (see above and Beaver and Clark 2008). That is, an expression that associates with a focus-sensitive operator (as in FOF and SOF structures) provides an answer to the QUD just as John answers the overt question in (15). Both types of answer count as ‘pragmatically important’. An important addition to this model is the constraint that contextually given referents are only unpredictable if they occur in a ‘new role’ as does John in (15b). In a parallel structure like the one in (16b), John would be predictable and thus not N-marked (see also Terken and Hirschberg (1994) and the effect of persistence of the same syntactic position discussed above).

(16)

Beaver and Velleman claim that both (un-)predictability and importance have an influence on the communicative significance of an expression, and relate this pragmatic significance to prosodic prominence in the following principle (Beaver and Velleman 2011: 1673): (17)

Page 9 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Second Occurrence Focus Together with the principle of competition for prominence (given in a slightly changed version in (18)), Beaver and Velleman suggest a pragmatic alternative to the FOCUSPROMINENCE rule in (8). (18)

(p. 492)

In order to determine the most prominent element in an ip, the degree of

communicative significance can be weighted in a cumulative manner. Unpredictability (Nmarking) and importance (F-marking) can be thought of as contributing one point each to an element’s prominence. This results in the following four possibilities: (19)

An element with two points will usually be realized as a (primary) nuclear pitch accent. If more than one element in an ip gets two points, the rightmost is primary, the others secondary (i.e. prenuclear). The prominence assigned to an element with one point can only be non-nuclear/secondary, that is, either prenuclear (generally realized as a pitch accent) or postnuclear (non-tonal prominence).8 If only one-point prominences are available in an ip, the rightmost one will be ‘upgraded’ and receive the nuclear accent.

(20)

Thus, the SOF example in (10) would be analysed by Beaver and Velleman as in (20), with John receiving the nuclear accent (‘important’ plus ‘unpredictable’ due to its new structural role) and thus winning the competition for prominence. Juice only bears a postnuclear prominence, since it is ‘important’ due to its association with the focus-

Page 10 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Second Occurrence Focus sensitive particle only but at the same time predictable from the context. Juice is predictable because it occurs in a parallel structure to the one in the context sentence, here even in a ‘proper SOF’ structure defined above as a string identical repetition.

24.2.3 Empirical evidence As mentioned several times above, the basic claim of all kinds of association with focus theories is that focus (or importance) is marked by prosodic prominence, and this claim has been challenged by SOF structures. In recent years a variety of production and perception studies on SOF—at least in West Germanic languages—were conducted to examine whether the theoretical claim of association with focus models can (p. 493) be empirically verified (e.g. Rooth 1996; Bartels 2004; Jaeger 2004; Beaver et al. 2007; Howell 2008 for American English; Féry and Ishihara 2009; Baumann et al. 2010 for German; van de Ven and Gussenhoven 2011 for Dutch). The studies clearly provide evidence for the fact that there actually is a phonological reflex of focus in SOF structures. However, SOF usually is not marked by fully-fledged (nuclear) pitch accents (i.e. by tonal movement), but by increased duration and intensity (compared to non-focus elements) reflecting secondary prominence. In other words, it could be shown that there is a correlation between focus and prosodic prominence but not between focus and (nuclear) pitch accent. The first, rudimentary experimental study was conducted by Rooth (1996) who acoustically analysed utterances spoken by himself, which contained SOF and Non-Focus elements. He found that SOF expressions were louder and longer than their Non-Focus counterparts, thus providing first empirical evidence in favour of weak (syntactic/ semantic) theories of focus sensitivity. In a more systematic production study in which pairs of American English speakers read sentences in a dialogue, Bartels (2004) investigated the acoustic differences between FOF and SOF but did not compare them with the prosodic characteristics of Non-Focus constituents. In fact, Bartels could confirm the assumption that SOF is not marked by pitch prominence, but she also found that SOF expressions are realized with shorter duration and intensity than FOF expressions. Furthermore, Bartels compared FOF and SOF with echo renditions and showed that a mere repetition does not always trigger reduced prominence. That is, her data suggest that echo NPs generally carry pitch accents (like FOF) but that they are often acoustically less prominent than first occurrence renditions. In comparison with SOF, echo NPs were found to display greater acoustic prominence. A larger-scale production and perception study by Beaver et al. (2007) concentrated on the prosodic difference between SOF and Non-Focus constituents. Twenty American English speakers read aloud short paragraphs like the ones in (21) and (22), with the (c) sentences being of particular interest. These sentences were composed of identical text

Page 11 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Second Occurrence Focus but differed in the position of the SOF and Non-Focus elements (here: Sid and court, respectively). (21)

(22)

The production data confirmed the tentative findings by Rooth (1996) in that a statistically significant increase in duration and relative energy (derived by multiplying duration by root-mean-square intensity)9 could be found for SOF in comparison with Non-Focus. Differences in fundamental frequency were only marginal. This result adds to the evidence in favour of association with focus theories, since it shows that SOF (p. 494)

elements are marked by a (slight) increase in prosodic prominence. The crucial next step was to find out whether the rather subtle phonological differences between SOF and Non-Focus are actually perceivable. Beaver et al. selected a set of minimal sentence pairs like (21c) and (22c) from the production experiment which were taken from the same speaker. These pairs—without the context they were produced in— were presented to subjects who had the task of deciding in which sentence a particular target word (e.g. Sid) was perceived as more prominent than another word (e.g. court). Subjects identified the SOF items as the more prominent ones in 63 per cent of the cases which led the authors to conclude that the prosodic marking of SOF is indeed perceivable. The empirical production data provided by Féry and Ishihara (2009) for SOF in German largely confirm the results for American English, in particular as far as the sentencemedial or ‘post-FOF’ position is concerned (i.e. no tonal movement but increased duration compared with non-focused items). Interestingly, Féry and Ishihara added conditions in which FOF, SOF, and Non-Focus elements occurred in sentence-initial position, as shown in (23) and (24).10

Page 12 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Second Occurrence Focus (23)

(24)

Crucially, the authors found that SOF items in sentence-initial position were marked by a fully-fledged prenuclear pitch accent—showing a categorical difference to the same items in sentence-medial position. Still, sentence-initial SOF (23c) was marked as less prominent than FOF (which received a nuclear accent; (23b)), but more prominent than Non-Focus (24b).11 It is claimed that this three-way difference can be explained by boosting and inhibiting factors of information structure: Focus leads to higher pitch, givenness to lower pitch. Since SOF is a combination of focus and givenness, its mean pitch value lies between FOF (higher) and Non-Focus (lower). Féry and Ishihara found similar effects for duration, with focused elements (i.e. FOF and SOF) being longer than Non-Focus words. However, there was no duration difference between FOF and SOF in sentence-initial position, although FOF was expected to be longer (focused, non-given, nuclear accent) than SOF (focused, given, prenuclear accent).12 Féry and Ishihara (2009: 307) suggest that this outcome may in fact be due to a phrase boundary after the SOF constituent (nur Peter in (23c)) triggering a lengthening effect which masks the givenness effect (i.e. shortening) of SOF. As the relevant domain for pre-boundary lengthening the (p. 495)

Page 13 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Second Occurrence Focus authors postulate the ‘prosodic phrase’ (p-phrase) which depends on syntax and may be embedded in an intermediate phrase. Thus, the (final) accent of a p-phrase may still be prenuclear (Féry and Ishihara 2009: 296). A similar observation has been made by Jaeger (2004) who found for American English that SOF elements may receive a pitch accent if they form an intermediate phrase of their own. In such a case, which is actually rather likely in slow speech, the pitch accent is in fact nuclear. If a nuclear accent on an SOF element is possible, this serves as even stronger evidence in favour of (direct) association with focus theories. The acoustic findings for SOF prosody in the reported studies could also be supported by articulatory data from German. By using an electromagnetic articulograph, Baumann et al. (2010) investigated not only the tonal and durational marking of sentence-medial constituents in three different focus conditions (Background (= Non-Focus), SOF, FOF) but also the modifications of the opening and closing gestures of the lips during the production of the target words. Results confirm that SOF is generally not marked by a nuclear pitch accent and show gradient but systematic adjustments of acoustic and articulatory parameters (word durations and temporal and spatial modifications of lip opening and closing gestures) leading to an increase in prominence from Non-Focus through SOF to FOF. Table 24.1 provides an overview of the strategies used, which are to some extent speaker-specific. The investigation of lip movements, that is, the degree of mouth opening, relates prominence to sonority, that is, the relative loudness of sounds, and shows its relevance as a distinguishing factor between focus types (which is in line with Beaver et al.’s findings for relative energy). Figure 24.1 provides averaged trajectories for the distance between the upper and lower lip during the production of the target word B/a:/ber. Low displacements indicate that the lips are closed for the production of the stop consonants. Going from Background (Non-Focus) to FOF, there is an increase in duration and lip aperture (displacement) corresponding to prosodic prominence. (p. 496)

Table 24.1 Summary of tonal, durational, and articulatory adjustments as prominence increases in the marking of target words in three different focus conditions in German

Page 14 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Second Occurrence Focus Source: Adapted from Baumann et al. (2010: 75). Summing up, the empirical studies—which did not try to predict a prosodic prominence but looked for experimental evidence— observed a secondary status of SOF elements in terms of their phonological marking. That is, they are usually not marked by (p. 497) pitch accents, and if they are, the accent is either prenuclear or is the only one in an intermediate phrase. In Figure 24.1 Averaged trajectories for lip opening postnuclear position, SOF and closing movements during the target word B/a:/ is marked by non-tonal ber for one speaker (DM) in three focus conditions (BG = Non-Focus, SOF, and FOF). prosodic parameters such as increased duration, intensity, and sonority. It has also been suggested that ‘postnuclear’ SOF prominence should be described in terms of L* pitch accents (which would by definition become nuclear; cf. Beaver et al. 2007: 28) or (low) phrase accents as defined by Grice et al. (2000) as edge tones with a secondary association to stressed syllables (cf. Baumann et al. 2010: 76). Although defined tonally, however, both concepts are mainly characterized by duration, at least in West Germanic languages.

24.3 Other second occurrence expressions There has been some discussion in the literature on the question of whether the characteristic prosodic pattern of proper SOF widely discussed in this chapter is also present in other, less restrictive structures. Thus, the question is whether only verbatim repetitions of first occurrence expressions are marked by secondary prominences, or also ‘quasi second occurrence expressions’ (Krifka 2004) such as cases of bridging (Clark 1977) or ‘implicit inferrability’ (Partee 1999: 216), which do not display segmental copies of first occurrence expressions. An example from Rooth (1996: 216) showing a case of implicational bridging is given in (25).13

Page 15 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Second Occurrence Focus (25)

Rooth claims that younger is coded by the same type of secondary prominence as, for example, juice in (10) above, provided that the inference the speaker wishes to express is that Susan and Harold are among the younger candidates. Krifka (2004: 203) gives a similar bridging example in (26) but claims that the inferrable information, namely that African-Americans are black, has to be highlighted by a pitch accent, even though it is ‘secondary’ in nature. Following the definitions introduced in the present chapter, however, this accent has to be interpreted as a—probably somewhat lowered in pitch, but still phonologically primary—nuclear pitch accent. (26)

(p. 498)

Both analyses are purely based on the authors’ (diverging) intuitions. However,

the different analyses do not seem to be justified by different degrees of inferability of the constituents in question, since both SOF phrases, younger candidates and black job candidates, are generic but accessible terms containing lexically new (younger, black) and given ((job) candidates) items. Furthermore, taking into account that persistence of grammatical function facilitates deaccentuation (Terken and Hirschberg 1994), the adjective black in (26) should be even less prominent than the adjective younger in (25), since the former replaces another adjective (African-American) while the latter replaces a noun phrase (Susan and Harold) as part of a comparative construction. Clearly, systematic experimental studies are needed to investigate cases like these. In fact, there is evidence from German corpus data (e.g. Baumann and Riester 2013) showing that postnuclear prominences do occur with phenomena which resemble proper SOF, for example epithets (cf. Clark 1977), defined as coreferential anaphors which consist of new lexical material. In the following example from a corpus of German radio news (DIRNDL, Eckart et al. 2012) the phrase der serbischen Provinz (‘of the Serbian province’) is an epithet to Kosovo mentioned in the previous sentence (Riester and Baumann 2013: 239f.). The RefLex annotation scheme for information status (Baumann and Riester 2012) is used here, which differentiates between various categories on a referential (R-) and a lexical (L-) level of givenness.

Page 16 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Second Occurrence Focus (27)

Epithets like these (on the word Provinz (‘province’)) are regularly realized by postnuclear prominence and exhibit a similar combination of boosting (newness at word level) and inhibiting (givenness at referential level) factors as second occurrence foci, which are combinations of focus/importance and givenness/predictability features. Thus, both types of expressions have a ‘hybrid’ information structural status, which is indicated by the same type of secondary prosodic prominence. Another widely discussed example potentially displaying ‘SOF phonology’ is Rooth’s (1992: 109) rice-grower example, given in (28) with a Büring-style analysis. (28)

Krifka (2004) proposes that there are contrastive pitch accents on grow and eat but no accent on the second occurrence of rice despite its association with only. Beaver and Velleman (2011: 1679) contradict this analysis and suggest that the second occurrence of rice in fact receives a secondary prominence just like proper SOF structures, since (p. 499) rice is important (= F-marked) but predictable (= not N-marked) information. In addition, the model generates secondary prominences on all unpredictable (= N-marked) elements (see (29)). (29)

However, the difference between the structure in (28)/(29) and the SOF structures discussed in previous sections is the intervention of a contrastive element between the focus-sensitive operator and the associated element, that is, its semantic focus. Whereas in ordinary SOF structures, the second occurrence of rice would be expected (and has been found) to bear a secondary, postnuclear prominence, elements in the direct vicinity of contrastive accents have been shown to be downplayed in order to increase the prominence of the contrasted item. Postfocal words in particular have been found to be marked by a compression of the pitch range in both tone and intonation languages (cf. Xu et al. 2004; Xu and Xu 2005 on Mandarin and American English). Another aspect supporting the reduction of the prominence of the final element may be the avoidance of a stress clash between eat and rice. In a similar case displaying an overt contrast, namely another famous example by Rooth (1992) given in (30), no model would predict a Page 17 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Second Occurrence Focus secondary accent on the second occurrence of farmer, since it is not associated with a focus-sensitive operator. An analysis along the lines of Beaver and Velleman (2011) should look like this: (30)

It is doubtful whether the degree of prosodic prominence of the second occurrences of rice in (29) and farmer in (30) are really different, since their (surface) structures are very similar. Again, empirical evidence is necessary to shed light on this controversial issue. So far we have only discussed examples of SOF and related structures in West Germanic languages. However, there are also comparable cases in other languages, for example Japanese. Ishihara (2003, 2007) found prosodic differences between clause types if they are embedded in a wh-phrase. He analyses all wh-phrases as prosodically defined focus domains (indexed FI for Focus Intonation) which display a pitch rise on the wh-element. If a wh-phrase is embedded in another one, as in (31) below (Ishihara 2007: 153), the embedded wh-element náni-o is both raised (due to focus) and lowered in pitch (due to embeddedness), thus surfacing on an mid pitch level. (31)

In an embedded yes/no-question, on the other hand, pitch is only lowered since a yes/noquestion is not interpreted as a domain of focus (analogous to the effect of postfocal compression mentioned above). Thus, embedded wh-phrases in Japanese show similar boosting and inhibiting effects as SOF structures in Germanic languages.

(p. 500)

24.4 Towards a unified approach

The empirical studies reported above provide abundant evidence from West Germanic languages that SOF is mainly marked by non-tonal phonetic parameters such as duration and intensity and not by pitch movement (in postfocal position). This finding supports the proposal put forward by a number of authors in similar ways (Rooth 1996, 2010; Beaver et al. 2007; Büring 2013), namely that metrical stress is the basic focus marker rather than pitch accent. Metrical stress can be regarded as the representation of postlexical stress in a metrical grid (see Truckenbrodt 1995; Selkirk 2006; Myrberg and Riad, this volume), which can be applied to prosodic phrases of various sizes (in the present proposal the relevant domain is the intermediate phrase, which contains at least one—the nuclear—pitch accent). A stress-first approach like Büring’s claims that SOF elements receive the strongest metrical stress in the scope of their operator and that their (lexical

Page 18 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Second Occurrence Focus and structural) givenness or predictability prevents them from receiving additional prosodic/metrical prominence by tonal movement. Note that non-tonal parameters are the primary markers of (postlexical) stress in stress accent languages (Beckman 1986) like the ones belonging to the West Germanic language family, but not in a pitch accent language like Japanese. Here, as we saw in the last section, the secondary prominence of SOF items is rather expressed by an intermediate pitch height. The metrical stress approach is compatible with the modified association with focus theories presented above and can thus serve as a unifying approach within this framework. In a broad combination of rules proposed by Büring (2013) and the weighting procedure suggested by Beaver and Velleman (2011) (see (19) above) we can assign the following rudimentary metrical grid to our proper SOF example in (4), repeated here as (32):

(32)

This pattern with two different prominence levels (1=secondary prominence, 2=primary prominence)14 can be derived in various ways. Following Beaver and Velleman, we assign level 1 prominence to each element that is either unpredictable (N-marked) or pragmatically important (F-marked). Level 2 prominence is added if a constituent is both F- and N-marked (cf. (33)).

(p. 501)

(33)

The same pattern follows from Selkirk’s (2008) analysis, if level 2 is directly assigned to F-marked elements and G-marked elements reduce F-marks by one level. A similar result also emerges from Büring’s analysis of the sentence, in that one prominence level can be assigned to each focus in its domain, be it associated or free (cf. (34)).

Page 19 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Second Occurrence Focus (34)

Let us see how these abstract (or at least underspecified) prominence levels are realized. Büring’s (2013: 7) STRESS-TO-ACCENT-RULE (in a modified version in (35)) determines the mapping of prominence levels to pitch accents: (35)

This rule ensures that the rightmost prominence at level 2 surfaces as a nuclear pitch accent (as the head of an ip), in (34) on John. Furthermore, the rule entails that a postnuclear prominence at level 1 can only surface as a secondary prominence, marked by increased duration and intensity (on juice in (34)). We may equate this basic level with metrical stress (see (36)) and claim that this prominence level is generally assigned to SOF items. (36)

The metrical approach also accounts for the empirical/perceptual difference between prenuclear accents on the one hand and both nuclear accents and postnuclear prominences on the other. These differences can be represented by an additional, medial level in the metrical grid, leading to three levels with a stepwise increase in phonological prominence: (p. 502)

(37)

Let us apply these metrical prominence levels to a second occurrence expression as the one in (38) (which is a repetition of (29)), analysed by Beaver and Velleman. If we assume a division into two intermediate phrases, we receive two nuclear pitch accents, on grow

Page 20 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Second Occurrence Focus and eat, as well as prenuclear pitch accents on people and generally. Furthermore, both occurrences of rice are marked by postnuclear non-tonal prominences. (38)

This simple incremental metrical model of the phonological realization of focus structures accounts for most of the empirical evidence for FOF and SOF, at least in Germanic languages. It is also compatible with many previous—and partly theoretical—proposals which suggest some sort of weighting procedure of boosting and inhibiting effects of focus and givenness on prosodic prominence. Additionally, an intermediate level of metrical prominence is proposed in order to distinguish between prenuclear and nuclear pitch accents, a difference which has been found to be relevant for a comprehensive description of SOF phonology (see Féry and Ishihara 2009). In general, the solutions found for the SOF challenge largely support association with focus theories, since SOF—as well as semantically related structures—has been found to be marked by prosodic prominence which is, however, non-nuclear and thus secondary in nature.

Notes: (1) Throughout this chapter, we will indicate a focus associated with a focus-sensitive operator by FOF and SOF, and a free focus (which is not associated with a focus-sensitive operator) just by F (unless the conventions used by specific theories are reported). (2) Capital letters indicate the presence of a pitch accent. (3) The superordinate prosodic domain is the intonation phrase (IP), which consists of one or more intermediate phrases (cf. Beckman and Pierrehumbert 1986). Each intermediate phrase contains one nuclear accent. (4) The semantic-pragmatic and prosodic analysis is restricted to the (b) sentences. (5) In fact, Büring (2013) suggests using the term ‘primary focus’ for focus expressions bearing the nuclear pitch accent, thus avoiding the potential inconsistency that a ‘First Occurrence Focus’ may contain a given element. However, defining a semantic-pragmatic category in terms of its prosodic realization runs the risk of circularity. (6) Secondarily prominent syllables are indicated by small capitals. The use of brackets and indices is taken from Büring (2013), with the exception of the convention introduced in footnote 1: associated foci are spelled out as FOF and SOF, free foci as F. (7) The symbols ~j and ~k indicate focus interpretation operators of different domains. Page 21 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Second Occurrence Focus (8) The structural difference between pre- and postnuclear prominences will be discussed in Section 24.2.3 and in the unified approach proposed in Section 24.4. (9) Beckman (1986) called this factor total amplitude, and found that it is the most relevant postlexical stress marker in English. (10) Only the relevant subscripts on the target words are indicated, as well as the pitch accents in the target sentences (by capitalization). (11) Nevertheless, Non-Focus items were marked by (prenuclear) pitch accents as well, which is a somewhat surprising result, since the Non-Focus target words in the experiment represent given information occurring in the same grammatical role as in the context question (cf. Terken and Hirschberg (1994) and the discussion in Section 2 above). (12) Nuclear accents usually show longer durations than prenuclear accents in West Germanic languages (see Silverman and Pierrehumbert 1990). (13) We will be using the subscript ‘SOF’ for all kinds of second occurrence expressions. (14) We disregard a potential ‘level 0’ for unstressed words here.

Stefan Baumann

Stefan Baumann is Assistant Professor at the Institute of Linguistics (Phonetics Department), University of Cologne. His main research interests include prosodic, semantic and pragmatic aspects of information structure, the phonetics and phonology of intonation, the role of prosody in neurocognitive language processing, and multi-layer annotation of spoken language.

Page 22 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy

Oxford Handbooks Online On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy   Sara Myrberg and Tomas Riad The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara Print Publication Date: Jul 2016 Subject: Linguistics, Morphology and Syntax Online Publication Date: May 2015 DOI: 10.1093/oxfordhb/9780199642670.013.41

Abstract and Keywords This chapter examines the phonological correlates of information structural focus in the metrical grid and in the prosodic hierarchy, with emphasis on how they cooperate to express focus in simple cases but seem to conflict in more complex cases. Two Germanic languages, English and Swedish, are compared to highlight the implications of so-called nested foci and Second Occurrence Focus (SOF) with respect to how the metrical grid and the prosodic hierarchy interact in phonology. After providing an overview of metrical grid and prosodic hierarchy, a short description of Swedish intonation is given. The chapter then considers nested foci and SOF. It offers some predictions for Swedish of the constraint STRESSFOCUS, which does not specify what the correlation between metrical grid marks and heads of the prosodic hierarchy should be in the case of focus. It suggests that focus is expressed in a more categorical fashion in Swedish than in English. Keywords: focus, metrical grid, prosodic hierarchy, English, Swedish, nested foci, Second Occurrence Focus, phonology, intonation, STRESSFOCUS

22.1 Introduction IN this chapter, we study the phonological correlates of information structural focus (as defined in Krifka 2008), in the metrical grid and the prosodic hierarchy. Our aim is to show how these two phonological structures cooperate to express focus in simple cases, but conflict in more complex cases. We compare two Germanic languages, English and Swedish, which are currently analysed as prosodically rather different. Where Swedish

Page 1 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy exhibits two separate prominence levels of accent (word accent, focus accent, Bruce 19771), English is taken to have only one (Pierrehumbert 1980). We show that this difference of analysis entails different predictions for so-called nested foci and Second Occurrence Focus (SOF), respectively, which have direct implications for typology and the understanding of how the metrical grid and the prosodic hierarchy interact in phonology. Specifically, Swedish seems to make less use of a purely relational notion of prominence for expressing focus (i.e. a prominence notion that refers solely to the strength of immediately surrounding prominences), than English. After the introduction, where the metrical grid and the prosodic hierarchy are presented, we turn to a short description of Swedish intonation in Section 22.2. Thereafter we discuss nested foci in Section 22.3, and SOF in Section 22.4.

22.1.1 The metrical grid The metrical grid represents linguistic rhythm, where rhythm refers to prominence patterns. In (1), the variation in prominence is represented as variable height of columns. (p. 442)

(1)

The grid indicates both rhythmic alternation (relative height of adjacent columns) and the hierarchical relation between prominences (relative column height within the whole expression). Example (1) is typical for Germanic rhythm, exhibiting an alternating weak/ strong pattern of adjacent columns, while prominence builds towards the right edge. The earliest versions of the metrical grid were constructed as binary branching trees (2), with nodes marked weak (w) or strong (s) (Liberman 1975; Liberman and Prince 1977). The numbers in the grid and the tree together specify the degree of prominence on linguistic units.

Page 2 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy

(2)

The metrical tree creates phonological domains which express grouping. However, this property was later taken to be at least partly superfluous.2 Prince (1983) and Selkirk (1984) argued that the location of the primary prominence (i.e. the one terminal strong node, which is dominated by the highest strong node in the tree) could be derived via reference to the edge of the domain under analysis, be it a word or a phrase. The weak/ strong distinction could be derived from general conditions on rhythmic alternation. The term eurhythm denotes rhythmic structures that fulfil these conditions. Languages (p. 443) often exhibit rules that serve to achieve eurhythm where illformed configurations arise, that is, when prominences are too close to each other (stress clash), or too far apart (stress lapse), at some level of the grid. The main arena for these processes in Germanic languages is phrasal phonology, cf. (3) and (4), from Hayes (1984: 46–8). (3)

Page 3 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy (4)

In (3) the clash is resolved by shifting the prominence on sipp to miss. The lapse in (4) is constituted by two equal word stresses, and better rhythm is achieved by adding a beat to Farrah.3 Note that this modification also leads to the addition of a beat on the last stressed syllable of the phrase, a measure driven by metrical culminativity (Hayes 1995: 24).

Page 4 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy

22.1.2 Focus in the grid In the Germanic languages, focus is typically realized as increased prominence in terms of duration, pitch, and intensity. In a grid-based analysis, focus is registered together with other types of prominences, as increased column height relative to the surrounding environment. A number of researchers have assumed various formulations of constraints called STRESSFOCUS or Focus Prominence to account for the effect of focus in the metrical grid (e.g. Truckenbrodt 1995; Schwarzschild 1999; Samek-Lodovici 2005; Féry and SamekLodovici 2006; Selkirk 2007; Büring 2010; Rooth 2010). In these accounts a focus must contain the most prominent element in the domain of that focus. Truckenbrodt’s (1995: 160) formulation of this constraint is given in (5) (though he calls it Focus).4 (p. 444)

(5)

The reference to focus domain in (5) should be understood in relation to Rooth’s (1992) insight that the semantic scope of a focus is not always the entire sentence in which that focus occurs. Instead, a sentence may contain multiple foci and multiple scopes. Truckenbrodt argues that the semantic scope of a focus (in Rooth’s sense) is also the phonological domain of that focus, the focus domain, within which the focus carries the strongest prominence. We return to some predictions of this claim below. It should be noted that in a typological view beyond the Germanic languages, StressFocus is not valid in any obvious way (although cf. Büring 2010 for an attempt to increase the validity of STRESSFOCUS via an extended notion of prominence). Across languages, focus is marked in several ways, for example by the insertion of a boundary after the focused constituent (Pierrehumbert and Beckman 1988) or by lengthening of a syllable (Downing 2008). Drawing on evidence from several languages, Féry (2013), argues that alignment with a phonological edge is more valid than prominence for a universal definition of the phonology of focus.

22.1.3 Bracketed grids The grid has proven suitable for expressing most word stress systems, as these are typically alternating, with primary prominence located at, or close to, one edge, according to rules like (3) and (4) (for an overview, see Kager 1995). Phrasal rhythm shares some of these properties, given the rhythm rule (3) and the typical edgeorientation of phrasal prominence (e.g. the English nuclear stress rule, cf. Zubizarreta, this volume).

Page 5 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy However, there are challenges to expressing prominence related phenomena in a gridonly representation. Many phonological and morphological processes in natural languages are bounded by prosodic domains or relate to specific types of prominences (e.g. Selkirk 1984; McCarthy and Prince 1986). The grid will not be helpful in describing, say, segmental phenomena that are related to stressed and unstressed syllables in particular positions of words. Nor will it serve to express the different quantitative behaviours of prominent syllables in so-called iambic and trochaic stress systems, where a stressed syllable may lengthen or shorten depending on whether it is grouped with an adjacent syllable to the left or to the right (Kager 1993). In response to these things, the notion of bracketing was introduced in the metrical grid (Halle and Vergnaud 1987). Bracketed grids, as in (6), express domains over which conditions can be stated and rules delimited. In the area of word stress, alternating rhythm has come to be modelled in terms of binary, headed feet, leading to a constrained typology of stress systems (Hayes 1995; Kager 2007). While we leave word stress systems aside here, the question of bracketing of the grid is relevant in that it moves away from a purely rhythmic representation (Prince 1983) and formalizes the (p. 445) relationship between rhythm and the categories of the prosodic hierarchy (Selkirk 1984).5

(6)

22.1.4 The prosodic hierarchy When bracketing is introduced in the grid there are a couple of ways to go. One is to use brackets to express computations made in the grid, to locate prominences in it, or to express other operations. Exponents for this tradition are Idsardi (1992) and Fabb and Halle (2008), the latter being an analysis of poetic metre. This approach to bracketing does not connect with prosodic domains in any basic way, and we will not develop the consequences of such bracketing for information structure. Another use of brackets is to formally connect the grid with the domains of the prosodic hierarchy. This view has become influential in works on prosody and information structure (cf. Truckenbrodt 1995).

Page 6 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy Arguments for the existence of a prosodic hierarchy as such were developed in the 1980s (e.g. Selkirk 1984; Nespor and Vogel 1986; Pierrehumbert and Beckman 1988). The prosodic hierarchy is a set of hierarchically ordered phonological domains, for example foot, word, phrase, and intonation phrase, created in part by alignment with syntactic and morphological edges. Phonological processes apply strictly within phonological domains, and can make direct reference neither to syntactic nor morphological structure (this principle was dubbed the indirect reference hypothesis by Inkelas 1989). Nespor and Vogel (1986) and other researchers have assumed the set of phonological domains in the hierarchy to be universal (although no strong consensus has, as of yet, been reached about the identity of these domains), whereas the phonological processes that apply in each domain have been assumed to be language specific. Already in the early prosodic hierarchies of Selkirk and Nespor and Vogel, each phonological domain was assumed to have a head, realized as some type of linguistic (p. 446) prominence. Selkirk (1984: 26, 146) as well as Nespor and Vogel (1986: 71) assumed that heads of phonological domains corresponded to marks in a metrical grid, so that each row in the grid corresponds directly to some category of the prosodic hierarchy (e.g. foot, word, phrase, or intonation phrase). This double structure is illustrated in (7), from Nespor and Vogel (1989: 71).6

Page 7 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy

(7)

According to Truckenbrodt (1995: 176) most researchers working on prosody and information structure have subsequently accepted that metrical grid marks are heads of the prosodic hierarchy. The formulation in (8) is due to Truckenbrodt (1995: 12, also 37), and we refer to it here as the one-mark-one-head principle. (8)

Other works where this correlation is explicitly assumed include Féry and Samek-Lodovici (2006), Féry (2010), and Büring (2010). As we will see in the next section, the one-markone-head principle has a constraining effect on the metrical grid, limiting column height, as a matter of principle, to the assumptions made for the prosodic hierarchy.

Page 8 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy (p. 447)

22.1.5 Recursion

The grids and prosodic hierarchies of Selkirk (1984) and Nespor and Vogel (1986, 1989) are characterized by a strictly limited height of their columns, as opposed to the original grids proposed by Liberman (1975). This is because they share two important assumptions. First they adhere to the idea that the number of grid marks in a column corresponds to a limited number of phonological domain types. Second, they both adhere to the so-called Strict Layer Hypothesis (SLH). According to this hypothesis, each domain type occurs exactly once in phonological structures like (7). There is no recursion of any domain type, and there is no skipping of domain types. The SLH was originally formulated by Selkirk (1984: 26). The general contention of the SLH is given in (9), from Ladd (2008: 291). (9)

More recently, an increasing number of researchers have argued that there is evidence for recursion as well as level skipping in phonology, and that the SLH is too restrictive (e.g. Ladd 1986, 1992; McCarthy and Prince 1993; Selkirk 1995; Itô and Mester 2006, 2007). This view has been accepted by many researchers today (among others Wagner 2005, 2010; Ladd 2008; Kabak and Revithiadou 2009; Selkirk 2009, 2011; Féry 2010; Itô and Mester 2011). The introduction of recursion into the prosodic hierarchy complicates the question of the relationship between the grid and the prosodic hierarchy, specifically with regard to the one-mark-one-head principle (8). If recursion is allowed in phonological representations, the number of layers increases, and if each layer of phonological structure corresponds to one grid mark in the column of a grid, then the number of grid marks also increases. This is pointed out by Féry (2010), who provides the illustration in (10) where a single domain type, P, is responsible for up to four grid marks in one column (for similar structures see Ladd 1986; Féry and Truckenbrodt 2005; Wagner 2005, 2010; Kentner and Féry 2013).7 (10)

Page 9 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy Structures like (10) also illustrate the fact that the connection between the prosodic hierarchy and the grid entails a link to syntax, as the prosodic domains are aligned or matched with syntactic categories (e.g. Ladd 1986; Selkirk 1986, 1995, 2009, 2011; Itô and Mester 2011). Several researchers have derived the rightward tendency of prominences from syntactic embedding, so that the column which corresponds to the deepest embedded syntactic constituent is higher than other columns (Zubizarreta, this volume). (p. 448)

22.1.6 Focus in the prosodic hierarchy In sentences with only one focus, the focus correlates with the highest grid mark, which is the strongest prominence in the sentence. This prominence is referred to as the nuclear accent (Jackendoff 1972; Selkirk 1984, see Zubizarreta, this volume). If metrical grid marks are heads of constituents in the prosodic hierarchy, as according to the one-markone-head principle, then the highest grid mark, that is, the nuclear accent, must be connected to the highest domain in the prosodic hierarchy. This would be the Intonation Phrase, or the Utterance (depending on model specific definitions of the domains in the prosodic hierarchy). In simple cases with only one focus per sentence, then, we can say that there is harmony between the requirements of STRESSFOCUS (5) and the one-markone-head principle (8). However, there is nothing inherent in the formulation of STRESSFOCUS that calls for a focus to correspond to a specific height of a metrical grid column. As long as no other prominence in the focus domain has more grid marks, it is immaterial whether a focus has three grid marks or five from the point of view of STRESSFOCUS. In the prosodic hierarchy this means that it does not matter if a focus is the head of a Prosodic Phrase or an Utterance, as long as no other head in the same focus domain belongs to a higher category. STRESSFOCUS, then, refers to the strength of prominences surrounding a focus, but does not refer to any absolute notion of prominence. In this sense STRESSFOCUS is an entirely relational definition of focus prominence. In the literature, the relation between the metrical grid and the prosodic hierarchy is often not made fully explicit beyond the harmonic state of affairs observed in simple sentences with one focus. Rather, researchers tend to pay more attention to either the metrical grid or the prosodic hierarchy for the understanding of some phenomenon at hand. However, there are areas where the relation between the two structures is complicated or challenged. In the remainder of this chapter, we address two such cases, namely embedded foci and second occurrence focus (SOF). In both cases we compare English with (less well-studied) Swedish. The interest of looking at Swedish here is that this language has pitch accents of two different prominence degrees, so that the perceived degree of prominence is largely determined by the shape of the pitch accent contour (in addition to correlates such as duration and intensity, e.g. Bruce 1977; Heldner 2001). In this respect, Swedish differs Page 10 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy from English. Although in English there are also several options for the shape of a pitch (p. 449) accent, these shapes are not strongly correlated with the perceived degree of prominence. Rather, the size of pitch excursions and the position within the phrase determine the degree of perceived prominence in English (Pierrehumbert 1980; Ladd 2008). Given the typological closeness of these languages, the intuition is nonetheless that the two languages should be given minimally different analyses (cf. also discussion in Ambrazaitis 2009: 64–65).

22.2 Intonational prominence in Swedish Swedish belongs to the North Germanic languages. It shares some of the basic properties of the intonation systems of the better-described West Germanic languages, such as the marking of focus with intonational prominence, and the marking of edges with tonal movements (boundary tones). However, Swedish also has (at least) two properties that set it apart from West Germanic. First, it has a binary lexical accent distinction between lexical accents 1 and 2. Words have either one of these lexical accents, depending on factors relating both to the lexical marking of morphemes and the phonological structure of the word (Riad 2009, 2012; Wetterlin 2010). Second, Swedish tonal prominences are divided into two categorically different prominence levels: focal accents and word accents.8 Here, our main concern is with the latter distinction between the two levels of prominence, as they regiment the realization of focus in Swedish, and thereby have implications for the relation between the metrical grid and the prosodic hierarchy. The lexical distinction between accents 1 and 2 is entirely orthogonal to the prominence distinction. In Stockholm Swedish, focal accents always contain a rise, whereas word accents contain a fall.9 Each accent type has two realizations: one for accent 1 and another for accent 2, cf. Table 22.1.10 A sentence has at least one focal accent. It appears on a narrow focus, otherwise on the sentence-final word (with some syntactically triggered exceptions that have direct parallels in the West Germanic languages, cf. Myrberg 2010). Sentences also frequently contain several focal accents, which may serve to phrase individual syntactic constituents or contrastive topics (Myrberg 2010, 2013). Most words which are not assigned focal accents carry word accents. These are generally perceptually less prominent than focal accents (e.g. Heldner 2001; (p. 450) Fant and Kruckenberg 2008).11 Word accents appear before and after the focal accent in a sentence, and between focal accents if there are more than one. Whereas focal accents are avoided on given material (especially following a focus), word accents are retained independently of information status.

Page 11 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy Table 22.1 The realization of focal accents and word accents, together with the lexical accents in Stockholm Swedish word accent

focal accent

lexical accent 1

HL*

L*H

lexical accent 2

H*L

H*LH (in compounds: H*L*H)

Note: The representations assumed here are based on Bruce (1998, see also 2005, 2007). However, for the focal accent 1 contour, and for the compound focal accent, we follow Riad (1998, 2006). To account for the distribution of the accent types in terms of prosodic structure, we assume, following for example Itô and Mester (2011), and Selkirk (2009, 2011), three structurally different levels of phrasing above the syllable: the prosodic word (ω), the phonological phrase (φ) (other names for similar phrase types include prosodic phrase, pphrase, and intermediate phrase), and the intonation phrase (ɩ). We follow Myrberg (2010, 2013), Myrberg and Riad (2013, 2015), Riad (2014) in assuming that the focal accent is head of φ, and that the word accent is head of ω. We thus assume one φ for every focal accent, and one focal accent for every φ in a sentence. Likewise, we assume one ω for every word accent and one (word or focal) accent for every ω.12 As the difference between the focal accent and the word accent is a categorical one, and the two prominence types have very different distributions, we assume that they must be heads of two different categories in the prosodic hierarchy. This assumption will be crucial for the discussion in Sections 22.3 and 22.4. It will, however, not be crucial which two categories the two prominence levels belong to. (Thus, the hypothetical claim that the focal accent is head of ɩ instead of φ would not substantially affect our claims in this paper. For discussion, see Section 22.3.2 and Myrberg 2010, 2013). The ɩ is defined by having edge marking tones. At the right edge there is a boundary tone, and at the left edge there is a so-called initiality accent, which has a contour that is phonologically similar to the focal accent, but which is functionally similar to a boundary tone (Roll et al. 2009; Myrberg 2010, 2013). Sentences can be phrased as one ɩ, containing one or more φ; thus with one set of edge markers at the left and right, and one or more focal accents. However, sentences can also be phrased as multiple ɩ; with multiple focal accents and multiple sets of edge markers (cf. Myrberg 2010, 2013). (p. 451)

In Sections 22.3 and 22.4, we contrast Swedish phrasing with that observed in English. In these comparisons, the notions nuclear accent, and pre- and postnuclear accent will be relevant (cf. Zubizarreta, this volume). These terms refer to positions within the ɩ, and will be helpful in comparing accents in different positions between Swedish and English. Page 12 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy We define the nuclear accent as the head of an ɩ, and we assume that this is the last focal accent (i.e. the last φ-head) in an ɩ. When we use the term ‘nuclear accent’ for Swedish, then, we always refer to a focal accent (although not all focal accents are nuclear accents). The area before the nuclear accent is the prenuclear area. This area can contain both focal accents and word accents in Swedish. The area following the nuclear accent is the postnuclear area. In all-new sentences, this area is usually very short, as a nuclear focal accent must appear toward the end of all-new sentences. In sentences with early narrow focus, the postfocal area is longer. The postnuclear area then contains material which is given and/or background to the focus. Such areas are often described as deaccented in the West Germanic languages (e.g. Ladd 2008: 102, 231). This is true in Swedish, too, insofar as there are no focal accents in the postnuclear area. However, word accents are retained on postnuclear material. In summary, we can distinguish prenuclear accents (either word accents, in which case they are ω-heads, or focal accents, in which case they are φ-heads), nuclear accents (focal accents and ɩ-heads) and postnuclear accents (word accents and ω-heads).

22.3 Relational and absolute prominence in nested foci In this section we show that the categorical distinction between word accents and focal accents in Swedish is incompatible with a relational notion of focus, such as that expressed by STRESSFOCUS (5). To that end, we review an analysis of sentences containing nested foci in English (Féry and Samek-Lodovici 2006). This analysis works well for English (Section 22.3.1), but makes the wrong predictions for Swedish (Section 22.3.2).

22.3.1 Nested foci in English Féry and Samek-Lodovici (2006) present their analysis of prosodic phrasing and pitch accent distribution as a set of interacting constraints within the framework of Optimality Theory (Prince and Smolensky 1993). The STRESSFOCUS constraint (cf. (5) above) calls (p. 452) for focused material to have the highest prominence within the focus domain (11). The constraint DESTRESSGIVEN requires that given material be non-prominent (12) (Schwarzschild 1999). In addition, there are constraints that call for prominences in prosodic structure to be aligned with the right edges of phonological phrases (13) and intonation phrases (14), and for each XP in syntax to have a prosodic prominence (15) (Truckenbrodt 1995). The constraints in (13)–(15) have been slightly reformulated, for the sake of exposition).13 Page 13 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy (11)

(12) (13)

(14)

(15)

With (11)–(15), Féry and Samek-Lodovici generate the prosodic output structures for sentences which contain so-called nested foci, that is, where one or more foci are contained within a larger focus, as illustrated in (16) (originally discussed by Rooth 1992) and (17) (originally discussed by Neeleman and Szendrői 2004). Foci are marked with square brackets. (16) (17)

In (16) American and Canadian are both focused on account of being contrasted with each other. Simultaneously, the whole sentence is focused, by virtue of being all-new. Within the sentence, however, the second instance of farmer is given. In (17), the whole clause Johnny was reading Superman to some kid is focused on account of being the answer to the father’s question What happened?. At the same time, the VP was reading (p. 453) Superman to some kid is focused on account of being contrasted with rather than doing his homework. Finally Superman is focused because it is contrasted with decent books. Tableau (18) illustrates how the constraints in (11)–(15) derive the prosodic structure of (16). The winning candidate a has one grid mark, corresponding to a φ-head, on American, and two grid marks, one corresponding to a φ-head and one corresponding to an ɩ-head, on Canadian. The φ-head on American satisfies STRESSFOCUS with respect to the focus on this word (note that there is no requirement that American has the highest prominence in the sentence, only in its focus domain, in this case the NP an American

Page 14 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy farmer). The ɩ-head on Canadian simultaneously satisfies STRESSFOCUS with respect to the focus on Canadian, and the focus on the whole sentence. (18)

Tableau (19) provides an analysis of the Superman-sentence from (17). Here, StressFocus is satisfied with respect to all three foci in the sentence via the single grid mark corresponding to the ɩ-head in candidate a. This grid mark provides the deepest embedded focus, F3, with an ɩ-level grid mark, which makes F3 the most prominent constituent within its focus domain. As F3 is embedded in F2 as well as F1, the grid mark in F3 simultaneously provides F2 and F1 with an ɩ-level grid mark, thus satisfying StressFocus for all foci. In addition to the ɩ-level mark on Superman, STRESSXP forces the insertion of a φ-head on each of the XPs Johnny and to some kid.

Page 15 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy (p. 454)

(19)

In the analyses in (18) and (19), respectively, STRESSFOCUS is satisfied by metrical grid marks corresponding to heads of two different categories in the prosodic hierarchy. On American STRESSFOCUS is satisfied by a φ-head. On Canadian and Superman, however, STRESSFOCUS is satisfied by ɩ-heads. This results from the interaction between STRESSFOCUS and the constraints in (12)–(15). It is possible because StressFocus is formulated in entirely relational terms. It is precisely this prediction, that the realization of focus can vary between φ-heads and ɩ-heads, which is interesting in relation to the categorical difference between focal accents and word accents in Swedish.

22.3.2 Nested foci in Swedish When we turn to nested foci in Swedish, the analysis must heed the categorical distinction between word accent and focus accent, and the entailed requirement that they belong to different categories of the prosodic hierarchy (as argued in Section 2). We consider first how Féry and Samek-Lodovici’s analysis would fare under the assumption that focal accents are φ-heads, as we assumed in Section 2. Under this assumption, the analysis provided in Tableau (18) makes correct predictions for Swedish farmer-sentences, as illustrated in (20). American and Canadian are φ-heads and get focal accents (marked with ‘FA’). Words which are ω-heads are predicted to have word accents (marked with ‘WA’). This follows from our assumptions that word accents are ω-heads and that most content words form ω (Riad 2012; Myrberg and Riad 2013, 2015).

Page 16 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy (p. 455)

(20)

However, the analysis makes wrong predictions with regard to the Superman-sentence in (19), as seen in (21). (21)

The empirically correct output structure should have only one focal accent, on Stålmannen, while Johan and unge should have word accents. However, STRESSXP forces one φ-head on every syntactic XP in the tableau in (19), resulting in three φs in the sentence, which is an erroneous prediction. To derive the correct output for Swedish Superman-sentences, one might be tempted to assume that only heads of ɩ have focal accents, and not heads of φ. Under that assumption, only one focal accent would be predicted, on Stålmannen, yielding the empirically correct result in (22), cf. Tableau (19).14 (22)

Page 17 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy However, the assumption that only ɩ-heads are focal accents makes erroneous predictions for the farmer-sentences, as it would predict only one focal accent in farmersentences, cf. (23). (p. 456)

(23)

In a correct analysis of Swedish farmer-sentences, amerikansk must get a focal accent. Thus, in order for Féry and Samek-Lodovici’s analysis to work for Swedish, φ-heads must sometimes be focal accents (farmer-sentences), and sometimes be word accents (Superman-sentences). This is unreasonable as there is no independent evidence for such a state of affairs, and it would push the analysis towards the circular. The null hypothesis, given the categorical distinction between focal accent and word accent (Table 22.1), must be that focal accents are the realization of a single constituent type, be it ɩ or φ. We favour the analysis where the φ-head is the focal accent, and therefore propose that the correct output structure for a Swedish Superman-sentence is the one given in (24), where the whole sentence is phrased into a single φ.15 (24)

In conclusion, Féry and Samek-Lodovici’s analysis works well for English sentences with nested foci, where a relational notion of focus prominence suffices, but fails to account for corresponding sentences in Swedish. For Swedish, focus is not only required to have a phonetically stronger prominence than a non-focus (increased duration and pitch excursion), but is also ‘locked’ to a specific constituent in the prosodic hierarchy, arguably the phonological phrase (φ). The prosodic hierarchy thus more clearly determines the expression of focus in (p. 457) Swedish, than it does in English. Whether

Page 18 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy this is a real typological difference between two closely related languages, or a result of analytical tradition, remains an open question.

22.4 Second Occurrence Focus (SOF) If the prosodic hierarchy plays a more important role for expressing focus in Swedish than in English, the question is raised whether a relational notion of focus prominence has any role to play in the analysis of Swedish. In this section we show that it does. Multiple foci also occur in so-called Second Occurrence Focus (SOF) sentences (Baumann, this volume). A SOF occurs together with another focus, the First Occurrence Focus (FOF), within the same sentence. SOF-sentences are different from those in Section 3, in that a SOF is contextually given and is embedded in the focus domain of the FOF (but not embedded in the focus itself). It has been observed that this causes the SOF to be prosodically less prominent than the FOF (Partee 1999; Beaver et al. 2007; Büring 2008; Féry and Ishihara 2009). Whereas the FOF receives a canonical focus realization with the strongest (nuclear) prominence in the sentence, the SOF receives a lower degree of prominence, which does not correlate with nuclear prominence. In this section, we show that for SOF-sentences, the relational notion of focus does seem to be important in Swedish, and our proposal will be that a mismatch between the grid and the prosodic hierarchy obtains precisely in SOF-sentences. This mismatch allows for both the SOF and the FOF to be represented as grid marks in metrical structure, in accordance with STRESSFOCUS, while simultaneously allowing the SOF to have less prominence than the FOF. An often cited SOF-sentence is given in (25) (Partee 1999: 215–16). Here, the SOF appears after the FOF, and we will refer to this position of SOF as postnuclear. There are also prenuclear SOFs, which hence appear before the FOF, as in (26) (Rooth 1996; cf. also Féry and Ishihara 2009). (25)

(26)

Page 19 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy In English and German, postnuclear SOF is usually marked with increased duration and in German also with increased pitch height as compared to non-focused words. (p. 458) Prenuclear SOF frequently receives a pitch accent (Rooth 1996; Beaver et al. 2007; Féry and Ishihara 2009; Baumann, this volume). In an experiment with Swedish data (Myrberg submitted), postnuclear SOF usually carried a word accent, but appeared with a focal accent 27 per cent of the time, (27b). Word-accented postnuclear SOF-words were shown to have longer duration than wordaccented postnuclear non-focused words. Prenuclear SOF was shown to typically carry a focal accent, but appeared with word accents 16 per cent of the time, (28b). Prenuclear focally accented SOF-words had longer duration than prenuclear focally accented nonfocused words.16 (27)

(28)

Page 20 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy The prominence on SOF in relation to non-focused items has been interpreted as the realization of metrical grid marks on SOF (e.g. Beaver et al. 2007). Thus, even though (p. 459) SOF does not receive nuclear prominence, the assumed grid marks are taken to show that SOF-structures conform to the generalization expressed by StressFocus, that is, that all foci correspond to grid marks in metrical structure (Beaver et al. 2007; Rooth 2010; Baumann, this volume). A metrical grid which would accurately reflect SOF prominence in Swedish compels at least one grid level for focal accents, and one grid level for word accents. We might also want to mark the final focal accent, a nuclear accent, as more prominent than other focal accents, requiring a third grid level. Moreover, SOF-prominences would motivate a fourth level, higher than any other non-nuclear accent. This is rather more than we have grounds for in the prosodic hierarchy. We have identified categories in the prosodic hierarchy where nuclear focal accents, nonnuclear focal accents, and word accents have canonical realizations in Swedish (Section 2). Nuclear focal accents are ɩ-heads, non-nuclear focal accents are φ-heads, and word accents are ω-heads.17 However, there are reasons to believe that SOF-prominences have no canonical match in any of the categories of the prosodic hierarchy. The cues which distinguish SOF prominences vary (within and across) speakers, and look different in the pre- vs the postnuclear area of a phrase. No single, stable correlate of SOF could therefore be used as a defining criterion for a separate prosodic category where SOF is realized. In addition, SOF is a marginal phenomenon, and its realization appears to be determined by two conflicting interests: the pressure for high prominence on focused material, and the request for low prominence on given material. It seems unreasonable to postulate a domain in the prosodic hierarchy for the sole purpose of expressing SOF. Instead of assuming a new prosodic category, we might consider the possibility that recursion of some domain is responsible for the extra grid mark on the SOF constituent, much as recursion is responsible for several levels in the grid in (10). Recursion of this type could in principle allow a grid mark on the SOF, which would be the head of a prosodic domain as required by the one-mark-one-head principle. In our view, such a solution is implausible, however. The problem is partially illustrated by the tree structure in (29). This tree shows a postfocal area containing an SOF (nallar). In order for this SOF to have one more grid mark than the other postnuclear ω-prominences (without projecting to the φ-level where it would receive a focal accent), the ω containing the SOF must project to a higher level of ω. In (29), only the single word which contains the focus prominence can project. If more than one ω would project, that would trigger compound accent, which is not correct here (cf. Riad 2012; Myrberg and Riad 2013, 2015).

Page 21 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy (p. 460)

(29)

However, the projection of the single word which receives the SOF prominence in (29) is difficult to derive non-circularly. It cannot be independently motivated, for example by the pressure for alignment with syntactic structure or, say, with the whole focus domain that contains the SOF. We therefore abandon the idea that recursion of a category in the prosodic hierarchy would be the source of the grid mark on SOF. Instead of assuming an accommodating structure like (29), we propose that SOFstructures violate the one-mark-one-head-principle (8). The violation results from the simultaneous and conflicting pressures for low and high prominence on SOF. In response to this conflict, the grammar creates a metrical structure which is deeper than that available within the prosodic hierarchy, in order to allow an expression of SOF which is equal neither to focused (highly prominent), nor to given (non-prominent) material. Under this view, the prosodic realization of SOF is a result of the mismatch between the deeper metrical structure and the more shallow prosodic hierarchy. The SOF sentences from (27) and (28) are given with prosodic representations in (30) and (31). The a-examples provide the representation with a focal accent on both the FOF and the SOF. This pattern was attested in 27 per cent of the postnuclear SOF, and 84 per cent of the prenuclear SOF (Myrberg submitted). The b-examples provide the representations with a word accent on the SOF and a focal accent on the FOF. The b-pattern was attested in 16 per cent of the prenuclear SOF, and in 73 per cent of the postnuclear SOF.

Page 22 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy (30)

(p. 461)

(31)

The mismatch between the prosodic hierarchy and the metrical structure raises several questions, which cannot be addressed here. We will however mention that mismatches as such are commonly noted in syntax–prosody alignment, as well as in form–meaning relationships in morphology.

Page 23 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy

22.5 Summary and conclusion This chapter has discussed some predictions for Swedish of the constraint StressFocus, which holds that a focus must contain the highest metrical grid mark in the domain of that focus. Researchers seem to agree that focus prominence should somehow be reflected in metrical structure, although the role of the prosodic hierarchy in this matter remains unclear. STRESSFOCUS refers to an entirely relational notion of focus prominence, expressed solely in terms of the column height in a metrical grid. A correlation between metrical grid marks and heads of the prosodic hierarchy is usually assumed, but STRESSFOCUS does not specify what this correlation should be in the case of focus. In theory, this constraint allows for focus to be expressed by any prosodic category. The two categorically different pitch accent types in Swedish—focal accent and word accent—are (partly) used to distinguish focused from non-focused information. We have shown that the categorical distinction between them poses a challenge to the relational notion of focus prominence implied by STRESSFOCUS. Féry and Samek-Lodovici’s (2006) analysis of nested foci, applied to Swedish, provided an illuminating example of this challenge. We concluded that the prosodic hierarchy plays a more important role for the expression of focus in Swedish, than it does for English. We have also shown that focus expression in Swedish cannot make do with just reference to the categories of the prosodic hierarchy. SOF structures provide a situation where the pressure for focus to correlate with a focal accent in Swedish is overridden. In the absence of a focal accent on the SOF, a relational notion of focus is in evidence. Thus, in spite of the general importance of the absolute notion of focus prominence that (p. 462) can be observed in the distribution of focal accents in Swedish, a relational type of focus prominence is employed, whenever the absolute type of focus expression has been put out of the running.18 Any analysis of focus prominence is likely to be influenced by the language studied, even when the aim is to arrive at a universalist account. No doubt, the theory of focus phonology is primarily based on observations made in English and some other Germanic languages (e.g. Jackendoff 1972; Reinhart 1995; Truckenbrodt 1995; Zubizarreta 1998; Gussenhoven 2008; Büring 2010). Looking at Swedish, we have found a number of points of divergence with analyses of English, leading us to assume, provisionally, that there may be fundamental differences in how the two languages establish focus prominence. For Swedish, at least, we have argued that the categories of the prosodic hierarchy play an important role in the expression of focus, although a relational notion of focus prominence does seem to exist alongside it. In general terms, focus appears to be expressed in a more categorical fashion in Swedish than in English.

Page 24 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy

Notes: (1) Bruce’s (1977) original term for the focus accent was ‘sentence accent’. (2) Ladd (2008) and Calhoun (2010) use a type of binary metrical tree representation with strong and weak nodes, which resembles the metrical trees of Liberman (1975). (3) The corresponding rhythmic adjustment can also be applied in (1). (4) See also Truckenbrodt (this volume). (5) A connection between prosodic categories and levels of the grid was actually also assumed by Prince (1983), who argued that prominences of foot structure should correspond to specific lines in the metrical grid (Prince 1983: 27). However, the relational property of his grid led to the proliferation of levels, undermining the correspondence between prosodic categories and the grid. (6) In Nespor and Vogel (1986), the symbols U, I, φ, ω, Σ, and σ denote respectively the phonological domain types Phonological Utterance, Intonation Phrase, Phonological Phrase, Phonological Word, Foot, and Syllable. (7) The proliferation of levels is in fact foreseen already in Prince (1983: 27), who acknowledges the possibility of bands of grid levels, corresponding to a given prosodic category. (8) In Myrberg and Riad (2015) these are referred to as big and small accents, respectively. (9) Dialectal differences in the realization of tonal prominences are discussed in Meyer (1937, 1954), Bruce and Gårding (1978), Fintoft et al. (1978), Riad (1998), Bruce (2007). (10) Below the tonal prominence levels, syllables divide into stressed and unstressed. This distinction is lexically determined, and phonologically distinguished in terms of syllable weight (Kristoffersen 2000; Riad 2014). The stressed–unstressed distinction is not immediately relevant to information structure, and we leave it aside here. (11) Word accents fail to appear on some verbs (auxiliaries in particular), some pronouns (which we consider lexically unstressed) and in lexicalized phrases (Myrberg and Riad 2013). (12) The ω canonically correlates with a morphological word. The presence of both stress (non-tonal) and word accents (tonal) warrants a sometimes more complex internal structure. This is not immediately relevant here, and we refer to Riad (2012, 2014) and Myrberg and Riad (2013, 2015) for detailed discussion. (13) (12)–(14) together amount to the Nuclear Stress Rule (Féry and Samek-Lodovici 2006: 134). Page 25 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy (14) While yielding the empirically correct prediction, this analysis leaves the φprominences with a structurally unclear status, being higher than ω-heads but lower than focal accents. There is no experimental evidence that word accents are distinguished in accordance with the metrical structure in (22), and our native speaker intuition does not indicate that this should be the case. We consider the divergence between the presence of metrical grid marks and the absence of phonetic prominence as additional evidence against the analysis in (22). (15) It is beyond the scope of this chapter to discuss further modifications of Féry and Samek-Lodovici’s analysis in order to make it work for Swedish. (16) Recall from Section 2 that longer sentences frequently contain multiple focal accents, independently of semantic focus. It should be noted in this connection, that the focal accents on SOF in the prefocal condition are not (merely) due to the length of the target sentences (which differs between the English and the Swedish experiment). Sentences of corresponding length as in (27) and (28), but lacking SOF, were tested for comparison, and exhibited much fewer prenuclear focal accents. See Myrberg (submitted) for further details. (17) In principle, the discussion here is applicable also if only ɩ-heads are assumed to be focal accents, leaving φ-heads as word accents. In order to make the discussion easier to follow, however, we consider only the option that φ-heads (and by inheritance also ɩheads) are focal accents. (18) One could explore other solutions to this situation than a phonological mismatch. Féry and Ishihara (2009) suggest that there may be phonetic factors that affect the realization of particular prosodic prominences, possibly without implications for the prosodic hierarchy or the metrical grid.

Sara Myrberg

Sara Myrberg is a researcher at the Department of Swedish Language and Multilingualism at Stockholm University. She defended her thesis The intonational phonology of Stockholm Swedish at Stockholm University in 2010, and primarily works on Swedish prosody and the interfaces between prosody, syntax, and information structure. Tomas Riad

Tomas Riad is professor of Scandinavian languages at Stockholm University. His research is concerned with prosody in North Germanic languages, in particular stress and tone accent in a historical and typological perspective. He also works on poetic meter, and the general relationship between meter and phonology.

Page 26 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

On the Expression of Focus in the Metrical Grid and in the Prosodic Hierarchy

Page 27 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Information Structure and Production Planning

Oxford Handbooks Online Information Structure and Production Planning   Michael Wagner The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara Print Publication Date: Jul 2016 Subject: Linguistics, Morphology and Syntax Online Publication Date: Nov 2015 DOI: 10.1093/oxfordhb/9780199642670.013.39

Abstract and Keywords Utterances are planned and realized incrementally. Which information is salient or attended to prior to initiating an utterance has influences on choices in argument structure and word order, and affects the prosodic prominence of the constituents involved. Many phenomena that the linguistic literature usually treats as reflexes of the grammatical encoding of information structure, such as the early ordering of topics, or the prosodic reduction of old information, are treated in the production literature as a consequence of how contextual salience interacts with production planning. This article reviews information structural effects that arise as a consequence of how syntactic and phonological information is incrementally encoded in the production process, and how we can tell these effects apart from grammatically encoded aspects of information structure that form part of the message. Keywords: production planning, topic, focus, incrementality, word order, prosody, prominence

Page 1 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of New England; date: 17 February 2019

Information Structure and Production Planning

27.1 Grammar and processing THE literature on production planning has a rich tradition of work which concerns phenomena that the linguistic literature would feature under the headline ‘information structure’, even if the perspective taken is quite different.1 Of central concern are the cognitive mechanisms that underlie certain choices in planning an utterance, and the incremental time-flow of the planning process. A model of production planning synthesizing insights from different strands of research (Garrett 1980, 1988; Levelt 1989, among others), and hence sometimes called the consensus model, differentiates the following stages in the planning of an utterance:2 (1)

(p. 542)

To make this more tangible, let’s consider an example:

(2)

These two utterances constitute two ways of conveying the same fact about the world, and will be interchangeable in many situations. Yet the choice is not arbitrary. Studies from the production planning literature have found that one crucial factor in such word order choices is that speakers tend to realize constituents encoding more active and accessible referents earlier in the sentence (Bock and Irwin 1980; Bock and Warren 1985). These results suggest that speakers should be more likely to use the passive than the active form in (2) than would be expected given the general preference for using active sentences. The reason is that Obama is human, animate, more imageable, and also frequently mentioned in the media (in 2016), and hence more available than the referent or linguistic form of the decision, and more easily planned constituents tend to be realized earlier. The premise of these studies is that the two sentences in (2) actually encode the same message—that is, the same conceptual intent. Given a particular message, accessibility-based word order choices are then necessarily mediated by the grammatical roles that the referents are assigned to. This decision is taken to happen at the functional level. A linguistic account might take a different stance, and posit that the

Page 2 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of New England; date: 17 February 2019

Information Structure and Production Planning choice depends on which constituent is construed as the sentence topic. Topicality and givenness are often assumed to be grammatically encoded and part of the message. The idea that there is a representation of a complete message corresponding to the meaning of a sentence that is devoid of linguistic structure may seem counterintuitive— isn’t the compositional linguistic structure itself the way to generate complex meanings in the first place? Since production planning is taken to be incremental at all levels, however, it is not necessarily assumed in this model that the entire message has to be completely planned before utterance formulation, or at least not in all models of speech production adhering to (1) (Brown-Schmid & Konopka 2008). The aspect of production theories of linguistic choices relevant here is that they usually take the informationstructural import that the choice between (2) has as a direct result of contextual availability on utterance planning, rather than positing that the difference is encoded more directly in the linguistic structure, for example by requiring the subject of a passive to be a sentence topic, as is sometimes assumed. Even if the choice between passive and active is indeed driven by extra-grammatical factors, it is surely a choice between two objects that differ substantially in their grammatical properties: The two arguments of the predicate will be realized with different grammatical functions, the form of the verb changes, an auxiliary is used, and one of the arguments is realized as an optional prepositional phrase. Consider now a different word order effect: Imagine you are asked to name ten politicians from the USA, as fast as possible. Chances are one of the first names you will list (at least in 2016) will be Obama. And yet this word order choice does not affect linguistic structure in the same way as choosing a passive over an active, and is usually not assumed to mark ‘Obama’ as topical by linguistic theories. In the production literature, word order in lists is often assumed to involve choices at the positional rather than the functional level. Both types of choices (p. 543) show accessibility effects, but they differ in interesting ways with respect to the range of factors that affect them (cf. Gleitman et al. 1996, for discussion and references).3 The literature on production planning provides empirical methods to test which planning level is involved in word order choices, a question that is clearly relevant to anyone working on the information structural import of word order variation. If planning considerations at the functional or positional level can account for a particular choice, then accounts which presume differences at the message level—for example the grammatical encoding of topicality—might be simply superfluous. The typical methodologies used in linguistics are often unable to directly pinpoint the underlying source of an effect. Intuitions about acceptability, for example, do not wear their provenance on their sleeves, and are affected by many non-linguistic factors (e.g. by how much a linguistic structure taxes memory resources) (cf. Chomsky 1965; Schütze 2011, among others), and these include the factors at play at the functional and positional planning level according to theories of speech production. Certain information-structural effects might be due to the interaction of memory resources and the planning of linguistic structure (under a speaker-oriented view), or maybe they reflect choices on the side of Page 3 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of New England; date: 17 February 2019

Information Structure and Production Planning the speaker that are intended to facilitate processing on the listener-side (under a listener-oriented view) (see Jaeger 2013, for a recent discussion of whether production biases are purely based on the need of the speaker). The promise of looking at the process of production planning is that it could provide a principled explanation of information structuring (for a look at evidence from perception and reading, see Kaiser, this volume). Two types of phenomena are addressed in this review, contextual effects on prosodic prominence, and choices between different word orders. These are both areas where encoding-level accounts (often assumed in the psycholinguistic literature) and messagelevel accounts (often assumed in theoretical linguistics) compete for the explanation of overlapping sets of data, and where quantitative studies of production studies have already established some of the basic facts and methods.

27.2 Prosodic prominence Linguistic constituents that encode information salient in the discourse are often prosodically reduced (cf. Féry and Ishihara 2010, and references therein). Consider the following contrast: (3)

The word bicycle is likely to be accented in (3a) and likely to remain unaccented in (3b). How exactly does this difference come about? (p. 544)

27.2.1 Anaphoricity or production facilitation? Focus theories such as the alternatives theory of focus (cf. Rooth, this volume) or theories of givenness (cf. Rochement, this volume) treat the lack of accentuation on ‘bicycle’ in (3b) as anaphoric. Just as pronouns are anaphors that require a linguistic antecedent of a certain type, so does deaccentuation require an appropriate antecedent, and this anaphoric requirement is encoded in the meaning of the focus operator whose phonological reflex is the shift in prosodic prominence. With respect to the model in (1), under this view prosodic prominence shifts are assumed to encode an aspect of the message. The semantics of anaphoric contrast is often modelled using alternative semantics (Rooth, 1985, 1992, this volume), which assumes that every constituent has two meanings associated with it, its regular denotation and a set of alternatives, called the ‘focus semantic value’. A prosodic prominence shift requires that the appropriate set Page 4 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of New England; date: 17 February 2019

Information Structure and Production Planning of focus alternatives are contextually salient. In focus theory, prominence shifts act like anaphors, and a pragmatic principle explains why focus and givenness marking are often obligatory when possible. Williams (1997) proposes a principle called Do not overlook anaphoric possibilities; Schwarzschild (1997) postulates a principle called Be attentive; Wagner (2005) and Sauerland (2005) assume the source of the obligatoriness is the principle MAXIMIZE PRESUPPOSITION (Heim 1991); and sometimes more specific constraints are assumed, for example a constraint on question–answer congruence (cf. Rooth 1992). And yet, it is not clear that examples like (3) warrant the introduction of such grammatical machinery. The word bicycle is repeated in the response, and the concept that the word bicycle refers to constitutes old information. This prior activation may more directly be the source of prosodic reduction. I will refer to accounts of this type broadly as ‘accessibility’ theories. A range of studies in the past 30 years have presented evidence that various factors increasing the accessibility of referents or of the linguistic expressions referring to them result in reduced prosodic prominence. Lack of accentuation has been linked to greater prior activation of referents in a speaker’s discourse representation (Nooteboom and Terken 1982; Terken and Hirschberg 1994), or to the activation, priming, or contextual predictability of linguistic expressions referring to them (Nooteboom and Terken 1982; Fowler and Housum 1987; Terken and Nooteboom 1987; Aylett and Turk 2004; Jaeger 2010; Watson 2010), both of which could plausibly be located at the level of lexical selection during functional processing rather than at the message level. Another source of prosodic reduction might be the prior activation of a motor-plan of a particular word or word sequence in cases of repetition (Terken 1984; Lam and Watson 2010; Kahn and Arnold 2012), a type of effect that could plausibly be attributed to the level of phonological encoding (see Arnold and Watson 2015, for a detailed overview of processing effects on prominence). The existence of accessibility effects on prosodic prominence at the encoding level is not controversial: There is general agreement that frequency, predictability, and repetition can have effects on prosodic prominence at least under certain circumstances, (and, in fact, also on word order, cf. Bock 1987). But is it a sufficient explanation for the effects usually discussed in the linguistic literature under the headlines ‘focus’ and ‘givenness’? A challenge for answering this question is that processing explanations and linguistic accounts often make similar predictions, and a lot of the evidence used in the relevant literature to motivate each account does not actually distinguish them (see Wagner 2015, for a detailed discussion). (p. 545)

Consider now the following example (cf. Wagner 2005, 2006; Büring 2008): (4)

Page 5 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of New England; date: 17 February 2019

Information Structure and Production Planning Once again, bicycle is repeated. But would it remain unaccented? If a repetition of bicycle sufficient to license its prosodic reduction, then we would expect a prominence shift, just as is likely in (3b); if a prominence shift does not occur here, it would suggest that it matters that expensive and new are not typically understood as alternatives to each other. Such an effect is unexpected if all that matters is the prior activation of the concept or the word bicycle, but it is expected if a prominence shift actually encodes a semantic contrast at the message level (cf. Wagner 2012a; Katzir 2013; Repp, this volume, for relevant discussions of the notion ‘contrast’). The comparison between (3b) and (4) thus offers a potential testing ground for which level of production planning contextual effects on prosody occur at.

27.2.2 An experiment on prominence shifts in English A production experiment was carried out looking at contrasts like (3b) vs (4), which to my knowledge have not been experimentally tested. What we are interested in here are the conditions under which speakers will spontaneously shift prominence in the target sentence to the adjective when producing these types of utterances. A set of 12 triplets similar to (3) and (4) were created. Participants were told to imagine saying them to a friend in a casual conversation. On each trial they had time to familiarize themselves with the materials. Every participant saw all conditions from each item. However, the presentation order was such we could analyse the data in a latin square design by only using the first 12 trials for each participant: The 36 sentences were partitioned into three playlists of 12 trials with one condition from each item, and four trials in each condition. The stimuli were presented in a pseudo-random order within each playlist such that the same condition was only repeated at most once, otherwise the order was random within each block. The order of the three blocks in turn was randomly varied between participants. The latin square design avoids repetition effects and (p. 546) prevents participants from guessing the purpose of the study. The fully-within-subjectsdesign offers greater power and allows us to test the development of the effect over time. Eighteen native speakers of North American English participated, with an age range between 19 and 31, 12 of whom were female. The recordings were hand-checked for whether speakers actually produced utterances according to script and checked for speech errors, which led to the exclusion of about 5 per cent of the trials. The recordings were forced-aligned using the HTK-based (Young et al. 2006) prosodylab forced aligner (Gorman et al. 2011), which provides a segment-by-segment and word-byword alignment. We then automatically extracted acoustic measures in Praat (Boersma and Weenink 1996) from words of interest. The adjective and the noun in each target sentence in particular were examined. The acoustic measures considered were the maximum intensity over the word, the maximum pitch, and the duration. These measures have proved to be reliable correlates of prosodic prominence in earlier studies (Breen et al. 2010; Wagner 2012b). We computed relative measures of prominence between adjective and noun. In particular, we looked at the difference in log duration (effectively a measure of the duration ratio between the two words), the difference in pitch in Page 6 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of New England; date: 17 February 2019

Information Structure and Production Planning semitones, and the difference in intensity. A shift in prominence toward the adjective would be reflected by an increase in the relative prominence measures. In addition to acoustic analysis, two research assistants were asked to annotate a prominence shift when the noun remained unaccented and the adjective hence carried the last accent of the utterance (Figure 27.1).4 They were only played the final part (e.g. a new bicycle) and hence did not have access to the context. Using the lme4-package in R (Bates and Maechler 2010), we fitted a mixed model regression with Context as fixed effect, and participants and items as random effects which included random slopes for Context (Table 27.1). All models report estimates and standard errors. The determination of p-values in mixed models is not trivial, since the degrees of freedom can only be estimated (Baayen et al. 2008). We report tables including p-values estimates generated with the R-package texreg (Hlavac 2013). The results show that there is a clear acoustic difference in terms of pitch and intensity between the Alternative condition and the other two conditions, but the acoustic measures used here did not detect a difference between the ‘no alternative’ condition in which the head noun was repeated but the adjectives were not mutually exclusive, and the control condition. The perceptual prominence annotation was more sensitive, however. A logistic mixed effects regression with random slopes and intercepts for Context for item and participants found a significant difference both between the Alternative condition and the other two, and between the No Alternative and the Control condition. The model estimates the odds for a prominence shift to be more than ten times higher in the Alternatives condition compared to the other conditions, and estimates the odds for (p. 547) (p. 548) a prominence shift to be four times higher in the No Alternative condition than in the control condition.

Page 7 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of New England; date: 17 February 2019

Information Structure and Production Planning

Figure 27.1 Measures of relative acoustic prominence between adjective and noun. On the left: acoustic measures (> means adjective is more prominent compared to the noun it modifies); on the right: research assistant annotation of prominence shifts to the adjective.

Table 27.1 Linear mixed effect regression model for relative intensity and relative pitch, and logistic mixed effects model for the occurrence of a prominence shift RelIntensity

RelPitch

ProminenceShift

(Intercept)

3.10 (1.38)*

18.93 (6.75)**

1.25 (0.36)***

ContextAlternative.vs.Other

−1.99 (0.43)***

−24.18 (6.86)***

2.51 (0.46)***

ContextNew.vs.NoAlternative

−0.11 (0.43)

−1.82 (6.17)

1.56 (0.79)*

(***) p < 0.001, (**) p < 0.01, (*) p < 0.05 Just as predicted by at least certain versions of the alternatives theory of focus, a prominence shift is much more likely to occur when a mutually exclusive alternative is salient. The results are incompatible with theories that try to explain all prominence shifts with accessibility effects that simply require a linguistic expression or its referent to be salient for it to remain unaccented. The smaller effect showing a tendency of deaccentuation in the case where the head noun was merely repeated but the noun Page 8 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of New England; date: 17 February 2019

Information Structure and Production Planning phrase was not contrastive might suggest that sometimes the adjective was interpreted as contrasting with the contextually mentioned adjective after all, or else it might indicate that there is a smaller but also significant pure repetition effect on accent placement.5

27.2.3 Prominence and production planning: summary The evidence presented here suggests that an account of contextually induced prosodic prominence shifts purely due to planning or retrieval facilitation cannot account for the accent placement pattern in all cases. At least some prominence shifts involve the encoding of pragmatic contrast between mutually exclusive alternatives, and hence form part of the message. But there is no general lesson that we can draw from this, other than that which planning stage(s) of an utterance are involved is an empirical question that we need to address individually for each phenomenon. Many of the factors that affect prosodic prominence also affect the choice of the lexical and morphological form of a referent. Whether a full noun phrase, for example, (p. 549) or a pronoun is chosen, depends on how salient the referent is and whether there are competing salient referents that would fit the features of the pronoun. Again, various processing factors have been identified that affect the choice of the form of referential expressions (Arnold 2010, for a review), many of which relate to the accessibility of the referent. And yet also here, similar to the case of accent placement, it is possible that the choice between a full noun phrase and a pronoun has further grammatical underpinnings. Consider the following observation (cf. Evans 1977): (5)

According to the marriage laws in 1977, the sentence John is married warranted the inference that John has a wife. And yet this inference is not (and was not then) quite sufficient to enable the pronoun she to pick up John’s wife as a referent in (5b). Evans argued that such examples show that there are syntactic conditions on pronominalization, and the choice of pronouns over full phrases might not be entirely reducible to accessibility factors. This type of restriction on pronoun use is an on-going question in current research in semantics (cf. Heim 1990; Schlenker 2011).

27.3 Word order More available or accessible constituents, that is, constituents that are easier to activate or to retrieve from the mental lexicon, have not only been argued to have a tendency to be prosodically reduced, they also tend to be realized earlier in an utterance than material that is less available (Ferreira and Dell 2000, and references therein). This propensity is often taken to be a result of efficient processing in the incremental production of utterances: Once a lexical lemma is activated, it is best to realize it at once, and more costly to keep it in memory while other less available material is processed. Page 9 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of New England; date: 17 February 2019

Information Structure and Production Planning Ferreira and Dell (2000) call this the principle of immediate mention: ‘Production proceeds more efficiently if syntactic structures are used that permit quickly selected lemmas to be mentioned as soon as possible.’ The relative order of constituents is also one of the most important issues in the study of information structure. Time and again it has been observed that given or topical information tends to precede new information, even if the terminology varies widely between different accounts (Halliday 1967; Daneš 1974; Clark and Haviland 1977; Lambrecht 1994, among others); a related observation is that shorter and simpler constituents tend to precede more complex constituents (Hawkins 1994, 2004). These preferences have been observed across many languages, although not without exceptions. How can we tell when such a word order effect is simply a result of the speed with which particular constituents can be planned, or whether it has become part of the grammatical encoding of information structure, as is often assumed for givenness- or (p. 550) topicrelated reordering in the linguistic literature (cf. Neeleman and van de Koot, this volume)? This section will review some of the relevant literature that bears on this question, with a special focus on coordinate structures.

27.3.1 Order in coordinate structures Coordinate structures are interesting in that they can include several referring expressions with the same grammatical and thematic role. Due to the parallelism between coordinates, a number of factors that might otherwise affect word order choices will not play a role here. And yet, the order of coordinates is far from random, and a complex set of tendencies have been observed in the ordering of coordinate structures, starting with Behaghel’s (1909) influential corpus study of coordinate structures. Behaghel (1909) made the important observation that longer and more complex constituents are much more likely to be ordered after shorter and simpler constituents, a tendency he called the Law of Increasing Constituents (‘Gesetz der wachsenden Glieder’). Behaghel established this pattern based on data from modern and historical German, and from Latin, and Ancient Greek, drawn from sources in prose and verse. As Behaghel observed, this tendency is often trumped by overriding factors, for example a particular order might make more sense for conceptual reasons, among which are the chronological order or causal relations between events. Furthermore, an anaphoric dependency to a constituent in another conjunct obligatorily requires a later ordering (cf. Behaghel 1909: 110, for the German equivalent): (6)

Why should more complex constituents be ordered later? Behaghel considered a number of ideas, all of which are still relevant today. He speculated that the later constituents are more easily remembered by a listener, and hence it is more beneficial to the listener if a Page 10 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of New England; date: 17 February 2019

Information Structure and Production Planning complex constituents comes last. But he also speculated, taking the speaker’s perspective, that ordering more complex tasks later might be a general principle of action planning. Behaghel furthermore emphasized the importance of rhythm, and observed that in coordinations of four elements there is a tendency to break it up into a grouping of two times two constituents, and within each a more complex constituent is ordered last. Cooper and Ross (1975) similarly speculate that there must be something ‘adaptive’ about particular orders of coordinates that makes them more likely to be chosen. They focus on ‘frozen’ word orders in idiomatic collocations (e.g. play cat and mouse), and identify a number of semantic and phonological factors that make it more likely that a particular order is observed in a frozen expression that includes considerations of complexity. More specifically, they propose that constituents that are easier to process are ordered first. This hypothesis is made plausible by the observation that the first conjunct is often the one that takes the more self-oriented perspectives (‘Me-first’) of the two (e.g. here and now), (p. 551) since these are comparatively more accessible. Similarly, the preferential ordering of up and related words before their antonyms like down correlates with a processing advantage for upward terms (Seymour 1969, cited after Cooper and Ross cf. Benor and Levy 2006). Such differences in processing complexity might reflect deeper properties of conceptual structure. Cooper and Ross (1975) note that some of the generalizations in freezes (e.g. vertical dimension before horizontal dimension) are mirrored by order preferences in prenominal adjectives and adverbials in sentences. Some other asymmetries observed in coordinate structures (e.g. humans > animals > inanimate entities) play an important role in the grammar of the agreement systems of many languages. And yet, at least for the semantic preferences in English, Cooper and Ross (1975) observe that there seem to be languages that show the exact opposite preference (e.g. in Finnish, distal terms precede proximal ones). Since it is usually assumed that the order of coordinates is free from the point of view of grammatical structure, and we can look at coordinate structures to establish some basic facts about processing-based constituent ordering. Behaghel’s law, for example, certainly seems to be operative outside of coordinate structures, (cf. Wasow 2003; Hawkins 2004).

27.3.2 Pronouns in coordinate structures In many languages, pronouns (and clitics) and other topical and given constituents tend to precede focused and new constituents, although again there are exceptions (see later in this section). This tendency led Behaghel (1932) to posit another law affecting word order (Behaghel’s Second Law): That which is less important (or already known to the listener) is placed before that which is important. Such word order effects are often analysed as the grammatical encoding of information structure, but in fact they might be a simple consequence of incremental production planning, based on the principle of immediate mention discussed in Ferreira and Dell (2000). Coordinate structures allow us to test this idea while holding other factors Page 11 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of New England; date: 17 February 2019

Information Structure and Production Planning constant, for example grammatical role. If greater accessibility leads to earlier production relative to less accessible constituents, pronouns, for example, should be ordered early in coordinations. This preference was indeed observed by Behaghel in his corpus study (Behaghel 1909: 111), and yet it cannot be very strong, since clearly both orders are possible in principle, at least in English:

(7)

Pronouns might be easy to plan and hence tend to be realized early because they do not involve a lot of phonological substance, and/or because they refer to given information. (p. 552) We can test which of these factors is more important by comparing them to full noun phrases that also refer to given information but have more phonological substance:

(8)

The preference predicted by Behaghel is also predicted by more recent theories of how referential expressions are realized, for example Gordon et al. (1999) argue that within complex NPs, constituents with more salient referents should be ordered first. A small production experiment was conducted to test whether there are indeed such ordering effects in coordinate structures, and whether pronouns differ from full noun phrases in this regard. One reason why pronouns and full noun phrases might differ, apart from their phonological content, is an effect called the ‘repeated-name-penalty’: realizing constituents as full noun phrases when the context would have warranted the use of a pronoun leads to a processing cost (Gordon et al. 1993). This effect is observed in subject position, but also in direct object position as in our examples, at least if the subject is not itself anaphoric (Gordon and Chan 1995: 228). Another additional factor is the prosodic realization of given constituents. At least according to Wagner (2006), a pronoun in a coordinate structure is typically accented, while in direct object position a pronoun is typically unaccented. Deaccenting a pronoun in a coordinate structure in fact requires a contrasting antecedent that provides an alternative coordinate structure: (9)

Page 12 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of New England; date: 17 February 2019

Information Structure and Production Planning While Gordon et al. (1999) found evidence that the repeated names penalty also applies in coordinate structures, the extent to which this interacts with word order and prosody has not been investigated. Intuitions about these sentences are subtle, and neither the order effect nor the prominence effect have been tested experimentally. The present experiment tested eight items of the form (7) and (8). These were compared to sentences with an uncoordinated direct object, realized again either as a pronoun or as a full name. The experiment thus had a 2×2×2 design. The order of the given constituent within the coordinate structure was manipulated between participants. The manipulation of referential form (pronouns vs full noun phrase) was within participant, as well as the manipulation of whether the direct object was coordinated or simplex. While we originally planned to run 20 participants in each group, we accidentally ran 26 participants in the first group (given constituent last in the coordinate structure) and 19 in the other (given constituent first in the coordinate structure). The reported results are qualitatively identical irrespective of whether they include the first 19 of the 26 participants to have matching numbers or all were considered. The figures and statistics reported here (p. 553) are based on all participants. Participants produced all conditions from each item in a pseudo-random order, separated by filler trials from an unrelated experiment. At the end of each trial, participants rated the naturalness of the sentence given the context sentence on a 5-point Likert scale. The data were hand-annotated by trained research assistants for whether the given constituent was accented or not.6 Our first set of questions relates to the acceptability of coordinate structures in which one constituent encodes given information. Figure 27.2 summarizes the result. Overall, the differences in ratings where not very big, indicating that as expected, all structures seemed relatively natural to the speakers, with some crucial variation depending on our manipulations. A mixed effects regression model with the ordering of the given constituent (early vs late), referential form (pronoun vs proper name), and the complexity of the NP (coordinated vs single), and the three-way interaction was fitted, which included random slopes for all main effects.7 There was a significant main effect of coordination, such that utterances with coordinated direct objects were rated as less natural than utterances with single direct objects. This could be because one conjunct was previously mentioned and the other was not, which might lead to some oddness, or simply because coordinations are more complex and difficult to process. Of more interest here is that the effect of coordination interacted with word order: when the pronoun was ordered last, the difference between single vs coordinated was significantly greater. This provides direct evidence for Behaghel’s observation that pronouns are preferentially ordered early in coordinate structures. There was no significant interaction between referential form and word order preference, but there was a clear trend for a bigger preference early ordering in the case of pronouns. A significant interaction would have been expected if it was true that accessibility effects (such as the givenness of the referent, that is, the factor relevant for Behaghel’s second law) exert their effect at the functional level via the assignment of grammatical role, while effects of complexity (such as the phonological difference Page 13 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of New England; date: 17 February 2019

Information Structure and Production Planning between pronouns and full noun phrases relevant for Behaghel’s first law) also show effects at the positional level (cf. Gleitman et al. 2007, for discussion, and some evidence to the contrary). There was a significant main effect of referential form, such that a full noun phrase was rated as a less natural realization of the given constituent than a pronoun. This effect interacted with whether or not the given constituent was in a coordination structure or not: using the proper name was significantly more natural relative to using a pronoun in the coordinated case. This provides the evidence that the repeated-names-penalty is significantly smaller in coordinations. This is a new observation, as a prior study looking for such an effect based on reading times failed to find a significant difference (Gordon et al. 1999: 361). The results of the statistical model are summarized in Table 27.2. (p. 554)

Figure 27.2 Acceptability ratings.

Table 27.2 Mixed effect regression model for acceptabilty ratings Coeff (SE) (Intercept)

0.00 (0.13)

PositionEarly.vs.Late

0.06 (0.17)

FormFull.vs.Pronoun

−0.41 (0.11)**

CoorCoordinated.vs.Single

−0.59 (0.14)**

Page 14 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of New England; date: 17 February 2019

Information Structure and Production Planning PositionEarly.vs.Late:FormFull.vs.Pronoun

−0.16 (0.17)

PositionEarly.vs.Late:CoorCoordinated.vs.Single

0.50 (0.17)*

FormFull.vs.Pronoun:CoorCoordinated.vs.Single

0.58 (0.15)**

PositionEarly.vs.Late:FormFull.vs.Pronoun:CoorCoordinated.vs.Single

−0.54 (0.29)

(***) p < 0.001, (**) p < 0.01, * p < 0.05 (p. 555)

Now turning to the prosodic realization of the given constituent: Wagner (2006) predicts that in the absence of a contrasting antecedent in the form of a coordinate structure, pronouns should be

Figure 27.3 Percentage of times prominence was shifted away from the given constituent. Only cases in which the given constituent came second are reported in this plot.

accented when they occur finally in a coordinate structure.8 This is indeed what the participants did

about 90 per cent of the time, both in the case of full proper names and pronouns. Figure 27.3 illustrates the results. Pronouns were essentially always unaccented when they occurred as single arguments, and essentially always accented when they occurred in coordinations. These results are compatible with the idea that the accentuation status of a pronoun is not intrinsic, but rather derived by the same principles that guide the accentuation of constituents more generally. A pronoun requires a unique antecedent matching in person and number features, and that antecedent has to be contextually salient. The conditions on their use will usually also satisfy the conditions of prosodic givenness marking, hence they tend to be unaccented. Pronouns in coordinations behave differently, arguably because of the contrast with their conjunct(s), or due to an interaction of focus marking and syntax: constituents cannot move out of coordinate structures, a fact which leads to interesting predictions about interactions with focus-marking (cf. Wagner 2006, for further discussion).

Page 15 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of New England; date: 17 February 2019

Information Structure and Production Planning Interestingly, even in the uncoordinated case full names referring to contextually salient referents were accented about two-thirds of the time.9 This difference in (p. 556) accentuation rate between given pronouns and given full names in the uncoordinated case lends credence to the idea that it is inherently unlikely for pronouns to be accented after all.10 On the other hand, it could be that proper names differ in the experiment because the choice of a proper name over a pronoun lead speakers to assume that a contrast to some other referent was intended (Table 27.3). Table 27.3 Logistic regression model for prominence placement Coeff (SE) (Intercept)

−0.74 (0.38)

CoorCoordinated.vs.Single

−5.17 (0.71)**

FormFull.vs.Pronoun

−3.44 (0.68)**

CoorCoordinated.vs.Single:FormFull.vs.Pronoun

4.29 (1.35)*

(***) p < 0.001, (**) p < 0.01, * p < 0.05 The choice of a full noun phrase where a pronoun might suggest to a reader that a contrast was intended. This potentially sheds new light on the earlier findings on the repeated-names-penalty: we found the repeated-names penalty to be weaker precisely in the case in which both a pronoun and a full noun phrase are usually accented. There is therefore a correlation between repeated-names penalty and accentuation status. This suggests that the additional processing time a full name incurred in self-paced reading studies (Gordon et al. 1993; Gordon and Chan 1995) might have been due to the processing of alternatives to that word (if indeed it was understood contrastively), or alternatively it might have been due to the processing of prosodic prominence itself, as a result of the implicit prosody assigned by a reader (Bader 1998; Fodor 2002; Kentner 2012). The former interpretation is more likely, since, as argued in Kennison and Gordon (1997), the processing cost associated with the repeated name penalty is not actually exclusively located where the referential expression occurs. To summarize: the experiment found evidence for a preference of pronouns to be ordered early in coordinate structures. Given the plausible account of preferential early ordering in terms of ease of planning of pronouns, no grammatical explanation in the narrow sense seems necessary to account for the word order preference observed in the acceptability judgements. It is clear though that other word order preferences related to pronouns Page 16 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of New England; date: 17 February 2019

Information Structure and Production Planning have become part of grammar in the more narrow sense, be it through grammaticalization (Hopper and Traugott 2003), or some other mechanism. In Romance (p. 557) languages, clitic pronouns often have to precede the verb; this has become part of the grammar of the language, not just because a violation of it leads to complete unacceptability, but because of complex interactions with other factors, for example clitics cannot occur in coordinate structures, and they cannot be dislocated or contrastively stressed. Such ‘grammaticalized’ patterns often mirror independently motivated processing preferences, just as phonetic tendencies that appear to recur across many unrelated languages (e.g. the lengthening of vowels before voiced obstruents) are sometimes phonologized into more categorical and exaggerated patterns that form part of the grammar of some languages (e.g, the lengthening of vowels preceding voiced obstruents in English). The word order preference observed here was rather small, which might or might not be a general property of accessibility-based word order effects. However, the case of coordination production might be special. Let us now turn our attention from coordinate and list structures to some of the recent literature on sentence production.

27.3.3 Word order in sentence production Bock and Irwin (1980), for example, report experimental findings that constituents in sentences are likely to be ordered early if either their referent or their lexical content is contextually given. The method used was sentence recall: participants were read a list of questions and then a list of answers, and afterwards were prompted again with each question and were asked to use the appropriate answer from the answer list. The dependent variable in this paradigm is whether participants use the same word order or a different word order in their response (e.g. by choosing a passive construction instead of an active). Participants tended to switch the order in their own response in order to pronounce given constituents first. While Bock and Irwin (1980) propose an accessibilitybased explanation, given-before-new order effects are often attributed to message-level mechanisms in the theoretical literature (e.g. Kučerová 2007, and references therein).11 The mechanism through which accessibility factors affect word order is often assumed to be attention-based. Accessibility factors might simply affect the likelihood that a perceiver attends to a particular referent at an early point in apprehending an event. This idea that attention is at the heart of accessibility effects was tested in a roundabout way in Tomlin (1997), who reports on several experiments in which the attention is guided by arrows to particular elements in a visually presented a scene they are asked to describe. Most famously, in one experiment participants were shown to choose passive or active descriptions depending on whether their attention was guided to a fish that was eating another fish, or to the one that was eaten, despite of an overall bias against the use of passives in English. Since the arrow in Tomlin’s experiments was visible to the participants, these results have sometimes been attributed to participant’s guesses about the intention (p. 558)

Page 17 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of New England; date: 17 February 2019

Information Structure and Production Planning behind the experiment. But Griffin and Bock (2000) found similar (if less dramatic) results with more sophisticated techniques to guide attention, that left participants unaware of the manipulation. Gleitman et al. (2007), looking at a range of linguistic alternations, showed strong support for Tomlin’s hypothesis using a similar method. Interestingly, Gleitman et al. (2007), also found that without any attention manipulation, early looks to certain elements in a scene was predictive of constituent order in a later utterance describing the event, compatible with the idea that attention to a particular element in a scene leads to earlier planning of correspondent linguistic description, which can then form the starting point for an utterance (but see Griffin and Bock 2000, for evidence to the contrary). The idea that word order choices are due to faster lemma activation of certain constituents (Ferreira and Dell 2000; Gleitman et al. 2007), rather than necessarily happening at the message level where the ‘gist’ of a thought corresponding to an utterance is determined, has an important conceptual advantage: it does not presuppose the potentially problematic idea that the basic message of an utterance can somehow be generated prior to the choice of the linguistic material encoding it, and seems more compatible with the plausible hypothesis that the message itself is only generated by way of incrementally computing the compositional linguistic structure to begin with. The effect of attention on word order led Tomlin (1997) to conjecture that notions such as ‘theme’ and ‘topic’ might become obsolete once the role of attention in utterance planning is properly understood. And yet, there are effects observed in other production studies that will not be reducible to attention alone. Attention has been shown to have an effect of figure–ground perception, for example Vecera et al. (2004) show that guiding a participant’s attention to one side of visually ambiguous stimuli determines which side will be perceived as figure and which as ground. But figure–ground effects have been argued to map to topic–comment structure in linguistic descriptions of scenes, for example it is often argued that subjects are the figure and the predicate the ground in simple SVO predications (Talmy 1978; Gleitman et al. 1996, 2007). Such figure–ground effects have been argued to be absent in lists (Gleitman et al. 1996, 2007). Compatible with the idea that figure–ground is at least sometimes encoded at the message-level after all, Cowles and Ferreira (2011) showed that while contextually given information is more likely to be produced early both in sentences and in lists, a contextually given constituent that is in addition linguistically marked as a topic in a contextual utterance is even more likely to be ordered early in a subsequent production— but crucially, this is only true in sentences, and not in lists. While such findings confirm the role of accessibility in early constituent-placement, Cowles and Ferreira (2011) attribute the additive effect of topicality to topic-marking at the message-level. So it can potentially be argued that message-level encoding of topicality and givenness is needed because sentences seem to differ from lists and coordinate structures in that the latter do not seem to single out constituents as topics. Various linguistic theories posit ‘functional projections’ that encode givenness, focus, or topicality that trigger word order changes (e.g. in ‘cartographic approaches’ Rizzi 1997; Cinque 1999). (p. 559) Constituents Page 18 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of New England; date: 17 February 2019

Information Structure and Production Planning in coordinate structures are usually not assumed to negotiate word order in this way. Evidence for such a difference between sentences and lists was also found in Bock and Warren (1985), who also used the sentence-recall paradigm and showed that constituents which are more easily accessible, in this case more imageable constituents, tend to be ordered early in sentences, but not in lists. Sentences, but not lists, distinguish constituents by their different grammatical role, and grammatical subjects (compared to objects) tend to be more animate, more concrete, more imageable, more salient, more given, receive more attention by the speaker, and tend to refer to entities that the speaker empathizes with to a greater extent (Bock and Warren 1985: 48), and all of these factors also make them more likely sentence topics (Givón 1983). The observed difference between sentences and lists does not necessarily speak for a message-level analysis as topicality marking. The explanation proposed in Bock and Warren (1985) is actually that accessibility-based word order choices are always mediated by grammatical role, and occur at the functional level of grammatical encoding: more accessible constituents are more likely to be mapped to grammatical roles that are ordered early, driving the choice between active and passive and between the doubleobject and the dative constructions in ditransitives. At the positional processing stage, where linear order is fixed and agreement morphology and other types of inflection are added, such accessibility on word order are absent, at least according to some models (cf. also Griffin and Bock 2000). If true, this predicts a difference between sentences and lists, without positing that givenness or topicality are part of the message itself. Under this view, the differences between coordinate structures and sentences is due to differences between different levels of planning rather than to a direct encoding of information structure in grammar. In contrast to Griffin and Bock (2000), other studies have actually found accessibility effects in purely positional word order changes (PratSala and Branigan 2000; Ferreira and Yoshita 2003; Gleitman et al. 2007), casting doubt on this hypothesis. These purely linear effects might, however, be less strong than the ones mediated by grammatical role assignment (Myachykov et al. 2011). This ongoing debate is of relevance to anyone interested in information structure and how it affects word order. The viability of processing explanations for word order variation can be further tested by looking at patterns across languages. A prediction of the incremental models of word order choices discussed here is that word order choices should make production planning easier since they allow the early ordering of a constituent where beneficial, while the alternative that such choices cause competition and hence processing costs predict the opposite. Sometimes, the prediction of the incremental view is borne out (Ferreira 1996, for the case of the English dative alternation), but evidence from Korean suggests this is not always the case (Hwang and Kaiser 2013). While early pronoun placement is a recurring tendency across languages, it is not universal, in Irish and other Gaelic languages, for example, pronouns are often postposed and in fact unacceptable if ordered early (Bennett et al. in press, and references therein). Likewise, Behaghel’s Law of Increasing Constituents, does not seem to hold in some languages that have been reported to show (p. 560) the opposite preference, for example Japanese (Dryer 1980; Page 19 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of New England; date: 17 February 2019

Information Structure and Production Planning Hawkins 1994), a fact often related to headedness (Hawkins 1994, among others). An alternative view is that speakers of these languages favour semantic ordering factors over processing-related ones, and order more content-full material first (e.g. Yamashita and Chang 2001). The division of labour between functional and positional processing levels in word order choices might also vary between languages. Certain languages have been characterized as being ‘non-configurational’ and allowing for much greater word order flexibility compared to others, following insightful work on Warlpiri syntax reported by Hale (1983). However, there has been relatively little work looking at production planning processes in flexible word order languages (but see Prat-Sala and Branigan 2000; Kaiser and Trueswell 2004; Myachykov et al. 2011), and less-described languages more generally (cf. Jaeger and Norcliffe 2009, for an excellent recent overview). It could be that what appear to be purely positional choices in word order in non-configurational (or less configurational languages) are in fact grammatical choices as well. This position has been defended in the syntactic literature by Legate (2001), among others. In some analyses, so called non-configurational languages are in fact discourse-configurational (cf. É. Kiss 1995), and notions such as focus and topic are represented in syntax through focus and topic operators. The relevant linguistic literature only rarely tries to rule out alternative production-based explanations of the information-structural import of word order choices, leaving much room for further studies that try to combine insights from both domains. In sum, whether or to what extent the information-structural import of word order variability is due to grammar or to processing considerations is often not obvious, but is by no means simply a terminological question, since the two views can make very different predictions. This also applies to the information-structural import of dislocation constructions (López, this volume; Neeleman and van de Koot, this volume). Does a particular type of left dislocation encode topicality, or does it simply correlate with, say, topicality because speakers use these constructions in order to be able to realize more accessible constituents first?

27.4 Looking forward This review focused on two particular aspects of utterances: prominence and word order. By looking at two specific aspects of utterance planning we saw that the questions of which level of planning is involved and whether grammatical structure is necessary to capture a phenomenon are not questions of taste or theoretical affinity, they are empirical questions. The literature on production planning provides insights and methods that help to tackle these questions. Most importantly for linguistic research, there is no a priori answer for what constitutes part of the message and what is a result of the process of encoding the message, and when trying to understand information structure, both aspects of an utterance will likely be relevant.

Page 20 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of New England; date: 17 February 2019

Information Structure and Production Planning Many areas of the study of production planning have not been addressed here, for example the study of speech errors (cf. Fromkin 1971; Garrett 1988), disfluencies (cf. Clark and Fox Tree 2002), turn-taking in spontaneous conversations (cf. Levinson 2013), or issues of the exact incremental time course of planning at different levels in the production model (cf. Wheeldon 2012). We also did not scrutinize certain assumptions built into the consensus model in (1), which would clearly merit extensive discussion, for example the precise relationship between message and linguistic structure. This review of how production planning relates to information structure should therefore be seen as a starting point for further exploration rather than a comprehensive overview. (p. 561)

Notes: (1) This chapter benefited tremendously from two insightful anonymous reviews and comments by the editors. The experimental results reported in this chapter were first presented at a conference on alternative semantics at Nantes in 2010; thanks for comments to the audience. Also thanks to members of prosody.lab for help in conducting the experiment and annotating the data, in particular Lizzy Smith, Lauren Mak, and Erin Olson. (2) It is important to note that message here is a technical term—of course choices at the functional level or positional level can be interpreted by listeners as meaningful and hence form part of the message in the intuitive sense of the word. (3) Linear order in coordinate structures plays an important role in the computation of presuppositions, but this fact is usually attributed to incremental interpretation rather than grammatical structure itself. (4) The annotation scheme is based on the assumption that the grammatically relevant aspect for prosodic focus marking is relative prominence, rather than accentuation of individual constituents (cf. Wagner 2005). (5) In order to control against learning or strategy effects, a regression model with trial order as a covariate was fit, not repeated here for space reasons, which did not show any effects other than a change of the effect on pitch over time. An analysis of the first third of the trials of all participants—effectively analysing the experiment under a latinsquare design in which every participant only sees one condition from each item—yielded qualitatively the same pattern as the results reported in 1. In other words, there is no hint in the data that the results might be an artefact of the experimental design. (6) I will not discuss acoustic measures here for reasons of space. (7) Interaction terms were excluded form the random slopes since the model did not otherwise converge. It is controversial whether data based on Likert scales should be analysed using parametric statistical methods. We also analysed the data using ordinal logistic mixed effects regression, and found the relevant comparisons to be significant as well. Page 21 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of New England; date: 17 February 2019

Information Structure and Production Planning (8) When they are ordered early, they are also predicted to be accented, but we focus on the cases where they are ordered last here, since the difference in placing an accent or not is much more salient for the final word. (9) The ratings for utterances with accented and unaccented uncoordinated proper names were not different from each other. There was actually an effect in the coordinated case, such that accented full noun phrases were rated as less natural. This is arguably due to the fact that, in this case, the very same lexical word was accented twice at the end of consecutive sentences, triggering a ‘givenness illusion’ effect (Wagner 2012b). (10) Brenier et al. (2006) conversely found that nouns tend to be accented, and that in fact most of the variability in accent placement in their corpus study was predicted by the likelihood of particular words to carry an accent. (11) An earlier ordering might also decrease prominence. Some theories directly relate prosodic prominence modulation and word order choices as two strategies to achieve similar goals. I will not discuss interactions between the two in this review.

Michael Wagner

Michael Wagner is Associate Professor in the Department of Linguistics at McGill University, and Canada Research Chair for Speech and Language Processing. His research interests encompass everything surrounding sentence prosody, the phonetic and phonological shape it takes as well as the semantic, syntactic, and pragmatic factors that it encodes. His contribution was made possible by fellowship from the Alexander von Humboldt Foundation during a stay at Goethe University Frankfurt.

Page 22 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of New England; date: 17 February 2019

Information Structure in First Language Acquisition

Oxford Handbooks Online Information Structure in First Language Acquisition   Barbara Höhle, Frauke Berger, and Antje Sauermann The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara Print Publication Date: Jul 2016 Subject: Linguistics, Language Acquisition Online Publication Date: Oct 2014 DOI: 10.1093/oxfordhb/9780199642670.013.011

Abstract and Keywords So far, research on the acquisition of information structure (IS) is still relatively sparse compared to other areas of first language acquisition research. A growing interest in this area has emerged with an increasing number of results indicating an asymmetry of an early production but late comprehension of linguistic means related to IS—which contrasts common findings in other areas of acquisition with comprehension skills typically being in advance to production skills and thus provides a challenge for theories of language acquisition. This chapter will give a review on recent findings on the acquisition of linguistic means to mark IS in first language acquisition focusing on studies looking at the production and comprehension of accentuation and word order. A further part will look at children’s production and comprehension of sentences with focus particles. Keywords: language acquisition, prosody, word order, focus particles

28.1 Introduction SINCE some emerging interest in the early 1970s of the last century, research on how children acquire linguistic means of encoding information structure (IS) has been neglected for almost 30 years.1 Hence, the current knowledge about the developmental trajectory of IS and the mechanisms that are relevant for its acquisition are rather fragmentary and incomplete. However, what available findings suggest is that—compared to other domains of language acquisition such as syntax, phonology, or the lexicon—the development of the production and comprehension of linguistic markers of information structure seems to be a rather long process in first language acquisition. The fairly slow developmental progress may be caused by the fact that producing and understanding Page 1 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 18 January 2019

Information Structure in First Language Acquisition linguistic means of encoding IS in a communicatively adequate way not only requires grammatical and pragmatic knowledge but is also based on an advanced system of socialcognitive abilities. Communication makes use of a common ground (Stalnaker 2002) between the interlocutors, the contents of which change within a discourse—thus the interlocutors have to keep track of which information they can assume to be shared by hearer and speaker and which information is new. This is an essential social-cognitive skill that children need to develop to adapt their language performance to a specific communication setting in an adequate way. Basic cognitive concepts of IS are available to children from very early on. All behavioural research on infants’ speech perception with habituation or familiarization techniques is based on the fact that even newborns respond differently to old and new information (cf. Schmitz and Höhle 2007). Children’s selection of the information to be explicitly expressed in their earliest multi-word utterances is also affected by IS as topics are typically missing in their utterances (Müller et al. 2009). Furthermore children create expectations about other persons’ behaviour based on their assumptions about the other persons’ knowledge state and can already adjust their own behaviour to the knowledge status of another person in their second year of life (O’Neill 1996; Onishi and Baillargeon 2005). However, the mastery of how IS is encoded in a given language occurs as a developmental challenge and seems to be a function of the complexity of the specific means that a language uses to mark IS, and their interactions with and interfaces to other modules of the linguistic and cognitive system. The acquisition of IS as one aspect of the acquisition of pragmatic skills seems to be characterized by an atypical developmental asymmetry: while in other areas of acquisition comprehension skills can typically be observed before children produce corresponding structures, pragmatic acquisition shows instances of production obviously preceding comprehension (for a review, see Hendriks and Koster 2010). An open question is whether these asymmetries indicate that the child’s underlying mental system is different for production and comprehension or whether these asymmetries—especially findings of a delayed comprehension—reflect methodological difficulties in assessing children’s comprehension abilities in the area of pragmatic competence (cf. the discussion in Brandt-Kobele and Höhle 2010). These production–comprehension asymmetries together with the observation that IS has early impacts on children’s verbal and non-verbal behaviour suggest that at least some categories of IS are available to the child from early on but that the specific instantiation of marking IS in a given language may pose problems in the acquisition process. Our chapter focuses on three areas that are subject to a recent growth in research activities: the acquisition of prosodic, syntactic, and lexical means that are related to the encoding of givenness, focus, and topic. Due to space restrictions, other areas of IS acquisition such as narrative competence and anaphoric reference are not considered in this chapter (for a recent review, see Berman 2009). (p. 563)

Page 2 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 18 January 2019

Information Structure in First Language Acquisition

28.2 Acquiring means of encoding information structure in child language 28.2.1 Acquisition of prosodic markers of information structure 28.2.1.1 Studies on children’s production Past work on prosodic marking of IS in child language has investigated phonological and phonetic means of marking focus, topic, and information status. So far, acquisition (p. 564) data from only a handful of languages (English, Dutch, German, Italian, Hungarian) are available. Most of the work has been done on children’s utterances that were elicited in picture description tasks or in sentence imitation tasks. Analyses of spontaneous speech are rather rare in this field. Wieman (1976) observed that 2-year-old English-speaking children already highlight a focused constituent by accentuation in their spontaneously produced two-word utterances. Hornby and Hass (1970) found that when describing a sequence of two pictures differing in only one depicted entity (e.g. a boy riding a bike and a girl riding a bike), 3- to 4-year-old English-speaking children frequently put an accent on the word referring to the new information in the description of the second picture (e.g. a GIRL is riding a bike). These results were supported in a further study by Hornby (1971) that showed that school-aged English-speaking children consistently used accentuation to mark new information with no further developmental progress between the ages of 6 and 10 years. MacWhinney and Bates (1978) compared English-, Italian-, and Hungarianspeaking 3- to 5-year-old children’s performance in a task in which they had to describe a series of three pictures in which one element always changed. They found that Englishspeaking children consistently accented the new information in their utterances while the Italian- and Hungarian-speaking children made less use of accentuation. Since Italian and Hungarian are languages with a much freer word order, accentuation may be less associated with IS in these languages than in English. Thus, the cross-linguistic differences observed in this study suggest that the language-specific ways of encoding IS are reflected in children’s utterances from early on. More recent studies have elaborated this early research in several aspects. First, different categories of IS, like focus and topic, and different types of focus have been included. Second, the types of accents and their phonetic realization have been analysed in more detail than in the previous studies, which mainly relied on impressionistic ratings of the accentuation patterns that children used. Chen (2009, 2011) investigated the accent types that Dutch-speaking children use for focus- or topic-marking in utterances that were elicited in a sentence imitation task in which the sentences to be imitated were presented as answers to subject- or object-questions by a robot voice. Word order was varied by using subject- as well as object-initial sentences. Most importantly, the stimulus Page 3 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 18 January 2019

Information Structure in First Language Acquisition sentences had been spliced together from words recorded in isolation such that the children were expected to impose their own sentence-level prosody on their imitations. The majority of the 4- to 5-year-olds most frequently accented both topic and focus by means of the same type of accent (H*L) in the sentence-initial position. However, in the sentence-final position focused constituents were more often accented than topics, but as in the initial position the types of accents for topics and focus did not differ. Older children (7- to 8-year-olds) again mostly accented both topic and focus in sentence-initial position but with different accent types. In contrast to the younger children, they mostly produced the sentence-final topic without an accent. According to Chen, the pattern of the older children was already closer to an adult pattern, but there were still differences in the kind of accents used—however data from adults using the same experimental procedure are lacking. De Ruiter (2010) compared the pitch accents on sentence-initial topics in German-speaking children’s and adults’ narratives elicited by a picture story-telling task. She found that adults mainly realized the topic with a rising accent. The 5-year-olds and 7-year-olds used the same inventory of accent types as adults but with a different distribution—especially with a greater amount of monotonal H* accents. Furthermore, the information status (new, given, accessible) had effects on accentuation in both adults and children. Across the ages, given information was mostly deaccented while new and accessible information was more often accented. Again, children mostly used H* accents for new and accessible information while rising accents were dominant in the adults’ utterances. Phonetic analysis showed two main differences across age: children’s utterances showed smaller pitch excursions, lower speed of pitch change and an earlier alignment of pitch minima. This suggests that the differences in the distribution of the accent types between children and adults may be related to constraints on children’s still maturing speech planning and control (c.f. Kent 1976). (p. 565)

Wonnacott and Watson (2008) studied the acoustic correlates of accentuation in subject nouns produced by 4-year-old English-speaking children in three IS conditions: new (i.e. not previously mentioned), given-non-shift (i.e. mentioned previously in the same grammatical role), and given-shift (i.e. mentioned previously but in a different grammatical role). They found that the nouns were produced with a higher maximal pitch and a higher intensity in the new and given-shift conditions than in the given-non-shift condition, but no differences between the new and the given-shift condition were found. This pattern mirrored findings from English-speaking adults by Watson, Arnold, and Tanenhaus (2005, reported in Wonnacott and Watson 2008), but in contrast to the adults, children did not show differences in durational measures across the conditions. Sauermann et al. (2011) analysed the phonetic realization of subjects and objects in sentences that were elicited by the imitation of answers to questions that elicited a broad, narrow or corrective focus on the crucial element. The stimulus sentences were presented by a robot voice with a flat intonation contour. German-speaking adults as well as 4-year-old children did not differentiate the objects with pitch or durational cues depending on the three conditions. Differences between the children and the adults Page 4 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 18 January 2019

Information Structure in First Language Acquisition emerged for the subjects: whereas in the children’s utterances narrow and corrective focus showed different pitch patterns than broad focus, the adults realized corrective focus with more enhanced pitch cues than broad focus and narrow focus. Thus for children, narrow and corrective focus patterned together and were set apart from broad focus, but in adults narrow and broad focus patterned together and were set apart from corrective focus. The available findings suggest that prosodic highlighting of new or contrastive information is present in children’s utterances from early on, that is, the cognitive competence to highlight information that is most relevant seems to be present. However, the fine-tuning to an adult-like use of different accent types and their phonetic realizations seems to be a developmental process that takes some time. Several factors may contribute to this. According to the current state of knowledge, there seems to be no one-to-one mapping between type of accentuation and IS category in the adult system: the (apparently) same IS category can be realized by different types of accents and the same type of accent can occur with different IS categories. Furthermore, even for adult language the (p. 566) relation of accentual patterns to different categories of IS is far from being clear even in the restricted set of languages that are covered by the existing developmental research. However, the lack of a model of the adult language renders the evaluation of the patterns observed in children impossible with respect to development. Even when control data from adults are available, sources of potential deviations in children may be manifold. In addition to the form–function mapping, the production of a coherent prosodic structure with different types of accents needs a fast and highly automatized system of motoric planning and control, which still may be constrained in young children. Future research is necessary to uncover a more detailed picture of the developmental trajectories and the interplay of different capacities in children’s prosodic marking of IS.

28.2.1.2 Studies on children’s comprehension Testing children’s processing of accentuation that goes beyond the ability to discriminate different prosodic patterns provides an even stronger methodological challenge than studying children’s production of prosody. Whereas the competence to perceive prosodic salience is present from the first months of life on (Sansavini et al. 1997; Schmitz et al. 2006), the use of accentuation in sentence interpretation seems to be hard for children and to lag behind their production skills. Some early studies (Hornby 1971; Cruttenden 1985) used picture selection tasks to test whether 6- to 10-year-old English-speaking children can exploit accent patterns to identify contrastive information or the topic– comment structure of a sentence. In both studies even 10-year-olds did not show the expected responses; however both studies are methodologically disputable as either no adults were included or sets of highly heterogeneous experimental materials were used. However, interestingly Hornby (1971) reports that the children in his study used accentuation to mark the topic–comment structure in their own productions despite the fact that they did not exploit the accent patterns for sentence interpretation.

Page 5 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 18 January 2019

Information Structure in First Language Acquisition To investigate the potential production–comprehension asymmetry, Wells et al. (2004) tested English-speaking 5- and 13-year-old children in the receptive as well as in the productive mode. In the production part of the study, sentences with a contrastive focus were elicited by letting the child correct an utterance from the experimenter. Overall, the children accentuated the focused constituents in the majority of their utterances without any developmental change across the ages. The receptive task required the identification of the referent of an accented constituent by pointing to a picture. Performance in the identification task was much lower than in the production study: the youngest children did not perform above chance level and a clear increase in performance occurred with increasing age. More recent studies using experimental methods that provide a measure of children’s processing of prosodic information during sentence comprehension indicate a much earlier capacity to make use of prosodic markers of IS. In an eye-tracking study, Arnold (2008) found an adult-like effect of accentuation as a cue for information status in English-speaking children as young as 4- and 5 years old. The children were presented with visual displays of four different objects. Two of the objects had labels with phonologically identical initial segments (e.g. bagel—bacon). The children were first instructed to (p. 567) move one of the objects and then they were instructed to move another object or the same object again (e.g. Put the bacon on the star. Now put the bagel/bacon on the square). The crucial word in the second sentence was either accented or non-accented. Remarkably, in the same-object condition adults as well as children fixated the target object (bacon in the example above) more when it was unaccented than when it was accented. This suggests that children and adults have the expectation that an unaccented word should refer to a referent that was previously introduced in the context. However, they did not find any effect of accentuation when the second sentence referred to the previously unmentioned competitor object (bagel). These results suggest that hearers do not necessarily take accented information as new but that they do take nonaccented information as given. Again using the eye-tracking technique, Höhle et al. (2009) provided evidence that 3- to 4year-old German-speaking children make use of accentuation to identify the sentence focus. They exploited the fact that in German, sentences with the additive focus particle ALSO placed after the finite verb are ambiguous with respect to the scope of the particle. In the test sentences used in the experiment this ambiguity was resolved by accentuation: if the particle itself carries an accent it associates with the sentence subject, if not it can associate with the sentence object by taking scope over the VP. The gaze patterns of the children to a visual display that depicted alternatives to the subject and the object of the sentences showed that children’s eye gaze was directed by the accentuation patterns with more looks to the respective alternative sets. Sekerina and Trueswell (2012) found evidence that Russian-speaking 5- to 6-year-old children can make use of accentuation to assign a contrastive reading to adjective-noun DPs (the RED fox vs the red FOX), again in an eye-tracking study. In contrast to adults,

Page 6 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 18 January 2019

Information Structure in First Language Acquisition children’s responses to prosodic information interacted with the syntactic construction: unlike adults, children only showed an effect of prosody in constructions in which the adjective and the noun occurred adjacent to each other. However, on-line tasks do not per se provide evidence that young children can exploit accentuation. Cutler and Swinney (1987) failed to find an adult-like advantage for detecting an accented word in a sentence (compared to its unaccented occurrence) in English-speaking children under the age of 6 years. The existing research only allows very limited conclusions on children’s abilities to exploit prosodic information in understanding IS. Results supporting the assumption of a late mastery stem from a time when more sophisticated methods that allow on-line measurements in children were not yet available. Given the recent experimental improvements, future research will hopefully produce a clearer picture of the assumed production–comprehension asymmetry.

28.2.2 Acquisition of syntactic markers of information structure IS may influence the syntactic realization of a sentence in terms of reordering of constituents (e.g. topicalization) or in terms of the choice/use of specific constructions (p. 568) (e.g. clefts). Studies on the production of word order variation revealed an irregular pattern of results: some studies provided evidence that 2-year-old children are sensitive to IS factors influencing word order (Penner et al. 2000; De Cat 2009) while others showed that even older children do not perform in an adult-like way (Barbier 2000; Schaeffer 2000; Narasimhan and Dimroth 2008; Dimroth and Narasimhan 2012; Mykhaylyk 2012). The discrepancy between children and adults has been discussed in terms of pragmatic (Schaeffer 2000) or cognitive limitations (De Cat 2009). In addition, comparing the outcomes of different studies on word order that are not identical with respect to their methodological approaches is challenging as a number of different factors have been identified as influencing word order, for example prosody, length of constituents, type of referring expression, and animacy. It is hardly possible to disentangle the strength of their impact and their potential interactions based on spontaneous speech data. Therefore, experimental approaches that restrict the type of utterances to be produced are extremely important in this area of research. Studies on children’s comprehension of word order variation have demonstrated that children usually have difficulties comprehending non-canonical constructions, especially when these constructions are presented without a context (Strohner and Nelson 1974; Lempert and Kinsbourne 1978, 1980; Dittmar et al. 2008). IS may ease the comprehension of these constructions (Gourley and Catlin 1978; Grünloh et al. 2011), but this may not apply to all constructions equally well (Gourley and Catlin 1978).

Page 7 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 18 January 2019

Information Structure in First Language Acquisition 28.2.2.1 Studies on children’s production Much research on word order in children’s language production has studied the impact of givenness but so far the results do not provide a coherent picture. While adults usually prefer to place given before new information (Narasimhan and Dimroth 2008; Stephens 2010), some studies have found that children have the same preference (Hickmann et al. 1996; Stephens 2010; de Marneffe et al. 2012; Anderssen et al. 2012a) while others have found a new-before-given preference (Bates 1976; Narasimhan and Dimroth 2008; Dimroth and Narasimhan 2012). This discrepancy in the children’s results may arise from the fact that word order is not only influenced by the givenness of a referent but also by the type of referring expression (e.g. pronoun vs lexical NP) used to refer to it (cf. MacWhinney and Bates 1978; Hickmann et al. 1996). Usually givenness and the choice of the referring expression are highly correlated (Chafe 1976; Gundel et al. 1993). Similarly to adults, 3-year-old children tend to use pronouns and definite expressions for given referents (Emslie and Stevenson 1981; Campbell et al. 2000), and to use different pronoun types to reflect differences in the givenness status (Gundel et al. 2007; Bittner et al. 2011). However, in contrast to adults, children overgeneralize pronouns and definite expressions to referents that are new or not uniquely identifiable by the hearer, and thus they do not necessarily adjust their choice of referring expressions to the common ground (Maratsos 1976; Campbell et al. 2000; Schaeffer and Matthewson 2005). Accordingly, the choice of the referring expression may provide (p. 569) the crucial cue for word order in both cases, when it is correlated with givenness and especially when it is not correlated with givenness. Elicited production studies investigating the dative alternation in Norwegian and English indicate that 4- to 5-year-old children are sensitive to the impact of givenness and the choice of the referring expression on word order (Stephens 2010; Anderssen et al. 2012a; de Marneffe et al. 2012). In both languages, children tended to use the prepositional construction which places the direct object (theme) in front of the indirect object if the direct object was given or pronominal (Peter gave [the book/it]given to her.). Givenness of the indirect object had a weaker effect on the choice of the construction: in Norwegian given indirect objects tended to be dropped while in English a restriction on the ordering of pronominal arguments (*/?Peter gave her it. vs Peter gave it to her.) already seemed to be operative in the children (Stephens 2010). The English dative alternation was also the subject of an analysis of spontaneous speech from 2- to 5-year-old children and their mothers by de Marneffe et al. (2012). By employing a statistical method to disentangle the effects of givenness and the choice of the referring expression, they found that the choice of the referring expression influenced word order in children and adults in the same way. Givenness had no significant effect on word order in adults, but in children givenness influenced the ordering of pronominal objects. In the children’s data pronominal direct objects tended to follow the indirect object when they were new (Peter gave her it) but preceded the indirect object when they were given (Peter gave it to her). This difference between children and adults may result from the observation that adults

Page 8 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 18 January 2019

Information Structure in First Language Acquisition in contrast to children rarely produced pronominal objects that were new and from the fact that the order (*Peter gave her it) is marked in adult language but not necessarily in child language (but see Stephens 2010). In contrast to these data, experimental studies that investigated the impact of givenness on the ordering of elicited full noun phrases in coordinated constructions (e.g. a spoon and an apple) revealed that 3- to 5-year-old German-speaking children follow a newbefore-given preference while 9-year-old children and adults follow a given-before-new preference (Narasimhan and Dimroth 2008; Dimroth and Narasimhan 2012). These results on the phrasal level contrast with those on the clausal level mentioned above. Accordingly, it is far from clear whether 3- to 5-year-old children have a given-before-new or new-before-given preference, whether these preferences differ at the clausal and phrasal levels, and how they interact with factors like the choice of the referring expression. The observation that children typically produce the comment but omit topics in their early multi-word utterances can be considered as evidence for an early sensitivity to the topic–comment structure of utterances (Gruber 1967). In experimental studies topicality has been found to influence children’s production of passive sentences. Elicited production studies demonstrated that English-speaking kindergarten and school-age children were more likely to produce passive sentences when the patient was the topic (Turner and Rommetveit 1967). Demuth (1989) could show that in Sesotho, 2-year-old children produce and comprehend passive sentences. In particular, they appear to be sensitive to the impact of topicality on passive and active sentences because, in Sesotho, (p. 570) the agent (subject) in active sentences has to be the topic and passive sentences are obligatorily required when the patient is the topic. Hornby (1971) investigated the impact of different devices for marking the topic– comment structure in comprehension and production in English-speaking 6- to 10-yearold children. Accent was the preferred means of marking the topic–comment structure across all age groups. However, with growing age topic–comment marking via the use of passives, clefts and pseudo-clefts increased. Hornby’s results suggest that syntactic devices may not be the ‘preferred’ means of marking topicality in English-speaking children. Other studies show that topicality also influences children’s choice of the referring expression, which in turn may influence word order. De Cat (2009) conducted an elicited production experiment to investigate the realization of topics and focus in Frenchspeaking children. Topic and focus were modified using a setting with two pictures. In topic contexts, the first picture introduced three agents involved in the same activity, and the second picture showed each agent performing a different action. Children were asked to state what the characters were doing in the second picture. In the focus condition, the first picture introduced a scene with an actor performing an action, whereas the second picture showed the same scene with a different actor. Children were asked to describe what was happening in the second picture. De Cat found that even 2;6-year-old children Page 9 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 18 January 2019

Information Structure in First Language Acquisition tended to realize topics as dislocated NPs, subject clitics or null subjects and focused referents as heavy pronouns or parts of cleft constructions. This indicates that children were sensitive to differences between topics and focus, even though they still overgeneralized clitics to referents in the topic context with three possible referents. This further indicates that children may not have problems with topics or the relation between pronouns and topic status but rather do not consider the presence of the other (salient) referents. Anderssen et al. (2012b) demonstrated that different pronoun types and their IS status influence the object shift in Norwegian-speaking children. In Norwegian, topical pronominal objects are subject to an obligatory object shift whereas nominal objects and contrastive, possessive, and indefinite pronominal objects are not shifted. 4- to 5- and 6to 7-year-old children never shifted non-topical pronominal objects. Whereas the older children shifted topical pronominal objects most of the time, the 4- to 5-year-old children did so only rarely. This developmental delay, however, may result from the fact that 4- to 5-year-olds may not have fully acquired the Norwegian object shift yet (see Anderssen et al. 2010). Research investigating the impact of focus on word order is sparse. Research with German-speaking children revealed that focus may influence the ordering of arguments relative to the position of focus or negation particles in the spontaneous speech of 2-yearold children (Penner et al. 2000). With respect to more complex constructions, that is, the ordering of double objects in ditransitive sentences, a recent elicited imitation study by Höhle et al. (2013) showed that 5-year-old German-speaking children repeated the sentences more often literally when the order of the objects followed the backgroundbefore-focus order with the focused constituent occurring in the default (p. 571) clausefinal position. Whether this is also an effect of the default nuclear-accent position in German is a question for further research. The preference to place the nuclear accent towards the end of a sentence may also play a role in the results of a study investigating the production of clefts in 4- to 10-year-old French-speaking children (Hupet and Tilmant 1989). Visual context was used to elicit corrective focus on either the agent or the patient of an action. Children produced more clefts that focused the agent than clefts that focused the patient. When corrective focus was on the patient, children preferred to use contrastive intonation rather than the cleft construction. Accordingly, the use of the agent-cleft may be seen as a way to avoid accent on the first constituent in the canonical word order.

Page 10 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 18 January 2019

Information Structure in First Language Acquisition 28.2.2.2 Studies on children’s comprehension The impact of IS-related word order on children’s sentence comprehension has hardly been studied. Early studies investigated the topic identification in sentences in which the topic–comment structure was marked syntactically or prosodically (e.g. Hornby 1971; MacWhinney and Price 1980). Hornby (1971) showed that 6-year-old English-speaking children are sensitive to the topic–comment structure in clefts. Topic identification in pseudo-clefts and passives increased with age and all sentence types tested provided a clear topic–comment effect in 10-year-olds. MacWhinney and Price (1980) used a similar procedure to Hornby (1971) but employed a more controlled setting to investigate the comprehension of topic–comment marking in cleft constructions. However, they did not find that 6- to 7-year-old English-speaking children correctly identified the topic in these sentences. Other studies have investigated the impact of IS on the comprehension of non-canonical word orders. In general, children have difficulties comprehending constructions in which the order of the thematic relations does not follow the agent-before-patient order (Strohner and Nelson 1974; Lempert and Kinsbourne 1978, 1980; Dittmar et al. 2008). Modification of the IS may ease the comprehension of non-canonical constructions (e.g. Gourley and Catlin 1978; Grünloh et al. 2011), but this may not always be the case (e.g. Paul 1985; see also Gourley and Catlin 1978). Gourley and Catlin (1978) investigated how givenness influences the comprehension of different sentence types (active, passive, agent-cleft, patient-cleft, dative alternation) in 4to 7-year-old English-speaking children by presenting the sentences in isolation or in an appropriate or inappropriate verbal context. The comprehension of passive sentences and patient-cleft sentences was enhanced when the constructions were presented in an appropriate context that met the givenness and newness restrictions of the constructions compared to constructions presented in an inappropriate context. That is, a passive sentence (Ernie was pushed by BERT) was more likely to be understood correctly following a context sentence that mentioned the patient (Ernie skated down the sidewalk) rather than following a context sentence that mentioned the agent (Bert skated down the sidewalk). Patient-cleft sentences (It was SUSAN that Gordon hit) were comprehended better when they followed a context that already mentioned the agent (Gordon hit someone) rather than the patient (Someone hit Susan). These results indicate that children (p. 572) were sensitive to the givenness/newness requirements of the constructions. However, IS did not influence the comprehension of the dative alternation with an overall lower performance for the double object construction than for the prepositional construction. This lack of a context effect seems to disagree with the results from the production studies that showed that givenness has an impact on the construction choice in children. Finally, Grünloh et al. (2011) demonstrated that contrastive pitch accent on the sentenceinitial direct object enhanced the comprehension of non-canonical OVS sentences in 5-

Page 11 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 18 January 2019

Information Structure in First Language Acquisition year-old German-speaking children. This indicates that children are sensitive to the (contextual and prosodic) factors that may license OVS sentences. Nevertheless, further research is required to investigate the contextual and possibly prosodic factors that may license or trigger the non-canonical word orders and in turn may ease the comprehension of non-canonical sentences. This research will also clarify whether there is a comprehension–production asymmetry for word order variation, especially with respect to the incoherent results from the production studies.

28.2.3 Acquisition of focus-sensitive particles Previous research in this area mainly addresses the acquisition of the restrictive focussensitive particle ONLY but also includes some studies on additive particles like ALSO, EVEN, and ANOTHER. Acquisition data exist for various languages such as Cantonese

(Lee 2003)

Dutch

(Drozd and van Loosbroek 1998; Bergsma 2002, 2006; Hulk 2003; Szendrői 2004)

English

(Crain et al. 1992; Crain et al. 1994; Halbert et al. 1995; Philip and Lynch 2000; Paterson et al. 2003; Paterson et al. 2006; Kim 2011)

European Portuguese

(Costa and Szendrői 2006; Santos 2006)

French

(Kail 1978, 1979; Hulk 2003; Benazzo et al. 2004; Gayraud 2004; Benazzo et al. 2012)

German

(Penner et al. 1999; Nederstigt 2003, 2006; Benazzo et al. 2004; Hüttner et al. 2004; Höhle et al. 2009; Müller et al. 2009; Müller et al. 2011a, 2011b; Berger and Höhle 2012; Müller 2012)

Japanese

(Endo 2004; Matsuoka 2004; Matsuoka et al. 2006; Ito 2012)

Mandarin

(Liu 2009; Notley et al. 2009; Zhou and Crain 2010; Zhou et al. 2012)

Norwegian

(Philip 1999)

Polish

(Benazzo et al. 2004)

Page 12 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 18 January 2019

Information Structure in First Language Acquisition Most of these studies, however, comprise the investigation of a single expression in a specific language, and are restricted either to its receptive or productive acquisition. When also considering the diversity of empirical methods used in these studies, the existing heterogeneity of data, results, and conclusions across studies is not surprising. (p. 573) Crucially, cross-linguistic comparisons regarding a specific particle as well as comparisons between children’s acquisition of different focus particles within one language can thus only cautiously be drawn at this point.

28.2.3.1 Studies on children’s production Across languages ALSO and ONLY tend to be part of children’s lexicons quite early in time. Spontaneous speech data from children speaking Cantonese (Lee 2003) and Japanese (Matsuoka et al. 2006) reveal that across these languages, children tend to start producing the additive particle ALSO around their second birthday, often at least a few months before their first production of the restrictive particle ONLY. Lee (2003) assumes this temporal precedence to reflect an ontogenetic priority of existential quantification, which he assumes to be involved in additive focus structures, over universal quantification, which is considered to be a way to analyse restrictive focus structures. The finding of an early production of ALSO is corroborated by spontaneous and experimental data from German-speaking children (Penner et al. 1999; Nederstigt 2003, 2006; Höhle et al. 2009; Müller et al. 2009) and spontaneous data from French- and Dutch-speaking children (Hulk 2003; Gayraud 2004). In addition, it holds true for other additive particles like EVEN, AGAIN, and ANOTHER in various languages (Nederstigt 2003, Gayraud 2004, Liu 2009).

28.2.3.2 Studies on children’s comprehension In order to arrive at an adult-like interpretation of a sentence containing a focus-sensitive particle, children must be able to consider lexical, syntactic, prosodic, discursive, and situational aspects of the utterance. Crucially, due to the interface character of focus interpretation there also has to be knowledge about how information in these different domains interacts. In most available comprehension studies children displayed patterns of performance that were different from those of adult controls. However, children’s performance in experiments that test the comprehension of a certain particle varies to a high degree across studies. Complicating things, it also seems to be the case that even adult controls’ performance is heterogeneous across these studies, indicating that performance in this field might be highly influenced by the specific design of the experimental tasks administered to participants.

(i) Children’s interpretation of sentences containing ONLY A first area of research concerns children’s interpretation of sentences in which ONLY takes scope over the subject or the VP—as prompted by different sentence positions of the particle. One of the earliest findings concerns children’s interpretation of English sentences containing ONLY in presubject position, as in Only the cat is holding a flag. Crain et al. (1992) conducted a sentence–picture verification task with 3- to 6-year-old English-speaking children. In this type of task participants have to decide whether or not Page 13 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 18 January 2019

Information Structure in First Language Acquisition a sentence is a true or false description of a picture that comes along with it. A subgroup of the children failed to interpret presubject ONLY-sentences in an adult-like fashion, but performed in a more adult-like way with sentences containing preverbal ONLY as in (p. 574) The cat is only holding a flag. Subsequently, this asymmetry has been observed in a wide a range of L1-studies covering various languages (e.g. for German: Müller et al. 2011b; Müller 2012; for English: Crain et al. 1994; Kim 2011; Philip and Lynch 2000; Notley et al. 2009; for Japanese: Endo 2004; for Mandarin: Notley et al. 2009; Zhou and Crain 2010). Most of the researchers agree with a syntactic explanation for this pattern, initially brought up by Crain et al. (1992). They assume that children misanalyse SVO-sentences with presubject ONLY as if the particle also took scope over the VP. More specifically, Zhou and Crain (2010) as well as Notley et al. (2009) claim that unlike adults’ grammar, children’s grammar would allow presubject ONLY to be a sentential adverb which ccommands not only the subject but all the rest of the sentence, too. This assumption gains strength from the fact that Zhou and Crain (2010) were able to show that 4-year-old Mandarin-speaking children give up this type of interpretation when tested on negated Mandarin sentences equivalent to Only the white dog didn’t climb up the big tree. Here, the intervening negation takes scope over the VP and is assumed to block the association of the particle to the VP. Notley et al. (2009) consider the VP/object as the default associate of ONLY in SVO-languages because it constitutes the canonical locus for new information and then bears the focal accent. A slightly different explanation of the same pattern of performance has recently been put forward by Müller et al. (2011b) and Müller (2012). They attribute German-speaking 4and 6-year-olds’ poor performance on interpreting SVO-sentences with presubject ONLY to the ambiguous status of the sentence subject. They argue that the (sentence-initial) subject is likely to be the topic of sentences whereas the (sentence-final) object is the focus of sentences (cf. Costa 1998; Molnár 1991). Adults have a preference for partitioning sentences this way (Frazier 1999). However, sentences with subjectassociated ONLY are in conflict with this default presumption because they contain a focused subject. Children are assumed to adhere to the topic default more strictly than adults and therefore do not consider the subject as an eligible associate of the focussensitive ONLY. Crucially, contrary to the first account described earlier, the latter one does not necessarily imply that children would interpret sentences with presubject ONLY as if ONLY associates with the VP/object. Müller (2012) argues that the meaning contribution of presubject ONLY might just as well be ignored. A third account put forth by Paterson et al. (2003) in fact places a strong emphasis on this possibility of children’s failure to interpret the meaning contribution of ONLY. More specifically, they argue that children have issues with instantiating the relevant set of alternatives unless it is explicitly introduced in the experimental task. As a result, sentences containing ONLY may be interpreted in the same way as sentences without the particle. To support their argument, Paterson et al. (2003) tested English-speaking children from age 4 onwards and adults on their interpretation of sentences such as Only Page 14 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 18 January 2019

Information Structure in First Language Acquisition the fireman is holding a hose and The fireman is only holding a hose as well as their counterpart The fireman is holding a hose without the particle. While adults accepted a more restricted set of pictures as matching a sentence with ONLY compared to a sentence without the particle, children until school age often accepted the same set of pictures as matching both sentence types. The account implies that interpreting a (p. 575) sentence with ONLY restrictively depends on the degree to which the contrastive set is made salient within the given context. Indeed, in various studies carried out in different languages such as German (Müller et al. 2011b; Berger and Höhle 2012; Müller 2012), Dutch (Bergsma 2002) and English (Gualmini et al. 2003; Kim 2011), children’s interpretation of sentences containing ONLY improved when a verbal context provided an explicit introduction of the distinct sets of alternatives (see also Clark and Amaral (2010) for a discussion of the contextual relevance of particles in experiments that test their comprehension). In depth investigations of children’s interpretation of preverbal ONLY in English sentences constitute the second research area. Without a non-disambiguating context, these sentences remain ambiguous, even in spoken form: with a focal accent on the object, ONLY can either associate solely with the object or with the entire VP. It has been observed that children’s interpretation preference for this structure deviates from adults’. Crain et al. (1994) showed that adults tend to associate ONLY with the VP-internal object in sentences such as The dinosaur is only painting a house. In contrast, in a sentence– picture verification task, 4-year-olds rejected the sentence by referring to the fact that the dinosaur in the picture was involved in another action in addition to painting a house, indicating that ONLY associates with the VP. Crain et al. (1994) argue that adults’ and children’s differing interpretations are motivated by adherence to different linguistic principles: unlike adults, children’s interpretation is constrained by the Semantic Subset Principle. When faced with ambiguous structures, this principle initially prompts them to always assign the reading that entails the other one, which in this case is the VP-reading. Data from Paterson et al. (2006) corroborates a VP-association of preverbal ONLY by 7- to 8-year-old English-speaking children, but in this study, adults’ pattern of responses also indicated a preference for the VP-reading. Paterson et al. take this to challenge the assumption that children’s interpretation of sentences with preverbal ONLY is initially guided by the Semantic Subset Principle. Since adults’ performance on ONLY-sentences is identical to that displayed by the children in this study, the authors assume it to be more likely that both are led by the same general processing principles. In summary, children’s as well as adults’ association constraints and/or preferences for English preverbal ONLY and the factors influencing them are far from being conclusively determined. A final area of research comprises the investigation of children’s interpretation of ONLY when it takes VP-scope in ditransitives, such as The farmer only sold a banana to Snow White. As the particle c-commands the whole VP, it could either associate with the direct object or the VP/indirect object. In spoken sentences, this ambiguity is resolved by prosodic means. In a task in which the children had to judge a statement about a Page 15 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 18 January 2019

Information Structure in First Language Acquisition previously verbally presented and simultaneously acted-out story (Truth Value Judgement Task, Crain and Thornton 1998), Gualmini et al. (2003) showed that English-speaking adults considered contrastive accent on one of the objects as an unequivocal cue that relates that object with ONLY. However, 4- to 5-year-old children in this study mostly associated the particle with the indirect object in sentence-final position, irrespective (p. 576) of accent placement. Gualmini et al. (2003) assume that children cannot use accentuation as a reliable cue to resolve semantic ambiguity and therefore resort to a default interpretation in favour of the indirect object. Szendrői (2004) and Costa and Szendrői (2006) also found that 3- to 5-year-old Dutch-speaking and 4- to 6-year-old European-Portuguese-speaking children interpreted sentences with contrastive accent on the direct object in a non-adult-like way and argue for a general VP-association of ONLY in ditransitive constructions. Although the results of these studies indicate a remarkably consistent pattern of non-adult-like performance, results of an eye-tracking study of Zhou et al. (2012) with 4-year-old Mandarin speakers indicate that children in fact do use focus accentuation in an adult-like way when processing a sentence with ONLY. Note however that the study investigated the association of presubject ONLY with the full subject-NP or just its modifier rather than preverbal ONLY in ditransitives.

(ii) Children’s interpretation of sentences containing ALSO and EVEN Children’s interpretation of ALSO and EVEN has only recently stimulated stronger research interest (Hüttner et al. 2004; Matsuoka 2004; Bergsma 2006; Matsuoka et al. 2006; Höhle et al. 2009; Liu 2009; Kim 2011; Berger and Höhle 2012; Ito 2012; for some earlier studies on French-speaking children, see Kail 1978, 1979). Studies on children’s interpretation of sentences containing EVEN are rare to this point (Kim 2011; Ito 2012) and mainly focus on the question of whether children are able to compute the scalar reading. Studies addressing children’s interpretation of ALSO pursue the question of whether children have difficulties with identifying the particle’s subject- or VP/object-associate. Several of these studies deal with Dutch or German. Note that in Dutch and German SVOsentences, the subject- as well as the VP/object-associated ALSO can appear in the same position (S V ALSO O). In this case they can only be disambiguated by different focal accent placements on either the particle or the object (e.g. Peter has ALSO a ball / Peter has also a BALL). Bergsma (2006) and Hüttner et al. (2004) investigated children’s interpretation of corresponding sentences in picture selection tasks, in which children had to choose a sentence-matching picture from a set of three. Bergsma observed that Dutch-speaking children at least until school age associated the additive particle in S V ALSO O with the subject or the object randomly. The author assumes this to be caused by children’s lack of competence in using the accent patterns as a reliable means of inferring the particle’s associate. Hüttner et al. (2004) in contrast found that 2- to 7-yearold German-speaking children often matched both kinds of ALSO-sentences with the picture which corresponded to the subject-association of ALSO. The authors assume this performance to reflect a default interpretation in favour of subject-associated ALSO. Interestingly, this default follows the opposite direction to that assumed for ONLY. Page 16 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 18 January 2019

Information Structure in First Language Acquisition However, neither Bergsma’s nor Hüttner et al.’s assumption was corroborated by data from an eye-tracking study in which German-speaking 3- to 4-year-old children just had to look and listen (Höhle et al. 2009, see section 28.2.1.2). Sentences containing subjectand object-associated ALSO both accompanied one and the same picture which matched with both types of test sentences. The picture displayed the two entities (p. 577) expressed by the subject and the object of the test sentence as well as a respective alternative for each of them. The gaze data revealed that in both kinds of ALSO-sentences the children considered the sentence-relevant set of alternatives. Corresponding sentences containing no particle did not trigger diverging looking behaviour. Thus, these results indicated that children must have generally been on the right track in the interpretation of both types of ALSO-sentences. A second line of research addresses the fact that children’s poor performance in traditional picture selection tasks and sentence verification tasks seems to persist even when the task becomes contextually enriched by adding a context story which guarantees the verbal introduction of the sets of alternatives (e.g. Matsuoka 2004; Bergsma 2006; Matsuoka et al. 2006; Liu 2009). Children’s poor performance in these studies is compatible with the claim that they neglect the contribution of the particle. Since 4- to 6year-old Japanese-speaking children in a Truth Value Judgement Task performed better on sentences with ONLY than with ALSO (Matsuoka et al. 2006), the cause for the poor performance on ALSO is likely to be related to a specific characteristic of the additive particle. In this connection it is interesting that the restriction of alternatives expressed by ONLY has an impact on a different sentence component than the addition of alternatives expressed by ALSO: ONLY contributes to the sentence assertion and therefore affects the truth conditions of an utterance, while ALSO merely gives rise to an additional presupposition and thus solely affects the felicity conditions of an utterance. Hence, Bergsma (2006) relates 4- to 6-year-olds’ massive failure to interpret sentences containing ALSO to the presuppositional status of the particle. In a similar vein, Liu (2009) attributes 4- to 6-year-old Mandarin-speaking children’s poor performance on sentences with ALSO in a Truth Value Judgement Task to children’s failure to accommodate the presupposition that is triggered by ALSO. Considering this argument, Berger and Höhle (2012) introduced a new experimental task to the field, which tests the interpretation of VP/object-associated ALSO-sentences without the need to accommodate the presupposition, as the presupposed content has been uttered in the preceding discourse and thus is part of the common ground. In this task the child was instructed to reward a toy animal when it had accomplished two tasks (e.g. eating an apple and a banana) but not when only one of the tasks had been accomplished. The rewarding decision could only be based on the animal’s answer to a question posed by the experimenter (EXP: You’ve surely eaten the BANANA, ANIMAL: I’ve (also) eaten the APPLE) as the child could not observe the animal’s actions. In addition, and in contrast to conventional sentence comprehension tasks based on acceptance or rejection of a sentence as a match to a picture, the task did not require the rejection of a sentence based on a presupposition failure while its assertion matches the picture. Furthermore, the children’s attention was implicitly drawn to the importance of the Page 17 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 18 January 2019

Information Structure in First Language Acquisition ALSO-triggered presupposition for solving the experimental task. Berger and Höhle (2012) showed that German-speaking 3-year-olds already interpret sentences containing VP/object-associated ALSO in an additive fashion, that is, in a less restrictive manner than their counterparts without a particle. Hence Berger and Höhle (2012) rule out that children up to school age generally tend to ignore the presuppositional meaning (p. 578) contribution of ALSO in sentence interpretation. Rather, they argue that, in order to test the comprehension of (the presupposition triggered by) ALSO in a felicitous, ecologically valid way, experimental tasks must give consideration to its presuppositional status and the characteristics and challenges that come along with it. Summing up, although production studies indicate an early mastery of both additive and restrictive particle use, results of comprehension studies suggest that children have issues when interpreting sentences containing them. Some researchers assume children’s problems to be related to the linking of the particle to a different associate than adults. Other researchers conjecture that children’s non-adult-like task performance is rather related to problems in considering the particles’ contribution to the sentence interpretation. For both accounts, various developmental explanations have been suggested, including linguistic knowledge as well as processing accounts. While hardly any of them address the issue, they all in fact need to come up with a developmental explanation of the apparent production–comprehension asymmetry, too. Alternatively, there is reason to assume that poor comprehension might be task dependent. Future research on a potential production–comprehension asymmetry should thus devote more attention to the differences between the conditions in which children can successfully demonstrate their understanding of sentences with focus particles and those in which they fail.

Page 18 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 18 January 2019

Information Structure in First Language Acquisition

28.3 Summary and outlook Even though our review was limited to three areas of the acquisition of IS, that is, the acquisition of prosodic and syntactic means of marking the discourse status of referents and the acquisition of focus particles, the findings do not yet provide a comprehensive picture of the developmental trajectory of IS—especially in the first two areas considered. A common observation in all three areas that we examined is that there is some concordant evidence that at least 3-year-olds use prosodic, syntactic, and lexical means that are related to IS in their language production but that comprehension of the same means is difficult to detect in children of the same age. There are several options to account for this apparent developmental gap. First, the linguistic competence underlying the early productive capacities can be overestimated. This approach has been taken by Cutler and Swinney (1987) with respect to the asymmetry of the early production of prosodic prominence and its late effect on children’s sentence processing. They consider the missing effect of focus intonation on young children’s word processing as evidence that the development of representations underlying the correct use of pragmatic means is a rather late process that continues far into school age. Based on a hypothesis put forward by Bolinger (1983), they suggest that children’s earlier ability to prosodically highlight important information in their language production does not reflect the availability of underlying pragmatic representations. Rather it results from a physiologically based (p. 579) mechanism that automatically leads the speaker to produce those parts of an utterance with a higher acoustic salience that he or she is excited about (not taking into account the common ground). Our review of children’s production capacities is compatible with this view as the findings confirm an early availability of basic means of highlighting information in children. However, the more detailed phonological and phonetic analyses reveal that the fine-tuning of the realization of prosodic prominence to the system of the specific language seems to be a longer lasting process. For comprehension, this fine-tuning may be required before the information provided by prosody can be integrated adequately into the processing of a sentence in a specific language. Future—especially cross-linguistic—research should aim at a more comprehensive picture of the contribution of potential universal mechanisms of highlighting important information and of their language-specific shaping during the process of language acquisition. A similar approach may also hold for the acquisition of focus particles. The early appearance of focus particles in children’s utterances—which so far has been mainly studied only on the basis of spontaneous speech data—does not allow for a conclusive picture of children’s knowledge underlying their use. It may be the case that the advanced skills that have been tested in most of the studies on children’s comprehension of focus particles go beyond the requirements that are needed to produce them in the short utterances that are typical for young children.

Page 19 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 18 January 2019

Information Structure in First Language Acquisition An alternative approach would suggest that the apparent developmental gap between production and comprehension is caused by an underestimation of children’s comprehension capacities due to the use of imperfect experimental paradigms in many studies (cf. Brandt-Kobele and Höhle 2010). Several indications from previous research point to this. Many of the typical off-line tasks used for studying language comprehension in children require skills such as action planning, decision making, and reasoning, skills that are subsumed under the heading of cognitive control, the maturation of which is still in progress in preschool children (for recent findings on a relation between cognitive control and sentence comprehension, see Minai et al. 2012). Outcomes of studies on language comprehension may be dependent on the maturation of these skills, which may obscure the linguistic skills of the children. Evidence for this assumption comes from the fact that studies using eye-tracking as a rather automatic response without much involvement of cognitive control provide evidence for the processing of prosody as a marker of information structure in much younger children than in studies that have used tasks like picture selection. The importance of task demands is also highlighted by the recent study by Berger and Höhle (2012), which showed the capacity to correctly interpret sentences with focus particles in children as young as 3 years by using an experimental paradigm that highlighted the necessity of taking the particle into account for a successful achievement of the task. Furthermore, comprehension of non-canonical structures can be enhanced in children when the sentences are embedded in a verbal context that renders them felicitous. This makes clear that the development of experimental paradigms to study the comprehension of IS is an especially challenging domain that needs further improvement. The observation that even adults—if they have been (p. 580) included at all—do not show homogeneous performance patterns in many experiments looking at IS is a largely overlooked fact in this area of research. However, recent developments in the experimental methodologies that allow for a more reliable comparison of adults’ and children’s data from the same experimental method offer hope for future advances.

Notes: (1) The work on this chapter has been supported by a DFG grant to Barbara Höhle as part of the Collaborative Research Centre 632 Information Structure: The linguistic means for structuring utterances, sentences, and texts.

Barbara Höhle

Barbara Höhle is a Professor of Psycholinguistics at the Department of Linguistics, University of Potsdam. Her field of research is first language acquisition with a main focus on early acquisition of phonology and syntax. Her work has appeared in numerous journals including Journal of Child Language and Lingua. Frauke Berger

Page 20 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 18 January 2019

Information Structure in First Language Acquisition Frauke Berger is a researcher at the ‘Information Structure’ collaborative research centre at the University of Potsdam. Her research includes experimental work on the acquisition of information structure and presupposition triggers. Her work has appeared in Journal of Child Language and Language Acquisition. Antje Sauermann

Antje Sauermann is researcher at the ‘Information Structure’ collaborative research centre at the University of Potsdam. Her research focuses on the acquisition and processing of information structure and word order. Her work has appeared in journals including Second Language Research and Language and Cognitive Processes.

Page 21 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 18 January 2019

Towards a Neurobiology of Information Structure

Oxford Handbooks Online Towards a Neurobiology of Information Structure   Ina Bornkessel-Schlesewsky and Petra B. Schumacher The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara Print Publication Date: Jul 2016 Subject: Linguistics, Discourse Analysis, Morphology and Syntax Online Publication Date: Jun 2016 DOI: 10.1093/oxfordhb/9780199642670.013.22

Abstract and Keywords This chapter reviews neurophysiological and neuroanatomical investigations of information structural notions, with a view to working towards a neurobiologically grounded perspective. It first considers components of a neurobiologically plausible theory of information structure and outlines candidate mechanisms for higher-order cognitive processing, namely prediction and mental modelling, attention orientation, memory, and inferencing. The chapter then proceeds to neuroscientific investigations of information structure, highlighting differences between sentence- and text-level processing and discussing findings for the information structural notions of givenness, focus, and topic, before presenting further insights from syntax-induced information structural effects. The chapter concludes with a discussion of neurobiological models of information structure processing. Keywords: event-related potentials, functional neuroanatomy, givenness, topic, focus, prediction, attention, memory, inference

NEUROLINGUISTIC

investigations have traditionally sought to situate cognitive

representations or mechanisms in the human brain.1 Accordingly, readers of this chapter will likely expect to learn about the ‘neural correlates’ of basic information structural (IS) notions—essentially, how these notions can be mapped onto the brain in terms of functional neuroanatomy (‘space’) and neurophysiology (‘time’). In part, this chapter will fulfil those expectations. It will, however, also disappoint them to some degree by arguing that we need to move beyond a ‘mapping’-based perspective if we are to truly understand how the human brain implements IS processing. What do we mean by ‘neural implementation’? Obviously, we cannot hope to trace the processing of IS to the level of individual neurons—or even circumscribed neuronal populations—and synapses. Nevertheless, we can attempt to understand how the complex Page 1 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: Freie Universitaet Berlin; date: 16 April 2018

Towards a Neurobiology of Information Structure information processing machinery that must underlie linguistic information packaging is grounded in more basic computational functions of the human brain. In this sense, the current chapter subscribes to the recent movement that strives to work towards neurobiologically plausible models of language in the brain (Small 2008). To rephrase the question: Can we ascribe a neurobiological reality to IS categories in the sense that they are processed by specific brain circuits or accompanied by specific neurophysiological processes? Probably not. But can we expect IS to manifest itself in some meaningful way in the workings of the human brain? Almost certainly. Here, we argue that IS impacts upon several fundamental neural mechanisms, including the (p. 582) prediction of upcoming input, the orienting of attention, and the updating of internal models about the world. In this sense, IS is intertwined with the neural implementation of linguistic and non-linguistic information processing in a number of basic and important ways. This chapter thus aims to undertake a first step towards a neurobiologically grounded perspective on IS. The remainder of the chapter is organized as follows. Section 29.1 discusses considerations for a neurobiologically plausible theory of IS and briefly describes the current state-of-the-art with regard to neuroscientific methods for studying higher-order language. Subsequently, we turn to differences between sentence- and text-level processing (Section 29.2) and basic IS notions: givenness (Section 29.3), focus (Section 29.4), topic (Section 29.5), and further notions at the IS–syntax interface (Section 29.6). Finally, Section 29.7 discusses current neurolinguistic and neurobiological models of IS processing.

29.1 Candidate neurobiological processing mechanisms and how to study them empirically In his seminal paper on basic IS notions, Krifka (2008: 243) notes: ‘The basic notions of Information Structure (IS), such as Focus, Topic, and Givenness, are not simple observational terms. As scientific notions, they are rooted in theory, in this case, in theories of how communication works’. A neurolinguistic theory of IS must necessarily go one step further: it must aspire to explain how communication is grounded in the workings of the human brain. In the following, we outline several neurobiologically plausible mechanisms which may underlie the neural processing of IS (Sections 29.1.1– 29.1.4). We then provide a brief discussion of how such mechanisms can be studied empirically (Section 29.1.5). Readers who are unfamiliar with neuroscientific methods such as functional Magnetic Resonance Imaging (fMRI) and Event-related Brain Potentials (ERPs) may find it useful to skip ahead to Section 29.1.5 prior to reading Sections 29.1.1–29.1.4.

Page 2 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: Freie Universitaet Berlin; date: 16 April 2018

Towards a Neurobiology of Information Structure

29.1.1 The perception–action cycle, internal models, and prediction In the neurosciences, the ‘purpose’ of the human brain is widely held to be that it allows us to successfully interact with the world around us. It enables us to perceive the world in several sensory modalities and to act within and upon it by means of motor commands (movement). In this sense, the brain continually implements a perception–action cycle (Fuster 1997): sensory input leads to behaviour and, conversely, behaviour—and the motor commands implementing it—predicts upcoming sensory input (e.g. if I move my vocal tract in a particular way, I will expect a speech signal as a sensory consequence of (p. 583) that movement). At a first glance, the perception–action cycle appears to have little to do with IS. Closer consideration reveals, however, that this cycle is central to communication: sensory input (e.g. an interlocutor’s speech input) leads to a particular action (one’s own response), which in turn leads to an expected self-generated sensory input and, possibly, to an expected response by the interlocutor. (For a detailed theory of dialogue in this spirit, see Pickering and Garrod (2013)). If we follow Krifka (2008) in assuming that IS notions should be grounded in theories of communication, the perception–action cycle constitutes a central notion with respect to its neural implementation. The perception–action cycle also provides the basis for predictive processing, which, according to a growing consensus in cognitive neuroscience, may constitute a basic, unifying principle of brain function (Schultz et al. 2005; Friston 2010). In this context, it is typically assumed that we rely upon an internal model of the world to predict upcoming sensory input. When the prediction does not match the input actually encountered, a ‘prediction error’ results and the model must be adjusted. A mechanism of this type was first proposed in the domain of motor control (Wolpert and Ghahramani 2000). Here, given a motor command (e.g. to move one’s hand and arm to grasp a glass), the context in which it is executed and the current state of the effector, an internal forward model predicts the upcoming effector state and its expected sensory consequences (e.g. the sensation of grasping the glass). The prediction error detected via sensory feedback is used to correct the current state estimate. A similar type of architecture has been proposed for action observation, social interactions, and even Theory of Mind (Wolpert et al. 2003; Koster-Hale and Saxe 2013), but also for language processing. As first suggested by Pickering and Garrod (2007, 2013), internal forward models may explain the rapidity of language comprehension. Here, we will argue that IS provides a powerful means of modulating predictions during language processing. The most well-known neural signature of predictive coding is reduced activity in sensory cortices in response to predicted stimuli (Blakemore et al. 1998). For example, activation in auditory cortex and early electrical/magnetic responses of the brain are reduced in response to self-generated speech in comparison to the same speech signal played back from an external source or distorted (Curio et al. 2000; Fu et al. 2006; Zheng et al. 2009). These characteristics also hold for higher-order language processing, for example for the prediction of particular word categories (Dikker et al. 2009, 2010), semantic classes such Page 3 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: Freie Universitaet Berlin; date: 16 April 2018

Towards a Neurobiology of Information Structure as animacy (Szewczyk and Schriefers 2013) and perhaps even individual words (DeLong et al. 2005). As we discuss in Sections 29.2–29.4, very similar correlates of predictive processing have been observed in the domain of IS.

29.1.2 Attentional orienting In spite of the centrality of the perception–action cycle, there is a limit to our uptake of sensory input: we cannot take in all of the information that is physically present in the world around us. Rather (e.g. while attending a visual scene), our expectations and current goals lead us to focus on some aspects over others. According to a prominent (p. 584) model (Corbetta and Shulman 2002), this top-down attentional selectivity is based on cognitive processes and demands and is controlled by a parieto-frontal brain system (the dorsal attention network, DAN). However, depending on the nature of the missed information (e.g. a snake under a log), attentional selectivity can prove dangerous. To provide for such eventualities, the brain contains a second, temporoparieto-frontal system (the ventral attention network, VAN), which can serve as a ‘circuit breaker’ to the dorsal attention system: ‘Unexpected, novel, salient and potentially dangerous events take high priority in the brain, and are processed at the expense of ongoing behaviour and neural activity’ (Corbetta and Shulman 2002: 201). Thus, while the DAN provides top-down orienting of attention, the VAN allows for bottom-up reorienting in the case of unexpected, but sufficiently salient stimuli. Attentional orienting constitutes a basic requirement for key aspects of communication. Intuitively, it is related to the notion of information packaging, defined as ‘ways in which a speaker accommodates his speech to temporary states of the addressee’s mind’ (Chafe 1976: 28) or ‘the tailoring of an utterance by a sender to meet the particular assumed needs of the intended receiver’, reflecting ‘the sender’s hypotheses about the receiver’s assumptions and beliefs and strategies’ (Prince 1981: 224). This focus on temporary states of the receiver/addressee—rather than long-term knowledge—parallels our attending of certain aspects of the world around us in accordance with current goals and strategies. At the same time, the notion that the sender/speaker accommodates to the needs of the receiver/addressee is indicative of the possibility to influence attentional orienting in accordance with current communicative aims. In this context, Krifka’s notion of common ground (CG) management, which Krifka views as central to an ecologically valid theory of CG also appears useful. It ‘is concerned with the way the CG content should develop’ (Krifka 2008: 246) and is shaped by communicative goals. In accordance with these considerations, and as described below, several IS-related neural correlates of language processing may be related to the notion of attentional orienting. A particularly important candidate in this regard is the P300 ERP component. The P300, a positive ERP deflection with a typical peak latency of approximately 300 ms, but also with much latency variability, is generally considered a component family. For language processing, perhaps the most relevant P300 instance is the parietallydistributed P3b, which was originally reported for unexpected or uncertain events (Sutton Page 4 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: Freie Universitaet Berlin; date: 16 April 2018

Towards a Neurobiology of Information Structure et al. 1965). However, it is also elicited by subjectively salient events such as hearing one’s own name (Gray et al. 2004) or one’s mobile phone ring tone (Roye et al. 2007). In language, the P3b has been reported for predictable, salient target events (e.g. the expected antonym in sentences like The opposite of black is white) (Roehm et al. 2007; Molinaro and Carreiras 2010; Vespignani et al. 2010). There has been substantial discussion as to whether the P600—a well-known language-related ERP response that was originally reported for structural anomalies and sentences requiring syntactic reanalysis (Osterhout and Holcomb 1992, 1993; Hagoort et al. 1993)—may be an instance of the P300 (Gunter et al. 1997; Osterhout et al. 1996; Coulson et al. 1998; Osterhout and Hagoort 1999; Frisch et al. 2003; Kretzschmar 2010; Bornkessel-Schlesewsky et al. 2011; Sassenhagen et al. 2014; Sassenhagen and Bornkessel-Schlesewsky, 2015). The P600 is (p. 585) relevant to discourse comprehension as it has also been reported for the processing of new referents and indirect anaphors (Burkhardt 2006; Schumacher 2009) and for non-literal utterances (Regel et al. 2011; Schumacher 2014). A recent, physiologically plausible theory of the P3 posits that the P3 results from a systemic Norepinephrine (NE) release by the Locus Coeruleus (LC; a brainstem nucleus) (Nieuwenhuis et al. 2005). The LC/NE theory of the P3 assumes that the NE release increases cortical gain—that is, increases the response to critical (currently relevant) stimuli and attenuates the response to non-critical stimuli—thus allowing for a transition into a cortical state that facilitates an adequate behavioural response to the current stimulus (e.g. facilitation of response execution, inhibition of a prepotent response, etc.). In the context of the LC/NE theory, the P3 can be viewed as a correlate of VAN reorienting (Sara and Bouret 2012).

29.1.3 Memory The cognitive notion of ‘memory’ is of central importance to a conceptualization of IS processing. For example, CG and givenness both refer to knowledge that can be attributed to one’s interlocutor, that is, knowledge that must somehow be represented in his/her memory. Depending on the nature of the knowledge under consideration, this representation may either be transient or long term. Similarly, following Reinhart (1981), Krifka characterizes the topic constituent as ‘identif[ying] the entity or set of entities under which the information expressed in the comment constituent should be stored in the CG content’ (Krifka 2008: 265). This definition presupposes the notion of ‘storage’, which is closely affiliated with memory in a cognitive sense. From a neurobiological perspective, however, memory and ‘storage’ are notoriously difficult to define. Consider, for example, the well-established finding that conceptual information is represented in a distributed fashion in the brain (Martin 2007; Patterson et al. 2007), that is, different aspects of meaning are associated with differing neuronal populations. As such, it is difficult to conceptualize storage in terms of a classical file-card model: in neural terms, it is not clear what would constitute the equivalent of a file card and how ‘access’ to such a card might be implemented.

Page 5 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: Freie Universitaet Berlin; date: 16 April 2018

Towards a Neurobiology of Information Structure Indeed, it has recently been proposed that the N400 event-related brain potential component, which is perhaps the most well-known, purported neurolinguistic correlate of access to a ‘mental lexicon’ or ‘semantic memory’, does not readily map onto the types of representations and processes (e.g. post-lexical vs sub-lexical processing) posited in traditional psycholinguistic frameworks (Kutas and Federmeier 2011). Rather, Kutas and Federmeier assume that the N400 reflects a temporal interval in which unimodal sensory percepts are transformed into multimodal conceptual associations. This claim is based partly on the observation that the N400 is highly stable in terms of peak latency—in contrast to other ERP components such as the P3, the latency of which is know to vary with stimulus complexity. Hence, in their view, the N400 does not reflect the activation of a word’s meaning or access to its lexical entry but ‘reflect[s]‌ the activity in a multimodal (p. 586) long-term memory system that is induced by a given input stimulus during a delimited time window as meaning is dynamically constructed’ (Kutas and Federmeier 2011: 640).

29.1.4 Inferencing Discourse relations clearly involve more than givenness, topic, and focus. A further important aspect is the need for coherence, possibly involving cohesive devices such as anaphoric relations or other connectors. Part of establishing cohesion lies in the ability to draw inferences. Krifka (2008) views these ‘extended’ aspects of IS as part of CG management. According to a recent neurobiological perspective, the perception–action cycle is driven by active sensory sampling of the world with the aim of maximizing evidence in favour of one’s current model of the world (and conversely minimizing surprise, i.e. prediction error) (Friston 2010). In this view, perceptual sampling is used to infer hidden states of the world. Moreover, since the observer has the ability to change the world and its hidden states by action, he/she can build up counterfactual representations, that is, beliefs about what he/she can infer about the world if it were to be sampled in a particular way (Friston et al. 2012). As an example, Friston and colleagues describe sitting in a garden and perceiving some fluttering in one’s visual periphery. If the brain encodes the hypothesis that this fluttering is caused by a bird, this minimizes surprise about the sensory input. In addition, it leads to sensory predictions that can be tested by moving one’s eyes to fixate on the bird, thus enabling the hypothesis generated by the internal model to be confirmed. This notion of ‘active inference’ could also provide a suitable neurobiological mechanism for establishing inferences during discourse processing.

29.1.5 Neuroscientific methods for studying cognitive mechanisms A myriad of techniques for examining cognitive mechanisms in the brain is now available (see Kaiser, this volume, for psycholinguistic findings and Wagner, this volume, for production data). For the sake of brevity, we briefly mention just two here: Page 6 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: Freie Universitaet Berlin; date: 16 April 2018

Towards a Neurobiology of Information Structure electroencephalography (EEG) and functional magnetic resonance imaging (fMRI). EEG, and particularly the EEG-based method of event-related brain potentials (ERPs), and fMRI currently constitute the two most commonly used neuroscientific methods for studying higher-order language processing. For a more in-depth introduction to these methods directed at a linguistic audience, see Bornkessel-Schlesewsky and Schlesewsky (2009a). ERPs are small changes in the spontaneous electrical activity of the brain that are temporally coupled to certain sensory or cognitive events (see Figure 29.1) and that can be measured non-invasively by means of electrodes attached to the surface of the scalp. The ERP signal can be isolated from background EEG activity (the ‘noise’), which is typically larger in amplitude by several dimensions of magnitude, by means of a simple averaging procedure. Participants are presented with multiple (in sentence or discourse level (p. 587) studies typically 30–40) stimuli of a particular type and the EEG signal is averaged from the onset of the critical stimulus onwards.2 This yields an ERP: a series of positive and negative potential shifts over time. ERPs must be interpreted by comparing different conditions. For example, in a study concerned with IS processing, we might compare sentences of the form ‘Peter washed Mary’s car’ in either a neutral context (‘What happened yesterday?’), a context that renders Peter given and the car in focus (‘What did Peter wash yesterday?’), a context that renders the car given and Peter in focus (‘Who washed Mary’s car yesterday?’), etc. In order to examine the processing of a given vs new vs focused referent, one could compare ERPs to ‘Peter’ in the target sentence as a function of each of the three contexts. ERPs are classified according to a number of dimensions: polarity (a negative or a positive deflection relative to a control condition), latency (time from stimulus onset to either onset (‘onset latency’) or peak (‘peak latency’) of the difference between a critical and a control condition) and topography (scalp distribution of the difference between a critical and a control condition, that is, electrode sites at which the difference was observed). Together with functional considerations (i.e. conditions under which an effect occurs), polarity, latency, and topography are jointly used to classify so-called ERP components, that is, particular ERP deflections that are ascribed a certain functional significance. The amplitude of ERP deflections, by contrast, is typically viewed as a measure of effect size. ERP component nomenclature is often defined (p. 588) via polarity and peak latency, with, for example, N400 denoting a negativity with a peak latency of 400 ms. Components may also be named functionally, for example the Mismatch Negativity (MMN), an early (approx. 100– 250 ms) negativity with a fronto-central distribution that is elicited preattentively by ‘oddball’ stimuli.

Page 7 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: Freie Universitaet Berlin; date: 16 April 2018

Towards a Neurobiology of Information Structure Language-related ERP components are often ascribed cognitive functions of varying specificity, for example, ‘semantic memory access’ for the N400, ‘syntactic reanalysis’ or ‘conflict monitoring’ for the P600. However, as described Click to view larger above, the neurobiological Figure 29.1 The physiological basis of event-related brain potentials. plausibility of functions (Bornkessel-Schlesewsky and Schlesewsky 2009a; such as these is adapted from Martin 1991 and Köster 2004). questionable. Here, we thus favour an approach that views neurocognitive correlates of language processing from a more general mechanistic perspective (Tune et al. 2014; Sassenhagen et al. 2014). Advantages of EEG-based methods are that they provide direct measures of brain activity with excellent temporal resolution (in the millisecond range). By contrast, the spatial resolution of the scalp EEG is severely limited due to volume conduction, including distortion of the electrical signal by the skull. Due to the so-called inverse problem, reconstruction of underlying sources from scalp EEG patterns is not uniquely possible, that is, there are always multiple solutions to the inverse problem. The current method of choice for high spatial resolution is fMRI, which, by contrast, does not offer very high temporal resolution. fMRI is based on the so-called Blood Oxygenation Level Dependent (BOLD) contrast. Hydrogen atoms absorb energy applied at a particular frequency in the presence of a magnetic field and subsequently emit energy at the same frequency. The rate of emission is affected by various factors, including the homogeneity of the surrounding magnetic field. The BOLD contrast takes advantage of the different magnetic properties of oxygenated and deoxygenated hemoglobin: the fMRI signal increases with higher proportions of oxygenated vs deoxygenated blood, since deoxygenated hemoglobin causes inhomogeneities in the local magnetic field. Neural activity consumes energy, the compensation of which requires oxygen. Hence, a short initial drop in the proportion of oxygenated vs deoxygenated blood in an active region is followed by a relative oversupply of oxygenated blood, thus leading to a higher fMRI signal (Heeger and Ress 2002). In contrast to EEG, fMRI is not a direct measure of neural activity. In addition, the BOLD signal is slow, taking 4–6 s to reach peak (even longer in some brain regions), thereby imposing limitations on the interpretation of language-related fMRI studies with respect to the timing of language comprehension processes. The spatial resolution of fMRI is very high—in the millimeter range—although this is tempered somewhat by the need for group analyses with most paradigms. When interpreting fMRI results, it must also be kept in Page 8 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: Freie Universitaet Berlin; date: 16 April 2018

Towards a Neurobiology of Information Structure mind that the ‘colourful pictures’ depicting fMRI results do not show actual brain activity but rather estimates of statistical significance. Thus, activation contrasts between different conditions should ideally be supplemented by some depiction of per-condition signal change. Figure 29.2 shows some of the brain regions relevant to the processing of language in general and IS in particular and will serve as a reference for the remaining sections of this chapter. (p. 589)

Click to view larger Figure 29.2 Regions of the brain related to language processing. (a) and (b) depict lateral views of the brain, (c) depicts a medial view of the brain, and (d) shows relevant subcortical structures. Numbers indicate Brodmann Areas. (Bornkessel-Schlesewsky and Schlesewsky 2009a; adapted from Bornkessel-Schlesewsky and Friederici 2007).

29.2 Information structure and text comprehension (p. 590)

Important distinctions have been observed during the processing of text as opposed to isolated sentences (Van Petten and Kutas 1990; van Berkum et al. 1999b; Bornkessel et al. 2003; Ferstl et al. 2008). One way to account for these differences is via the contribution of IS. During text comprehension, predictive processing relies on rich representations from the ongoing discourse, and meaning construction is supported by multi-modal associations and inferences as the system strives for optimal coherence. Here, we briefly mention three pieces of evidence that demonstrate such differences. First, comparing repeated and first-mentioned words in lists and isolated sentences reveals more Page 9 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: Freie Universitaet Berlin; date: 16 April 2018

Towards a Neurobiology of Information Structure pronounced N400-effects for new information and a late positive complex for given information (Rugg 1987; Besson et al. 1992). In texts, the N400-difference is also observable, but now new entities evoke a positivity (Van Petten et al. 1991). This reversal of the positivity effect can be explained by the distinct attentional demands associated with word and text processing. In text, repetition functions as a coreference device and facilitates the computation of coherence relations (see Section 29.3 on givenness). Second, word order variation in the German middlefield evokes robust effects for marked over unmarked linearization in isolated sentences (Bornkessel and Schlesewsky 2006b): the marked object-before-subject order engenders a negativity between 300 and 450 ms (termed scrambling negativity) relative to subject-before-object order. Yet, presented after a focus question, focused middlefield-initial objects and subjects no longer differ from each other and evoke an early positivity between 280 and 480 ms post-onset relative to unfocused arguments (Bornkessel et al. 2003). Again, text comprehension exerts distinct demands from sentence comprehension due to an additional layer of IS (focus; see also Section 29.4). Third, sentence-initial arguments in the German prefield show no processing differences for unambiguously marked objects vs subjects in isolated sentences (Frisch et al. 2002).3 In context, however, the initial argument’s information status influences comprehension, reflecting distinct demands during the dynamic construction of meaning as a function of contextual predictability (Schumacher and Hung 2012). Neuroimaging data suggest that text comprehension is subserved by a processing network of fronto-temporal regions, reflecting semantic and textual processes, and a network including the medial prefrontal cortex implicated in social cognition but also in inferencing and attentional processing (Ferstl and von Cramon 2002; BornkesselSchlesewsky and Friederici 2007; Ferstl et al. 2008; Mar 2011). Meta-analyses indicate (p. 591) a widespread network for semantic knowledge. Multimodal associations and meaning construction have primarily been linked to bilateral activation of the anterior temporal lobe. The network that is involved in coherence computation includes the precuneus and prefrontal cortex, but also posterior temporal and parietal areas. Inferences are further subserved by fronto-medial and medial parietal areas. There is still an unresolved debate over whether text comprehension is subserved by right hemisphere regions (Mar 2004; Ferstl et al. 2008; Johns et al. 2008). In addition to functional differences associated with information packaging, the contrast between processing sentence- and text-level information becomes evident when considering the predictive power of context information. Co-textual information generates expectations about upcoming words during incremental processing. Information is thus evaluated with respect to the preceding text or shared knowledge. If a hearer who has evidence that Tim is very quick encounters an utterance like ‘Tim was very slow’, its unexpectedness engenders a prediction error, reflected in an N400 (van Berkum et al. 1999a). Similarly, if a comprehender reads a story about a singing peanut, a cartoon-like character, he is surprised to read ‘The peanut was salted’, but will more likely anticipate Page 10 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: Freie Universitaet Berlin; date: 16 April 2018

Towards a Neurobiology of Information Structure an utterance like ‘The peanut was in love’; again, irrespective of lexical-semantic associations, contextually unexpected utterances evoke an N400 as a prediction error (Nieuwland and van Berkum 2006). These examples illustrate that contextual information modulates language comprehension above and beyond mere lexical-semantics (see e.g. Schumacher 2012 for a comprehensive overview of context effects, including the co-text, world knowledge, mutual knowledge, and interlocutor-specific knowledge). Context effects can generally be accounted for in terms of coherence satisfaction. The processor computes expectations for upcoming words with the aim of maximizing coherence. Consequently, coreference relations are preferred (hence less costly; Burkhardt 2006), causal relations facilitate processing (St. George et al. 1997; Kuperberg et al. 2011), and events are more easily processed in their natural order (Münte et al. 1998). When coherence is not met, prediction errors occur and processing costs accrue, reflected in an N400 for unexpected continuations and a sustained negativity for enhanced memory demands or activation in the inferior precuneus, prefrontal, and posterior cingulate cortex (Ferstl et al. 2008). Cohesive devices are also computed incrementally, reflected in activation in left inferior prefrontal regions and anterior negativities (Ferstl and von Cramon 2001; Robertson et al. 2000; Streb et al. 1999; Anderson and Holcomb 2005). In the following sections, we look more closely at the processing of particular IS notions—givenness (Section 29.3), focus (Section 29.4), topic (Section 29.5), and other cartographic functions (Section 29.6).

29.3 Givenness Givenness (information status) refers to a continuum of cognitive accessibility with brand new information at one end of the scale and textually given information at the other (Prince 1981). Information status contributes to the perception–action cycle by (p. 592) modulating predictive processing on the one hand (given information is more predicted than new information) and requiring mental model updating on the other (introduction of a new referent results in updating). An interesting test case for the neurocognitive correlates of givenness are therefore entities that figure somewhere in-between, such as inferrables, that is, expressions referring to an entity that is not directly given in previous discourse but is accessible via an inference (see Sæbø, this volume for this kind of accommodation). Behavioural research has demonstrated a comprehension advantage for given over new and inferred entities (Haviland and Clark 1974; Yekovich and Walker 1978). ERPs provide a more differentiated view: comparing given (‘Tobias visited a conductor in Berlin. He said that the conductor was very impressive.’), inferred (‘Tobias attended a concert in Berlin. He said that the conductor was very impressive.’), and new entities (‘Tobias talked to Henriette. He said that the conductor was very impressive.’), ERPs suggest that information status contributes to two discrete processing steps, reflected in a biphasic N400—late positivity pattern (Burkhardt 2006; Hirotani and Schumacher 2011): the amplitude of the N400 increases with decreasing predictability Page 11 of 20

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: Freie Universitaet Berlin; date: 16 April 2018

Towards a Neurobiology of Information Structure (accessibility; given FocP) such that the double in the CLLD-construction occupies the specifier position of the higher projection (TopP) and the fronted constituent the specifier position of the lower one (FocP; see Agouraki 1990; Tsimpli 1990, 1995; Georgiafentis 2004; see Aboh as well as Bocci and Poletto, this volume, about cartographic approaches to syntax). The alternative hypothesis is that the hierarchical structure hosts purely structural configurations such as the double constituent involved in CLLD and the fronted constituent in fronting, as schematically presented in (9b)3 (see the ‘non-configurational approach’ in Alexopoulou 1999 and the ‘syntax-information structure underdeterminacy hypothesis’ in Haidou 2012). The order ‘topic > focus’ is not accounted for by the constituent structure in these accounts. It results from independent generalizations of the phonological form: the topic is realized within a clausal layer that is higher than the clausal layer that contains the focus, which predicts that topics can only precede foci in the left periphery.

Page 8 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 23 October 2018

Information Structure in Modern Greek (9)

Research on Modern Greek syntax has established an extensive body of knowledge about the properties of the left periphery. In general, the fronting construction is an instance of A-bar movement, whereas CLLD shows mixed properties of A-bar movement (e.g. sensitivity to island constraints) and base generation (e.g. obviation of Weak (p. 693) Crossover effects) (Iatridou 1995: 13f.). The relevant question for the present section is whether these phenomena are sensitive to the information structural features. Binding facts show an asymmetry between the CLLD and fronting: fronted constituents have the same binding properties as the basic configuration, as shown in (10a,b), while CLLD creates new possibilities, as shown in (10c) (Iatridou 1995: 13; Alexiadou and Anagnostopoulou 1997; Alexopoulou 1999, 2009). (10)

Crucially, the facts in (10) are not sensitive to information structure. The properties of focus fronting in (10b) are independent of focus. Example (11) is a paraphrase of the ‘topicalization’ construction in (8) with an intonational nucleus in the post-verbal domain. The topicalized constituent cannot bind into the possessor of the subject, that is, it is in a position that does not take scope over the core clause. Hence, the binding properties of

Page 9 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 23 October 2018

Information Structure in Modern Greek fronted topics differ from the binding properties of CLLD in (10c) and are identical to the properties of fronted foci in (10b). (11)

The independence of the syntax of CLLD from information structure is also supported by the binding properties of wh-questions. Wh-fronting shares the same binding properties as focus fronting; compare (10b) and (12a). However, the CLLD question has the same properties as the CLLD example in (10c) (Alexopoulou 1999: 111, 2009, Alexopoulou and Kolliakou 2002). Hence, binding possibilities pattern with the syntactic construction and not with information structure. (p. 694)

(12)

Page 10 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 23 October 2018

Information Structure in Modern Greek Further support for the independence of syntax from information structure comes from the scopal relations between quantifiers, as illustrated in (13) (Alexiadou and Anagnostopoulou 1998: 505; Alexopoulou 1999: 111f.). The canonical configuration in (13a) is scopally ambiguous. The fronting construction in (13b) also has both readings, although there is a preference for the narrow scope reading. This effect is irrelevant for the scopal properties of the hierarchical structure: focus evokes the exclusion of alternatives, hence it motivates an inference that a particular individual is at issue (Fox and Sauerland 1997). The scopal asymmetry between subjects and objects is independent of the focus domain of (13b), as has been shown experimentally by Baltazani (2002: 165– 199). Crucially, the scopal relations are different in CLLD, where the subject cannot have scope over the left-dislocated constituent; see (13c). (13)

The same facts hold for fronting and CLLD questions, as illustrated in (14a,b). The question with the fronted wh-pronoun retains the scope ambiguity of the basic configuration, which replicates the observation for the fronting construction in (13b). However, the CLLD question has only a possible reading in which the universal quantifier is under the scope of the wh-pronoun; see (14b). (p. 695)

Page 11 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 23 October 2018

Information Structure in Modern Greek (14)

In sum, the scopal asymmetries between fronting and CLLD are independent of information structure. The presence of a clitic in CLLD gives rise to new scopal possibilities that are not influenced by the contextual trigger of the construction (topic or focus). This implies that, from the point of view of the hierarchical structure, the four information structural options are subsumed under two left-peripheral hierarchically ordered configurations. A final note is due concerning preverbal subjects in SVO. Some authors assume that subjects share the same position with left-dislocated objects (Alexiadou and Anagnostopoulou 1998: 501).4 As regards the information-structural possibilities, it is (p. 696) clear that the discourse properties of preverbal subjects and left-dislocated objects are completely different; as discussed in Section 34.1 the latter but not the former are contextually restricted. A closer examination of their syntactic properties shows that preverbal subjects crucially differ from left-dislocated objects. Left-dislocated and fronted constituents are islands for extraction (Tsimpli 1995: 182), whereas this is not the case for subjects (see Spyropoulos and Revithiadou 2009 for a detailed discussion about subjects).

34.2.2 Post-verbal domain The crucial question in the post-verbal domain is the syntactic status of clitic-doubled objects; see (5). Do these objects occupy an argument position, or are they adjoined elements (see summaries in Alexiadou and Anagnostopoulou 2000 and PhilippakiWarburton et al. 2002; see also Tsakali 2008 on cross-linguistic variation)? Constraints on extraction suggest the latter view (see further evidence and detailed discussion in Androulakis 2001; Philippaki-Warburton et al. 2002; Spyropoulos and Revithiadou 2009: 9), as illustrated in (15): extraction is possible out of objects in situ, but not possible out of left-dislocated or fronted objects; see the contrast between (15a) and (15b). Crucially,

Page 12 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 23 October 2018

Information Structure in Modern Greek extraction is not possible with clitic-doubled post-verbal objects, which contrasts with the properties of objects in situ; see (15c). (15)

5

Assuming that material in adjoined positions is an island for extraction (see López, this volume), the implications of the facts in (15) are straightforward. Adjunction is possible at several layers of the clause structure (vP, CP), and post-verbal clitic-doubled objects exploit several possibilities in different constructions (Androulakis 2001; Philippaki-Warburton et al. 2002). For the purpose of understanding the information structural possibilities, the relevant issue is that the double is an adjoined constituent, as indicated in (16b), and this contrasts with an object in situ, as in (16a). (p. 697)

Page 13 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 23 October 2018

Information Structure in Modern Greek

(16)

34.3 Prosodic properties Modern Greek is an intonational language with a repertoire of tonal events comprising pitch accents (L*+H, L+H*, H*+L, H*, L*), phrase accents (H-, L-, and !H-), and boundary tones (H%, L%, and !H%); see Greek ToBI (Tones and Break Indices) in Arvaniti and Baltazani (2005). Two issues are particularly relevant for the study of information structure: (a) the delimitation of prosodic domains (see Section 34.3.1), and (b) the role of accentual contrasts (Section 34.3.2).

Page 14 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 23 October 2018

Information Structure in Modern Greek

34.3.1 Prosodic domains Figure 34.1 illustrates an SVO utterance with neutral intonation (left panel), and the same order with a clitic-doubled object (right panel); see gloss in (17).6 The prosodic realization of the SVO utterance in the left panel is felicitous as an answer to the question ‘What happened?’; see (18). The default prenuclear accent in Modern Greek is a rising pitch accent (p. 698) (L*+H) (Arvaniti and Ladd 1995; Arvaniti et al. 1998, 2000; Arvaniti and Baltazani 2005). The right edge of the preverbal subject is demarcated with a high phrase accent (H-); this tonal event is not bi-uniquely associated with a particular discourse function (such as topic), but occurs with a wide range of non-final prosodic constituents, for example left-dislocated material, lists, and parentheticals (Baltazani and Jun 1999), and non-final clauses (Botinis et al. 2004). The most frequent type of nuclear accent in declaratives involves a high target associated with the starred syllable (either H* or L+H*; see Section 34.3.2; see Baltazani 2002 for further options). The low target at the right edge of the intonational phrase is a boundary tone (Arvaniti and Baltazani 2005; see Arvaniti and Godjevac 2003 for evidence that this tonal event is independent from declination). The clitic doubling version in the right panel could be an answer to the question ‘What happened with the race?’. The nuclear accent is aligned with the starred syllable of the verb in this Click to view larger example; another possible Figure 34.1 Canonical configuration vs clitic tonal structure would be a doubling. nuclear accent on the subject (depending on focus). One possibility is categorically excluded though: the cliticdoubled object cannot bear the nuclear accent (Iatridou 1995; Alexiadou and Anagnostopoulou 1997). The double is in an adjoined position, as presented in (16b), and adjoined material is extrametrical, that is it cannot bear nuclear stress; see (16) (Revithiadou and Spyropoulos 2008: 43). Additional evidence comes from the fact that sandhi rules do not apply at the left boundary of the clitic-doubled constituent (see the detailed discussion in Revithiadou and Spyropoulos 2008, Spyropoulos and Revithiadou 2009).

Page 15 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 23 October 2018

Information Structure in Modern Greek (17)

(p. 699)

(18)

The contrast between the left-peripheral configurations is illustrated in Figure 34.2, which presents the sample sentence in (19) pronounced in the contexts in (20) (the pronominal clitic occurs in the CLLD construction in the left panels). The properties of the figures in the top left and in the bottom right panel are reported in a large number of studies, but the intonations of the CLLD construction with an initial focus and the fronting construction with a final focus have not yet been studied with instrumental phonetic methods. The pitch tracks in Figure 34.2 reveal a major contrast between final foci (top panels) and initial foci (bottom panels). The focus hosts the nuclear accent and the postnuclear material is deaccented (Baltazani and Jun 1999; Baltazani 2002, 2003b; Arvaniti and Baltazani 2005; Gryllia 2008). This property is reflected in the non-final foci in the bottom panels: the focus bears a nuclear accent H* and the postfocal domain is generally deaccented. The CLLD utterance with initial focus (bottom left panel) is slightly different: the clausal core must bear a secondary stress (compare with the deaccented version in the bottom-right panel). The tonal structure of the final foci in the top panels is richer: the left side of the focus domain of the utterance is demarcated by an intonational boundary (see H- target in Figure 34.2, top panels). A further correlate of the alignment of the left edge of the focus with the boundary of a phonological phrase is the fact that some sandhi rules do not apply at this position (Revithiadou 2003). There is no evidence for tonal events demarcating the right edge of a focus domain. Clitic left-dislocated material is demarcated with a high phrase tone at its right edge; see top left panel. The fronting construction in the top right panel has a final focus aligned with a left boundary: the prefocal material is realized as a single prosodic phrase with two prenuclear accents (Baltazani 2002; Revithiadou 2003).

Page 16 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 23 October 2018

Information Structure in Modern Greek

(19)

(20)

(p. 700)

Click to view larger Figure 34.2 Prosodic structure and left-peripheral configurations.

Page 17 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 23 October 2018

Information Structure in Modern Greek

34.3.2 Nuclear accent contrasts The accentual contrast in the realization of the nuclear accent correlates with different types of focus in several languages (Alter et al. 2001: 65 and Steube 2001: 233 for German; Face 2002: 34 for Madrid Spanish; see several Italian dialects in Grice et al. 2005: 364). Studies on Modern Greek syntax frequently assume a concept of ‘contrastive stress’ that may encode information structural differences independently from word order (see, e.g. Georgiafentis 2004). Following Greek ToBI, such a contrast applies between the nuclear accents H* and L+H*. The H* accent corresponds to an f0-peak that is preceded by a small rise and signals broad focus (Arvaniti and Baltazani 2005: 87; see also Baltazani and Jun 1999; Baltazani 2003a). The bitonal L+H* corresponds to an f0-peak that is preceded by a noticeable dip and aligns with the middle of the accented syllable; this accent is reported for narrow focus contexts (Arvaniti et al. 2006). The contrast between these tonal realizations is illustrated in Figure 34.3, which presents the sentence in (17) under the contexts in (21). The critical difference lies in the realization of the nuclear accent, which is a high tonal target (H*) in the left panel (followed by a fall towards the low (p. 701) boundary tone), and a rise (L+H*) in the right panel. The latter tonal event involves an initial dip, such that the H-target is aligned with the middle of the stressed syllable.

Click to view larger Figure 34.3 Nuclear accent contrast.

(21)

Experimental studies on the interpretation of these accentual patterns do not reveal a significant correlation between accent and type of focus (Gryllia 2008). Crucially, both accentual realizations are possible in contrastive contexts (e.g. answers involving correction) and non-contrastive contexts (e.g. answers to wh-questions). However, there is evidence for a difference with respect to the focus domain: both varieties of the nuclear accent have a local interpretation, that is, the focus domain may be the phonological word that bears this accent. However, only the H* may be projected to higher layers of the metrical structure: for example an H* (and not the L+H*) accent on a final O may be Page 18 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 23 October 2018

Information Structure in Modern Greek interpreted as O-focus, VP-focus, and all-focus. This property is independent of focus type, that is, the L+H* does not project to a contrastive VP-focus domain (see the experimental results in Georgakopoulos and Skopeteas 2010). Given these facts, the relevant contrast is the distinction between an underspecified nuclear accent and a nuclear accent that is associated with its phonological host and cannot be projected to higher domains of the constituent structure, that is, it is not interpreted as the nuclear stress of the VP or the clause, as outlined in (22). This contrast only applies if the projection of the nuclear accent is made possible by the nuclear stress rule (at the right edge of a prosodic domain); otherwise (e.g. in clause-initial contexts) the two realizations of the nuclear accent alternate in free variation (Georgakopoulos and Skopeteas 2010).

(p. 702)

(22)

34.4 Interpretational properties 34.4.1 Focus fronting The introduced syntactic and prosodic assumptions open an array of information structural possibilities that can be straightforwardly accounted for through the rules for assigning the phrasal stress that independently hold true. Greek is a head-initial language, which implies that phrasal stress targets the rightmost constituent that is eligible for stress within the stress domain (see Georgiafentis 2004; experimental evidence in Keller and Alexopoulou 2001 and Georgakopoulos and Skopeteas 2010; see also Zubizarreta, this volume). Beyond the local interpretation of the nuclear accent, which is always available (under the tonal conditions determined in (22)), a final accent also allows for projected interpretations. This asymmetry gives rise to different focus sets, that is, sets of possible focus domains (Reinhart 2006); compare (23a) and (23b).

(23)

7

According to the generalizations about phrasal stress, fronted constituents bearing a nuclear accent form a narrow focus domain, which implies that the discourse properties of fronted foci are already predicted by generalizations about the phonological form.

Page 19 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 23 October 2018

Information Structure in Modern Greek Furthermore, fronted constituents are expected to be interpreted as foci only if they bear the nuclear accent. Fronted constituents that do not bear the nuclear accent are interpreted as topics; see (8). The question is whether this contrast captures the entire range of data in Greek. If there are interpretational properties beyond those that are independently predicted by the prosodic structure, then we must assume that a feature with semantic content is associated with the constituent structure. A good candidate for such a property is the identificational interpretation of focus fronting (see É. Kiss 1998, 2009). If a fronted constituent excludes alternatives while in situ focus does not, this would imply a semantic contrast that is not captured by the options in (23a,b) and has to be accommodated (p. 703) by additional assumptions (such as an identificational operator triggering the fronting operation). Some research on focus in Modern Greek argues that the fronting construction involves identificational properties: Greek focus fronting is classified as [+exhaustive, +contrastive] (É. Kiss 1998: 270; based on Tsimpli 1995; see also Georgiafentis (2004) for a detailed examination of interpretational and distributional evidence). This generalization is established through comparisons of sentences with a fronted focus vs canonical sentences. However, canonical word order is ambiguous between a broad focus and several narrow focus readings in Greek depending on the accentual structure. Hence, the role of syntax can only be disentangled from the role of prosody by a comparison between fronted focus and narrow focus in situ as expressed with a non-projectable nuclear accent; see (22). The empirical facts do not indicate a contrast between these two possibilities of narrow focus. The basic facts are illustrated in (24) with a test proposed in É. Kiss (1998) for Hungarian (see detailed discussion on contrastive and exhaustive properties of Greek foci in Gryllia (2008: 7–55) and Haidou (2012: 126–197)). In example A1 (broad focus), the object constituent does not necessarily identify the exhaustive set of relevant referents for which the event is the case. In examples A2 (ex situ focus) and A3 (narrow focus in situ), the referents at issue are exhaustively identified. The empirical evidence for this contrast is that the presence of a further referent for which the proposition holds true contradicts A2 and A3 but not A1 (see appropriateness of the negation in the continuation B).

Page 20 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 23 October 2018

Information Structure in Modern Greek (24)

There is evidence that the identificational interpretation of focus-fronting is a defeasible inference. First, this construction is compatible with expressions that contradict an exhaustive identification, for example an even-phrase; see (25a). Furthermore, the (p. 704) exhaustive interpretation does not arise if there is another trigger for focus fronting: if focus fronting is motivated by the non-predictability of the referent, the interpretation that the focus referent denotes the exhaustive subset of referents for which the presupposition holds true does not apply (see experimental evidence that Greek differs from Hungarian in this respect in Skopeteas and Fanselow 2011). The direct object in (25b) denotes a referent that is non-predictable in this context. If non-predictability accounts for focus fronting in a particular context, the inference of exhaustive interpretation does not arise (world knowledge suggests that the ‘power outlet’ is only a subset of the entities washed by the grandfather in (25b)).

Page 21 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 23 October 2018

Information Structure in Modern Greek (25)

The focus options of the fronting operation in Greek are accounted for by the rules determining phrasal stress. Fronted foci evoke alternatives under particular contextual conditions and do not differ from narrow focus in situ in this respect. This interpretational property arises through a conversational implicature: the intuition that the proposition does not hold for the subset of non-asserted relevant alternatives can be explained by the quantity maxim (Spector 2005: 229). This intuition arises with any type of focus (also instances of broad focus) and is stronger if the relevant alternatives are easier to retrieve, which is the case for narrow focus.

34.4.2 CLLD and clitic doubling CLLD is not triggered by simple givenness asymmetries, that is, it requires a stronger contextual trigger than scrambling in Germanic (see cross-linguistic evidence from speech production in Skopeteas and Fanselow 2009). The appropriate context for CLLD is contrastive topicalization: the left-dislocated constituent identifies a referent among a set of relevant alternatives in the common ground for which the asserted information (in its complement) is the case (see an account in terms of ‘linkhood’ in Alexopoulou and Kolliakou 2002; see also López, this volume). This opens up an array of interpretative variants such as implicational topics, shifting topics, partial topics, etc. (see Büring, this volume), which are all possible with CLLD in Greek. The interpretation of CLLD is illustrated by the minimal pair in (26). The answer with an object in situ in (26a) is congruent to the VP question. The CLLD answer in (26b) evokes the intuition that there is further contextually relevant information that is not explicitly available in the presented context. In particular, the clitic left-dislocated object in (26b) evokes a set of alternative referents that are presupposed by the speaker as part of the Common Ground, for example {Harry Potter, The Hobbit}, and for which the asserted information (‘I read x’) is not the case. (p. 705)

Page 22 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 23 October 2018

Information Structure in Modern Greek (26)

CLLD is also possible with an intonational nucleus on the left-dislocated constituent; see Figure 34.2, bottom left panel. The interesting issue is the difference between fronting and left dislocation in this case. Starting with wh-questions, the default construction involves a fronted wh-pronoun as in (27a). This question presupposes that there is some x (x = pupil), who is called by the director and expresses a request to specify this referent. The question with the clitic in (27b) is contextually restricted. This question has a reading in which it is presupposed that the addressee knows the referent that instantiates the variable (‘Can you tell me again, who was the pupil who was called by the director?’). With this reading, example (27b) may occur as a quiz question. The effect of CLLD is that the clitic left-dislocated wh-phrase is interpreted D-linked (=discourse-linked), that is related to a referent that is already available in the common ground (TheophanopoulouKontou 1986–87; Iatridou 1995: 27). Furthermore, the same example (27b) can be used with a generic interpretation that does not presuppose a specific referent (Alexopoulou 2009); this interpretation can be supported by an adverb such as poté ‘ever’. The closest paraphrase of this type of question in English is ‘which pupil is such, that the director called him?’. This reading could be used in a rhetorical question implying that there is no such pupil that the director may have called. This reading cannot be accounted for by a D-linking account of CLLD. Example (27b) is interpreted as specificational predication: the double denotes a set and the specificational predicate characterizes this set by denoting a property that holds true for its members (compare the analysis of É. Kiss 2006 on Hungarian focus).

Page 23 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 23 October 2018

Information Structure in Modern Greek (p. 706)

(27)

Turning to focus, the contrast is similar: the constructions in (28) uttered as answers to the wh-question in (27a) have different interpretations. The congruent answer is (28a); the CLLD answer in (28b) with an intonational nucleus on the left-dislocated material evokes the interpretation that the answer is not exhaustive, that is, it is not excluded that other referents are the case. This interpretation comes from the fact that the double in CLLD matches the pronominal variable of its complement. The utterance ‘Janis is such that the director called him’ is a partial answer to the question ‘Who did the director call?’ The speaker gives an under-informative answer: s/he asserts that the restrictor of the question specifies the referent of the left-dislocated object, which motivates the implicature that s/he does not have enough evidence to list the exhaustive subset of referents for which the restrictor of the question holds true. This implicature does not arise if the CLLD answer is uttered as an answer to the CLLD question in (27b): in this case, the utterance ‘Janis is such that the director called him’ is a congruent answer to the question ‘Who is such that the director called him?’.

Page 24 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 23 October 2018

Information Structure in Modern Greek (28)

CLLD is also possible with quantified noun phrases as doubles; see (10c) and (29). The double of these examples is clearly not a ‘contrastive topic’ and it is not D-linked. The CLLD utterance in (29) presupposes that the proposition ‘some x (x = pupil) exists such that x was called by the director’ is available in the Common Ground, which does not imply that the speaker commits to the truth of this proposition. In terms of specificational predication, the double denotes an empty set (‘no pupil’), and its complement (i.e. the specificational predicate) characterizes this set by denoting a property that applies to its members. (p. 707)

(29)

Clitic doubling does not occur under the same conditions as CLLD. Clitic-doubled objects cannot bear nuclear stress (Iatridou 1995), which is an expected correlate of rightdislocation; see (16b). This implies that only constituents that do not need to be accented may occur with clitic doubling. The counterparts of the constructions in (10c) and (29a) with quantified constituents are categorically excluded in clitic doubling; see (30).

Page 25 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 23 October 2018

Information Structure in Modern Greek (30)

The condition that the double cannot be stressed in clitic doubling has repercussions on information structure, that is, it must be background information such that it can be deaccented; see (31). Givenness is not a sufficient condition for clitic doubling: the utterance in (31) would not be felicitous in the context of a question in which the double is given but not background information, for example ‘Did you read The Jungle Book or The Hobbit?’ (see also account in terms of a ‘prominence condition’ in Anagnostopoulou 1999: 777). The difference to the contrastive topics in CLLD is that clitic doubling does not evoke alternatives (Valiouli 1993). Similarly to CLLD, the utterance in (31) is a partial answer to the question ‘What did you do in the summer?’; see (26). However, a cliticdoubling answer to this question does not evoke alternatives, it rather suggests a context in which the question ‘Did you read The Jungle Book?’ is part of the implicit Common Ground. (p. 708)

(31)

There are some intensively discussed counterexamples with indefinites in clitic doubling which show that the clitic-doubled constituent is not necessarily given information; see (32) (Kazazis and Pentheroudakis 1976; Anagnostopoulou 1999; Androulakis 2001). These examples come with additional requirements: the indefinite double must be an afterthought and constitutes an intonational unit that is definitively not part of the intonational entity encompassing the predicate, that is, example (32) is not possible with a nuclear stress on the indefinite double (the preferred prosodic option for this example involves an intonational nucleus on the adverb). Hence, these examples do not challenge the generalization that the double cannot bear the intonational nucleus of the sentence. As previously discussed in Kazazis and Pentheroudakis (1976), these examples are cases

Page 26 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 23 October 2018

Information Structure in Modern Greek of presupposition accommodation: they evoke a question that makes the indefinite object available in the implicit Common Ground. (32)

34.5 Summary Modern Greek is a language with flexible word order as well as intonational variation— both influenced by information structure. Two different syntactic configurations appear in the left periphery, that is fronting and clitic left dislocation. The fronting construction is predominantly used to express narrow focus, in which case the fronted constituent bears the nuclear accent of the utterance. Clitic left dislocation is a construction typically used for contrastive topics. However, it also occurs with an accented double, in which case it displays the interpretational properties of specificational predication. In the post-verbal domain, there is a contrast between objects in situ and clitic-doubled objects. There is syntactic evidence that clitic-doubled objects are right dislocated. As is generally the case for right dislocation, clitic-doubled objects cannot be accented, which has repercussions on their possible functions in discourse: only background information can occur with clitic doubling.

Notes: (1) I am grateful to Zoë Belk and Martina Schüler for redactional assistance. (2) Abbreviations (glosses): ACC accusative; PFV perfective; PST past; DEF definite; F feminine; GEN genitive; INDF indefinite; M masculine; N neuter; NEG negation; NOM nominative; PL plural. For the sake of simplicity, unmarked inflectional categories (singular, active, indicative, present) are not given in the glosses. (3) The properties discussed in this section do not apply to hanging topic left dislocation, which differs both in its contextual as well as in its syntactic properties (Anagnostopoulou 1997; Grohmann 2000). (4) The basic evidence for this view is that preverbal indefinite subjects can only have wide scope over universally quantified objects, while post-verbal subjects (in VSO) may not. However, there are counterexamples to this generalization (Giannakidou 2001: 3.2.3).

Page 27 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 23 October 2018

Information Structure in Modern Greek (5) These extraction facts hold true for instances of alienable possession (inalienable noun phrases may be accompanied by a zero pronoun that allows for more possibilities of extraction) and for verbs that exclude an ambiguity between extracted DP subconstituent and V-adjuncts (e.g. verbs allowing an experiencer genitive or several types of embedded prepositional phrases). Examples that do not fulfil these requirements have different properties, which are independent of the syntactic asymmetry under discussion. (6) The pitch tracks in this section are illustrations—not empirical proof; the evidence for the discussed phenomena is found in the cited literature. (7) FP stands for ‘functional projection’ underspecified for discourse features.

Stavros Skopeteas

Stavros Skopeteas is Professor of General Linguistics at the University of Bielefeld. His interests include syntax, phonology, information structure, experimental fieldwork and language typology. His research focuses on Greek as well as on several Mesoamerican and Caucasian languages.

Page 28 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 23 October 2018

Topic and Focus Marking in Chinese

Oxford Handbooks Online Topic and Focus Marking in Chinese   Yiya Chen, Peppina Po-lun Lee, and Haihua Pan The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara Print Publication Date: Jul 2016 Subject: Linguistics, Morphology and Syntax Online Publication Date: Oct 2015 DOI: 10.1093/oxfordhb/9780199642670.013.34

Abstract and Keywords This chapter reviews how the two important notions of information structure—topic and focus—are encoded in Chinese. It first describes the properties of syntactic constructions (e.g. shi/shi . . . de, lian–dou/ye) and semantic particles (e.g. cai, jiu, dou, and zhi) for focus marking. It then discusses how Chinese, a topic-prominent language, conveys topical information via different structures such as base-generated topics, dangling topics, and moved topics. Finally, it provides an overview of how prosodic structure and prominence cues (e.g. pitch register raising, pitch range expansion, and lengthening) complement and enhance focus and topic marking in a variety of Chinese dialects, and how such prosodic reflexes of information structure are constrained by their characteristic sound structures. Lacunae for the studies of focus and topic in Chinese are also identified as important questions for future research. Keywords: information structure, topic, focus, semantic particles, dangling topics, prosodic structure, prominence cues, shi and shi . . . de constructions, lian . . . dou/ye constructions

Page 1 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese

36.1 Introduction THERE is a general consensus that, in speech communication, information flow is structured so as to allow the interlocutors to regulate and control the way information is transferred in discourse. Languages of the world differ in the ways that they package information in discourse. This chapter will review the range of linguistic devices which have been reported to mark various information structure notions in Chinese. While the term information structure (IS) goes back to Halliday (1967) concerning the partition of information units within a sentence, there has been a long tradition of labelling information units as dichotomies such as theme–rheme, topic–comment, background–focus, presupposition–focus, or given–new (see Kruijff-Korbayová and Steedman (2003) for a summary of the terminologies). Concerning these dichotomies, there have also been heated debates on which IS notions are essential for modelling IS packaging and whether additional notions such as contrast/Kontrast are needed (e.g. Vallduví and Engdahl 1996; Umbach 2004). To avoid the terminological quagmire, we will adopt the IS notions laid out in Krifka (2008). Due to space limitations, particular attention will be paid to two important notions of IS: Focus and Topic. We will review how semantic particles and syntactic structures conspire to encode these notions (Sections 36.2 and 36.3) and how prosody, mandatory or optional, complements and enhances the relevant linguistic expressions in Chinese (Section 36.4). In the following, ‘Chinese’ and ‘Standard Chinese’ will be used interchangeably, referring to Mandarin or Putonghua, the official language of the People’s Republic of China, as well as Guoyu, the official language of the Republic of China.

36.2 Semantic and syntactic marking of focus (p. 734)

Page 2 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese

36.2.1 Different types of focus in Chinese Liu and Xu (1998), in their pioneering study on focus in Chinese, attempted to provide a finer classification of focus from a grammatical perspective. They argue that there are three types of focus, namely informational focus (1), contrastive focus (2), and topical focus (3), which can be classified using the features [±prominent] and [±contrastive]. A focus is [+prominent] when it stands out against the rest of the clause as background, and it is [+contrastive] when it stands out against something outside of the clause as background. Informational focus (referred to as natural focus by Liu and Xu) can thus be defined as [+prominent, –contrastive], contrastive focus as [+prominent, +contrastive] and topical focus as [–prominent, +contrastive]. As for ‘topical focus’, it is [–prominent] as it conveys given information, and [+contrastive] as it also evokes alternative readings. Within the framework of Krifka (2008), topical focus would be classified as contrastive topic. (1)

1

(2)

(3)

Page 3 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese Liu and Xu propose that for every sentence, the sentence-final position in Chinese is by default the position where new information is located, hence the syntactic structural position for informational focus. Wuhu in (1a) and sanshi nian in (1b) therefore can be interpreted as the new information of their respective utterance. In contrast, Zuotian xiawu ‘yesterday afternoon’ in (2) marks a contrastive focus, indicating that the time when he went to city/downtown was yesterday afternoon, not any other time. Lao Zhang in (3) is defined by Liu and Xu as a topical focus, as it is marked by the topic particle me and provides a contrastive reading with the NP Lao Wang in the preceding clause. (p. 735)

Xu (2004: 277) further states that ‘Chinese is a language which exhibits a reverse relationship between syntactic positioning and phonological prominence of focus’, different from West-Germanic languages. For example, Xu argues that informational focus is always located in the final position in Chinese, whereas in languages such as English, focus has a systematic manifestation via pitch accent, regardless of whether the focused element is situated in the syntactically favoured focus position or not.

(4)

So, when an English sentence like (4) is uttered out of the blue, either the subject, as in (4a), or the predicate, as in (4b), may be pronounced with salient prosodic prominence2 to signal focus. In Chinese, however, once a focus element takes the default syntactically most-embedded position, the phonological marking of focus seems less prominent and has therefore been described by Xu (2004) as optional.

(5)

Page 4 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese In (5b), as an answer to the preceding question (5a), Meiguoren (Americans) appears in the sentence-final position and the prosodic prominence is not required to be salient. To signal contrast (as in (5c)), however, prosodic prominence is necessary to invoke the alternative reading. Furthermore, when focus is not realized at the default structural focus position (as in (5d)), prosodic prominence is salient. Despite the well-recognized contributions of Liu and Xu (1998) and Xu (2004) to our understanding of focus in Chinese, it is important to note that when syntactic devices are employed to convey focus, prosody is typically present, not optional. Section 36.3 will show that evidence abounds from well-controlled experiments that prosody is present even in well-known focus constructions. For example, Jia et al. (2009) reported robust acoustic cues used to mark the focused element in the ‘shi-construction’, regardless of the presence of the immediately neighbouring focus particle shi ‘be’. (p. 736)

36.2.2 Focus marked by shi and shi … de constructions Chinese boasts quite a range of morphological markers and syntactic constructions for focus-marking, which have aroused much interest in the literature. In example (2) above, zuotian xiawu ‘yesterday afternoon’ is marked as a contrastive focus by the copula shi ‘be’, which introduces contrast with the background ‘he came into the city some other time’. Following Paul and Whitman (2008), this pattern is called ‘bare shi focus construction’. The following examples are from Hole (2012).3 (6)

Page 5 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese (7)

Note that shi has to immediately precede the focus phrase here. Examples (6a) and (6b) have the contrastive focus preceding the VP, while in (7a) and (7b), it is inside the VP. (p. 737) The ungrammaticality of (7c) shows that shi cannot occur further to the right than at the left edge of the VP. The ‘bare shi focus construction’, however, is not the canonical cleft construction of Chinese. Hole (2012) argues that the canonical clefts in Chinese should be the ‘shi … de construction’ (Hole 2012). (8)

Sentence (8a) is an example from Hole (2012), with its linear syntax diagrammed in (8b). Cheng (2008) discusses a variant of the ‘shi … de cleft’—‘the bare de construction’, which is realized in two different forms, as exemplified below.

(9)

Page 6 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese

(10)

In (9), the de marks the whole proposition in focus, without perceptually salient prosodic marking, which is comparable to the English construction ‘it is the case that’. The question in (10a) elicits focus on its answer only, as indicated by the salient phonological prominence on Zhangsan in (10b), an in-situ focus which typically requires phonological prominence. Moreover, the ‘shi … de construction’ is generally exhaustive, which presupposes the falsity of all alternative sentences with non-entailed focus values (Hole 2012), different from the ‘bare shi focus construction’. This contrast is exemplified in (11a,b), with the exhaustive reading of ‘shi … de clefts’ shown in (11a) and the non-exhaustive reading of the ‘bare shi sentences’ in (11b) (Paul and Whitman 2008). (11)

Hole (2012) also mentioned that when preceding the VP, the copula shi can be stressed to express focus in Chinese, which is referred to as ‘verum focus’, as illustrated in (12). (p. 738)

Page 7 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese

(12)

Hole has further defined Mandarin verum foci as belonging to the same class as answers to canonical yes/no-questions and to the special kind of tag question frequently found in Chinese. Along such a line, we think that when the emphasized copula shi signals verum focus, what is confirmed/emphasized is the truth of the whole proposition. Here, shi is not a focus particle associated with any specific phrase in the sentence and should be differentiated from the shi in the aforementioned ‘bare shi focus construction’ and the ‘shi … de construction’.

36.2.3 Focus marked by lian … dou/ye construction Another widely discussed focus-marking construction is lian … dou/ye ‘even … all/also’, as exemplified in (13b).

(13)

Page 8 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese (14)

Whether the phrase marked by lian … dou is a topic or a focus has long been a controversial issue. Xu and Liu (1998) argued that phrases marked by lian … dou are topical (p. 739) focus while Chu (2003) proposed that they are contrastive topics. It is sufficient to say that lian … dou invokes a [+contrastive] meaning (in the sense of the focus features in Liu and Xu 1998). Since the [+contrastive] feature is generally found in focus but not topic, Chu used (14) to further support the claim that topics can be contrastive, although further studies are needed to tease apart these two proposals. Note that there is a difference between a contrastive focus appearing in the topic and the one in the comment, namely that a contrastive focus appearing in the topic part may be referred to as a contrastive topic, as in Chu (2003), while a contrastive focus appearing in the comment is still regarded as a contrastive focus.

36.2.4 Restrictive focus particles In Chinese, there are a few particles (i.e. jiu, cai, zhi, and dou) which are sensitive to focus and affect the truth condition of a sentence, known as focus-sensitive particles (e.g. Jackendoff 1972; Rooth 1992, 1996). Given their effects on the truth conditional meaning of a sentence, they are also known as semantic focus particles.

(15)

Examples of jiu as a restrictive focus particle are given in (15). Sentence (15a) conveys the meaning that Lisi only speaks German, but not any other languages, while (15b) means that Lisi only speaks German, but not reads, listens, or writes, under the assumption that the alternative set is triggered by the focus [speak]f which may include {listen, read, speak, write}. The interpretation of focus with the focus particle jiu above is generally termed as ‘focus association’ in Partee (1991, 1999). Further discussions on the

Page 9 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese characteristics of this particle can be found in Biq (1984, 1988), Lai (1995, 1999), and Paris (1987). A close counterpart of jiu in Chinese is cai, also sharing the restrictive function, as exemplified below.

(16)

Cai and jiu are seemingly interchangeable in (16a) and (16b) as English ‘only’, with their restrictive function indicated by marking the asserted quantity to be less than expected. Although much research has been conducted on cai and jiu, there is no unified account regarding their semantics. Complications come from their polysemous nature, as gleaned from their four uses: (i) the temporal use indicating ‘immediate past’ or ‘close to’ meaning; (ii) the parametric use; (iii) the limiting or restrictive use; and (iv) the emphatic use (e.g. Biq 1984, 1988; Lai 1995, 1999; Hole 2004, 2006; Paris 1987, as discussed in Hole, 2004). Various proposals have been made, including their being (i) focus particles marking exclusive focus (Biq 1984) or denying-expectation focus (Biq 1988); (ii) connective elements with the discourse function to establish a relation between two units (Paris 1987); and (iii) as scalar particles (Lai 1995, 1999). (p. 740)

Another restrictive focus particle in Chinese is zhi, which behaves similarly to the counterpart of English ‘only’, a typical focus-sensitive operator. In addition, Mandarin Chinese has another adverbial focus particle, dou ‘all’, which can be associated with a contrastive focus in both the topic and the comment. When in the topic, the contrastive focus only introduces an alternative set to serve as the domain of dou quantification, with no exclusive interpretation, like English ‘always’ or ‘all’. When in the comment, the contrastive focus serves to deliminate the scope of quantification so that the sentence excluding the focus functions as the domain. In this use, dou is just like English ‘only’, inducing an exclusive interpretation on the focused element, as exemplified in (17):

(17)

Page 10 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese While contrastive focus in the topic can only serve as the domain of quantification and does not induce an exclusive interpretation, the one in the comment (as in (17)) has to serve as the scope of quantification and does induce an exclusive interpretation. For details, see Pan (2006) and Jiang and Pan (2013).

36.3 Semantic and syntactic marking of topic According to Li and Thompson (1981), what sets Chinese apart from many other languages is that in addition to the grammatical relations of ‘subject’ and ‘object’, ‘topic’ is an indispensable notion for the description of Chinese. Topic, which typically appears at the beginning of the sentence, denotes what the sentence is about and is generally realized as definite or generic noun phrases. In line with Li and Thompson, Xu (2007) considers Chinese to be a topic-prominent language, or a topic configurational language (following É. Kiss 1995), in contrast with English which is a subject-prominent language.

(p. 741)

36.3.1 Topics and subjects

When talking about topic in Chinese, it is important to discuss the relationship between topic and subject. Three views have been summarized in Shi (2000). The first view is a modified version of the tradition started as early as in Mashi Wentong (Ma 1898), which maintains that subject–predicate is the fundamental relationship between VP and preverbal NP(s). If there are two or more preverbal NPs, the VP and the NP closest to it will first form a predicate and this predicate can take another NP as its subject (Lü 1986). Such a view basically denies the grammatical status of ‘topic’, suggesting that Chinese is unique in the sense that there could be as many subjects in a sentence as the number of preverbal NPs. The second view reinstates the dichotomy between topic-prominent languages and subject-prominent languages (Li and Thompson 1976) by claiming that syntactic notions such as subject and object are not grammaticalized in Chinese. The argument is that IS partitioning rather than syntactic structuring is essential in the grammar of Chinese and only topic and focus are grammaticalized in Chinese (LaPolla 1990, 1993). There could be more than one topic in a sentence, if there are two or more NPs in front of the verb (with the ones following the left-aligned initial topic generally considered sub-topics). The third view represents a symbiotic view of the two aforementioned ones, which, according to Shi, has been adopted by the majority of linguists working on Chinese. It states that both topic and subject co-exist in the same sentence in Chinese as separate grammatical notions (e.g. Li and Thompson 1976, 1981; Tsao 1979, 1990; Huang 1982; Li 1990; Jiang 1991; Tan 1991; Xue 1991; Ning 1993; Qu 1994; Shyu 1995).

Page 11 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese Despite the divergent views, Tsao (1979, 1990) gave the following six properties of Chinese topics. (18)

Under the general assumption that topics are grammaticalized in Chinese, it is natural that word order should be a good indicator of topics. Chinese allows variations of SOV (p. 742) and OSV, on top of its canonical order of SVO. Examples below are from Huang et al. (2009), with the original translation from the authors.

(19)

(20)

Page 12 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese With the marked word orders in the above examples, there have been considerable debates on whether topic structures are derived by movement or base-generated, which will be further discussed in the following section.

36.3.2 Moved topics and base-generated/dangling/gapless topics in Chinese Approaches which take topics as base-generated are generally characterized by an aboutness condition (e.g. Chao 1968; Chafe 1976; Li and Thompson 1981; Xu and Langendoen 1985), which is also the traditional characterization used to capture the relation between a topic and its comment in Chinese. But Chinese has also been argued to have moved topics (e.g. the so-called ‘instance topics’ as in Chen (1996)). One approach that has generated much debate in the literature is the dangling topic, also known as Chinese-style topic (Chafe 1976). Examples of dangling topics are given in (21) and (22) (see also, Chao 1968; Li and Thompson 1976, 1981; Tsao 1977).

(21)

Two ways have been proposed to account for dangling topic or ‘gapless’ topic structures. Shi (2000) adopts the movement approach and claims that Chinese does not have dangling topics and maintains that all the so-called base-generated topics can be derived by (p. 743) movement and subsequent deletion, as shown in (22). In this way, he argues for a unified account of both gapped and base-generated topics in Chinese. (22)

Since dangling topics are not always subcategorized by the predicate in question, they may not be related to a syntactic position in the comment, which is the required basis of the movement approach. Shi’s approach thus also amounts to claiming that dangling topic sentences that are not derivable by movement are not topic–comment but are subject–predicate sentences. Pan and Hu (2008) and Hu and Pan (2009), however, argue that all the dangling topic sentences should be treated as real topic sentences, not subject–predicate sentences. They propose an alternative analysis to unify both moved and dangling topics in Chinese. They start their analysis by reviewing the aboutness condition which is generally

Page 13 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese considered to be a condition that requires the comment to be related to the topic. They argue that this aboutness condition in fact fails to account for the contrast between (23a) and (23b). (23)

The acceptability difference in (23a) and (23b) cannot be accounted for by aboutness or relatedness, as proposed in the literature. Although the aboutness condition may correctly predict the unacceptability of (23a) by claiming that the comment is not about the topic, the same account would wrongly exclude the acceptable (23b). The acceptability contrast between (23a) and (23b) shows that it is quite difficult to predict under what specific condition a comment may bear an aboutness relation to the topic, due to the reason that the precise nature of aboutness has never been made clear in the literature. Hu and Pan (2009) provide a more explicit characterization of the aboutness condition. Their proposal is that topics can be licensed either in syntax or at the semantic–pragmatic interface. When being licensed in syntax, there is a syntactic position (which can be filled by either a syntactic gap or a resumptive pronoun) in the comment related to the topic in question. When being licensed at the semantic–pragmatic interface, there (p. 744) is a semantic variable in the comment. Hu and Pan propose a unified account to both cases and it consists of two conditions: a licensing condition and an interpretation condition. The licensing condition stipulates that a topic is licensed if there is a semantic variable in the comment and the set generated by this variable produces a non-empty set when intersecting with the set denoted by the topic. The interpretation condition requires that an element inside the comment forms a subject–predicate relation with the topic. The two decomposed components of the aboutness condition have the advantage of unifying both gapped and gapless topic structures within a single account. Notice that such a licensing condition for dangling topic can avoid the aforementioned problems with the aboutness condition. Example (23a) is unacceptable because the empty possessor of erzi ‘son’ can be easily identified as the relevant variable. However, if the empty possessor, that is, the fathers of sons, is the variable, the set thus generated produces an empty set when intersecting with the set of the topic which consists of all the children in the kindergarten. Sentence (23b) is acceptable because the intersection of the Page 14 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese topic set with the set projected from the focus phrase erzi ‘son’ is not empty, as the alternative set to the focus is not the set of fathers in (23b), but all the entities which are compatible to erzi. Hence the relevant intersection produces a non-empty set, which is why (23b) is grammatical. Hu and Pan’s decomposition analysis is argued to be a more precise characterization of the aboutness condition, with the advantage of unifying both the moved and base-generated topic sentences.

36.3.3 Preposing objects and Topic in Chinese Although Chinese has a canonical order of SVO, object preposing is common in Chinese, leading to an alternative word order SOV. Consider (24) below, cited from Ernst and Wang (1995).

(24)

For sentences like (24a), the commonly adopted analysis is double topicalization (e.g. Xu and Langendoen 1985; Lee 1986; Tang 1990; Lin 1992), in which the object first adjoins to IP, producing the sentence in (24b). The subject wo may then subsequently topicalize across the object, giving (24b). Ernst and Wang (1995) argue against the double topicalization analysis, and propose that sentences like (24a) are derived by adjoining the object directly to VP. One major argument comes from the distribution of adjuncts. (p. 745) Specifically, for adjuncts which adjoin only to VP, double topicalization would predict that objects will always precede these adjuncts. On the other hand, if the preposed object is assumed to be directly adjoining to VP, it would predict that these adjuncts may precede the preposed object. Ernst and Wang appeal to sentences like (25) to support their claim.

Page 15 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese

(25)

Example (25a) is used to show that the adverb yizhi ‘continuously, always’ must occur in VP, and (25b) shows that it may precede a preposed object. Based on (25), Ernst and Wang argue that sentences with preposed object should not be considered as double topicalization, and the preposed object is in fact directly adjoined to VP, although whether preposing objects should be considered as topic or not remains debatable and calls for further research.

36.4 Prosodic marking of focus and topic

Page 16 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese

36.4.1 Focus Most work on the prosodic realization of IS in Chinese has been concerned with focus. An earlier consensus on focus in West-Germanic languages is that it is encoded via specific pitch accents, which, in turn, are signalled mainly via fundamental frequency changes (e.g. Jackendoff 1972; Selkirk 1984, 1995; Pierrehumbert and Hirschberg 1990). This has led to the question of whether prosody is at all used to signal IS in tonal languages like Chinese, where changes of fundamental frequency (hereafter f0 or perceptually speaking, pitch) over a syllable encode lexical meaning. The first phonetic study on focus in Chinese dates back to Gårding et al. (1983) which shows that in Standard Chinese, under focus, lexical tones are realized within an expanded pitch range grid and out of focus, compressed range grid. This pattern has been repeatedly observed in many subsequent studies on focus in Standard Chinese (e.g. Shih 1988; Jin 1996; Xu 1999; Chen 2003; Chen and Braun 2006; Chen and Gussenhoven 2008; Chen et al. 2009; Greif 2012). Listeners also show sensitivity to such a partition of f0 range in focus perception (Xu et al. 2004). Xu and Xu (2005) therefore made an explicit proposal that focus is directly encoded via the tri-zone pitch (p. 746) range control: Expansion under focus, compression after focus, and little or no change before focus. Chen (2003, 2010) and Chen and Gussenhoven (2008) took a closer look at the mechanism of lexical tone modification under focus in Standard Chinese and argued that the observed f0 adjustments lead to an enhanced implementation of the tonal contours that cannot be reduced to mere f0 range expansion. Specifically, on-focus tonal gestures are implemented with magnified and more distinct f0 contours, and exhibit less coarticulatory influence from neighbouring tones, comparable to the distinctive realization of segmental contrasts under focus-induced prominence in West-Germanic languages (e.g. de Jong 1995; Erickson 2002; Cho and McQueen 2005). Such a prominence-marking view of focus (in contrast to the f0-range marking view of focus) makes it possible for a unified account of focus-induced prosodic encoding across languages with typologically different prosodic systems. Under this view, focus introduces abstract prosodic prominence. Languages then only differ in their different instantiations of focal prominence in terms of their available phonological entities and phonetic cues to encode focus. For example, English employs intonational tonal events as a salient cue while Standard Chinese manifests focal-prominence more in the distinctive realization of lexical tones, since the addition of pitch accents is prohibited. In addition to f0, prior research on Standard Chinese has also reported that words under focus generally have longer duration (Jin 1996; Xu 1999; Chen 2006) and higher mean intensity (Shih 1988; Chen et al. 2009; cf. Jin 1996, for the variability of this parameter in focus encoding). With durational data, Chen (2006) showed that acoustic modification under focus is constrained by prosodic boundaries. For a quadri-syllabic monomorphemic word in Standard Chinese, focus on the initial syllable of the word results in robust lengthening of not only the target syllable but also the following non-focused syllable. Page 17 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese When the second syllable is focused, however, foot boundary clearly attenuates the magnitude of such spill-over lengthening over the third syllable which shows a limited amount of lengthening. Given the various acoustic cues involved in focus-marking, new studies are yet to be carried out to compare the relative cue weighting of the different acoustic parameters in signalling focus. Within the Sinitic family, both cross-dialect similarities and differences have been reported on the prosodic encoding of focus. In Shanghai Chinese, a Wu dialect spoken in the metropolitan city of Shanghai, focus is realized with prosodic prominence, which generally leads to constituent lengthening and f0 range expansion (Selkirk and Shen 1990). When a syllable contains a short vowel, however, focus fails to introduce robust lengthening; when a high-register rising tone is under focus, f0 expansion also becomes restricted (Chen 2009). This is probably because the potential cues for focus expression, that is expansion of f0 range over a high-register rising tone and lengthening of a short vowel, would make the targeted focused items easily run afoul of lexical neutralization to items with low-register rising tone and long vowel, respectively. (For more details of the lexical tonal system and segmental structure of Shanghai Chinese, see Chen and Gussenhoven, 2015.) Of note is that focus does enhance the distinctiveness (p. 747) of both lexical tonal contours (Chen 2009) and vowel formant patterns (Chen 2008a), as predicted by the prominence-marking view of focus. In Cantonese, a Yue dialect spoken mainly in Guangdong and Hong Kong, focus has been reported to introduce both lengthening and higher intensity, but only a variable effect on f0 (Man 2002; Wu and Xu 2010; Gu and Lee 2007). Furthermore, different from the trizone f0 manipulation often observed in Mandarin, Cantonese shows no post-focus f0 compression regardless of the identity of the preceding on-focus tone. The Min dialect is another Sinitic language which shows a robust effect of lengthening under focus but lacks consistent focus-related f0 range manipulation. For example, Pan (2007) shows that in Taiwanese, a Southern Min dialect spoken in Taiwan, f0 raising/expansion is clearly present in the HH and HL tones but not so evident in the ML and MM tones. Taiwanese therefore parallels Shanghai Chinese in that focus-related f0 modification is consistent only when it does not entail sacrifice in the distinctive realization of lexical tonal contrasts. In the post-focus position, Xu et al. (2012) further shows that Taiwanese lacks f0 compression, just like Cantonese. Taken together, the high functional load of pitch register for tonal contrasts in both Cantonese and Taiwanese seems to play an important role in forbidding post-focus pitch range compression. Future studies, preferably with more tonal combinations and varied syntactic structures, are desirable to substantiate the current findings and further examine how various factors dictate the prosodic expression of focus. It is well-known that focus may differ in the size of its domain (or breadth) depending on the context, which ranges from words (known as ‘narrow focus) to syntactic phrases or full sentences (known as ‘broad focus’) (e.g. Schmerling 1976; Ladd 1980, 1996; Gussenhoven 1983; Selkirk 1984). Much phonological work has been done to investigate the prosodic reflex of focus with varying breadth. It is generally held that the domain in Page 18 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese which focus attracts the strongest stress as its prosodic reflex is the semantic scope of the focus (e.g. Truckenbrodt 1995, 2012; Féry and Samek-Lodovici 2006). A common assumption of this body of work is that prosodic prominence (i.e. stress) for focus entails the presence of an accent in West-Germanic languages (Selkirk 1995, 2002, 2007), a phonological property which is absent in tonal languages such as Standard Chinese. Phonetically speaking, however, Chinese seems to be just like Germanic languages (e.g. Breen et al. 2010). When an SVO sentence in Standard Chinese is produced with different degrees of broad focus, the most prominent element tends to be on the object, comparable to the case of narrow focus on object although a moderate amount of durational increase and/or f0 expansion can also be observed on the preceding constituents (Jin 1996; Chen 2008b). Some Chinese dialects exhibit an interesting constraint on the minimal domain for the prosodic reflex of focus when the size of an expression for focus domain narrows down beneath the lexical level. This is different from West-Germanic languages where focus can be expressed over a domain below the syllable level (van Heuven 1994; Sluijter and van Heuven 1995). A case in point here is Wenzhou Chinese, where a disyllabic word serves as the domain for tone sandhi. Scholz (2012) shows that when only one syllable within a bi-syllabic sandhi domain is contrastively focused, f0 range expansion and durational lengthening are quite uniformly distributed over the entire disyllabic (p. 748) domain, suggesting the constraint that the tone sandhi domain serves as the minimal domain for focus expression and focus cannot break up its internal structure (Chen 2009 for further discussion on Shanghai Chinese; cf. Chen 2000 on the effect of focus in sandhi domain formation). These findings lend further support to the prominence-marking view of focus, which takes focus-induced f0 range and duration increase as a consequence of more abstract, prosodic prominence brought about by focus, whose expression is contingent upon the prosodic structure of the focused constituent. Not only may the breadth of focus vary, an utterance may also vary in the number of constituents that are under focus. Jia et al. (2010) show that when an utterance contains two pragmatic/semantic foci, both are prosodically expressed as equally prominent. Under a multiple-focus condition (i.e. with three foci), however, only the rightmost focused element is realized with prominence. Kabagema-Bilan et al. (2011) report a different pattern. In their double-focus condition, only the second of the two foci shows phonetic f0 effects, while the first is neither cued via f0 nor duration. Further research is necessary to investigate the discrepancy of the two studies. What can be drawn from the studies is that a one-to-one mapping between focus and its prosodic expression is unviable, lending further support to the need for an abstract and indirect mapping relationship between focus and its prosodic expression. Two other findings are worth noting. Chen and Gussenhoven (2008) show that more emphasis under corrective focus leads to significant durational lengthening and little f0 range expansion, suggesting a constrained role of focus-induced f0 modification. The flexibility that Standard Chinese enjoys in lengthening is probably due to the lack of length contrast at the segmental level in the language. In the face of the different degrees Page 19 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese of ‘expandability’ of f0 range and duration under different degrees of emphasis, the lexical tonal f0 gestures (in particular of the rising and falling tones) are nevertheless implemented in various ways so as to remain distinctive under the robust lengthening of the tone-carrying syllable. This again suggests the importance of prominence-marking, instead of mere f0 range expansion, as the prosodic reflex of focus. Liu and Xu (2005) report that when all target lexical tones are high level, focus is typically marked in a similar way in questions as in statements despite the fact that questions are cued mainly with f0 range modification. Their perception test reports the lowest identification rate for neutral focus (i.e. focus over the whole sentence) in questions and final focus in statements. While more lexical tones should be included for future research, the results of this study suggest that the parallel nature of the f0 encoding schemes for focus and interrogativity (for high-level tone) can be compromised/ constrained when there are competing demands for f0-range adjustments for different communicative functions.

Page 20 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese

36.4.2 Topic Over the last decades, research devoted to the study of topicalization in Chinese has mainly been concerned with semantic and syntactic issues (as reviewed in Section 36.3). Relatively little has been done on the prosodic encoding of topic. (p. 749)

Chen (2009) examined the prosodic realization of bi-syllabic nouns in Shanghai

Chinese, produced as non-contrastive topic, contrastive topic, and focus, in a question– answer reading task. Focus induces significant lengthening over the noun phrase in question, compared to the contrastive topic condition, which, in turn, is significantly longer than the non-contrastive topic condition. f0 manipulation, however, is more varied and depends, to a large extent, on lexical tonal identity. Details aside, focus ensures a magnified and distinctive tonal gesture over the target noun phrase (and therefore may involve lowering of the f0 minimum), while contrastive topic raises the whole f0 contour. Furthermore, the verb following the target noun shows a significantly lowered f0 register post focus, as compared to the two topic conditions, which suggests a more local effect of prosodic marking for topic. Wang and Féry (2010) investigated the prosodic encoding of split noun sentences (i.e. OSV constructions with left object dislocation) in Standard Chinese. The object vs. noun was elicited with different preceding contexts which the authors classify as topic vs. focus. Their results show that focus introduces lower f0 maximum in the base part and shorter silence after the object noun than topic. No difference, however, was found on the dislocated object. The results thus provide further evidence for the global vs. local prosodic modification for focus vs topic. Wang and Xu (2011) is another study on prosodic realization of noun phrases in Standard Chinese with different information status. Given the way the data were elicited, it is difficult to pin down, within the framework in Krifka (2008), the specific IS notions they investigated. A general pattern to be noted is that focus raises on-focus f0 and lowers post-focus f0 while topic raises the f0 register at the beginning of the sentence and allows f0 to drop gradually afterwards, in line with the findings in Chen (2009) for Shanghai Chinese and Wang and Féry (2010) for Standard Chinese. These studies jointly reveal that contrast and newness are not directly encoded as pure f0 range manipulation. Needless to say, given the limited number of studies on topic and prosody, well-controlled experiments remain a must to substantiate these observations and to further understand the range and nature of the prosodic marking of topicalization in Chinese dialects.

36.4.3 Lacunae for studies on the prosodic expression of focus and topic Despite the progress that has been made over the last decades on prosody and IS marking in Chinese and various Chinese dialects, much more is yet to be done. Five

Page 21 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese lacunae have been identified which should serve as important questions on the agenda of the research community. First, an important aspect that has been neglected in all works of Chinese focus expression is the necessity to explicitly distinguish givenness and focus. Crosslinguistically, it has been argued that givenness should be treated as a fundamental concept in information structuring which is marked not only lexically (e.g. Ariel 1988; Gundel et al. 1993) and syntactically (e.g. Kučerová 2012), but also prosodically (e.g. Pierrehumbert (p. 750) and Hirschberg 1990; Kohler 1991; Ladd 1996; Féry and SamekLodovici 2006; Selkirk 2008). Changes in a referent’s level of givenness (or accessibility) are reflected in the corresponding adjustments in its prosodic marking (Baumann 2006) and can result in different updating costs of on-line discourse representation, as shown in a recent reading experiment that measures eye movements (Chen et al. 2012). Readers of Chinese are sensitive to the difference between focus and givenness: a critical word was read more slowly when it was the name of a previously-unmentioned individual (i.e. new) than when the name had been used earlier (i.e. given). This, as argued further by Benatar and Clifton (2014) with English data, suggests a strong influence of givenness as a core IS notion even in silent reading. Previous experimental work on focus in Chinese, however, has treated the prosodic effect of focus heuristically, without teasing apart the expression of focus from the prosodic encoding of givenness over contextually entailed constituents within the same utterance. For example, a pertinent question here is to what extent the observed compression of pitch range after a focused element should be accounted for as the effect of focus expression or as the prosodic encoding of givenness (see further discussion on post-focus compression in Chen 2010). Second, most prior studies did not explicitly differentiate the prosodic effects of different types of focus. Under the cover term focus, the methods employed to elicit data on focus in Chinese often include making corrections over (part of) a word (e.g. Chen 2006) or answering WH-questions set up in the preceding discourse (e.g. Jin 1996; Xu 1999). The potpourri of their results are often discussed and compared under the cover term focus. Within the framework of Krifka (2008), the former can be classified as expression focus while the latter denotation focus which has no consequence on the truth condition of the target utterances (i.e. involving only the pragmatic use of focus). An interesting question here is whether and how a language with a rather complex lexical prosodic system (such as complex lexical tonal inventory) allows for relatively fine-grained distinctions in the prosodic packaging of IS. Chen and Braun (2006) is the first study, to our knowledge, to compare the effects of various types of IS notions on the prosodic realization of bi-syllabic names in Standard Chinese. The study adopts the framework of Vallduví and Vilkuna (1998) and Steedman (2000) and tests whether in a tonal language, IS operates along two semantic dimensions with a primary distinction between theme and rheme and a secondary distinction of focus and background. Translating the results into the framework in Krifka (2008), we note that Page 22 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese utterances that are new in discourse (i.e. informational focus with a broad scope) need to be differentiated from given information. The prosodic marking of alternatives to provide new information over a sub-constituent (i.e. narrow informational focus) is also different from that to correct information (i.e. corrective focus). This has been echoed in a few more recent studies. For example, Wang and Xu (2011) shows that when newness is independently controlled from focus, a new word is only slightly longer than a given word. Comparable results have been reported in Ouyang and Kaiser (2015). Further support for contrast and newness to have different information status comes from Chen et al. (2012), an eyetracking experiment, and Chen et al. (p. 751) (2014), an event-related brain potential study. Both reveal differences in processing patterns between focus and newness, and suggest that they are different concepts that relate to different aspects of cognitive processing. Third, there is a general lack of care in data elicitation which simulates natural-occurring speech communication. By now, we all recognize that the basic notions of IS should be rooted in theories of how communication works. Krifka (2008) motivates the choice of IS notions within a model of speech communication where the interlocutors continuously update their common ground (CG) in terms of both CG content and CG management. Little, however, has been done on the possible effects of interlocutors’ mutual belief or CG on the prosodic marking of focus in Chinese. One happy exception is Greif (2012) who shows that we need to differentiate different types of corrective focus, one with the corrigendum being foregrounded assertion (A-COR) and the other presupposed background (P-COR), as compared to the baseline neutral information focus (NIF). P-COR differs from both A-COR and NIF with expanded f0 range and increased duration while ACOR shows a slight lowering of f0 than NIF. These results, yet to be replicated, suggest that studies of IS marking need to tune into perspectives of both speakers and listeners. A related problem is that most studies on the prosodic encoding of IS in Chinese have elicited data via reading aloud scripted dialogues. A few studies have emerged recently which elicited semi-controlled data by asking participants to describe visual scenes according to different contextual questions (e.g. Greif 2012; Ouyang and Kaiser 2015). It has also become clear that the more natural the task becomes, the more linguistically complex the stimuli are, and the more variability we observe in the data. We thus not only face the challenge of how to elicit naturally-occurring data which reflect genuine patterns of information exchanges between interlocutors but also need to address the issue of how to make sense of the variability observed in the prosodic encoding of information structure notions. A fourth issue to be addressed is to understand further the underlying mechanisms that account for the prosodic encoding of various IS notions. While there is a general consensus that focus induces prosodic prominence, various proposals have been made on how to pursue the idea further (see Rooth 1992; Truckenbrodt 1995; Selkirk 2002, 2007; Féry and Samek-Lodovici 2006; Büring 2009 for different proposals). Such an indirect approach of focus-marking helps to explain the way various factors (such as tonal register and tone sandhi domain) condition the way lexical tones are modified to signal Page 23 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese information structure, in addition to other acoustic cues, as discussed earlier (e.g. Chen 2003; Chen and Gussenhoven 2008; Chen 2010; Scholz 2012). This view also accounts for recent findings on the constrained prosodic expression of multiple foci in Standard Chinese (Kabagema-Bilan et al. 2011). The prominence-lending view of focus also raises the question of whether in addition to focal prominence, focus inserts a new prosodic boundary and introduces prosodic (re-)phrasing (e.g. Shih 1997 for Standard Chinese; Selkirk and Shen 1990 for Shanghai Chinese; and Chen 2000 for Wenzhou Chinese). There is some evidence which suggests that focus does not directly introduce rephrasing; rather, focus-induced acoustic modifications are sensitive to and respect (p. 752) the prosodic structure of the utterance (Chen 2004, 2010; Scholz and Chen 2014). More studies, however, are certainly needed to understand the interaction between the prosodic marking of focus and the general prosodic marking of utterance phrasing. Last but not least, we believe that another important question on the agenda of the research community is the way the IS of an utterance is planned during speech production. Little is known about this process. Ganushchak et al. (2014) reported an event picture description experiment where different focused referents (as a function of different preceding discourse contexts) induce different eye gaze shift patterns. This suggests that discourse, which in part determines the IS partition of an utterance, affects the processing of different focused referents. Furthermore, there is a cross-linguistic difference between Dutch and Chinese speakers, probably due to the different linguistic structures of the two languages (in particular with regard to their different WHplacements in the preceding discourse). Much more work, however, is needed to understand the formulation of a linguistic message with different patterns of information packaging.

36.5 Concluding remarks This chapter has reviewed how two important notions of information structure—topic and focus—are encoded in Chinese, a lexical tone language. Chinese is known as a topicprominent language with a great variety of topic structures. It has also been classified as a wh-in-situ and focus-in-situ language (e.g. Huang 1982; Soh 2006; Hole 2012), with rich syntactic constructions for focus-marking. We have shown that there is a wide range of syntactic constructions, semantic particles, as well as a conglomerate of prosodic cues that speakers of Chinese have at their disposal to structure information so as to achieve effective communication. However, various issues are yet to be further explored. For example, it is debatable whether preposing objects should be considered as focus or topic. Both topical focus and contrastive focus bear the [+contrastive] feature. In what way can relevant syntax–semantics mapping account for their differences? Further investigation of different topic constructions across Chinese dialects can also be a fruitful enterprise (Xu and Liu 1998). We have also reviewed various debates and experimental findings concerning the underlying mechanisms of the prosodic encoding of various IS Page 24 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese notions. While much progress has been made over the last decades, it is also apparent that much more experimental and theoretical work needs to be done (and preferably done together) in the future for us to gain a deeper understanding of IS marking in Chinese.

Notes: (1) Abbreviations used in this chapter include: CL: classifiers; EXP: experiential markers; PERF: perfective markers; PROG: progressive markers; and SFP: sentence-final particles. (2) Prosodic prominence is marked in boldface in this chapter. (3) Contrastive foci are marked with the notation []f in this chapter.

Yiya Chen

Yiya Chen is a Lecturer at Leiden University Center for Linguistics (LUCL) and senior researcher at Leiden Institute for Brain and Cognition (LIBC). Her research mainly focuses on prosody and prosodic variation, with particular attention to tonal languages. Her work has recently appeared in Journal of Phonetics and Phonetica. Peppina Po-lun Lee

Peppina Po-lun Lee is Associate Professor at the Department of Linguistics and Translation, City University of Hong Kong. Her research interests lie primarily in Chinese and Cantonese syntax and semantics, and she has worked on a variety of research topics, including quantification, focus and information structure, negation, aspect and eventuality. Her publications include Cantonese Particles and Affixal Quantification (Springer, 2012) and about 60 research papers in edited books and international journals such as Lingua, Journal of Pragmatics, Linguistics, and Language and Linguistics. Haihua Pan

Haihua Pan, Ph.D (Texas) is Professor of Linguistics, Department of Linguistics and Translation, City University of Hong Kong, and Changjiang Scholar-Chair Professor (2012-2015), offered by the Ministry of Education of China, associated with Beijing Language and Culture University. His research interests include Syntactic Theory, Formal Semantics, Corpus Linguistics, and Computational Linguistics.

Page 25 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Topic and Focus Marking in Chinese

Page 26 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 04 November 2018

Information Structure in Japanese

Oxford Handbooks Online Information Structure in Japanese   Satoshi Tomioka The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara Print Publication Date: Jul 2016 Subject: Linguistics, Semantics Online Publication Date: Feb 2016 DOI: 10.1093/oxfordhb/9780199642670.013.42

Abstract and Keywords Japanese is known to have rich encoding of information structure in its grammar. Focus and givenness can be prosodically marked, and there are syntactic structures that are sensitive to information structure, such as cleft sentences, right dislocation, and a variety of phonologically silent structures. This chapter introduces core empirical facts surrounding Japanese information structure, which form the basis of numerous theoretical endeavours. Special attention is paid to the properties of wa-marked phrases. While the particle wa is closely tied to the notion of ‘aboutness’ topic, it has other uses that are not obviously connected to aboutness. The grammar of wa-marking also figures prominently in the discussion of a few additional issues that the chapter addresses, namely contrastiveness and its possible link to scalar implicature, and the recursively/ embeddability of topicality in the context of ‘embedded root phenomena’. Keywords: information structure, Japanese, grammar, focus, givenness, topicality, wa, recursivity, embedded root phenomena, wa-marking

THIS

chapter presents a compact overview of the information structure of Japanese,

focusing on three major concepts of information structure; focus (Section 37.2), givenness (Section 37.3), and topicality (Section 37.4).1 It is by no means a comprehensive and complete portrayal of all the relevant facts, as the richness of encoding information structure in the language makes such a task practically impossible to accomplish. There are inevitably some non-trivial issues that are left out or are only touched upon briefly, and I will make reference, whenever appropriate, to other sources from which more information can be obtained.2 On the other hand, I have chosen a few controversial or frequently misunderstood issues for more rigorous examination, as they provide important clues for understanding the universality and the cross-linguistic variability of information structure. The recursivity of information structure (Section 37.5) and the

Page 1 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese relation between contrastiveness and scalar implicature (Section 37.3.2) are among such topics.

37.1 Expressing focus 37.1.1 Focus prosody In the majority of natural languages, focused items are assigned prosodic prominence of various kinds (e.g. stress, higher pitch), and Japanese is no exception. The manifestation of focus prosody in Japanese differs, however, depending on whether foci are contrastive (p. 754) or not. Differences of this sort are not unknown. Katz and Selkirk (2011) report, for instance, that English speakers make a systematic phonetic distinction between contrastive focus and non-contrastive discourse-new focus.3 In the previous literature on Japanese prosody, much more attention has been paid to contrastive focus, presumably because the effects are more noticeable. There are two specific prosodic phenomena that have been reported in connection to contrastive focus in Japanese. One is ‘focal f0 rise’, in which the pitch (f0-peak) of a focused item in a sentence is raised. The second process is ‘post-focal reduction’, a process which reduces the f0-peaks of the material that follows the focus. At the descriptive level, these phenomena are universally acknowledged. Opinions differ, however, on how these effects should be characterized theoretically. The most popular analysis connects these prosodic effects with modifications of phonological phrasing (e.g. Pierrehumbert and Beckman 1988; Nagahara 1994; Truckenbrodt 1995; Selkirk 2006). Nagahara (1994), inspired by Pierrehumbert and Beckman (1988), proposed the following mechanism for the two prosodic effects of focus. (1)

There are some empirical objections to this ‘rephrasing’ analysis of focus prosody. First, the idea of phrase boundary insertion before a focused phrase has been challenged by Shinya (1999) and Kubozono (2007), whose experimental studies show that focused phrases are subject to downstep when they are preceded by accented words/phrases. If Page 2 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese the f0 boost on a focused phrase is a result of the insertion of a phrase boundary which results in the cancellation of downstep, as stated in (1b), a focused phrase is expected to be free from downstep effects, contrary to Shinya’s and Kubozono’s findings. The mechanism of post-focal reduction under the rephrasing analysis has also encountered challenges. Sugahara (2003) claims that the post-focal deletion of phrase boundaries takes place only when the post-focal material represents given information. When it represents new information, on the other hand, f0-lowering occurs, but the phrase boundaries are retained. The preservation of phrase boundaries following focus is replicated in other experimental studies, such as Ishihara (2007) and Féry and Ishihara (2010). The rival theory to the rephrasing analysis (e.g. Poser 1984; Shinya 1999; Féry and Ishihara 2010; Ishihara 2011) argues that the focus prosody materializes as a result of (p. 755) a pitch register manipulation that is independent of phonological phrasing. Ishihara (2011) furthermore presents a phonological account of this alternative approach by modifying the optimality-theoretic analysis proposed by Truckenbrodt (1995). He proposes to re-rank some of the constraints and to permit multi-headedness in a prosodic representation, which leads to a preference to avoid rephrasing (both boundary insertion and boundary deletion) in Japanese. As for non-contrastive focus, the effects are arguably weaker. For instance, Vermeulen (2012) reports, citing Deguchi and Kitagawa (2002) and Ishihara (2007), that in a question–answer pair, the f0-rise of the answer constituent is typically not as sharp as the one observed for the wh-phrase in the question. It remains to be seen whether the difference is categorical or is a matter of gradable scale. It has also been observed (e.g. Ishihara 2001) that an item that comes immediately before a verb receives the main prosodic prominence by default in Japanese. Therefore, the immediately pre-verbal position is the ideal location for discourse-new focus. It is also the position from which ‘focus projection’ (cf. Selkirk 1984) takes place. In (2), the entire VP corresponds to the main locus of the answer, and the primary prominence (indicated by the ACCENT ´) must be placed on the direct object. Placing it on either of the other constituents leads to infelicity.

(2)

37.1.2 Syntax of focus

Page 3 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese There are some syntactic operations that are frequently identified across languages as focalization strategies. The cleft and the pseudo-cleft constructions are among such strategies, and, not surprisingly, Japanese also has a similar construction, which takes the following form: it begins with a clausal structure with a gap, followed by the particles no and wa in that order, and it ends with the focused expression, which corresponds to the gap in the preceding clause, and the copula da. (3)

The focused items, those that are attached to the copula –da, are interpreted exhaustively. In (3a), for instance, it is understood that (the) three apples are the only relevant (p. 756) items that Mari ate. More controversial is the syntax of this construction. The particle no is identified as a nominalizer, and with it, this construction is usually analysed as a biclausal structure with an embedded clause buried within a nominal structure. Thus, (3a) is paraphrased as ‘the things Mari ate are (the) three apples’, which is more or less identical to the conventional understanding of what the pseudo-cleft construction means. However, Hiraiwa and Ishihara (2012) argue that the nominal analysis does not uniformly apply to the Japanese cases. First, following Hoji (1987) and Fukaya and Hoji (1999), they categorize the construction into two sub-types, based on the presence vs absence of case morphology in the focus position. If the focus has no case particle (e.g. (3a)), the sentence is indeed a pseudo-cleft, a bi-clausal structure with no acting as a nominalizer. When the focus retains a case particle, on the other hand, it has an entirely different structure. Hiraiwa and Ishihara note (i) that the morpheme no cannot be replaced by a more semantically contentful noun, such as mono ‘thing’ or hito ‘person’, and (ii) that the nominative–genitive conversion, a signature property for a clause contained within a nominal structure, is not available when the pivot retains its case morphology. Importantly, these two nominal properties are observed with caseless foci. Hiraiwa and Ishihara (2002, 2012) propose a mono-clausal analysis of clefts. A summary of their analysis is given in (4).

Page 4 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese (4)

Example (3b) has the following derivation, starting with (5b). (5)

Under this model, the Japanese cleft construction does not resemble the bi-clausal structure that is customarily associated with clefts across languages. While the new analysis accounts for the unexpected facts listed above, it presents its own mystery. In the pseudo-cleft structure, the exhaustive meaning for the pivot can be attributed to the definite-description-like interpretation of the nominalized clause. No such strategy is available in the new analysis, however, and Hiraiwa and Ishihara (2012) attribute the exhaustivity to the focus movement of the pivot. The problem with this strategy is that there is no independent criterion to identify such a movement, as the examples below (p. 757) illustrate: (6a) shows that a dative NP with the additive particle mo ‘also’ can be fronted within the no-da structure. However, the clefted version of this sentence is ungrammatical, as shown in (6b), presumably because the lexical meaning of mo as the additive particle is incompatible with the exhaustive meaning associated with this construction.

Page 5 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese (6)

Hiraiwa and Ishihara (2012) would regard the fronting operation in (6a) as an instance of scrambling, rather than focus movement, since the additive meaning is incompatible with the exhaustivity associated with the latter movement. Such an analysis is, however, simply a statement of the facts, rather than an explanation of them, and what is lacking is an independent and reliable way to distinguish a focus movement from other movements. As a matter of fact, there has been no solid evidence that confirms the existence of focus movement in Japanese, whether it elicits exhaustivity or is based on a broader definition of focus. Focused items can scramble, but it is far from clear that such a movement is triggered by focus. Long distance scrambling is perhaps the best candidate for focus movement, as Miyagawa (2006) claims that it obligatorily leads to a contrastive interpretation. While contrastiveness is much more routinely associated with longdistance scrambling than local scrambling, it remains a mere tendency, however. In the following example, in which the embedded clause has a wh-phrase, the scrambled object does not seem to elicit any sense of contrast.4 (7)

Scrambling can affect the focus interpretation by moving un-focused items from the default focus position (cf. Neeleman and Reinhart 1998; Ishihara 2001). As mentioned earlier, the immediately preverbal position receives the default prominence, which (p. 758) makes it a suitable position for a discourse-new item. The example below is the scrambled version of (2), in which the locative PP is now in the immediately preverbal position and is interpreted as discourse-new focus.5

Page 6 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese

(8)

37.2 Expressing givenness There are multiple strategies to indicate the given status of linguistic expressions. Such strategies include prosodic reduction, ellipsis/empty structure, right dislocation, and wamarking. As mentioned in Section 37.2.1, the prosodic manifestation of contrastive focus has two components: the pitch boost on a focused item, followed by the compression of the pitch range of the material that follows the focused phrase. Prosodic reduction is commonly conceived as one effective strategy to express backgrounded/given information, and the ‘post-focal reduction’ in Japanese naturally presents itself as an ideal environment for discourse-old expressions to appear. In connection with the post-focal reduction, scrambling once again plays an interesting role. If a focused expression scrambles, the post-focal reduction domain expands, as previously pre-focus material now follows the focus after the movement. Therefore, scrambling of a contrastively focused phrase, such as a wh-phrase, has the effect of extending the domain of given material.6 Phonologically silent structures, either silent anaphora or ellipsis, are sometimes regarded as the ultimate form of reduction and, in this sense, only given material can become silent. The issues of how missing elements are represented and how their meaning can be recovered still divide the field (cf. Winkler this volume), and there have been extensive theoretical endeavours on ellipsis and null anaphora in Japanese. First of all, Japanese allows verbal arguments to go missing quite frequently. It is categorized as a ‘discourse pro drop language’ (cf. Huang 1984), in which the familiarity in the discourse is the primary condition for silent arguments. Hasegawa (1984/85) argues that they are pronouns with no (p. 759) phonological substance whereas Otani and Whitman (1991) propose that a silent internal argument is a case of disguised VP ellipsis that takes place after a verb raises out of a VP. Hoji (1998) challenges the VP ellipsis approach of Otani and Whitman (1991) based on a wide variety of empirical facts, some of which became the foundation for the most recent trend in the syntax of silent arguments, namely the argument ellipsis analysis (e.g. Saito 2007; Takahashi 2008b, 2008a). The main advantages of analysing silent arguments as instances of nominal ellipsis are twofold: (i) silent arguments yield so-called sloppy interpretations while their overt counterparts do not, and (ii) some silent arguments can be quantificational. These facts can be accounted

Page 7 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese for by assuming that silent arguments are not pro-form expressions but are instances of ellipsis and have a full-fledged structure at LF. However, those merits are not necessarily deterministic. As pointed out in Tomioka (2003), sloppy interpretations are not banned for overt pronouns. The distribution of overt sloppy pronouns is more restricted than their silent variants, but exactly how it is constrained is still poorly understood. It is also unclear whether the comparison between a silent argument and its overt pronominal counterpart is a reliable test. In Kurafuji (1999), null and overt pronouns are not regarded as semantically homogeneous, and such a hypothesis makes the null–overt contrast in sloppy readings hard to evaluate theoretically. The quantifier argument also encouters some challenges. In particular, downward monotone quantifiers cannot seem to be antecedents for null arguments. Prenominal downward monotone quantifiers are quite rare in Japanese, but X-paasento miman-no ‘less than X%’ is semi-acceptable. When it is used as an antecedent of a silent argument, however, it produces a very dubious result. (9)

The first sentence is certainly not the most natural way to express the proposition, but the dubiousness of the second sentence is far beyond the matter of naturalness. It utterly fails to mean what it is intended to mean.7 The ellipsis analysis does not predict this fact.8 Another type of ellipsis that has been frequently discussed is sluicing. Takahashi (1994) initiated the debate by proposing that sluicing in Japanese is overt wh-movement (p. 760) followed by ellipsis of the clausal structure. Nishiyama et al. (1995), Kuwabara (1996), and Kizu (1998) present an alternative analysis in which sluicing is a reduced form of a cleft sentence with the renmant wh-phrase occupying the focus position, noting that there are some close correspondences between the two constructions. Furthermore, Fukaya and Hoji (1999) notice that case-marked and caseless wh-remnants in Japanese sluicing exhibit a number of different properties which are duplicated with cleft sentences, including island-sensitivty. Their observations played a key role in the aforementioned

Page 8 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese analysis of the cleft construction by Hiraiwa and Ishihara (2012), who also endorse their own version of the ‘sluicing-as-reduced-cleft’ approach. While Japanese is often considered to be strictly verb-final, it is actually not uncommon in colloquial speech to find some constituents placed after the matrix verb. This ‘right dislocated’ structure is arguably sensitive to givenness. Tanaka (2001), for instance, claims that the post-verbal position in right dislocation cannot host new information. However, it is more accurate to call it an ‘anti-narrow focus structure’, since it cannot host such items as a wh-phrase (10a) or an answer to a wh-question (10b), but a discourse new item can appear in the dislocated position, as pointed out by Shimojo (2005). For instance, (10c) can be uttered at the beginning of a conversation. (10)

Finally, the particle wa is often associated with giveness, but its meaning is much more multi-layered. In particular, the particle is tied very closely to topicality, a notion that many theorists find difficult to define. In the following section, we will review its basic properties and the many theoretical puzzles and challenges that it presents.

37.3 Wa-marking It is generally believed that there are at least two distinct uses of the particle wa. We will follow this tradition in this chapter and first examine the two well-known subtypes; (p. 761) thematic wa and contrastive wa. Additionally, wa in negative sentences, which seems neither thematic nor contrastive, will be discussed.

Page 9 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese

37.3.1 Thematic wa The thematic use of the particle is frequently associated with the notion of topicality. A number of authors, including Mikami (1963), Kuno (1972), Noda (1996) and Portner and Yabushita (1998), identify wa as a topic marker. It is hard to determine whether such an analysis gives the correct characterization because we still do not know exactly how to define topics. Despite the considerable effort from theoreticians of diverse disciplines, topicality remains a slippery notion, as documented in some of the review articles on this issue (e.g. Roberts 2011; Büring, this volume). At the descriptive level, the differences between wa and the canonical nominative marker ga are extremely instructive, and comparing the two particles has become a popular methodological tool in studies of wa. However, I will not follow this practice in this chapter since Heycock (2008) provides detailed and informative portrayals of the wa–ga contrasts and their theoretical implications. The strategy I adopt is to go over typical attributes that topics are claimed to have and examine how wa-phrases behave in connection to those attributes.9 During this process, a few additional issues will be raised that deserve closer examination. (11) lists the three major concepts that are most frequently associated with wa. (11)

The idea of ‘aboutness topic’ (Reinhart 1982) is probably the most popular characterization of topicality. Krifka and Musan (2012) define it as in (12) within the file card semantics of Heim (1982) or some variant of it. (12)

There are a few empirical facts that support this characterization. First, thematic waphrases are limited to nominal or quasi-nominal categories (e.g. NP, PP, and CP), and adverbial topics, discussed in Roberts (2011: 1908–1909), are not compatible with wa. Presumably adverbial expressions do not correspond to file cards, which makes them incompatible with (p. 762) wa. In addition, certain expressions, such as wh-based quantifiers (e.g. daremo ‘everyone, anyone’) and a disjunctive NP, cannot be marked with wa. In Tomioka (2007), the anti-topicality of those expressions is attributed to their inability to specify particular file cards.

Page 10 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese The notion of frame-setting in (11b) is what Chafe (1976) advocates for the definition of topicality. There has been some scepticism expressed (e.g. Jacobs 2001), however, that the frame-setting function is not always distinguishable from aboutness topic. While this criticism is applicable to wa-phrases as well, there are a couple of distinct types in Japanese that may be unambiguously characterizable as frame-setters: (13)

(14)

Unlike aboutness topic constituents, which are usually nominal or quasi-nominal categories, frame-setting wa-phrases can be verbal. In (13), for instance, wa is attached to the gerundive form of the verb kans- ‘to concern’. (14) is an instance of the ‘conditional topic’ construction. As noted by Tateishi (1990) and Kuroda (1992), the function of the waphrase in (14) mimics that of an if-clause in a relevance conditional (cf. Derose and Grandy 1999; Siegel 2006). While this is not a prototypical instance of a frame-setting topic that Chafe (1976) had in mind, its pragmatic function is clearly distinct from that of aboutness wa. The correspondence between topics and conditionals goes even further. (13), for instance, can be minimally altered by replacing the first wa-phrase with the ifclause, sore-ni kansite-i-eba ‘if (we) speak of it’, as shown in (15a), without affecting the overall meaning of the sentence. In addition, reduced conditionals of the form XP-nara/ dattara ‘XP-be-if/be.past-if’ are sometimes interchangeable with XP-wa (cf. Munakata 2006), as in (15b).

Page 11 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese (15)

The association between wa and givenness (and relatedly, definiteness) has been mentioned frequently (e.g. Kuno 1973; Martin 1975; Makino 1982 among many others). It is hardly surprising because the connection between topicality and givenness has also been frequently pointed out, particularly in the Prague School tradition (e.g. Daneš 1970). However, Krifka and Musan (2012) warn that the correlation is just a tendency, providing the following example in which the indefinite topic introduces a new discourse referent. (p. 763)

(16)

The concept of discourse-new topics has been suggested to account for wide-scope/ specific indefinites (e.g. Cresti 1995; Endriss 2009). As for wa–phrases, they can refer to new discourse referents as long as the referents are inferable. For instance, a proper name can bear wa even when the name is mentioned for the first time in the conversation, provided that the participants can all identify its referent. However, discourse novelty with indefinites is a different matter. In the Japanese translation of (16), the indefinite subject cannot appear with wa. Although it is tempting to connect this impossibility of indefinite wa-phrases to the popular view that givenness is an essential, and possibly the defining attribute of wa (e.g. Martin 1975; Makino 1982; Fiengo and McClure 2002), there are reasons to be cautious with this approach. First of all, not all given elements are wa-marked, and the aforementioned anti-topical expressions can be discourse-old but are still unable to have wa. The giveness-based theory of wa also offers very little once we consider the distribution of wa-phrases in the embedded contexts discussed in Section 37.4.

Page 12 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese In addition to some theoretical disagreements in analysing wa, we occasionally find descriptive discrepancies as well. Multiple wa-marking is one of such controversial phenomena. Kuno (1973: 48) states that ‘a given sentence can have only one thematic wa: if there is more than one occurrence of wa, only the first is thematic: all the rest … are contrastive’. Krifka and Musan (2012) also mention that a sentence typically has only one topic, which can be accounted for with the ‘file card’ analogy of aboutness topics in (12). On the other hand, Kuroda (1988) goes so far as claiming that wa is essentially iterative, which is more in line with Rizzi’s (1997) observation on topics in Italian (i.e. clitic-leftdislocated phrases). The truth is somewhere in the middle but probably closer to Kuroda’s characterization. First of all, multiple wa-phrases are quite natural when a temporal or a locative phrase is involved.10 (p. 764)

(17)

Multiple argument wa-phrases are more likely to be construed as Kuno described: the first wa is thematic, and the rest are contrastive (cf. Heycock 2008). For instance, (18a) fits this generalization. However, the multiple thematic interpretation becomes much more natural if what follows the multiple wa-phrases becomes ‘heavier’ or more descriptively loaded, as in (18b).11

Page 13 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese (18)

To sum up, multiple thematic wa-phrases are indeed possible and perhaps more prevalent than standardly assumed. A proper theoretical interpretation of this fact is still unclear. Krifka and Musan (2012) maintain that having one topic per sentence makes it easier to explain wa-marking in terms of the ‘file card’ metaphor. At the same time, they also acknowledge that multiple topics are possible when a relation between two files cards is commented on. Their example (Krifka and Musan 2012: 29) is As for Jack and Jill, they married last year, but the multiple wa cases shown above do not necessarily fit the profile of this kind of example. It remains to be seen how the notion of ‘a relation between two file cards’ can be extended to the Japanese multiple wa-phrases.

(p. 765)

37.3.2 Contrastive wa

The contrastive use of the particle wa is generally considered a distinct type from the thematic wa since they have a number of differences that are not easily resolved. (19)

Page 14 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese The following two examples are representative of the kinds of environments in which the particle is used contrastively. (20)

(21)

In (20), the two wa-phrases are contrasted with each other, and pragmatically, they seem to function as the ‘sorting keys’ in answering a multiple-wh question (cf. Kuno 1982). (21), on the other hand, does not furnish an overt expression that contrasts with the waphrase. It is often reported for a case like (21) that the wa-phrase contrasts with implicit alternative candidates and indicates that the other candidates would make the sentence false. However, this characterization makes the contrastive wa indistinguishable from the focalizing strategy with the canonical nominative ga, which gives, in the context described above, a complete and exhaustive answer. While the wa-version in (21) sometimes elicits the same exhaustivity effect, its meaning is much more loaded, as pointed out in Hara (2006) among others. It can indicate that the speaker does not know about the outcomes of the other students, and this is confirmed by the fact that the waversion of (21) can be followed by such statements as ‘but I don’t know about the others’. The possibility of a weaker interpretation than (p. 766) the ga-version presents a good basis for a lexicalist analysis of contrastive wa, in which the ‘at least’ meaning is conventionally encoded in the particle.12 The advocates of this type of analysis include Hara (2006), Schwarz and Shimoyama (2011), and Sawada (2013). The basic meaning of the wa-answer in (21) is, according to the lexicalist view, ‘(at least) Mari passed, and the others may or may not have passed’. It can be strengthened in some cases to the exhaustive meaning that Mari came and the others did not. However, the lexicalist view leaves a few matters unresolved. First, the correlation between the two uses of the particle is obscured. Despite the differences, there are some indications that they are, at some level, interconnected. Recall that reduced conditionals

Page 15 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese with X-nara/dattara can function as thematic topics. With focal accent, they exhibit a scalar meaning that is very similar to that of contrastive wa. (22)

There is also a cross-linguistic factor to consider. Some languages have designated positions for topics. Hungarian, for instance, is believed to have such a position in the left periphery of a clause (cf. É. Kiss, this volume). It has been claimed that Hungarian uses this same position for contrastive topics that yield semantics/pragmatic effects that closely mirror those of contrastive wa (cf. É. Kiss 2002; Gyuris 2009). The lexicalist analysis would have virtually nothing to say about these facts. It should also be noted that the scalar interpretation of contrastive wa is very similar to a ‘weak’ scalar implicature. According to what Geurts (2010) calls ‘the standard recipe’ for generating scalar implicatures: (23)

The statement in (23d), called ‘Competence Assumption’ in van Rooij and Schulz (2004), is an extra step needed to generate the ‘strong’ implicature (e.g. the ‘not all’ meaning for some or most, the ‘not both’ meaning for or).13 Without it, the implicature remains weak: The speaker is unsure whether the stronger alternative ψ is true or false. (p. 767) This uncertainty meaning is very much like the scalar meaning of contrastive wa, but the lexicalist analysis cannot offer a good explanation for this similarity since it takes the second meaning to be conventionally encoded in wa. A possible alternative to the lexicalist analysis is a ‘mixed’ approach with both conventional and conversational meaning involved. The ‘at least’ scalar meaning of wa itself is not a part of its conventional meaning. Rather, it is a result of some inferences derived from a more general convention of the particle. Imagine, for instance, that wa conventionally signals the avoidance of the Competence Assumption, as in (24).

Page 16 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese (24)

Without the Competence Assumption, the hearer is invited to entertain a variety of possible reasons for the speaker’s not making stronger statements. The speaker’s ignorance as to the stronger alternatives and her belief that such alternatives are false are both good possibilities. A non-epistemic reason is also possible. The speaker might have avoided talking about the other alternatives for reasons of politeness. The other two did not pass, and the speaker may not have wished to advertise their failure. Since the avoidance of the Competence Assumption does not mean the application of the Incompetence Assumption, the stronger meaning (the exhaustive interpretation) is not automatically ruled out either. Thus, (24) can match the interpretive possibilities of contrastive wa. This strategy is adopted in Tomioka (2010), and it is further argued that the effect of (24) is derived via the wide-scope property of a wa-marked phrase. Fox (2007) endorses a ‘grammatical theory’ of the Competence Assumption in which the strong implicature is derived by the syntactic presence of the Exhaustive Operator, Exh, that operates at the level of proposition. The role of wa is to ensure wide scope, a function shared by the thematic use of the same particle (cf. Jacobs 1984; Krifka 2001; Endriss 2009), which allows a contrastive wa-phrase to escape the exhaustification process.14

37.3.3 Negative wa Relatively lesser known is the peculiar fact concerning the appearance of wa-phrases in negative sentences and sentences with inherently negative predicates, such as kirai-da ‘dislike’ or nigate-da ‘be poor at’. McGloin (1987) reports that wa is used much more liberally in negative sentences than in affirmative sentences, and its meaning is not easy to pin down, as it seems neither thematic nor contrastive. The following pair was presented in Kuno (1973: 31) to highlight the affirmative–negative contrast. (p. 768)

(25)

Page 17 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese The affirmative version is quite awkward unless the second wa-phrase is contrastive with proper focus prosody (as is often the case for multiple wa-phrases). The negative counterpart, on the other hand, sounds perfectly natural without focus prosody. As noted by Oshima (2011), the lack of focus prosody presents a big challenge to the most common analysis of ‘negative’ wa: Kuno (1973), McGloin (1987), and Noda (1996) endorse the view that it is an instance of contrastive wa licensed by negation, which these authors assume intrinsically induces a sense of contrast. Oshima (2011), on the other hand, proposes a givenness-based analysis of the negative wa, as part of his hybrid analysis of wa-marking. It is indeed true that a negative wa-phrase tends to be discourse-old. Consider (26). (26)

The clause under because represents new information in this context, and the wa-marking in it makes it a less natural answer than its ga counterpart. However, the effect is rather weak, which is surprising since the wa/ga contrast in a discourse-new/old context is usually much stronger. The true nature of negative wa is therefore still a mystery. It is neither contrastive nor thematic and is only mildly givenness-oriented.

37.3.4 Summary We have reviewed three types of wa-phrases. They do not appear to share many properties. Indeed, differences among them are much more noticeable than similarities. The most pressing question is, therefore, whether it is still possible to find a common thread that binds the three uses together. Some are more optimistic on the prospect than others, but it is certainly a worthy research project.

Page 18 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese

(p. 769)

37.4 Recursivity of information structure

Recursivity is one of the most fundamental properties of natural language, and syntactic embedding and the integration into compositional semantic processes are the key manifestations of it. Where does information structure stand in this respect? Among the key concepts of information structure, the embeddability of focus has not been raised as a debatable issue. A matrix-level focus (often called a sentence focus) can appear in an embedded clause, and an embedded focus can be locally ‘resolved’ by the presence of a focus-sensitive operator at the embedded level. The following is one such example. The focus on the wh-phrase is bound off by the concessive morpheme mo at the embedded level, and there is a higher focus (on karee, ‘curry’) which functions as a matrix level focus. (27)

Givenness expressed by prosodic reduction or by ellipsis/null anaphora applies both at the matrix and the embedded level. Perhaps, embeddability is not even a relevant question for givenness, as this notion seems to be superimposed at a higher level onto syntactic embedding and compositional semantics.15 Therefore, the most intriguing question concerning the recursivity of information structure is topic. In the present context, it translates into the question of whether a wa-phrase can be embedded. It has often been reported that thematic wa is a root phenomenon and therefore cannot be embedded (e.g. Roberts 2011) or that it can be embedded only in the complements of some attitude predicates (e.g. Heycock 2008; Vermeulen 2012; Maki et al. 1999). Such clauses are identified as ‘statement-making’ (Kuroda 2005) or ‘complements of nonfactive bridge verbs’ (Maki (p. 770) et al. 1999). Miyagawa (2012) goes beyond the criterion of ‘statement-making’, but his generalization would not allow wa-marking in clauses under the nominal koto. However, these characterizations are not entirely accurate. For instance, Wasureru ‘forget’ is a factive non-statement making verb that selects a koto-clause but can nonetheless host wa fairly comfortably, as in (28a), and (28b) shows that a question in the subject position can also have wa with a predicate that cannot possibly select a speech act or a statement.

Page 19 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese (28)

Examples of these sorts are abundant, and although the nominative ga may be preferred in many cases, the wa option is definitely available.16 Although not acknowledged often, there are adjunct clauses that can host wa-phrases. One such linguistic environment is because-clauses, as discussed in Ueyama (2007), but Tomioka (2015) notes that not all because-clauses license wa, as the following examples demonstrate.17 (29)

Page 20 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese The difference between the two cases is rooted in the semantics of causation. In (29a), the because-clause specifies the basis for the speaker’s decision to send plates to Mari. This exemplifies Davidson’s (1963) notion of causal explanation, and wa-phrases can appear in this type of because-clause. (29b), on the other hand, corresponds to Davidson’s singular causal statement, where the first event caused the second event without provoking any sense of conscious decision-making processes. (p. 771)

Other typical adjunct clauses, such as temporal adjuncts and conditionals, are generally believed to be incapable of licensing wa. (30)

Surprisingly, however, there is a certain type of conditional that is more comfortable with embedded wa (cf. Tomioka 2015). The form is either no-nara or n(o)-dattara, which contains the no-da ‘NML+COPULA’ component in its structure. This type of conditional can be used in a very specialized situation. Someone asserts p. The addressee is, however, still not 100 per cent certain about p and makes a statement of the form if (indeed) p, then q.18

Page 21 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese (p. 772)

(31)

This fact conforms with the observation by Haegeman (2002, 2010) that this type of conditional, which Haegeman calls ‘peripheral conditional’, can host such root phenomena as argument fronting, long-distance adverbial fronting, and the presence of high modal markers. So far, three embedding contexts have been identified as wa-friendly: (i) complement clauses of attitude verbs, (ii) because-clauses of the causal explanation type, and (iii) noda-based conditionals. Tomioka (2015) suggests that the crucial characteristic that sets (i)–(iii) apart from other embedded clauses is the local presence of a point of view (cf. Tenny 2006). It is easy to imagine that a complement CP of an attitude verb furnishes a point of view that is identified with the attitude holder ( = the matrix subject). When a because-clause is used as a causal explanation, the agent in the matrix clause event believes the complement of because. In (29a), for instance, the sender of the wedding gift ( = the speaker) believes that Mari likes cooking. The presence of such a propositional attitude is not entailed in the other type of because. A no-da conditional is often used when the proposition within the if-clause has just been asserted by someone. In (31), the conditional can be paraphrased as ‘if Kei got divorced, as you said’, where the embedded proposition is presented with the addressee’s point of view. Needless to say, a root clause has a point of view, which is most frequently identified as the speaker of the sentence. Those who claim that wa is a root phenomenon consider wa-embedding clauses to be ‘root-like’. The presence of a point of view is what makes an embedded clause ‘root-like’ enough for wa, although the new generalization of embedded wa makes it unclear whether it makes sense to maintain the ‘wa as a root phenomenon’ label.19

Page 22 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese

(p. 773)

37.5 Conclusion

The readers of this chapter should not be overly disappointed to learn that very few issues connected to Japanese information structure are settled. This state of affairs certainly does not reflect the lack of interest or effort. On the contrary, it proves the popularity of this intriguing research area and the diversity of the research endeavours engaged in the field. The Japanese language has also been a major player in the theorization of information structure in general, particularly in the area of topicality. There are clear indications that studies of information structure will continue to evolve and expand. Japanese is hardly an understudied language for information structurerelated studies, but it still offers a rich empirical ground for exploration. It is my hope that the inclusion of this chapter in the handbook will further stimulate research interest in this exciting area.

Notes: (1) I am grateful to the editors of this volume, Caroline Féry and Shin Ishihara, for comments and encouragement. I also benefited greatly from many insightful comments and suggestions by two anonymous reviewers. Special thanks to Anne Peng, who helped me prepare the manuscript. (2) The recent review articles on information structure in Japanese include Heycock (2008) and Vermeulen (2012). (3) What counts as a contrastive focus is not a settled issue. In Katz and Selkirk (2011), focused phrases associated with only are regarded as contrastive. In Tokyo Japanese, the clearest cases of contrastive focus prosody are found in corrective expressions and whphrases. The inclusion of wh-phrases in this category may be subject to dialectal variations. Smith (to appear), for instance, claims that the focus prosody and the whprosody are distinct in Fukuoka Japanese. (4) In (7), there is no clear post-focal reduction on the wa-marked subject, which would be the expected pattern if the scrambled object were contrastively focused. If there is any discourse effect of the scrambling in the example, the fronted NP is more topical than focal. (5) There are a few focus-related issues that are left out in this chapter. One is association with focus with focus-sensitive particles such as –dake ‘only’, –sae ‘even’ and mo ‘also’. Focus effects in wh-questions are also not included. The reader is referred to Vermeulen (2012: section 37.1) for an overview of these phenomena.

Page 23 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese (6) The expansion of reduction domains is an important part of the analysis of intervention effects in Tomioka (2007). It has been observed (e.g. Hoji 1985; Beck 1996; Beck and Kim 1997) that a wh-phrase cannot take scope over a certain quantifier or a focused expression (= intervener) that c-commands it, and that the ill effects disappear when the wh-phrase is scrambled past the intervener. Tomioka (2007) argues that the ill effect of the intervention structure is due to the focused status of the intervener. The leftward movement of the focused wh-phrase places the intervener in the post-focal reduction domain, and the intervener is considered given. (7) The sentence sounds really odd and is not easily interpretable. If it has any meaning at all, it means that the same set of students who passed last year also passed this year as well. (8) More convincing evidence for ellipsis would be extraction out of elided structures. Such evidence is difficult to come by for the case of silent arguments, however, as extraction out of a nominal structure is believed to be unacceptable. (9) One of the popular characterizations of topicality that has not been embraced in the Japanese literature is the idea of topic as a question-under-discussion (cf. Roberts 1998, von Fintel 1994). Portner and Yabushita (1998) argue that the ‘topic-as-an-entity’ analysis suits wa-marking better for wa-phrases. (10) Multiple frame-setting wa-phrases are possible and can appear in addition to an aboutness wa, as in (i). ((i))

(11) Vermeulen (2013) presents some instances of multiple wa structures in which a noninitial wa is not a topic but merely a given expression (‘discourse anaphoric’ in her term). In her examples, all of the non-initial, discourse-old wa-phrases have already been marked with the same particle in the previous contexts. Therefore, they can be described as ‘second occurrence topics’, a term inspired by ‘second occurrence focus’ (cf. Baumann, this volume). (12) The scalar meaning of the contrastive topic strategy is observed in many other languages. Büring (this volume) gives an informative review of the phenomenon, focusing primarily on data from English. (13) (23d) has other names such as ‘Experthood Assumption’ (Sauerland 2004) and ‘Authority Assumption’ (Zimmermann 2000). (14) See Wagner (2012) for a similar proposal. Page 24 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese (15) Interestingly, however, the right dislocation construction, discussed in Section 37.3, is strictly a root phenomenon. In many cases, right dislocated structures accompany sentence-final discourse particles, such as yo and ne, and those particles are not easily embedded in non-quotative contexts. One marker that can be used in both root and embedded clauses is the proposative marker (y)oo. Even with this marker, however, the embedding of a right dislocated structure is not possible.

((i))

((ii))

(16) Other non-statement-making predicates that license wa include … to gokai-suru ‘misinterpret that … ’, … ka wakaru ‘know whether … ’, … koto-ga hanmei-suru ‘(the fact) that … becomes known’. In fact, there are not too many attitude predicates that clearly ban wa. Kuroda (1992) lists … koto-o zannen-ni omou ‘think of it regrettable that’ as one of the predicates incapable of licensing wa. The crucial point may not be the meaning but the form, … koto-o … ni omou. For instance, … koto-o ikan/husigi/kooun-ni omou ‘think of it regrettable/strange/fortunate that … ’ does not license wa within its complement. (17) Larson and Sawada (2012) also discuss root phenomena in because-clauses, but the distinction based on the two Davidsonian notions is not considered in their analysis. Similarly, the ‘central vs peripheral’ distinction in adverbial clauses proposed in Haegeman (2012) does not correspond to the Davidsonian categorization. (18) This is what Iatridou (1991) calls factual conditional. See Bhatt and Pancheva (2006) for more discussion on factual conditionals and related phenomena. (19) One exception to the new generalization is relative clauses. In many cases, they do not license wa even when the point of view condition is met. To account for this exception, Tomioka (2015) proposes a ban on semantically meaningful abstraction over wa, appealing to Kuroda’s (1992) notion of predication. The embeddability of contrastive wa is not too different from its thematic counterpart. It can appear in complements of attitude predicates, and Hara (2006) notices the same contrast in the two types of because-clauses for contrastive wa. Other embedding environments are less certain. Kuroda (2005), for instance, claims that a conditional can host contrastive wa, but this judgement is not shared by everyone (e.g. Hara 2006). Hara (2006) also claims that relative clauses fail to host contrastive wa, but there have been some grammatical examples reported. Tomioka (2015) maintains that a relative clause like [NP [CP e EEgo-WA hanas-eru] hito] ‘a person who can speak (at least) ENGLISH’ is quite acceptable. Oshima (2011) presents a relative clause example which embeds a conjoined sentence in which each conjunct has the

Page 25 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Japanese subject marked with contrastive wa. Negative wa is more widespread. It can appear in a conditional or a relative clause (e.g. [NP [CP e eego-wa hanas-e-nai] hito] ‘a person who cannot speak English’). See Tomioka (2015: appendix) for more discussion.

Satoshi Tomioka

Satoshi Tomioka is Associate Professor of Linguistics and Cognitive Science at the University of Delaware. His research interest is in semantics, syntax, pragmatics and their interfaces. He has published articles in journals such as Lingua and Journal of East Asian Linguistics, and has contributed to edited volumes published by OUP, CUP, and de Gruyter.

Page 26 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use. Subscriber: University College London; date: 08 May 2018

Information Structure in Asia: Yongning Na (Sino-Tibetan) and Vietnamese (Austroasiatic)

Oxford Handbooks Online Information Structure in Asia: Yongning Na (SinoTibetan) and Vietnamese (Austroasiatic)   Alexis Michaud and Marc Brunelle The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara Print Publication Date: Jul 2016 Subject: Linguistics, Pragmatics Online Publication Date: Sep 2014 DOI: 10.1093/oxfordhb/9780199642670.013.28

Abstract and Keywords The languages of Asia are highly diverse. Rather than attempting a review about information structure (IS) in this huge linguistic area, this chapter provides observations about two languages that differ sharply in terms of how they convey IS. Yongning Na (Sino-Tibetan) is an example of a language with abundant morphemes expressing IS, which stand at different points along the grammaticalization path: some are exclusively used for the marking of IS, others (such as demonstratives) are equally common as IS markers and in another function, and others still are used secondarily to indicate IS, in particular particles indicating the relationship that a noun phrase bears to a verb. Vietnamese (Austroasiatic) makes little use of such morphemes, and relies greatly on word order and on a range of passive-like structures. Along with key morphosyntactic facts, this chapter addresses the issue of how intonation contributes to foregrounding and backgrounding strategies in these two tonal languages. Keywords: information structure, Yongning Na, Vietnamese, referential density, discourse particles, intonation

38.1 Introduction THE present chapter is intended to be complementary with the overview of information structure (IS) in Chinese by Chen et al. (this volume).1 Neither of the two languages presented in this chapter, Yongning Na and Hanoi Vietnamese, belongs to the small set of ‘canonically exotic languages’ (Matisoff 1973: xlv), such as Japanese, which are often discussed in the literature, so their inclusion in a textbook can be seen as a modest contribution towards broadening the empirical basis of IS research. Moreover, these two

Page 1 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Information Structure in Asia: Yongning Na (Sino-Tibetan) and Vietnamese (Austroasiatic) languages differ sharply in the way they convey IS, providing an illustration of the great diversity found among languages of Asia. ‘Information structure is a vast topic of research that has been pursued within different theoretical frameworks’ (Krifka 2008: 244), and with different objectives in view. While we have adhered to the framework set by Krifka (2008), we are keenly aware that crosslinguistic categories can only capture a small part of IS (see Matić and Wedgwood 2013, and the general discussion by Haspelmath 2007); we have accordingly stated our observations in terms that we hope can easily be interpreted by (p. 775) readers working within different theoretical frameworks. Our view of IS is that it is not to be viewed as a separate ‘module’, but as part and parcel of language as communicative activity. IS ‘cannot be described without referring to the strategies used to narrate events or make arguments’ (Krifka 2008: 272; see also Karcevskij 1931: 191; Kohler 2009; Horn, this volume). The chapter consists of two sections looking in detail at the two languages Yongning Na and Hanoi Vietnamese. Each of the two sections deals with topics, focus, and givenness.

38.2 Yongning Na Yongning Na is a language of the Naish subgroup of Sino-Tibetan (Jacques and Michaud 2011). It has typologically salient characteristics, many of which are shared with other Sino-Tibetan languages.

38.2.1 Topics 38.2.1.1 Word Order The following generalization about Qiang proposed by Lapolla and Huang (2003) also applies to Yongning Na: The structure of the clause is to some extent affected by pragmatic factors, but this only applies to the order of noun phrases in the clause. The utterance-initial position is the unmarked topic position (though secondary topics can follow the primary topic), while the position immediately before the verb is the unmarked focus position, and so the focused element will generally appear there. The verb always appears in final position; there is no possibility for the actor of a clause to appear in postverbal position, even if it is focal. The only exception to this is the occasional afterthought clarification of a noun phrase that was omitted or expressed as a pronoun in the clause. (LaPolla and Huang 2003: 221)

Page 2 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Information Structure in Asia: Yongning Na (Sino-Tibetan) and Vietnamese (Austroasiatic) IS in Na has accordingly been described as ‘topic–comment’, extending an observation made by Chao Yuen-ren about Chinese: ‘the grammatical meaning of subject and predicate in a Chinese sentence is topic and comment, rather than actor and action’ (Chao Yuen-ren 1968: 69; Shi 2000; LaPolla 2009). The primary information structure in Na is topic/comment rather than subject/ predicate…. a topic can be a nominal argument, about which the rest of the sentence will comment upon, but the topic can also be an adverbial, an independent clause, or a dependent clause. (Lidz 2010: 296) Word order thus plays an important role in the structuring of IS in Na. Example (1) provides an illustration.2 Sentence (1) is translated with a conditional, illustrating the proximity between topics and conditionals (Haiman 1978). (p. 776)

(1)

3

Page 3 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Information Structure in Asia: Yongning Na (Sino-Tibetan) and Vietnamese (Austroasiatic) 38.2.1.2 Discourse Particles Conveying IS Yongning Na is similar to a number of other Sino-Tibetan languages in possessing a wealth of discourse particles that contribute to the structuring of information. Pinpointing the exact range of uses of particles can be a thorny task, given their various connotations and rich compositional potential. Grammars that address these issues in detail are available for Lahu (Matisoff 1973), Lalo (Björverud 1998), and Qiang (LaPolla and Huang 2003). Particles stand at different points along the path towards use as ‘pure’ information-structure markers. In Na, a few particles are exclusively used for the marking of IS, such as the topic marker /-dʑo˩/; others are about as commonly used as (bleached) information-structure markers and in another function, such as demonstratives; others still are used secondarily to indicate IS, in particular particles indicating the relationship that a noun phrase bears to a verb (thematic relations). In (1), the topic marker /-dʑo˩/ does not appear; a plausible reason is that in this context (after a verb preceded by the accomplished morpheme /le˧-/) /-dʑo˩/ would be interpreted as the progressive, which is homophonous with the topic marker. In the second sentence in (2), on the other hand, the topic marker /-dʑo˩/ appears as expected: (p. 777)

(2)

In the first sentence in (2), ‘robbers’ is strongly brought to the fore through the combination of /-ʈʂʰɯ˥/ and /-ɳɯ/. These two particles help to indicate that the preceding noun has the status of topic; on the other hand, /-ɳɯ/ and /-ʈʂʰɯ˥/ are often associated with the topic marker /-dʑo˩/, highlighting the fact that they are not yet fully grammaticalized as topic markers. The marker /-ʈʂʰɯ˥/ was grammaticalized from a proximal demonstrative, also used as a 3rd-person pronoun, with which it remains formally identical. The postposition /-ɳɯ/ is an ablative marker, which has developed into an agent marker (Lidz 2011:53; for observations on similar developments in neighbouring languages: LaPolla 1995; on the use of case markers to convey IS, see also the alternation between ergative and absolutive in Newari: Genetti 1988). A close examination of cases of

Page 4 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Information Structure in Asia: Yongning Na (Sino-Tibetan) and Vietnamese (Austroasiatic) use of /-ɳɯ/ as an agent marker reveal that it is fairly rare for there to be true ambiguity as to semantic roles, however, see Lidz (2011: 55). Cases like (2) are common: the agent marker functions prominently as an IS indication. From a lexicographic point of view, it is an issue whether to set up a distinct subentry. It appears appropriate to set up three subentries for the morpheme /-ʈʂʰɯ˥/, since its uses as a proximal demonstrative, as a third person pronoun and as a topic marker are by now relatively well-differentiated. The three glosses used are DEM.PROX, 3SG, and TOP(/ DEM.PROX): the latter combines an indication of the particle’s IS function (topic marker) with an indication of its origin as a proximal demonstrative. On the other hand, the use of /-ɳɯ/ as an IS marker does not currently appear highly grammaticalized; omission of this particle from (2) would not modify the IS in any spectacular way, even though it would lend less salience to the topic. It therefore does not appear warranted to set up a distinct subentry; the three uses of /-ɳɯ/—as ablative, agent, and topic marker—are therefore indicated side by side in the gloss, in the hypothesized order of diachronic development: ABL/A/TOP. In addition to /-dʑo˩/, /-ʈʂʰɯ˥/, and /-ɳɯ/, there exist three less frequently occurring topic markers, whose origin remains unclear. One is an archaic word, /-læ˧/, appearing only once in the recorded data. The two others are /-no↿/ and /-se/, which can only (p. 778) apply to a noun phrase, whereas the other topic markers can bear on a noun phrase or an entire clause. Examination of the contexts of appearance of /-no↿/ and /-se/ brings out the presence of strong emphasis on the topicalized element in contrast to others: a possible translation is ‘as for X/as concerns X’. These two morphemes can be described as contrastive topic markers (Krifka 2008: 267).

38.2.2 Focus Focus in Yongning Na is conveyed by local phenomena of intonational emphasis. The discussion progresses from emphatic stress to less extreme cases of intonational emphasis.

38.2.2.1 Emphatic Stress Emphatic stress is found in Yongning Na. It appears to have essentially the same function as in English and French—calling the attention of the listener to a particular word—hence the choice of this label (proposed by Coustenoble and Armstrong 1937). In the languages where emphatic stress has been experimentally investigated, it has been found that prototypical realizations involve supplementary activity of the expiratory muscles, resulting in a sudden increase in subglottal pressure during the articulation of a consonant (Benguerel 1973; Carton et al. 1976; Ohala 1978; Fant et al. 1996), hence the term ‘force-accent’ used by Kohler (2003). Acoustically, this exerts an influence on spectral slope (Rossi 1971: 143; Glave and Rietveld 1975; Gobl 1988; Fant and Kruckenberg 1995; Sluijter et al. 1997; Heldner 2003).

Page 5 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Information Structure in Asia: Yongning Na (Sino-Tibetan) and Vietnamese (Austroasiatic) Emphatic stress has been somewhat neglected in intonation studies, as researchers focused their attention mostly on the acoustic parameter of fundamental frequency. But it is an important component of intonation, in Na as well as in a number of other languages. Its linguistic functions range from the attitudinal and emotional to the pragmatic. It is most often encountered in toned-down versions, physiological effort at a subglottal level being mimicked through such strategies as f0 excursions and consonant lengthening. Like other linguistic phenomena, emphatic stress comprises important language-specific and speech-style-specific dimensions: its frequency of use varies greatly from language to language, from speaker to speaker, and from style to style; its stylistic effect is inversely proportional to its frequency of use. An ‘up’ arrow ↑ to mark intonational emphasis is adopted for the Na data, following Mazaudon (2004). The arrow is placed to the left of the emphasized syllable, as in (3).

(3)

In many contexts, emphatic stress appears on a constituent that can be predicted to receive normal focus prosody. For example, in (3), we would expect the focus to be in the immediate preverbal position—the usual unmarked focus position for verb-final languages. Emphatic stress can be considered an extreme form of focus prosody; it is an extreme along a continuum: there is no hard-and-fast boundary between emphatic stress and milder realizations of focus prosody. When annotating recordings, it is sometimes an issue whether to add an ‘up’ arrow ↑ or not. (p. 779)

The phonetic realization of emphatic stress includes effects on the articulation of vowels and consonants and also partakes in the realization of emphatic stress: for instance, the second syllable of the verb /dʑɤ˩↑bv˥/ ‘to play’ is realized in Caravans.231 with a much stronger trilling of the /b/ than is found in non-emphatic contexts. (About ‘articulatory prosodies’, see Niebuhr 2009, 2013; Kohler and Niebuhr 2011.) Emphatic stress is phonetically located on one syllable only, but from the point of view of interpretation, there is ambiguity of focus-marking. This phenomenon is extensively studied in the literature on focus projection (e.g. Selkirk 1995); Lambrecht uses the terms ‘broad focus’ vs ‘narrow focus’ (Lambrecht 1994). In its most vehement manifestations, emphatic stress intrudes into a sentence’s intonation, wreaking havoc on tonal contrasts. Example (4) is a case in point. Page 6 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Information Structure in Asia: Yongning Na (Sino-Tibetan) and Vietnamese (Austroasiatic) (4)

The usual pronunciation is /pʰv˩-tɕæ˩ɻæ˥/ ‘very white’. In (4), the second syllable is realized phonetically with extremely high fundamental frequency on the syllable /tɕæ˩/, which is considerably lengthened. From a phonetic point of view, its phonetic L tone is conspicuously disregarded. One way of looking at this modification would be to describe it as due to an intonational overlay: functionally, one could consider transcribing it as / pʰv˩-↑tɕæ˩ɻæ˥-gv˩/, where the arrow ↑ indicates emphatic stress on the second syllable, and the underlying tonal string is unchanged. The lowering of the syllable string /ɻæ-gv…/ would then be ascribed to post-focal pitch range compression. Another interpretation is that this forcible intonational modification interacts with the phonological tone string of the tone group (on this notion: see Section 38.2.3.2). If the modification of the second syllable in /pʰv˩-tɕæ˩ɻæ˥-gv˩/ only took place on an intonational level, and the underlying tonal string remained unchanged, the third syllable would retain its phonological H tone. Discussions with the consultant who produced this story suggest that the third and fourth syllables in (4) are lowered to L, however, yielding /… ɻæ˩-gv˩/, as expected if the second syllable carried H tone: the lowering of all tones to L after an H tone is an exceptionless tone rule of Yongning Na. At present, no truly decisive evidence on this issue can be offered; if the speakers could write their language, it would be interesting to see which solution they would prefer, but so far there (p. 780) has not yet been any opportunity to work with a consultant who had metalinguistic awareness. This phenomenon is provisionally analysed as involving a categorical tone change, from a L.L.H sequence, /pʰv˩-tɕæ˩ɻæ˥/, to a L.H.L sequence, / pʰv˩-tɕæ˥ɻæ˩/.

38.2.2.2 Focalization Through a Dipping Contour In Yongning Na, one means of emphasizing a word within a sentence consists in a rapid dipping contour, as in example (5). Focalization through a dipping contour is compatible with emphatic stress, but formally distinct from it.

Page 7 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Information Structure in Asia: Yongning Na (Sino-Tibetan) and Vietnamese (Austroasiatic) (5)

The phenomenon marked as ‘F’, for ‘intonational Focalization’, is an intonational device to set part of the utterance into relief. It is realized through a dip in fundamental frequency, accompanied by lengthening and formant movement towards a central vowel. The realization of focalization is sufficiently specific—involving a movement in fundamental frequency, and a change in the vowel: a difference in the time course of formant frequencies—to avoid interference with lexical tone. Emphatic stress is likewise identifiable as such, from cues other than fundamental frequency. This greatly limits the possibility of a misperception of lexical tone caused by these intonational phenomena. (An example is found in Healing.55.)

38.2.3 Givenness 38.2.3.1 Referential Density One of the means for backgrounding information in Yongning Na consists simply in not making any mention of it. This holds true of any language, but is especially salient in Yongning Na. Referential density—the ratio of overt to possible argument NPs (Bickel 2003)—is low in many languages of the Tibeto-Burman area. The active referent, known from the earlier context, serves as given, without being mentioned (Mazaudon 2003 on Tamang). Yongning Na does not require pronouns to stand in place of the pragmatically inferable noun phrases, as illustrated by (1); this can be referred to as a case of ‘ProDrop’ (for Pronoun Dropping; Chomsky 1981), but as pointed out by Launey (1994: 45, 93), the perspective could just as well be reversed, considering languages that require overt NP arguments as a special case (widely represented in Western Europe: Haspelmath 2001). There is a limit to economy in referential density, however. For instance, in example (2), one could be tempted to push referential economy further and—in (p. 781) addition to ellipsis of the subject argument—to avoid the repetition of the object /qæ˧do˧/ ‘lumber’. However, it is not acceptable to remove the second /qæ˧do˧/ ‘lumber’: when this manipulation was attempted, the result, shown in (2′), was rejected by the consultant.

Page 8 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Information Structure in Asia: Yongning Na (Sino-Tibetan) and Vietnamese (Austroasiatic) (2’)

This probably has to do with several factors. First, there exist numerous homophonous roots, following monosyllabicization and phonological erosion (Michaud 2012); lexical identification can be difficult for a monosyllabic verb without an object. A second factor that may be at play is rhythm and the overall symmetry of the utterance (what see, what rob; lumber see, lumber burn). A third factor, certainly not the least important, concerns IS: ‘(they) burn lumber’ is the focal information, so it would be inappropriate to give it a less complete representation than other components of the utterance. Cross-linguistically, there is a closer syntactic integration of object and verb, as compared with subject and verb; this goes a long way towards explaining why all languages that allow Object ProDrop also allow Subject Pro-Drop, whereas the reverse is not true.

38.2.3.2 How Prosodic Phrasing Reflects Information Structure: The Division of the Utterance into Tone Groups A salient characteristic of Yongning Na is the interaction of IS and morphotonology. Yongning Na possesses abundant tonal morphology, comparable in its extent to the segmental morphology found in the (very distantly related) Kiranti languages (Jacques et al. 2012; Jacques 2012). For instance, the determinative compound /ʐwæ˧zo˧-gv˧dv˥/ ‘colt’s back’ has a final H tone, whereas the coordinative compound ‘father and mother’, which has the same input tones, is /ə˧dɑ˧-ə˧mi#˥/, with a floating H tone (a tone which can only associate to a following syllable; it is represented as #˥). The floating H tone on ‘father and mother’ is not by itself a marker of the morphological status of the compound as being coordinative rather than determinative: not all determinative compounds share the same tone pattern, any more than coordinative compounds do. The output tone depends on the input tones, but different rules apply in determinative compounds and in coordinative compounds. The tone rules that apply when combining an object with a verb are likewise different from those that apply when combining a subject with a verb. A systematic study of numeral-plus-classifier determiners brings out no less than nine categories with different tone patterns (Michaud 2013). The tone group is the domain within which these abundant morphophonological tone rules apply: each utterance consists of one or more tone groups; tonal computation takes place separately in each tone group. The phenomena that refer to this domain are exclusively tonal: they (p. 782) consist of tone rules, in particular the following seven rules: (i) L tone spreads progressively (‘left-to-right’) onto syllables that are unspecified for tone; (ii) syllables that remain unspecified for tone after the application of Rule 1 receive M tone; (iii) H and M are neutralized to M in tone-group-initial position; (iv) a syllable following an H-tone Page 9 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Information Structure in Asia: Yongning Na (Sino-Tibetan) and Vietnamese (Austroasiatic) syllable receives L tone; (v) all syllables following an HL or ML sequence receive L tone; (vi) LH and LM contours on a tone-group-final syllable are neutralized to LH; and (vii) if a tone group only contains L tones, a post-lexical H tone is added to its last syllable. As illustrated by examples (6)–(9), different divisions into tone groups have different implications in terms of prominence of the various components. Tone groups may be considered as constituting one phonological phrase each; changes in phonological phrasing serve to highlight a certain element, lending it focal prominence. (The term ‘tone group’ is nonetheless used here in preference to ‘prosodic phrase’; the defining characteristic of this phonological unit is that it serves as the domain of tonal processes.) A tone group boundary is always found after topicalized phrases; but apart from this hard-and-fast rule, speakers generally have several options for dividing the utterance into tone groups. They may choose to integrate large chunks of speech into a single tone group, resulting in a stronger integration; or they may divide the utterance into a number of tone groups, with the stylistic effect of emphasizing these individual components one after the other. The latter option is illustrated by (6): (6)

The noun phrase /dzɯ˧-di↿/ ‘food’ and the negated existential verb /mə˧-dʑo˧/ ‘there isn’t’ can be separated into two tone groups, as in (6). This has the effect of emphasizing the two noun phrases, /dzɯ˧-di↿/ ‘food; thing to eat’ and /ʈʰɯ˩-di˩/ ‘drink; beverage’. The following sentence in the story repeats the statement ‘There was no food’, continuing the same strategy of bringing out the noun phrase ‘food’, this time with the topic marker /dʑo˩/: /dzɯ˧-di↿ ǀ -dʑo˩, ǀ mə˧-dʑo˧-ɲi˥-tsɯ˩ ǀ -mv˩/ ‘As for food, it’s said that there was none!’ (Seeds2.68). Then the narrator recapitulates, and moves on: (7)

Page 10 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Information Structure in Asia: Yongning Na (Sino-Tibetan) and Vietnamese (Austroasiatic) At this juncture, ‘there was no food’ is integrated into a single tone group, and followed by the topic marker /-dʑo˩/. This provides an exemplary illustration of the integration of larger chunks of information into a single tone group as this information changes its status from new to old and backgrounded. (p. 783)

Conversely, insertion of a tone-group boundary highlights the word or phrase that precedes. Even function words can be emphasized in this way. Consider (8): (8)

A simpler formulation would be /ɬo˧pv˥-ti˩-kv˩-tsɯ˩-mv˩/, integrating the passage into a single tone group. The formulation in (8) emphasizes the reported-speech particle. This particle is used over and over again by the consultant when telling narratives: it is used whenever the speaker has only indirect knowledge of the situation at issue. But in the context of (8), the emphasis laid on this particle is one of the manifestations of the speaker’s efforts to adhere to truthfulness and precision: the narrator never witnessed the ritual that she describes. The particle could be paraphrased, in this context, as: ‘at least, that’s what they say’, or ‘but I can’t personally vouch for it: this is just hearsay, you know’. As a final example, consider (9): (9)

In (9), the noun phrase ‘dog meat’ is set into relief by constituting a tone group on its own. Despite the absence of a morphemic indication that it is topicalized, it clearly has the status of topic. In this context, tonal integration with a following verb would not be stylistically appropriate. This is parallel to prosodic grouping in English: the phrase ‘the Massachusetts Institute of Technology’ contains three lexical stresses, but someone who is familiar with the place is likely to pronounce it (when not using the acronym ‘MIT’) as one single prosodic phrase, integrating the three stressed words and two grammatical words into a single

Page 11 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Information Structure in Asia: Yongning Na (Sino-Tibetan) and Vietnamese (Austroasiatic) fundamental frequency contour. (On a similar phenomena in French, see, among other descriptions, Vaissière 1975; Rossi 1999; Vaissière and Michaud 2006; Martin 2009.)

(p. 784)

38.3 Vietnamese

38.3.1 Topics Like Yongning Na, Vietnamese tends to favour a topic+comment order (Thompson 1965: ch. 10; note that what the author calls ‘focal elements’ corresponds to Krifka’s ‘topic’; see also: Cao 1992; Clark 1992, 1996). The relevance of word order for topic-marking can be illustrated by a variety of passive-like structures. As these structures are described in more detail elsewhere (Simpson and Hồ 2013), a single example is given in (10). While in (10a) police is the topic, in (10b) it is Hoa that is topicalized. (10)

Page 12 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Information Structure in Asia: Yongning Na (Sino-Tibetan) and Vietnamese (Austroasiatic) Another frequent topicalization strategy is pause insertion, as in (11). This strategy is not necessarily associated with a special focus, although such an interpretation is not ruled out if focus is marked intonationally (see Section 38.3.2). A sentence such as (11) could not be expressed in Na without using information-structure morphemes:

(11)

A strategy very similar to pause insertion, if more rare, is the use of the equative copula là to mark a neutral type of topic in sentences like (12). Here, the fact that everybody likes mangosteens is merely indexed under the entry mangosteen in the common ground, without any special emphasis. Once again, focus could be intonationally assigned to the topic, but this is not a common strategy.

(p. 785)

(12)

More interestingly, Vietnamese has a contrastive topic marker, thì, which, as in (13), serves a role comparable to the topic marker /-dʑo˩/ in Na. This topic marker has the effect of contrasting the topic with its alternatives, which Clark (1992) analysed as an attributive meaning.

(13)

Thì can also be used after an entire sentence, but in that case, it does not clearly serve to rule out alternative interpretations. In (14), for instance, thì is used to set a piece of information as a common ground explaining the following clause (and is glossed as ‘so’). As pointed out by Cao (1992: 141), there does not appear to be a clear cut-off point between the contrastive and non-contrastive uses of thì.

Page 13 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Information Structure in Asia: Yongning Na (Sino-Tibetan) and Vietnamese (Austroasiatic) (14)

When the topic is co-referential with the subject, the use of an anaphoric pronoun in the subject position can also be used to focus it. This is illustrated in (15): (15)

Other markers can be used for contrastive topicalization in Vietnamese, such as còn ‘as for, and lại ‘on the other hand’. In the same way as their English equivalents, these two (p. 786) expressions are only possible if alternatives have explicitly been set into the common ground in previous clauses. Vietnamese sentences can have multiple topics as in (16). While we would not analyse là as a topic/theme marker, but rather as an equative/identification copula (contrary to Cao 1992, but following Clark 1996), the first two constituents are nonetheless both topicalized. (16)

38.3.2 Focus Experimental work on the phonetic realization of information structure in Vietnamese is limited to the realization of pragmatic focus (Jannedy 2007) or corrective focus (Michaud and Vu-Ngoc 2004; Michaud 2005). These studies reveal that these two forms of in-situ Page 14 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Information Structure in Asia: Yongning Na (Sino-Tibetan) and Vietnamese (Austroasiatic) focus are realized by most, but not all, speakers as an f0 increase on the target constituent (Michaud 2005; Jannedy 2007) and a hyperarticulation of voice quality (Michaud 2005). Duration and intensity are even more speaker-specific. From this, we can conclude that Vietnamese does not have a single grammaticalized means of marking contrastive focus, but that speakers pick and choose from a pool of hyperarticulation strategies. Jannedy (2007) also shows that although listeners can identify focused constituents, their performance is generally poor, and even more so for broad scope focus (entirely VP or entire sentence). The only other explicit proposal for a phonetic marking of IS in Vietnamese is Thompson (1965). In his section 12.5, Thompson describes a three-way stress system in which: By far the great majority of syllables are accompanied by medium stress; it is deviation from this in either direction which marks a particular form as conveying an especial heavy or light load in conveying information. In general, weak stress signals information already known or obvious in the context; heavy stress signals new or contrastive information. (Thompson 1965: 287) More details are given in Thompson’s following subsections (12.51 and 12.52), but he does not explicitly address the issue of the phonetic nature of stress in Vietnamese, and many (if not most) of his examples of stress notation fail to coincide with native speaker’s intuitions (regardless of the dialect). In the end, a close look at Thompson’s examples suggest that he is conflating several types of prominence: 1. Grammatical words seem to have reduced phonetic prominence, while lexical words do not. (p. 787) 2. As discussed above, contrastive focus seems to be marked by means of increased prominence. The existence of a prosodically-marked new information focus (phonetically identical or distinct from contrastive focus) is less clear. 3. Like many languages (Beckman and Edwards 1990; Turk and Shattuck-Hufnagel 2000), Vietnamese shows phrase-final lengthening. The nature of the phrasal domains that condition this lengthening is still unclear, as is the exact amount of phonetic lengthening involved. In any case, it is likely that this final-lengthening effect provides cues to prosodic phrasing, which in turn plays a role in conveying of IS (see the examples of pause insertion in Section 38.3.1). A systematic description of the intonational marking of focus thus remains to be undertaken. However, besides in-situ intonational focus, two other strategies are used to mark pragmatic focus. The first one is the use of contrastive topicalization (with marker thì) or of a topic followed by a resumptive pronoun, already described in Section 38.3.1. The second is the use of cái as a focus marker (Nguyễn 2004, 2013; Simpson and Hồ 2013). Although cái is normally used as a classifier, it also serves as a focus marker, and as such, always precedes a classifier or a mensural noun. Although cái usually conveys

Page 15 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Information Structure in Asia: Yongning Na (Sino-Tibetan) and Vietnamese (Austroasiatic) the idea of a negative emphasis, as in (17), Nguyễn (2004) also gives examples devoid of this negative connotation. (17)

Vietnamese also has focus-sensitive particles like thậm chí/đến ‘even’, cả ‘also’ and chỉ/ mỗi ‘only’ that indicate semantic focus (Hole 2013). Each of these focus-sensitive particles is associated with another particle introducing information that is already part of the common ground. In the case of thậm chí, đến ‘even’ and cả ‘also’, the cũng particle is mandatory, as in (18). In the case of chỉ ‘only’, the particle mới is optional, as in (19). A full discussion of these focus-sensitive particles can be found in Hole (2013), from which we take up the notion of background particle, which we label BG. Another focus-sensitive particle with a more limited scope is sentence-final không ‘only’ (Brunelle et al. 2012). (18)

(p. 788)

(19)

38.3.3 Givenness Like Yongning Na, Vietnamese allows frequent ellipsis of arguments; corpus studies will be necessary to quantify its referential density relative to a broad sample of other languages. Illustrative examples of subject and object pro-drop are given in (20) and (21). As in Yongning Na, the active referent is usually inferred from previous sentences, but it Page 16 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Information Structure in Asia: Yongning Na (Sino-Tibetan) and Vietnamese (Austroasiatic) is not uncommon to find sentences in which the referent is inferred from common situational knowledge. For instance, in (20) the subject could be interpreted as ‘we’ or ‘I’, depending on context. Argument ellipsis is often used as a strategy for avoiding an overt argument, especially in contexts where the proper use of pronouns is difficult to assess— for instance when the relative age or hierarchical status of interlocutors is unclear. Inversely, it tends to be avoided in formal situations, especially when the argument is a pronoun or an address term. (20)

(21)

As in Yongning Na, leaving out an argument is a way of backgrounding it; probably for the same reasons as in Yongning Na, object ellipsis seems rarer than subject ellipsis.

38.4 General Conclusion The overview in this chapter offers elements of answer to the question of the relative roles of morphosyntax and intonation in conveying IS. In Na, which has a host of (p. 789) particles conveying IS, intonation plays a lesser role in conveying IS, even though Na has emphatic stress as well as a specific intonational device for focalization. In Vietnamese, intonation seems to play a more important role, although experimental findings reported so far in the literature are largely limited to focus, and cover a relatively narrow range of speaking styles. The nature of the language’s tone system also appears to play a role. The multiplicity of phonetic cues to tone in Vietnamese allows for greater malleability: for instance, a tone that has final glottalization can have its overall f0 register raised to convey intonational information (Michaud and Vu-Ngoc 2004; Hạ and Grice 2010) without a major threat to lexical identification, so long as the phonation-type characteristics that are part of the definition of this tone are present. On the other hand, in a language where tones are specified solely in terms of pitch levels, such as Yongning Na, intonational

Page 17 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Information Structure in Asia: Yongning Na (Sino-Tibetan) and Vietnamese (Austroasiatic) modification of the f0 curve is more difficult for the hearer to tease apart from the effect of the tones themselves. On a methodological note, the findings reported above suggest that it is useful to maintain a clear functional distinction between lexical tone, on the one hand, and intonational modifications (reflecting boundaries/junctures and IS), on the other. A fully fledged study of the respective contributions of the various components of prosody (lexical tone; phrasing; prominence; speaker attitude …) in shaping f0 curves remains a task for the future. Research of this type has been attempted for Qiang, a Sino-Tibetan language (Evans et al. 2010); general proposals have been put forward for languages with lexical stress (Mishra et al. 2006).

Notes: (1) Many thanks to the Na and Vietnamese consultants and friends, and to the editors and reviewers. Financial support from Agence Nationale de la Recherche (HimalCo project, ANR-12-CORP-0006, and LabEx EFL, ANR-10-LABX-0083—Investissements d’Avenir) is gratefully acknowledged. (2) Unless specified otherwise, the examples provided in the chapter have been collected by the authors or coined and double-checked with native speakers. Apart from example (1), all the Yongning Na examples are drawn from a set of twenty narratives; the recordings are available online with time-aligned transcriptions through the Pangloss Collection (Michailovsky et al. 2014). The reference to these texts is provided in the following format: .. For instance, Housebuilding.259 refers to sentence 259 of the narrative ‘Housebuilding’, and Seeds2.67 refers to sentence 67 in the second version of the narrative ‘Seeds’. The numbering of sentences is indicated in the online display of the texts. Glossing follows the Leipzig Glossing Rules, with the following additions to the core set of standard abbreviations: ACCOMP for ACCOMPLISHED; ADVB for ADVERBIALIZER; CERTITUDE for an epistemic value of the copula in Yongning Na; and EXIST for EXISTENTIAL verbs. (3) The vertical bar ǀ indicates a juncture between tone groups; about the division of utterances into tone groups in Yongning Na, see Section 38.2.3.2. Tone is indicated by means of International Phonetic Alphabet tone letters placed after the syllable at issue: ˥ for High, ˧ for Mid, ˩ for Low, and the combinations for Low-to-High and for Mid-toHigh; for a presentation of the Yongning Na tone system, see Michaud (2013).

Alexis Michaud

Page 18 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Information Structure in Asia: Yongning Na (Sino-Tibetan) and Vietnamese (Austroasiatic) Alexis Michaud is Research Scientist at CNRS. His main research topic is phonetics/ phonology – especially the study of tone – but his commitment to 'all-out' linguistic fieldwork goes some way towards counterbalancing this narrow specialization. Marc Brunelle

Marc Brunelle is Associate Professor in the Department of Linguistics at the University of Ottawa. His research focuses on phonetics, phonology, and language contact, with an emphasis on tone and intonation in Southeast Asian languages, especially Vietnamese and Cham.

Page 19 of 19

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Freie Universitaet Berlin; date: 22 January 2019

Information Structure in Sign Languages

Oxford Handbooks Online Information Structure in Sign Languages   Vadim Kimmelman and Roland Pfau The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara Print Publication Date: Jul 2016 Subject: Linguistics, Signed Languages Online Publication Date: Apr 2014 DOI: 10.1093/oxfordhb/9780199642670.013.001

Abstract and Keywords This chapter demonstrates that the Information Structure notions Topic and Focus are relevant for sign languages, just as they are for spoken languages. Data from various sign languages reveal that, across sign languages, Information Structure is encoded by syntactic and prosodic strategies, often in combination. As for topics, we address the familiar semantic (e.g. aboutness vs. scene-setting topic) and syntactic (e.g. moved vs. base-generated topic) classifications in turn and we also discuss the possibility of topic stacking. As for focus, we show how information, contrastive, and emphatic focus is linguistically encoded. For both topic and focus constructions, special attention is given to the role of non-manual markers, that is, specific eyebrow and head movements that signal the information structure status of constituents. Finally, aspects that appear to be unique to languages in the visual-gestural modality are highlighted. Keywords: sign language, topic, focus, non-manuals, modality

40.1 Introduction IN this chapter, basic Information Structure (IS) notions will be applied to sign languages (SLs), that is, languages in the visual-gestural modality of signal transmission (as opposed to the oral-auditory modality of spoken languages). It will be demonstrated that IS notions are relevant for SLs, just as they are for spoken languages—as is expected, of course, given that SLs are fully-fledged natural languages with complex structures on all levels of linguistic description (Sandler and Lillo-Martin 2006).

Page 1 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University College London; date: 06 July 2018

Information Structure in Sign Languages At first sight, it may appear a little awkward to discuss SLs as a group within a single chapter. After all, one would not devote a chapter to ‘spoken languages’, but rather to a single language or a language family (as in Chapters 33–39). Comparative studies on SL structure have revealed that SLs differ from each other structurally and that, to a large extent, they differ along similar lines as spoken languages (Perniss et al. 2007; Pfau 2012). Still, in certain areas of grammar, SLs also show striking similarities, some of which are likely to be motivated by the visual-gestural modality (Meier 2002). As the following discussion will reveal, there are also interesting similarities in the realm of IS— and it may therefore not be so awkward after all to discuss SLs as a group in this chapter instead of focusing on a single SL. In Section 40.2, we will set the stage by providing some background information on SL structure and notation. Subsequently, the structure of the chapter reflects the well-known basic distinctions Topic–Comment (Section 40.3) and Focus–Background (Section 40.4). We will show that, across SLs, IS is encoded by syntactic and prosodic strategies, sometimes in combination, while morphological markers, the third strategy commonly found in spoken languages, are not attested. In Section 40.5, we conclude (p. 815) by briefly addressing modality effects: features in the expression of IS that are unique to SLs, such as the use of space and body leans. Throughout the chapter, we take a comparative perspective; that is, an effort will be made to provide and compare data from different SLs.

40.2 Background on sign language structure Space does not allow us to provide a detailed overview of aspects of SL structure. Therefore, we will focus on two aspects that will turn out to be relevant to the discussion of IS in Sections 40.3 and 40.4: the grammatical use of the signing space and of nonmanual markers. In most SLs investigated to date, the signing space in front of the signer’s body can be employed for referential purposes. This space can be used, for instance, to localize nonpresent referents. In the SL of the Netherlands (Nederlandse Gebarentaal, NGT) example in (1a), the signer talks about his brother, who is not present in the discourse setting. The lexical sign BROTHER is followed by an indexical (pointing) sign towards location 3a (forward right) in the signing space, thereby localizing the referent at this arbitrary location. Subsequently, the same location can be used for pronominalization, that is, the second instance of INDEX3a in (1a) is interpreted as ‘he’. Interestingly, some verbs, the so-called ‘agreeing’ or ‘directional’ verbs, can be modulated such that they target loci in signing space—be it the locus of a present referent or a locus established by means of INDEX. In (1a), the verb VISIT, which in its citation form moves forward from in front of the signer’s body, is directed from location 3a towards the signer, thereby agreeing with the subject and object (‘he visits me’; see Lillo-Martin and Meier (2011) for a recent

Page 2 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University College London; date: 06 July 2018

Information Structure in Sign Languages discussion of this phenomenon). The example is visualized by the video stills in Figure 40.1 (for VISIT, the beginning and end locations of the movement are shown).1 (p. 816)

Click to view larger Figure 40.1 Video stills illustrating example (1a). © Nederlands Gebarencentrum (Dutch Sign Centre; NGC 2002); stills used by permission.

(1)

Example (1a) also illustrates the use of a non-manual marker, namely a brow raise accompanying a topic NP. The topic constituent is followed by a prosodic break (indicated by the comma and realized as a pause and a change in non-manual activity). Actually, across SLs, non-manual markers—that is, mouth configurations, head and body movements, and facial expressions—may fulfil functions at all levels of grammar (see Pfau and Quer (2010) for an overview). Here, we only provide one more example of a syntactic marker. Across SLs, wh-questions are commonly marked by furrowed brows, as illustrated by the American SL (ASL) example in (1b) (Petronio and Lillo-Martin 1997: 26). While both markers in (1) can be argued to fulfil a syntactic function, it has also been claimed that at least some non-manuals are prosodic in nature and can thus be compared to intonational contours in spoken languages (Sandler 2011). Under this assumption, the wh-marker in (1b) could be considered the prosodic reflex of a [+wh]-feature.

40.3 Topic–comment distinction

Page 3 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University College London; date: 06 July 2018

Information Structure in Sign Languages Traditionally, topic constructions have received more attention in the SL literature than focus constructions. On the one hand, topics (especially their frequent use in (p. 817) ASL) have been subject to investigation since the early days of sign language research (Friedman 1976; Ingram 1978); on the other hand, they have been studied for various unrelated SLs, such as ASL (Aarons 1994; Todd 2008), Finnish SL (FinSL: Jantunen 2007), Hong Kong SL (HKSL: Sze 2008b, 2011), Israeli SL (ISL: Rosenstein 2001), NGT (Coerts 1992; Crasborn et al. 2009), and Russian SL (RSL: Kimmelman 2015). The data reveal that across SLs, topic (backgrounded) information tends to occupy a left-peripheral position in the clause; also, it is often accompanied by a non-manual marker (e.g. raised eyebrows) and followed by a prosodic break. It has thus been argued that topichood in SLs is commonly syntactically and prosodically marked. In addition, some SLs have even been claimed to be topic-prominent, for instance, ASL (McIntire 1982), British SL (Deuchar 1983), and ISL (Rosenstein 2001).2 In this section, we will address a number of complexities concerning the use and interpretation of topics. First, in Section 40.3.1, we will present and discuss different types of topics, focusing on semantic and syntactic distinctions that are well-known from the study of spoken languages (see Büring, this volume). We then turn to the issue of nonmanual marking, and the occasional absence thereof, in Section 40.3.2. Finally, in Section 40.3.3, we investigate possible combinations of topics, that is, topic stacking.

40.3.1 Types of topics 40.3.1.1 Semantic distinctions Topic constituents can be linked to the preceding discourse in various ways and may include different types of information. Here we adopt the basic semantic distinction between ‘aboutness’ and ‘scene-setting’ topics (Jacobs 2001), which is, amongst others, applied by Sze (2008b, 2011) in her study on HKSL. Although not everyone would agree that both aboutness and scene-setting topics belong to one general category of topics, the data discussed below (from HKSL and RSL) suggest that this a reasonable analysis for SLs, as both types of topics are similarly marked. It is generally assumed that an aboutness topic represents what the sentence is about. It includes information which is either familiar to the interlocutors or identifiable from the preceding context. Based on these characteristics, Sze (2011: 137) identifies the fronted object in (2a) as the topic of the sentence; this NP is followed by a prosodic break but it is not non-manually marked (see Section 40.3.2 for further discussion). The NGT example in (2b) contains a topicalized subject, which is non-manually marked and the articulation of which is prolonged by means of a final hold (as indicated by the horizontal line) (Coerts 1992: 223). For both examples, we must assume that the topicalized NP has been mentioned in the preceding discourse.

Page 4 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University College London; date: 06 July 2018

Information Structure in Sign Languages (p. 818)

(2)

In contrast to aboutness topics, scene-setting topics provide a spatial or temporal context within which the main predication holds (Chafe 1976). They may thus involve either locative expressions or temporal information (be it expressed by adverbials, NPs, or subordinate clauses). While aboutness topics usually include discourse-old information, scene-setting topics may include both discourse-new or -old information. Two examples from FinSL, in which a locative expression (3a) and a temporal adverbial (3b), respectively, occupy the topic position, are provided in (3); both topics are marked nonmanually (Jantunen 2007; ‘ew’ = eyes wide). (3)

At the same time, the examples in (2) and (3) also illustrate an important distinction related to the thematic roles assigned by the verb. Simplifying somewhat, aboutness topics are generally arguments of the verb (object in (2a), subject in (2b)), while scenesetting topics are adjuncts (see also Crasborn et al. (2009) for the distinction of ‘argument’ and ‘spatio-temporal’ topics). Sze (2011), however, points out that, occasionally, adverbials may also represent what the sentence is about. A third semantic type mentioned in the literature is contrastive topics (see Büring, this volume). Just like aboutness topics, contrastive topics refer to previously mentioned information and are arguments of the verb; unlike aboutness topics, however, they create a contrast between the topic constituent and a previously mentioned constituent (see example (5) below for further discussion).

40.3.1.2 Syntactic distinctions

Page 5 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University College London; date: 06 July 2018

Information Structure in Sign Languages Besides semantic characteristics, topics are also commonly distinguished based on their syntactic properties or, to be more precise, the extent to which they are (p. 819) syntactically integrated into the rest of the sentence. The crucial distinction, which has first been explicitly discussed for SLs by Aarons (1994), is that between ‘base-generated’ and ‘moved (fronted)’ topics. According to Aarons, the topic VEGETABLE in the ASL example (4a) must be base-generated in topic position, as it does not constitute an argument of the verb. Commonly, in these cases, an argument of the main clause bears a semantic relationship to the topic—in (4a), a relation of class membership (Aarons 1994: 152). Note that the example also illustrates that topics need not be specific and definite. (4)

Aarons also assumes that structures like (4b) involve a base-generated topic. While MARY —just like VEGETABLE in (4a)—is not an argument of the verb, it is co-referential with the pronominal INDEX in object position (cf. the NGT example (1a), which illustrates the same phenomenon for a subject). Constructions in which a topic NP is co-referential with a (resumptive) pronoun within the clause are often referred to as ‘left dislocation’ (López, this volume). The example in (5) displays different properties, as the clause following the topic NP would be ungrammatical by itself (Aarons 1994: 154; also see example (2b) above). According to Aarons, we therefore have to assume that the object NP MARY moved to the topic position (SpecTopP) leaving behind a trace (‘t’) in its D-structure position.3 Structures which involve movement of a noun phrase to the left periphery of the sentence are labelled ‘topicalization’. (5)

Interestingly, Aarons points out that topicalization either selects one specific member from a set or is used for contrastive focus. Therefore, what we are dealing with is actually focus (see the discussion of (18) in Section 40.4.1.2; also see Neidle (2002), who assumes that MARY in (5) actually occupies the specifier of a Focus Phrase). Wilbur (1997), however, argues that (5) might well be uttered in a context like the following: There are three girls in John’s class, Jane, Mary, and Sasha—Jane, he hates, but Mary, he (p. 820)

Page 6 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University College London; date: 06 July 2018

Information Structure in Sign Languages likes. In this context, MARY would indeed be a contrastive topic. The different possible functions of fronted constituents are attested in other languages as well, including English (Prince 1986). A third type that is often distinguished in the literature are ‘hanging’ topics (sometimes also referred to as ‘Chinese-style’ topics; e.g. Todd 2008; Sze 2011). In contrast to left dislocation, a hanging topic is not co-referential with an argument of the clause following it; in contrast to topicalization, the topic NP is not an argument of the verb. Sze (2011: 137) categorizes the topic constituent in (6) as a ‘hanging “aboutness” topic’. (6)

As for the syntax of topic constructions, Crasborn et al. (2009) observe that in NGT, a clause-final indexical sign may optionally be used, which is co-referential with the topic constituent. Crucially, this sign is different from the (resumptive) pronouns attested in examples like (1a) and (4b) above, as it does not occupy the position of the respective argument (NGT is an SOV language). The example in (7a) actually contains three pointing signs targeting the same position: the first one localizing the topic NP GIRL, the second one being a resumptive pronoun in subject position (indicative of left dislocation), and the third one appearing in clause-final position (Crasborn et al. 2009: 359). In previous studies on ASL (Padden 1988) and NGT (Bos 1995), it has been argued that such clausefinal pronominal signs always refer to the subject (hence the name ‘subject pronoun copy’). Crasborn et al., however, provide examples in which the final INDEX refers to an object (7b) or a locative expression. They argue that the final INDEX actually refers to the topic and therefore label the phenomenon as ‘topic agreement’. (7)

(p. 821)

40.3.2 Non-manual marking

Ever since the early studies on IS in SLs, it has been observed that topicalized constituents are commonly accompanied by specific non-manual markers. A marker that is mentioned in basically every study addressing this issue is brow raise; in addition, head movements are often described. Ingram (1978: 204), for instance, observes that ‘ASL Page 7 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University College London; date: 06 July 2018

Information Structure in Sign Languages topics may be marked with the raising of the eyebrows, the tilting of the head, the twitching of a shoulder’ and goes on by pointing out that it is ‘not yet known whether the choice of topic marker is determined by individual preference…, by sociolinguistic influences, by grammatical context, or by a combination of these’. Similarly, Liddell (1980) mentions brow raise and a slight backward head tilt for ASL, while Coerts (1992) concludes that ‘eyebrows up’ is the only non-manual feature characteristic of topics in NGT, as it is the only feature in her data that accompanied more than 50 per cent of all topics (in fact, this feature accompanied 90.6 per cent of all topics).4 Other markers that one may come across in the literature include ‘eye gaze at addressee’ and ‘head nod’ (see Sze (2011: 126f) for a convenient overview of non-manual markers as identified in different studies). In other words: the abbreviation ‘top’ as used in some of the above examples is often a cover term for a set of non-manual markers. Aarons (1994) was the first one who attempted to establish a correlation between topic type and non-manual marking. She distinguishes three subtly different non-manual markers, glossed as ‘tm1’, ‘tm2’, and ‘tm3’, which systematically accompany different types of topics in ASL. Here we only consider the former two, the form of which Aarons (1994: 156) describes as follows: tm1 —raised brows; head tilted slightly back and to the side; eyes widened; head moves down and forward → moved topics tm2 —large movement of head back and to the side; eyes very wide; head moves down and forward → base-generated topics She observes that ‘tm1’ accompanies moved topics like the one in (5) while ‘tm2’ occurs with base-generated topics of the type illustrated in (4). Note that it may well be the case that these (and other) markers are compositional in the sense that different components (e.g. brow position, head position) contribute distinct meanings. All of the non-manual markers discussed so far are domain markers in that they extend over a syntactic/prosodic domain. In addition, there are boundary markers. Leftperipheral topics generally constitute an intonational phrase which is followed by (p. 822) a prosodic break and which may also be manually marked by a phrase-final hold (as in (2b)). A non-manual marker which is frequently observed in this context is an eye blink (‘bl’) coinciding with the final hold or following the topic constituent, thus marking the prosodic boundary (Sze 2008a; Herrmann 2010), as illustrated in the German SL (Deutsche Gebärdensprache, DGS) example in (8) (Herrmann 2010: 23).5 (8)

Page 8 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University College London; date: 06 July 2018

Information Structure in Sign Languages All of Aarons’ examples include a non-manual (domain) marker accompanying all types of topics (moved and base-generated, arguments and adjuncts). Similarly, Jantunen (2007: 165) points out that in his data, topics ‘were always layered with some sort of a position of the eyes and eyebrows’. Later research on other sign languages, however, suggests that SLs may differ from each other when it comes to non-manual marking. Rosenstein (2001), for instance, claims that topics in ISL are not marked by brow raise—at least not in spontaneous discourse. Sze (2008b, 2011) observes that in HKSL, only a very small number of aboutness topics—including hanging topics, left-dislocated topics, and topical (clause-internal) subjects—are accompanied by brow raise and/or a specific head position (see e.g. (2a) above). Also, they are not consistently followed by an intonational break. In contrast, scene-setting topics in her data (with the exception of temporal adverbials) are much more likely to be accompanied by either brow raise, a forward head tilt, or both. In (9), the NP SECONDARY-TWO, which sets up a temporal domain, is marked by brow raise and forward head tilt (‘fht’) (Sze 2011: 14). (9)

In contrast to HKSL, in NGT and RSL both aboutness topics (10a) and scene-setting topics (10b) can be marked by eyebrow raise (Kimmelman 2015). However, neither of the two types of topics is obligatorily marked; in fact, the majority of topics remain unmarked. Furthermore, with respect to aboutness topics, the NGT and RSL data (p. 823) reveal another contributing factor: only shifted aboutness topics in these two SLs can be marked by eyebrow raise. (10)

Taken together, the data from different SLs reveal (i) that topics are commonly, but not consistently, marked by non-manual domain and boundary markers, (ii) that different types of topics may be accompanied by subtly different domain markers, and (iii) that SLs may differ with respect to the form and frequency of the markers employed.

Page 9 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University College London; date: 06 July 2018

Information Structure in Sign Languages

40.3.3 Topic stacking In his study on ASL topics, Ingram (1978) points out that topic NPs may combine with temporal and/or locative phrases to form what he calls the ‘theme’ of the sentence. Put differently, he may have been the first one to observe that argument and adjunct topics may be stacked, as shown in (11) (adapted from Ingram 1978: 204).6 (11)

Aarons (1994) further investigated the possibilities of topic stacking in ASL. She finds that there is a maximum of two topics adjoined to the sentence (CP) and also describes some interesting combinatorial constraints. First, she observes that two base-generated topics of the types presented in (4) may be combined, as illustrated in (12) (Aarons 1994: 176). However, while in (12a), the order of the two topics, neither of which is coreferential with an argument of the verb, is free, in (12b), reversal of the two topics would lead to ungrammaticality; that is, a topic which is co-referential with an argument of the clause (JOHN) must precede a topic which is in a class-member relationship with an argument of the clause (VEGETABLE). (p. 824)

(12)

Second, grammaticality judgements by native signers indicate that two moved topics can never co-occur, irrespective of order, as shown by the ungrammaticality of (13a) (Aarons 1994: 179). Third, when it comes to the combination of a base-generated and a moved topic, only the sequence of base-generated topic followed by moved topic appears to be marginally acceptable (13b), while the reverse order is judged ungrammatical (Aarons 1994: 177).

Page 10 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University College London; date: 06 July 2018

Information Structure in Sign Languages (13)

Crasborn et al. (2009) discuss topic stacking from a different perspective, focusing not on the co-occurrence of moved and base-generated topics but rather on the combination of argument and spatio-temporal (scene-setting) topics. They observe that in NGT, argument topics precede spatio-temporal topics and within the latter group, time specifications typically precede place specifications, as illustrated in (14) (Crasborn et al. 2009: 359). In Aarons’ system, the topic constituent [IXrt PERSON] would be classified as a basegenerated topic, as it is resumed by a pronoun in the clause. Note that the left-dislocated topic is not non-manually marked (‘neutral’ expression) while the scene-setting topics are marked by head tilt—which is in line with the observation made by Sze (2011) for HKSL. Crasborn et al. also claim that an argument topic always constitutes a prosodic unit of its own (an intonational phrase) while spatial and temporal topics may be merged into a single prosodic unit, as is the case in (14). (14)

Clearly, in a phrase structure model like that of Rizzi (1997), which assumes multiple topic positions within the left periphery, topic stacking can be accounted for (see Aboh, this volume). Lillo-Martin and de Quadros (2008), however, further assume that the higher topic position (which they label ‘Topic-Comment Phrase’) hosts base-generated topics while the lower one hosts moved topics—with a Focus Phrase sandwiched between the two topic positions, as illustrated in (15a) (also see (16b) below). (p. 825)

Page 11 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University College London; date: 06 July 2018

Information Structure in Sign Languages (15)

Similarly, Puglielli and Frascarelli (2007) claim—based on spoken and sign language data —that aboutness topics occupy a topic projection in the left periphery (‘ShiftP’) above contrastive (i.e. moved) topics (15b). They further assume a third topic position for familiar topics, which they situate highest in the tree (note that these are the topics, which, according to Aarons (1994), are marked by ‘tm3’, and which we excluded from the discussion in Section 40.3.2). While this order of dedicated positions allows for the derivation of the (marginally acceptable) ASL example (13b), it seems that one further has to assume that the projection for aboutness topics (‘T-C P’ or ‘ShiftP’) is recursive in order to account for the examples in (12).7

40.4 Focus–background distinction Although the focus–background distinction in SLs has been explored in less detail than topics, the expression of focus has been relatively well studied for ASL (Wilbur 1994, 1996; Neidle 2002), and some data is available for other sign languages, including Brazilian SL (Língua de Sinais Brasileira, LSB: Nunes and de Quadros 2008), NGT (Van der Kooij et al. 2006), and DGS (Herrmann 2013); for an overview, see also Wilbur (2012). Based on the available data from these languages, we first address (p. 826) types of focus in Section 40.4.1; this will be followed by a discussion of non-manual marking of focus in Section 40.4.2, while the association with focus will be reviewed in Section 40.4.3.

40.4.1 Types of focus

Page 12 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University College London; date: 06 July 2018

Information Structure in Sign Languages It has been shown that different types of focus, including information focus, contrastive focus, and emphatic focus, each have a different means of expression across SLs (Wilbur 2012). Crucially, just like topics, all three types of focus can be expressed both syntactically and non-manually (see Section 40.4.2 below). Before addressing the three types of focus, we should briefly address the relation between focus and stress in SLs. As in many spoken languages, the focused constituent in SLs generally has to be stressed (Wilbur 1994, 1999). Despite a disagreement among researchers as to how exactly stress is realized in different SLs, the most common features of a stressed sign are longer duration, larger movement trajectory, and higher velocity of the movement (Wilbur 1999; Van der Kooij et al. 2006; Crasborn and Van der Kooij 2013). As Crasborn and Van der Kooij (2013) point out, there is no research on potential analogues of focus projection in sign languages (see Arregi, this volume).

40.4.1.1 Information focus As for the first type of focus under consideration, information focus is often not marked syntactically (Lillo-Martin and de Quadros 2008), which means that the constituent expressing information focus remains in situ, as in (16a), which could be an answer to the question ‘What did you read?’. However, in ASL, the focused constituent can optionally be moved to a clause-initial position—SpecFocP in Neidle (2002) and Lillo-Martin and de Quadros (2008). The example in (16b) illustrates the co-occurrence of a left-peripheral information focus (BANANA) with a base-generated topic (FRUIT). As pointed out by Lillo-Martin and de Quadros (2008: 169), this sentence could be uttered as a reply to the question ‘As for fruit, what does John like?’ (‘t-c’ = topic-comment topic; see (15a)). (16)

It should be noted that a different analysis of the ASL data has been proposed by Petronio (1991) and Wilbur (1997), who claim that stress in ASL—unlike in English—has a fixed position: the right edge of the clause. As a consequence, the constituent in focus must (p. 827) either undergo rightward movement or be doubled, that is, appear once in its original position and once in the clause-final one (see Section 40.4.1.3 for discussion).8 Should the focus be expressed syntactically, one of the main means to do so in SLs is the so-called ‘wh-cleft construction’, which has been described for many SLs, including Australian SL (Johnston and Schembri 2007), NGT, and RSL. Wilbur (1996) argues that ASL constructions like (17a) should be analysed as wh-clefts, where the first clause expresses the topic and the second clause expresses the focus of the sentence (Wilbur Page 13 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University College London; date: 06 July 2018

Information Structure in Sign Languages 1996: 210).9 She provides several syntactic arguments—for instance, embedding, interaction with negation, and ellipsis—for the claim that these constructions are single sentences, and not a combination of a rhetorical question and an answer, as had previously been argued by Baker-Shenk (1983). In particular, she argues that the first part of the construction is a reduced relative clause, while the second part is a constituent smaller than a clause, which conveys new information. (17)

Recently, however, Caponigro and Davidson (2011) have claimed that these constructions are not syntactically and semantically parallel to wh-clefts in spoken languages, but have instead analysed them as clausal question–answer pairs unique to SLs. According to them, the function of these constructions is indeed connected to Information Structure, but cannot be subsumed under the traditional functions of wh-clefts. While they agree with Wilbur in treating the whole construction as a single declarative sentence, they provide arguments against her analysis of the two parts of the construction. As for the first (question) part, they argue that it is an embedded interrogative and not a relative clause, as ASL relative clauses never contain wh-words. In addition, they show that the question part can also be a polar question. As for the second (answer) part, they provide evidence that it actually is a partially elided declarative clause. For illustration, consider the example in (17b), which includes a polar question in the first part and the obligatory negative reply NO along with the optionally omitted material (HE NOT HAVE MOTORCYCLE) in the second part (Caponigro and Davidson 2011: 336). (p. 828)

40.4.1.2 Contrastive focus

With respect to the second type of focus, contrastive foci in SLs are also often marked by topicalization, as has already been mentioned in the context of example (5). Aarons (1994), for instance, has shown that in ASL, topicalization can be used to express contrast, as illustrated in (18) (Aarons 1994: 159; example slightly adapted). (18)

Page 14 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University College London; date: 06 July 2018

Information Structure in Sign Languages Lillo-Martin and de Quadros (2008) also claim that in ASL and LSB, the constituent which is in contrastive focus is moved to the focus projection in the left periphery of the sentence. As pointed out previously, however, Wilbur (1997) has suggested that a leftperipheral contrastive constituent can either be a contrastive focus or a contrastive topic. This implies that topicalization as a syntactic mechanism is used to express a variety of meanings, and not just contrastive focus. Interestingly, syntactic marking of contrast can also be achieved by certain modalityspecific means, in particular, by making use of the signing space or the two manual articulators. As for the former, two referents that are contrasted can be localized in opposite locations in the signing space, and this strategy alone can yield a contrastive reading in SLs. Moreover, contrasted referents may trigger the use of a so-called ‘dominance reversal’ (Frishberg 1985). Dominance reversal is a phenomenon connected to the fact that signers usually use one hand more actively than the other, so that one-handed signs, for instance, are signed with this dominant hand (‘dh’). However, sometimes the roles of the hands are switched with the non-dominant hand (‘ndh’) becoming more active. According to Frishberg, this strategy can be applied in order to separate a topic (signed with one hand) from the comment (signed with the other hand), as in the Jordanian SL (Lughat il-Ishaara ilUrdunia, LIU) example in (19a) (Hendriks 2007: 250), but it may also serve the purpose of contrasting referents, as in the NGT example in (19b) (adapted from Crasborn and Van der Kooij (2013)). The latter use implies that referents localized contrastively (to the right and to the left of the signer) can be signed with two different hands, leading to the dominance reversal. (19)

In addition to syntactic marking, contrastive focus or, more generally, contrast is usually marked non-manually. The non-manual marking associated with it can be the (p. 829)

same as the one that is associated with information focus (Lillo-Martin and de Quadros 2008), but contrastive focus appears to be marked more consistently (and other special means are also available, see Section 40.4.2).

40.4.1.3 Emphatic focus

Page 15 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University College London; date: 06 July 2018

Information Structure in Sign Languages Turning to the last type of focus, emphasis in SLs is usually connected to the phenomenon of doubling, which means that a constituent referring to the same object/activity appears twice within a single clause, as illustrated for the modal verb CAN in the LSB example in (20) (de Quadros 1999; cited in Nunes and de Quadros 2008: 178). (20)

This phenomenon has been described for many SLs, including ASL, LSB, and HKSL (Sze 2008b). According to Nunes and de Quadros (2008), doubling in ASL and LSB is used to express emphasis, and it surfaces when two copies of a moved constituent are overtly realized. Alternative analyses of doubling have been proposed by Petronio (1991), who connects doubling to all types of focus, and by Fischer and Janis (1990), who argue that doubling is motivated by morphosyntactic factors. Recently, Kimmelman (2012) has proposed a unified account of doubling in RSL and NGT. Based on the observation that different types of constituents—from pronouns to full clauses—can be doubled, he claims that doubling in these two sign languages serves for foregrounding of both topical and focal information, both at the syntactic and discursive level. Although prominent across SLs, doubling as a grammatical phenomenon is not unique to the visual-gestural modality. Kandybowicz (2007) has shown that it occurs in many spoken languages from different language families, and that it fulfils similar functions related to IS: emphasis, contrast, or polarity.

40.4.2 Non-manual marking Similar to topics, foci in SLs can be marked non-manually. Interestingly, non-manual marking of foci is in many respects similar to non-manual marking of topics. For instance, according to Lillo-Martin and de Quadros (2008), different types of focus (information, contrastive, and emphatic) in ASL and LSB are consistently marked by eyebrow movement and a backward head tilt; judging by pictures provided in their article, contrastive focus is also marked by a forward head tilt. In other SLs, similar non-manual markers have been observed. In DGS, for instance, replacing and corrective foci are marked by raised eyebrows and head tilts and movements, while information and (p. 830) selective foci are not consistently marked (Waleschkowski 2009). In addition, Crasborn and Van der Kooij (2013) claim that information and contrastive foci in NGT can be marked by a large number of non-manuals. In particular, all kinds of foci are characterized by eye contact between the signer and the addressee and mouth actions; in addition, focus can be marked by body lean, head tilt, head nod, and brow raise.

Page 16 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University College London; date: 06 July 2018

Information Structure in Sign Languages One can notice that non-manual markers that are used to mark focus (such as raised eyebrows) also commonly accompany topics. In addition, in some SLs both topics and foci can be fronted. These similarities could be taken to suggest that these markers actually mark something else; Wilbur and Patschke (1999), for instance, claim that raised eyebrows in ASL consistently mark A′-positions. On the other hand, there are also numerous differences in syntactic and non-manual marking of topics vs foci in different SLs (see also below). Further research on more SLs is required to disentangle how syntactic position and (possibly subtle) differences in non-manual marking correlate with the IS status of a constituent. Until recently, it was largely unknown how exactly different non-manual markers interact with each other. Lately, however, the importance of this question has begun to be understood. For instance, Crasborn and Van der Kooij (2013) discuss the possibility that in NGT, raised eyebrows, open eyes, and backward head tilts are all realizations of an abstract phonological feature [open up!]. It would therefore be insightful to study the features that can have multiple phonetic realizations and find out their IS-related functions. In addition to non-manual markers that are used to mark both topics and foci, there are non-manual markers associated specifically with focus and contrast, namely body leans and contrastive spatial localizations. Wilbur and Patschke (1998) have found that forward and backward body leans in ASL serve to mark inclusion vs exclusion in general, and, more specifically, that they may also mark stress, accompany the focus particles EVEN and ONLY, and signal focus and contrast. According to Van der Kooij et al. (2006), body leans fulfil similar functions in NGT: in particular, the focus particles ONLY and ALSO are accompanied by backward and forward leans, respectively. Unlike ASL, however, in NGT contrast is expressed by left- vs rightward body leans, as shown in (21), where two actions are contrasted (Van der Kooij et al. 2006: 1607, slightly adapted). (21)

Non-manual contrast marking by sideward body leans is probably a manifestation of a more general strategy, namely localizing contrasted referents in opposite areas in the signing space, as discussed in Section 40.4.1.2. By leaning towards these contrastive locations, the signer further emphasizes the contrast. Actually, the same contrastive opposition is also manifest in dominance reversals. The fact that three related strategies—body leans, contrastive localization, and dominance reversal—are used to express contrast in SLs may be an argument in favour of considering contrast a separate notion of IS, and not a subtype of focus (see Repp, this volume). However, Crasborn and Van der Kooij (2013) have suggested that in longer (p. 831)

Page 17 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University College London; date: 06 July 2018

Information Structure in Sign Languages stretches of discourse, these strategies can be used to mark focus (without contrast) as well. Therefore, this question requires further research, in particular, the analysis of data from various SLs.

40.4.3 Focus particles Focus-sensitive particles, such as only and also (Beck, this volume), also exist in SLs. These particles have, for instance, been claimed to exist in ASL (Wilbur 1994) and NGT (Van der Kooij et al. 2006). Herrmann (2013) investigates and compares the use of focus particles in NGT, DGS, and Irish SL, describing both similarities and differences between the three languages. Manually expressed particles are usually associated with nonmanual marking. Focus particles are often accompanied by non-manuals connected to focus, such as eyebrow raise and head tilt, as well as by additional markers. For instance, the manual sign ONLY in ASL and NGT is often accompanied by a backward body lean, while the particle ALSO is usually accompanied by a forward body lean (Wilbur and Patschke 1998; Van der Kooij et al. 2006). Moreover, these non-manuals alone can express the relevant meaning, that is, the manual sign is optional, as is evident from example (22) (Van der Kooij et al. 2006: 1603). (22)

According to Herrmann (2013), particles expressing meanings similar to only and also are present in many SLs; however, a particle that is functionally equivalent to even has not yet been found in any SL. The emphatic meaning connected to this latter particle in spoken languages can only be expressed non-manually in SLs, or by a non-manual combined with a manual particle with a different function. For example, in DGS, the emphatic non-manual expression can be combined with the sign ALSO, yielding the emphatic meaning (23) (Herrmann 2013: 261, slightly adapted; PF stands for a particle, ‘g-pu’ for a palm-up gesture). (23)

This fact is interesting from a typological perspective, as it provides evidence for the universality of the restrictive and additive particles, but not the emphatic ones. While the status of the non-manual marker that can serve to express emphatic meaning is not (p. 832)

Page 18 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University College London; date: 06 July 2018

Information Structure in Sign Languages clear yet, its use in (23) suggests that the relevant meaning can be expressed compositionally.

40.5 Modality effects While spoken languages exist in the oral-auditory modality, sign languages exist in the visual-gestural modality. Furthermore, SLs use several partially independent articulators to transfer information, namely the two hands, the torso, the head, the lips, and the eyebrows, among others. Therefore, it is always important to determine whether and, if so, how the modality influences the structure and functioning of a sign language. Separating modality-specific from modality-independent properties of languages is also crucial to the quest for linguistic universals and our understanding of the linguistic capacity in general. For this reason, we will briefly summarize the possible modality effects identified in the domain of IS in SLs. The first major modality effect that appears in connection to IS is the crucial role of nonmanual marking. It has been claimed that non-manual marking in SLs is in many respects parallel to intonation (pitch movement) in spoken languages (Pfau and Quer 2010). However, despite the parallels, non-manual marking is formally much more elaborate than intonation due to the availability of multiple articulators. These different articulators are commonly used to express several separate bits of information simultaneously; for instance, eyebrows can mark information focus while a body lean may express contrast at the same time. In addition, several non-manual features can be a realization of a more abstract phonological feature (see Section 40.4.2 above). Furthermore, non-manuals interact with manual signs in a non-trivial way: they can double the meaning of the sign (as in the case of non-manual marking accompanying ONLY or ALSO); they can have a separate meaning (e.g. a sideways body lean that expresses contrast), or they can create a meaning in combination with the manual sign (as in the case of emphatic non-manual marking combined with the sign ALSO yielding the meaning ‘even’). As a result, in SLs, the non-manual channel can be employed to express elaborate semantic distinctions in the domain of IS. Another group of effects of modality on IS is connected to the presence of two hands as partially independent articulators, which in principle would allow for expressing two streams of information simultaneously. Due to processing difficulties, the hands are generally not used independently: it is almost impossible to produce or understand two separate utterances at the same time. However, the hands do not always act as one articulator either, and signers can exploit their partial independence to structure information. For instance, dominance reversal can be used to express contrast and also to separate the topic, signed with one hand, from the comment, signed with the other hand (Frishberg 1985). Moreover, the use of the non-dominant hand in general is probably (p. 833) connected to IS: it may convey less important or backgrounded information and

Page 19 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University College London; date: 06 July 2018

Information Structure in Sign Languages maintain a continuous topic in discourse (Vermeerbergen et al. 2007). Although a few studies that consider this phenomenon are available, the role of the two hands in relation to IS awaits further research. The third group of effects is connected to the use of space, a hallmark of the grammar of SLs. As mentioned previously, referents are often localized in the signing space, and this localization is connected to topicality (Barberà 2012) and contrast. Furthermore, localization can be realized not only directly (through pointing signs) but also indirectly by means of body leans and dominance reversals. As has already been suggested, these mechanisms may be relevant for the discussion of the universality of IS notions (Section 40.4.1.2). Finally, another interesting difference between SLs and spoken languages has surfaced in the discussion of structures that have been analysed as wh-clefts, a construction also attested in spoken languages. Recently, however, it has been claimed that these structures actually constitute question–answer pairs of a type not attested in spoken languages (Caponigro and Davidson 2011). It is not clear how this phenomenon can be directly connected to the visual-gestural modality of SLs; however, if it is confirmed that only SLs have such pairs, a modality-specific explanation would be welcome. To sum up, the discussion in this chapter reveals that across SLs, IS is expressed by syntactic and prosodic strategies that are similar to those that have been described for spoken languages. Crucially, the basic IS notions, such as topic and focus, which have been developed based on spoken languages alone, can also be applied to SLs. However, modality, as manifested in multiple simultaneously used articulators and the use of space, also has its effect on the encoding of IS, as it allows SLs to employ means of expressing IS that are not available in spoken languages. (p. 834)

Notes: (1) Following standard conventions, SL examples are glossed in English in SMALL CAPS. Obviously, these glosses do not provide any information about the phonological form of a particular sign. Subscript numbers (in some examples, subscript letters) indicate points in the signing space used in pronominalization and verbal agreement, whereby ‘1’ is used for a locus close to the signer, ‘2’ for a locus close to the addressee, and ‘3’ for established loci and loci close to present third-person referents. INDEX/IX stands for a pointing sign (usually with index finger extended) and POSS for a possessive pronoun. The convention SIGN-SIGN is used when two English words are necessary to gloss a single sign; SIGN^SIGN indicates that two signs are combined in a compound. Lines above the glosses indicate the scope (i.e. onset and offset) of a particular non-manual marker. Note that abbreviations used may refer to the form (e.g. ‘br’—brow raise) or the function (e.g. ‘top’—topic) of a particular marker. Other abbreviations used in this

Page 20 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University College London; date: 06 July 2018

Information Structure in Sign Languages chapter include: ‘neg’—negative headshake and ‘wh’—wh-question marker (further abbreviations will be explained in the context of specific examples). (2) Slobin (2013) goes further by claiming that all SLs are topic-prominent, based on Li and Thompson’s (1976: 466) observation that ‘[i]‌n topic-prominent languages, there will be a surface coding for the topic, but not necessarily for the subject’. (3) Aarons (1994: 154) also provides the corresponding example with the topicalized subject JOHN (JOHNi, ti LOVE MARY). In this case, only the prosody (non-manual marking and pause) indicate that we are dealing with a topic construction, as the word order is SVO. Note that it is crucial that the verb in (5) is a plain verb, as it is commonly assumed that agreeing verbs may license empty arguments (pro) in subject and object position; that is, in the presence of an agreeing verb, it cannot be determined whether the topic is moved (trace in argument position) or base-generated (pro in argument position). (4) Janzen (1999) argues that the non-manual topic marker (in particular, raised eyebrows) is grammaticalized from a communicative questioning gesture commonly used by speakers. This non-manual co-speech gesture entered the grammar of ASL as a yes/no question marker and then developed further into the topic marker (for grammaticalization in SLs, see Pfau and Steinbach 2011). (5) Liddell (1980) and Aarons (1994) further observe the presence of a head nod (‘hn’) on the subject of the main clause in case of VP topicalization; see example (i) (Liddell 1980: 30).

((i))

(6) The example actually comes from a study on pauses in ASL by Grosjean and Lane (1977). Ingram does not provide the non-manual marker in his gloss, but points out that the constituent GIRL SMALL carries a non-manual topic marker. The commas reflect pause durations as measured by Grosjean and Lane. (7) With respect to SL, Puglielli and Frascarelli (2007) do not specify the hierarchical position of the Focus Phrase, but only point out that ‘Focus is realized in a position in the C-domain that is higher than the subject (but lower than a Topic)’ (2007: 155). For spoken language, they establish the sequence ShiftP > ContrP > FocP > FamP > IP. Moreover, they explicitly claim that only FamP is recursive. They base their arguments on the data provided by Aarons (1994), but it has to be pointed out that (i) they neglect the fact that examples like (13b) are only marginally acceptable and (ii) they wrongly claim that the combination of two base-generated (aboutness) topics is ungrammatical in general (see (12b)). (8) For a recent discussion of different analyses, see Wilbur (2012). Page 21 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University College London; date: 06 July 2018

Information Structure in Sign Languages (9) See Wilbur (1997) for a discussion of other types of clefts attested in ASL.

Vadim Kimmelman

Vadim Kimmelman is a postdoctoral researcher at the Amsterdam Center for Language and Communication (ACLC) and he is also teaching at the Department of Literary Studies and Linguistics at the University of Amsterdam. His research interests are in sign linguistics, including morphology, syntax, semantics, and pragmatics of sign languages. Roland Pfau

Roland Pfau is Associate Professor at the department of General Linguistics at the University of Amsterdam, where he teaches in the sign linguistics program. In his research, he focuses on morphosyntactic and syntactic aspects of sign languages, (sign) language typology, and grammaticalization.

Page 22 of 22

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University College London; date: 06 July 2018

Information Structure in Bantu

Oxford Handbooks Online Information Structure in Bantu   Laura J. Downing and Larry M. Hyman The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara Print Publication Date: Jul 2016 Subject: Linguistics, Morphology and Syntax Online Publication Date: Mar 2015 DOI: 10.1093/oxfordhb/9780199642670.013.010

Abstract and Keywords For some 40 years, the role that information structure (IS) plays in the grammatical structure of the ca. 500 Bantu languages has been the topic of considerable research. In this chapter we review the role of prosody, morphology and syntax in expressing IS in Bantu languages. We show that prosodic prominence does not play an important role; rather syntax and morphology are more important. For example, syntactic constructions like clefts and and immediately after the verb position correlate with focus, while dislocations correlate with topic. Among the morphological properties relevant to IS are the “inherently focused” TAM features (progressive, imperative, negative etc.) and the “conjoint-disjoint” distinction on verbs, as well as well as the presence vs. absence of the Bantu augment on nominals. Finally, we consider a range of tonal effects which at least indirectly correlate with IS (tonal domains, metatony, tone cases). Keywords: Bantu languages, focus particle, IAV, conjoint, disjoint, topicalization hierarchy, subject inversion, dislocation, focus prosody, inherent focus

39.1 Introduction THERE are some 500+ Bantu languages, making it the largest subgroup within the Niger-Congo language phylum, which is in turn the largest in the world, comprising approximately 1,500 languages. Bantu languages are spread over most of Africa south of a line extending from eastern Nigeria in the west to southern Somalia in the east. This language family has provided important insights into the marking both of topic and focus, which have captured the attention of general linguists from at least the 1970s on.

Page 1 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu In this chapter we present an overview of the major means by which different aspects of information structure are marked in representative Bantu languages. After a brief background section which sketches the basic grammatical properties of Bantu languages, the two main sections of the chapter present a brief survey of the some of the ways that focus and topic structure the grammars of particular Bantu languages.1

(p. 791)

39.2 Background information

39.2.1 Phonology Both Proto-Bantu (PB) and most Bantu languages exploit tone for both lexical and grammatical contrasts (Meeussen 1967; Kisseberth and Odden 2003). By convention high (H) tone is marked with an acute accent and low tone (L) is either unmarked or is indicated by a grave accent. In many of the eastern and southern Bantu languages that have lost the Proto-Bantu vowel length contrast, lengthening of the penultimate vowel is a cue to prosodic phrasing, as we shall see (Downing and Pompino-Marschall 2013; Hyman 2013).

39.2.2 Morphology A hallmark of Bantu languages is their rich noun class agreement system. A head noun takes a prefix indicating its agreement class, and modifiers in construction with it take an appropriate agreement prefix. As seen in the Swahili (G41–43) examples in (1), agreement is also marked on verbs, on the obligatory subject prefix and an optional object prefix:

(1)

Page 2 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu Verbs also contain tense–aspect affixes, as shown in the maximal verbal template in (2), cited from Nurse (2003: 90) and illustrated in (3) with an especially complex Kinande (J/ D42) example, both cited from Nurse and Philippson (2003: 9): (2)

(3)

In these examples, ≠ marks a stem boundary and = marks a clitic boundary. In canonical Bantu, the verb stem consists of a root (e.g. ≠king-), possible verb extensions marking causative (-is-i̹), applicative (-ir-), etc., and an obligatory final inflectional ending, for example /-a/ in the example in (3). Tense–aspect and agreement markers typically precede the stem. (p. 792)

39.2.3 Syntax The basic word order in most Bantu languages is: (Subject) Verb (Object1) (Object2) (Oblique) (Heine 1976; Bearth 2003). Although there is variation, canonical Bantu languages allow multiple objects, with a non-theme (e.g. benefactive) object generally preceding the theme object. Zulu (S42) provides an example where this order is rigidly enforced (apostrophes indicate vowel elision; note the penult lengthening in the phrasefinal word):

(4)

2

(Adjuncts follow objects, as shown in (5a), below.) In (4) and throughout this chapter parentheses are used to indicate phrasal domains. We thus note in the above example that under broad focus, a Zulu sentence can be parsed into a single prosodic phrase whose main cue is penultimate lengthening, indicated by a doubled vowel.

39.2.4 Preview of the effects of information structure

Page 3 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu Information structure has consequences for prosody, morphology, and syntax in many Bantu languages. This is again illustrated with data from Zulu, in (5), where the resumptive class 6 object marker is in bold and the focus is underlined: (5)

In this exchange IS has influenced the syntax (word order), morphology (verb morphology), and prosody (vowel length) of each utterance. As was stated in §2.3 the (p. 793)

basic word order in Bantu is for an object to precede an adjunct, as in the first sentence of (5), which represents ‘broad focus’. However, in Zulu, as in many Bantu languages, (most) focused constituents, including WH-elements, occur in the immediate after verb (IAV) position, while non-focal information commonly occurs in peripheral positions. As a result, the remaining sentences in (5) do not have canonical word order: that is, the objects do not precede the adjuncts. Instead, the WH-adjunct ngááni ‘how’ precedes the object in the question, and the object is sentence initial (A1) or sentence final (A2) in the answers. Object marking on the verb (embolden) obligatorily resumes an object that does not occur in its basic position. Focused IAV elements are followed by a prosodic phrase break that correlates with penultimate vowel lengthening. The next two sections discuss in detail these kinds of correlates of, first, focus and then topic (or out of focus) in representative Bantu languages.

39.3 Focus marking Page 4 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu As Hyman and Watters (1984: 238) have observed: ‘Focus distinctions are typically realized in one of three ways: prosodically, morphologically, and syntactically’. As these authors show, it is not just Bantu, but many African languages which exploit these three types of marking. However, as Hyman (1999: 152) points out, features in Bantu languages that appear to be focus-conditioned are often instead conditioned by certain grammatical configurations which in turn only imperfectly correlate with focus. This point will be amply demonstrated in this section.

(p. 794)

39.3.1 Syntactic marking of focus: IAV position

Significantly, many Bantu languages exploit the immediate after verb position (IAV) for focus-marking, especially of non-subjects. This was illustrated for Zulu in (5), above, and further exemplified by the Luganda (J/E15) exchange in (6):

(6)

The IAV requirement on non-subject focused elements is found in a number of other Bantu languages, including Kimatuumbi (P13) (Odden 1984), Makhuwa (P31) (van der Wal 2006, 2009), Matengo (N13) (Yoneda 2011), Tswana (S31) (Creissels 1996a, 1996b, 2004, 2012), and Tumbuka (N21) (Downing 2012). It is also found in more distant Bantoid languages such as Noni (Hyman 1981), Naki (Good 2010), and Mambila (Perrin 1994, cited in Güldemann 2007) and in other African language families, for example Chadic (Tuller 1992). Since most Bantu languages are canonically SVO, exploiting IAV position is consistent with the observation of Harries-Delisle (1978) that the object position is often the locus of focus marking, hence IAV in VO languages vs immediate before verb (IBV) position in OV languages. The Western Grassfields language Aghem has been widely cited as exploiting the IAV for focus (Watters 1979; Hyman 1985, 2010; Hyman and Polinsky 2010), as in the following examples:3

Page 5 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu (7)

The broad focus word order S-AUX-V-O-X is shown in (7a). In (7b) the focused temporal adjunct ‘today’ is in IAV position with the backgrounded object ‘fufu’ appearing after it. The same effect is achieved in (7c), where defocused ‘fufu’ is instead in preverbal position, thereby isolating ‘today’ as the focused IAV element. Example (7d) shows that the subject may also be focused in IAV position, requiring an expletive subject. Only some canonical Bantu languages that exploit the IAV position for focus allow this option for focusing the subject. (See the discussion of subject inversion in Section 39.4.1.1.) (p. 795)

As already seen in the Zulu examples (4) and (5), it is not only explicitly focused constituents that occur in IAV position, but also WH-elements, which are inherently focused. As a result of their proximity to the verb, they often develop into enclitics, as in Luganda:4

Page 6 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu (8)

Interestingly, it is possible for the WH-element to occur in post-IAV position in an echoquestion (Hyman and Katamba 2011: 70). Thus, contrast (9a) with (8b). (9)

In addition, (9b), a subject relative, which here requires the determiner-like augment morpheme a-, shows that WH elements are not attracted to the IAV in relative clauses. Instead, the WH element occurs in situ, as in (9c), as it would in an echo question. In (10a) we see that multiple post-verbal WH-markers cannot co-occur: (p. 796)

(10)

(Luganda otherwise allows strings of more than one enclitic, e.g. y-a-gí-téék-á =múù =kí ‘what did he put in it?’, lit. he-PAST-it-put-TAM =in =what). Since subjects are questioned in situ, the multiple WH-question in (10b) is grammatical, as is (10c), which involves a cleft construction marked by the relativizer (REL) gwè. Since Aghem allows multiple WH elements to follow the verb, while languages like Luganda do not, whether a WH element can occur post-verbally outside the IAV is a parameter of variation across Bantu

Page 7 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu languages. Interesting in this context, Basaá (A43) shows attraction of most non-subject WHs to the IAV, but speakers prefer canonical word order in the corresponding answers (Hamlaoui and Makasso 2011: 55): (11)

Hamlaoui and Makasso conclude that there must be a reason other than focus that WHelements, but not answers, are attracted to the IAV in Basaá, and accounting for this distinction is a topic of current research. While the IAV focus word order strategy is quite widespread, many Bantu languages do not exploit it. An alternative focusing structure is the cleft, also available in Luganda: (12)

(13)

In (12) evidence for the cleft structure is that the same marker (gwè) is used as in relative clauses, while in (13) note that the answer uses a copular proclitic, yèè= (Hyman and Katamba 1990: 6–7). As also seen in (13), in Luganda subject questions optionally take an augment (a determiner-like element), but cannot be clefted. Similarly, adjunct WH-words cannot be clefted, although their answers can (Walusimbi 1996: 67, 71; Hyman and Katamba 2011: 76).5 (p. 797)

39.3.2 Morphological marking of focus As in many languages in the world, some Bantu languages have specific focus markers that appear before or after the targeted constituent. In Abo (A42), when the focus marker ndi occurs after a pre-verbal constituent, for example the subject in (14b), it focuses that

Page 8 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu constituent. When it occurs immediately after the verb, as in (14c), it ambiguously marks focus on the verb, the object, or the verb phrase (Hyman and Lionnet 2012: 11):

(14)

While marking focus with particles or adpositions is relatively common crosslinguistically, what makes Bantu (and certain other African languages) special is a focusdependent morphology affecting either TAM marking or nominal complements of the verb. Originally surveyed under the heading of ‘auxiliary focus’ (Hyman and Watters 1984), TAMs may either exhibit allomorphic distinctions correlating with the focus structure of the utterance, or they may show ‘inherently focused’ effects on the rest of the utterance. In both cases the effects typically involve differences in phrasing that affect the prosody, especially tone. We survey these possibilities in the following sections. (p. 798)

39.3.2.1. Conjoint/disjoint verb morphology

Originally introduced into Bantu by A. E. Meeussen, the terms ‘conjoint’ vs ‘disjoint’ refer to contrasting allomorphs of the same tense–aspect category which (partially) encode different focus structures in main clauses. These are found in languages such as Kirundi (JD62) (Meeussen 1959; Ndayiragije 1999), ChiBemba (M42) (Sharman 1956; Givón 1975), ChiTonga (M64) (Carter 1962; Meeussen 1963), Tswana and Northern Sotho (S31) (Creissels 1996a, 1996b; Zerbian 2006), Zulu (Buell 2005, 2006; Doke 1961; van der Spuy 1993), and Makhuwa (P30) (van der Wal 2009). (See also the papers in van der Wal & Hyman, in press.) As the terms indicate, verbs with conjoint tense allomorphs are in a close syntactic relationship with what follows, while verbs with disjoint tense marking are

Page 9 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu not. Focus—either on the verb or what follows—partially correlates with the distribution of conjoint/disjoint verb morphology. We illustrate this first with the following oft-cited sentences from Chibemba (Sharman 1956: 30). (15)

In (15a) the prefix -la- marks the disjoint form of the present tense on the utterance-final verb, which is clearly in focus, since the question is about smoking, not some other activity. In the answer in (15b), the conjoint form of the present tense lacks this prefix, since the focus is now on ‘cigarettes’ (as opposed to, say, cigars). Example (15c) shows that the disjoint form can be used when the verb is not final, but is included within the scope of focus. The following table of Chibemba verb forms from Hyman and Watters (1984: 251), based on Sharman and Meeussen (1955), Sharman (1956), and Givón (1972), provides an idea of how pervasive conjoint/disjoint allomorphy can be in a Bantu language: As seen in Table 39.1, the disjoint verb often has a fuller marking than the corresponding conjoint form. The progressive forms in Table 39.1 show that the conjoint/disjoint distinction need not involve different TAM prefix marking on the verb, but rather can be strictly tonal. Verb forms marked [+TS] (for ‘tone spreading’) spread their last underlying H tone to the end of the word (Sharman and Meeussen 1955: 395). [±TS] indicates +TS in the disjoint and –TS in the conjoint form. The following shows the [±TS] alternation on the final vowel of the verb -toba ‘break’ in the Past1 tense (Sharman 1956: 40): (16)

In (16a) focus is on úmutóndó ‘the pot’ (contrasting with múnsupa ‘calabash’). In (16b) focus is on breaking the pot; the verb is included in the focus and occurs in the disjoint form. (p. 799)

Page 10 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu Table 39.1 Chibemba conjoint/disjoint forms Non-progressive conjoint Present/

Progressive disjoint

:

-Ø-

-a

-la-

-a-

[±TS]

Past1/ Future1

:

-á-

-a

-áa-

-a

[±TS]

Past2

:

-ácí-

-a

(complex)

Past3

:

-á-

-ile

-álii-

-a

Past4

:

-a-

-ile

-alí-

PresLinge r

:

-Ø-

-ile

PastLinge r

:

-a-

Future2

:

(-ka-

-lée-

-a

[±TS]

hab.

-ácíláa-

[±TS]

[±TS]

-álée-

[±TS]

-ile

[+TS]

-alée-

[±TS]

náa-´

-á

[-TS]

-á

-alí

-a

[+TS]

-a

-ka-

-a)

[+TS]

-kalée-

[±TS]

Page 11 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu We illustrate some of the cross-Bantu range of conjoint/disjoint TAM morphology by comparing Chibemba with Zulu. Zulu only shows a conjoint/disjoint distinction in two tenses, the present and the perfective (Doke 1961). In (17a,b), we see that it is ungrammatical to use the disjoint present tense prefix -ya- if a post-verbal constituent is focused. When the disjoint verb form occurs before post-verbal constituents, as in (17c), it necessarily indicates that the verb is focused: (17)

In contrast with Chibemba, the Zulu disjoint morpheme cannot be considered a focus marker, in spite of examples like (17c). This is illustrated by the Zulu example in (18), where the verb has disjoint morphology (-ya-) because it is an intransitive verb, canonically final in this SV clause, not because it is in focus (Cheng and Downing elicitation notes): (p. 800)

(18)

As work like Buell (2005, 2006), van der Spuy (1993), and Zerbian (2006: 96–104) have argued, the disjoint marker found in Southern Bantu languages like Zulu predictably occurs when the verb is final in a major syntactic constituent (vP/IP/CP), and this

Page 12 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu syntactic position only sometimes correlates with verb focus. (Note in (17c) that object marking on the verb indicates the object is external to vP, so the verb is vP final.)

39.3.2.2 TAMs with intrinsic focus In this section we show that focus-marking can be triggered by specific TAMs and negation. We shall first illustrate this from the Grassfields Bantu language Aghem. Recall from (7a,b) that Aghem exploits the IAV for focus-marking; note in (19a,b) that the position of ‘today’ varies according to its focus status: (19)

Note that when the direct object kɨ́-bɛ́ ‘fufu’ occurs in the IAV in (19a) it appears in its citation form. In (19b), however, the bare noun cannot occur. Instead a ‘dummy determiner’ is required (which, like most other modifiers, conditions the deletion of the prefix on ‘fufu’). Since it would take considerable discussion to distinguish between the following two interpretations (see Hyman 2010; Hyman and Polinsky 2010), let us simply assume that the isolation form kɨ́-bɛ́ ‘fufu’ does not occur in (19b) either because it is ‘out of focus’ or because it is not in the IAV position. Now consider the use of the ‘dummy determiner’with ‘fufu’ in the sentences in (20). (20)

In (20a) another form of the same TAM (hodiernal past tense) has been used, which indicates that the truth value is included within the scope of focus. Perhaps because the auxiliary máà is explicitly focused, the determiner kɔ́ is required. We might first propose that this is because ‘fufu’ is pragmatically out of focus (as the gloss implies). However, in (20b,c) we observe the same marking of the object after a negative or imperative verb. In this case we cannot refer directly to the focus of the utterance: (20c) would be the appropriate response not only to a question ‘what should I do with the fufu?’ (‘EAT (the) fufu!’), but also ‘what should I eat?’ (‘eat FUFU!’) or ‘should I eat rice?’ (‘no, eat FUFU!’), where ‘fufu’ would clearly be in focus (and possibly marked by (p. 801)

Page 13 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu the focus marker nô). Instead, the hypothesis that works is that ‘fufu’ is not in the IAV focus position in (20a–c), even though it occurs right next to the verb: there is a major break between the verb and what follows. However, for this to go through, we have to say that negatives and the imperative require the same disjoint phrase break after the verb as the one we find in (20a), independent of actual focus structure. Indeed, as Hyman and Watters (1984) show, ‘marked’ tense, aspect, mood and polarity (negation) behave as if they are intrinsically focused in many Bantu and other African languages (cf. Marchese 1983 for additional evidence from Kru, and Güldemann 2003 for further discussion). Further evidence that grammatical marking only imperfectly lines up with actual pragmatic focus considerations is seen in Table 39.2, which summarizes the intrinsically focused [+F] vs unfocused [–F] TAMs in Haya (J/E22): The above [±F] marking is required to account for a process of H tone reduction. All of the H tones of a [–F] verb are deleted, when it is followed by a constituent within the same clause, and with which it forms a tonal domain. [+F] verbs are unaffected by the process. H tone reduction is thus observed in the [–F] present habitual in (21a), but not in the [+F] progressive (21b), which is identical except for the addition of the initial (p. 802) ni- prefix (an historical focus marker in fact!). Example (21c) shows that negatives are also intrinsically [+F]: (21)

Table 39.2 Intrinsic focus of TAMs in Haya (cf. Hyman and Watters 1984: 260).

Page 14 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu

(In all three examples /-jun-á/ is realized -jún-a phrase-finally.) Although this may appear to be a tonal reduction conditioned by lack of focus, again it has more to do with generalizations concerning phrasing: a verb’s H tones are deleted only if it forms a tonal domain with what follows, as in (21a). This is what gives the impression that [+F] TAMs have disjoint and [–F] TAMs have conjoint properties. In (22), we summarize the canonical properties of the conjoint/disjoint distinction. As we have seen, not all of them will be present at one time: (22)

We have illustrated all of them in this section so far except for (22f). Many Bantu languages have different TAM allomorphy in main (MC) vs subordinate, for example relative clauses (RC), or in the affirmative vs negative. This is seen in the following Haya data:

Page 15 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu (23)

Since relative clauses are ‘backgrounded’ at the discourse level, essentially identifying referents rather than advancing the story line, they often fail to undergo innovations that occur in main clauses. (See Hyman and Watters (1984) for additional examples and discussion.)

(p. 803)

39.3.3 Prosodic marking of focus

In familiar European languages, IS roles such as focus and topic often directly condition prosodic prominence and related phrasing. (See Gussenhoven 2004; Ladd 2008 and various contributions to this volume for detailed discussion.) As far as we know, focus does not directly condition the analogous use of obligatory prosodic prominence or prosodic restructuring in Bantu or other African languages.6 (See Downing 2012 for a recent survey.) Instead, as we have already seen, prosodic marking of focus in Bantu languages involves the establishment of focus-related phrasal domains that are only indirectly conditioned by focus. A common phenomenon is for the verb to phrase with a following constituent that is in (narrow assertive or contrastive) focus. This was illustrated for Zulu, above, where the evidence for the prosodic phrasing is the distribution of penult lengthening. As Cheng and Downing (2009, 2012) show, the phrasing is syntactically motivated, however, not directly conditioned by focus. In other languages, a Verb + X sequence may form a single prosodic domain within which tones are modified, as we saw, for instance, in Haya in (21a). A very similar example comes from Luganda, where instead of H tone deletion, a [–F] affirmative verb which has a H to L drop in it will lose its L and plateau with the H of the following word (Hyman et al. 1987), as in (24a), where y- marks the subject, -a- is the general past prefix, and temarks negation): (24)

Page 16 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu The [+F] negative verb keeps its H to L drop in (24b). An additional complicating factor is that the post-verbal constituent must not begin with an augment (e.g. the e- in (25a)). (25)

(p. 804)

As was seen in (24a), the lack of an augment corresponds with post-verbal focus,

suggesting a conjoint relationship with the verb, while the presence of the augment in (25a) corresponds with broad focus, suggesting a disjoint relationship with the verb. However, this is not predictive. When the IAV constituent begins with a word that does not take an augment, for example a proper noun as in (25b), L tone deletion and H tone plateauing are automatic and hence independent of focus. It is thus not only a [+F] verb, but also a following word occurring with the augment, that will block H tone plateauing between the verb and what follows. For this and other reasons Hyman and Katamba (1993) conclude that focus does not directly determine the phrasing that is needed for H tone plateauing. The nominal augment in focus-marking has been shown both to be implicated in focusmarking and to affect tonal realizations in other Bantu languages as well. In Makhuwa the first H of a nominal is deleted after a conjoint verb (van der Wal 2009: 128; cf. Stucky 1979: 191): (26)

As seen in (26a), the noun ‘grass’, whose H in its citation form maláshi comes historically from the Proto-Bantu *H augment, becomes all L after a conjoint verb (although another rule assigns it a final H tone if occurring before a pause). No such change occurs in (26b), where the corresponding disjoint verb is not as tightly bound to the object. Van der Wal (2009: 32) clearly shows that it is only nouns that were marked by an historical augment that undergo this change. As currently in Luganda, we can hypothesize that the H tone augment was historically lacking in post-verbal focus constructions such as (26a). Since the augment has since dropped out of the language, the impression one gets is that focus is marked by tone lowering.

Page 17 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu

39.4 Topic marking, topicality, and givenness There has been general linguistic work on topic-marking in Bantu dating back to the 1970s (Givón 1976). Topic, topicality, and givenness, like focus, have been claimed to have an effect on the prosody, morphology, and syntax of many Bantu languages. We discuss each of these in turn, beginning with the morphosyntactic marking of topic.

39.4.1 Morphosyntactic effects of topic marking and topicality 39.4.1.1 Subject as topic While focus is mainly expressed on the verb or by post-verbal position, as shown in the preceding section, preverbal position is often reserved for topics (or ‘out of focus’). The (p. 805) Zulu example in (5), repeated below for convenience, provides an example where the given/topic object amá-thaanga ‘pumpkin’ occurs in preverbal position. (27)

The correlation between preverbal position and topic has been proposed to account for why subjects are canonically preverbal in Bantu languages (see Section 39.2.3, above): subjects are inherently topical (as in SVO languages in general).7 The inherent topicality of subjects is traditionally appealed to in order to account for the fact that, in many Bantu languages, subjects cannot be focused in situ. They must either be clefted or occur in post-verbal position when focused. This phenomenon is discussed in, for example: Bearth (2003); Bokamba (1976); Bresnan and Kanerva (1989); Cheng and Downing (2007); Downing (2012); Maxwell (1981); Mchombo (2004); Moshi (1988); Muriungi (2003); van der Wal (2009); and Zerbian (2006).

Page 18 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu Tumbuka (N20; Downing 2012) provides an example of a language where subjects must be clefted when they have new information focus (i.e. when they occur as WH-questions or -answers): (28)

When a subject has presentational focus, it appears to be common cross-Bantu for it to occur in post-verbal position. As Bresnan and Kanvera’s (1989) study of Chicheŵa locative inversion shows, in this construction a focused subject occurs immediately after the verb, while the preverbal locative triggers subject agreement on the verb (29a). The obligatory (p. 806) focal (or non-given status) of the post-verbal subject is illustrated in (29c), which shows that locative inversion is infelicitous if this sentence is a response to (29b) because the postverbal subject a-lendóo-wo has been previously mentioned in the discourse. (In (29c) the locative answer to the indirect question is focused by means of clefting.)

Page 19 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu (29)

8

Strikingly, in other languages where focused subjects occur in post-verbal position—for example, Dzamba (H10), Rundi (J60), and Rwanda (J60)—even preverbal logical objects can trigger subject agreement on the verb (Bokamba 1976; Ndayiragije 1999; Morimoto 2000). This phenomenon—called subject–object reversal—is illustrated with the Rundi data in (30). In this dataset, the canonical SVO word order is illustrated in (30a). The subject–object reversal order (OVS) is illustrated in (30b). As Ndayiragije (1999) shows, OVS order implies a contrastive focus reading on the post-verbal logical subject (underlined), while the preverbal object is out of focus. Notice that a disjoint marker -raoccurs on the verb in the canonical order (30a); it cannot co-occur with a post-verbal focused element (30b).

Page 20 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu (30)

The attentive reader will have noticed that the examples in this section do not necessarily show that subjects correlate with topics in the information structure sense. They merely show that preverbal position is incompatible with narrow focus on the grammatical subject. While appealing to the inherent topicality of subjects might account for this, Zerbian’s (2006) thoughtful study shows that preverbal subjects in Northern Sotho—a language which also bans in situ focus on subjects—are not required to display defining topic properties, like definiteness or aboutness. This raises the issue of what is meant by inherent topicality, which is addressed in the next section. (p. 807)

39.4.1.2 Topicality hierarchies and object status Another important influence of topicality on the syntax of many Bantu languages is that intrinsically topicality of object NPs correlates with an array of morphosyntactic effects. Intrinsic topicality is defined as having the properties ranked high in the semantic hierarchies in (31):

(31)

These properties help account for asymmetries often found in the morphosyntactic status of the two object NPs in a ditransitive construction. The object NP which is higher on the topicality hierarchy is, first, more likely to be able to become the subject of a passive construction based on a ditransitive. Note that this property is related to the general topicality of subject position discussed in the previous section. It is also more likely to trigger object marking on the verb when the overt object is deleted due to its given discourse status. (More on object marking and given status in Sections 39.4.2 and 39.4.3) And, paradoxically, it is also most likely to be licensed to occur in immediate post-verbal

Page 21 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu (IAV) position in VPs under broad focus. (Recall from Section 39.3.1 that IAV is a privileged position for elements in narrow focus in many Bantu languages.) Hyman and Duranti’s (1982) analysis of ditransitive objects in Sesotho (S30), based on Morolong and Hyman (1977), illustrates the role of the topicality hierarchy—and especially the human factor—in accounting for asymmetries in their morphosyntactic properties. These authors show that the complete analysis of Sesotho objects involves a rather complex interaction between the hierarchies in (31a,b). We can see in (32) that when a non-human circumstantial ‘object’ of an applicative (APP) co-occurs with a human patient ‘object’, it lacks the following properties that the human patient object in the same sentence has. (The human patient baná and its object marker referent are emboldened in the data.) • It cannot occur immediately after the verb (32a). • It cannot be passivized (32b). • It cannot be referred to with an object marker (32c).

Page 22 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu (p. 808)

(32)

The role of such syntacticized hierarchies of intrinsic topics in accounting for object asymmetries has also been demonstrated for Shona (Hawkinson and Hyman 1974), Haya (Duranti and Byarushengo 1977), and other Bantu languages. (See Givón 1988, among many others.)

39.4.2 ‘Dislocation’ of non-focused constituents and its prosodic consequences

Page 23 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu The requirement that topics be preverbal while focused elements occur in IAV position leads to displacement (informally, ‘dislocation’) of non-focused nominals from their canonical positions. The displaced nominals (if they are objects) require resumptive pronominal marking on the verb in many Bantu languages: for example Chicheŵa (Bresnan and Mchombo 1987), Haya (Tenenbaum 1977), Luganda (Hyman and Katamba 2010), Northern Sotho (Zerbian 2006) and Zulu (Cheng and Downing 2009)—see (17), above. While both left and right dislocation are widely attested in Bantu languages, asymmetries in the information structure status of left vs right dislocation has been reported for some languages. For example, to return to the Zulu example in (27), above, we can see that it is felicitous to left dislocate amá-thaanga ‘pumpkins’, the object repeated from the question. However, as Cheng and Downing (2009) show, right dislocating ‘pumpkins’ would be an infelicitous answer. Both contrastive topics and aboutness topics must be preverbal in Zulu. It has been suggested since Byarushengo et al.’s (1976) study of Haya that right dislocations have an afterthought function, rather than a true topic function. Zerbian (2006) finds for Northern Sotho that right dislocations are almost non-existent in her natural language corpus, and proposes that this is due to their information structure status as afterthoughts. However, Tenenbaum (1977) demonstrates that in Haya right dislocations are also used to connote surprise, emphasis, or contrast. She proposes that this follows from the fact that both utterance-initial and utterance-final positions are perceptually salient. The asymmetries in the information structure of left vs right dislocations found in some languages therefore require further explanation. (p. 809)

Left vs right dislocation also has different prosodic consequences in different Bantu languages (Hyman and Byarushengo 1984; Hyman 1999, 2010; Zerbian 2006, 2007; Downing 2011). In languages like Northern Sotho (S30), left dislocations phrase with what follows, while right dislocations phrase separately. (Resumptive object marking on the verb is emboldened):

Page 24 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu (33)

In languages like Luganda, in contrast, left dislocations (and subject NPs) phrase separately, while right-dislocations join in the same phonological phrase with what precedes:

Page 25 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu (p. 810)

(34)

And in languages like Chichewa (N30), both left and right dislocations phrase separately:

Page 26 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu (35)

Downing (2011) suggests that the different phrasings might well provide cues to the different syntactic status of the dislocations (clause-internal vs clause external) in different languages. Clause internal dislocations, for example, could be more likely to phrase with surrounding material. The syntax and the information structure status of left vs right dislocations and their influence on prosodic phrasing clearly require further research.

(p. 811)

39.4.3 Reduction of given information

While given elements can be dislocated, it is also common for them to either be deleted or pronominalized in Bantu languages, as in many other languages. We have seen several examples of resumptive object marking with an overt given NP object above. The examples from Northern Sotho (Zerbian 2006), below, illustrate how answers to questions in natural speech typically delete discourse-old information from the question (36a) or pronominalize arguments which cannot be grammatically deleted (36b):

Page 27 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu (36)

As Zerbian (2006) demonstrates, while the result of these reduction processes, along with dislocation, is to place focused elements in a particular syntactic position in Northern Sotho (in this case, IP/VP-final), crucially, it is not focus which triggers these processes, rather the given status of certain elements. (See Buell (2009) and Cheng and Downing (2009, 2012) for a similar account of IAV ‘position’ in Zulu.)

Page 28 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu

39.5 Further considerations and conclusion In the above sections we have surveyed some of the most prominent structures and issues which arise in the study of information structure in Bantu. Most of our attention (p. 812) has been on the ways in which focus and topic are marked, rather than on their semantics or actual use in discourse. While we have attempted to give a representative sampling of how syntactic, morphological, and phonological markings correlate with differences in information structure, we are only just beginning to understand their full range and nature—including cases which appear to go against the general trends. For example, while Bantu and many other African languages exploit the IAV position to mark focus, Bostoen and Mundeke (2011: 75–6) show that although Mbuun (B87) has the expected unmarked SVO word order, as in (37a), it exploits the Immediately Before Verb (IBV) position for focus-marking, as in the Q/A exchange in (37b,c). (37)

While such use of the IBV for focus frequently correlates with OV, not VO, (37d) shows that topical information occurs initially, as expected. On the other hand, Tunen (A44), one of the rare Bantu OV languages, does not exploit the IBV for focus (Mous 1997). What this means is that much more attention needs to be paid to the syntax of Bantu languages in the northwest zones. Concerning morphological and prosodic marking, we have only scratched the surface in the above presentation. In addition to conjoint/disjoint verb marking and the presence vs absence of the augment on objects, a number of Western Bantu languages have been reported to have what Schadeberg (1986) has called ‘tone cases’. In such languages starting in the north with the B40 languages (Blanchon 1999) and going through the H10 Kikongo group (Daeleman 1983; Blanchon 1998) as far south as (Otji) Herero (R31) (Kavari et al. 2012), nouns take different tonal forms in different contexts, some related to information structure. A particularly rich such system is presented in (38) from (Gi)Phende, from unpublished work by the second author with Mwatha Ngalasso:

Page 29 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu (37)

As seen, the above distinctions go beyond what is traditionally identified as ‘case’. While the correlations are not perfect, case 1 seems to be topic-like and cases 2 and 5 (p. 813) focus-like. Case 4 represents how an object is realized with broad or VP-focus. This raises the broader question of how the various tonally implicated markings are related to each other: conjoint/disjoint verbal morphology, augment nominal morphology, tone cases, and, ultimately, ‘metatony’ (Meeussen 1967: 111), a phenomenon by which a subset of verb forms are realized H before an object, L otherwise, as in the following examples from Songye (L23) (Stappers 1964, cited by Dimmendaal 1995: 32 and Schadeberg 1995: 176): (40)

In some languages the subset of verb forms end High when followed by any element, including a non-object, and hence end with Low tone only when phrase-final, suggesting that the High tone is an IAV focus marker. While conjoint/disjoint verb forms, the nominal augment, and tone cases undoubtedly bear an historical, if not synchronic relationship to information structure, Hyman and Lionnet (2012) show that proposals to relate metatony to focus are fraught with difficulties. Such overextensions are perhaps understandable, given that focus-marking is so extensive in Bantu—and many other African languages. As these final remarks make clear, the overview we have presented in earlier sections certainly underestimates the considerable diversity that still awaits discovery in the many Bantu languages in which information structure and its various markings have not been studied in depth.

Notes: (*) We would like to thank the editors and two anonymous reviewers for their thoughtful comments and suggestions on the original manuscript. (1) Throughout the paper, the first citation of a language indicates the noun class prefix in parenthesis as well as its updated Guthrie referential code (Maho 2009). Each code begins with a letter which indicates the approximate geographical location, e.g. Abo (A42) in the Northwest, Kikongo (H16) in the west, Luganda (J/E15) in the east, and Zulu (S42) in the south. While these codes sometimes suggest approximate subgrouping, for

Page 30 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu our purposes what will be important is to recognize that languages in the northwest zones A–C often diverge typologically from more canonical eastern and southern Bantu, as do some of the languages in zone D. Subsequent references adopt the usual citation form of the language, which may or may not include the prefix, e.g. Luganda vs Zulu. In the morpheme glosses, numbers indicate noun class agreement, following the standard Bantu glossing system. The following abbreviations are used: AUG = augment (a determiner-like morpheme); CL = noun class marker; COP = copula; DJ = disjoint verbal affix; INF = infinitive; LOC = locative; NEG = negative; OBJ = object marker; REL = relative; SUBJ = subject marker; TAM=tense–aspect marker. (2) It is beyond the scope of this chapter to discuss a theoretical syntactic analysis of the data presented. The interested reader should consult Cheng and Downing (2009) and references cited therein. (3) In this chapter we refer to languages which have maintained the traditional agglutinative structure of the verb in (2) as ‘canonical Bantu’. This structure tends to break down in Northwest Bantu (roughly, Cameroon and neighbouring countries) and especially in other Bantoid languages, such as Grassfields Bantu, also spoken in Cameroon. For example, while prefixal object marking on the verb provides cues to information structure in canonical Bantu, as shown in the Zulu example in (5), such marking is absent in much of Northwest Bantu and all of Grassfields Bantu and Bantoid. (4) The Bantu augment (AUG) is an article-like morpheme whose function varies from one language to another. In Luganda its (non-) appearance often correlates with focus structure (see (24) and (25) below). (5) In other Bantu languages, in contrast, a cleft is obligatory for questioning the subject (Downing 2012; Zerbian 2006); see section 4.1.1 for discussion. (6) The only African language, to our knowledge, where it has been claimed that focus directly leads to prosodic restructuring is Chicheŵa (Kanerva 1990; Downing et al. 2004). However, Downing and Pompino-Marschall (2013) demonstrates that there is, in fact, no obligatory focus prosody in Chicheŵa. What had been reported as obligatory focus prosody is instead best interpreted as optional emphasis prosody. (7) See, e.g., Givón (1976), Morimoto (2000), van der Wal (2009) and Zerbian (2006) for discussion. (8) The class 17 prefix ‘ku’ is a locative prefix. The complementizer kuti and the word for ‘where’ are near homonyms.

Laura J. Downing

Page 31 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018

Information Structure in Bantu Laura J. Downing is Professor of African Languages at the Institute for Languages and Literatures, University of Gothenburg, Sweden. Her research is concerned with the phonology of Bantu languages, and in particular with the role of prosody in conditioning or revealing morphosyntactic structure in domains such as prosodic morphology and the phonology-syntax-information structure interface. Larry M. Hyman

Larry M. Hyman is Professor of Linguistics and Executive Director of the FranceBerkeley Fund at the University of California, Berkeley. He has worked extensively on phonological theory, tone, and other aspects of language structure — particularly as concerns the history and description of the Niger-Congo languages of Africa, especially Bantu.

Page 32 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Michigan; date: 25 June 2018