From Cognitive Semantics to Lexical Pragmatics: The Functional Polysemy of Discourse Particles [Reprint 2013 ed.] 9783110828641, 9783110168761

Table of contents :
1. Introduction: The domain
1.1. Aims
1.2. Definition
1.2.1. Semantic properties
1.2.2. Functional properties
1.2.3. Form-related properties
1.3. Corpora
1.3.1. The German corpora
1.3.2. The English corpora
1.4. Methods
1.4.1. Interpretative methods
1.4.2. Quantitative and computational methods
1.4.3. Linguistic models
1.5. The structure of the following
2. Contexts and categories: Functional interpretation
2.1. The functional spectrum of German ja
2.2. Category assignment
2.2.1. The descriptive inventory
2.2.2. Classification in artificial neural networks
2.3. Consequences for lexical representation: Constructions
3. Conceptual background frame: Evidence from extra-linguistic variables
3.1. The variable communication partner
3.2. The variable speaker’s gender
3.2.1. Äh and ähm in human-to-human communication
3.2.2. Gender-related functional shifts in human-computer interaction
3.3. Consequences for lexical representation: Conceptual background frame
3.3.1. The relation lexeme – function
3.3.2. A frame of communicative domains
4. Lexical analysis
4.1. Semantic relations
4.1.1. Translation equivalents
4.1.2. Semantic fields
4.2. Semantic decomposition
4.2.1. Methodological considerations
4.2.2. Semantic tests for discourse particles
4.2.3. English oh
4.2.4. Tests for the features of oh
4.2.5. Further English discourse particles
4.3. Consequences for lexical representation: Invariant meanings
5. Lexical representation
5.1. A unified model of the meanings and functions of discourse particles
5.1.1. The contextual meanings of discourse particles
5.1.2. From contextual meanings to discourse functions
5.1.3. The different word classes
5.1.4. The general function of discourse particles
5.2. Aspects of the lexicon
5.2.1. General properties of linguistic lexica
5.2.2. The structure of lexical entries
5.2.3. Types of lexical information
5.2.4. Linguistic generalisations in ILEX/DATR
5.3. A frame- and construction-based lexicon for discourse particles
6. Conclusion and prospects
6.1. From cognitive semantics to lexical pragmatics
6.2. Automatic processing of discourse particles
References
Appendix A: Questionnaire
Appendix B: DATR Program
Index

Citation preview

From Cognitive Semantics to Lexical Pragmatics

W G DE

From Cognitive Semantics to Lexical Pragmatics The Functional Polysemy of Discourse Particles

by

Kerstin Fischer

Mouton de Gruyter Berlin · New York 2000

Mouton de Gruyter (formerly Mouton, The Hague) is a Division of Walter de Gruyter GmbH & Co. KG, Berlin.

Gedruckt mit Unterstützung der Deutschen Forschungsgemeinschaft. D 361

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

Library of Congress Cataloging-in-Publication Data Fischer, Kerstin, 1966— From cognitive semantics to lexical pragmatics : the functional polysemy of discourse particles / by Kerstin Fischer, p. cm. Includes bibliographical references and index. ISBN 3-11-016876-6 1. Discourse analysis. 2. Polysemy. 3. Pragmatics. I. Title. P302.F548 2000 401'.41-dc21 00-061631

Die Deutsche Bibliothek -

CIP-Einheitsaufnahme

Fischer, Kerstin From cognitive semantics to lexical pragmatics : the functional polysemy of discourse particles / by Kerstin Fischer. - Berlin ; New York : Mouton de Gruyter, 2000 Zugl.: Bielefeld, Univ., Diss., 1998 ISBN 3-11-016876-6

© Copyright 2000 by Walter de Gruyter GmbH & Co. KG, D-10785 Berlin All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Printing: Hubert & Co, Göttingen. Binding: Lüderitz & Bauer GmbH, Berlin. Printed in Germany.

Acknowledgements The main part of this work has been carried out within the graduate program 'task-oriented communication' at the University of Bielefeld, financially supported by the Deutsche Forschungsgemeinschaft, which is gratefully acknowledged. There are many people who contributed to this study, in one way or another. Most importantly, I want to thank Andreas Dommes, my friends and my family for being with me during the hard times. Then there are my academic supervisors, Prof. Dr. Elisabeth Gülich, Prof. Dr. Hans Strohner, and Prof. Dr. Dafydd Gibbon. Furthermore, I got much inspiration and encouragement from Prof. Charles Fillmore and Prof. Paul Kay during my stay at UC Berkeley (made possible by a research grant from the German Academic Exchange Service (DAAD)) and afterwards. I am also very grateful for the exciting discussions with Dr. Ulrich Dausenschön-Gay and Dr. Ulrich Krafft, and for Prof. Dr. Helge Ritter's advice on artificial neural network classifiers. Finally, I would like to thank my colleagues in Bielefeld, Berkeley and the Natural Language Systems group in Hamburg for their moral, technical, and academic support, and Dr. Christie Manning and Dr. Julie Berndsen for correcting my worst grammatical errors; however, I am solely responsible for any errors and inconsistencies that may remain in the text.

Contents

1

Introduction: The domain

1

1.1

Aims

1

1.2

Definition

13

1.2.1

Semantic properties

16

1.2.2

Functional properties

18

1.2.3

Form-related properties

23

1.3

1.4

1.5 2

Corpora

27

1.3.1

The German corpora

29

1.3.2

The English corpora

34

Methods

35

1.4.1

Interpretative methods

36

1.4.2

Quantitative and computational methods . .

55

1.4.3

Linguistic models

58

The structure of the following

60

Contexts and categories: Functional interpretation

65

2.1

The functional spectrum of German ja

68

2.2

Category assignment

97

2.2.1

99

The descriptive inventory

vili

Contents

2.2.2 2.3

3

Consequences for lexical representation: Constructions

110 120

Conceptual background frame: Evidence from extra-linguistic variables

125

3.1

The variable communication partner

126

3.2

The variable speaker's gender 3.2.1 Äh and ähm in human-to-human communication

157

3.2.2 3.3

4

Classification in artificial neural networks .

Gender-related functional shifts in humancomputer interaction

157 166

Consequences for lexical representation : Conceptual background frame

175

3.3.1

The relation lexeme - function

176

3.3.2

A frame of communicative domains . . . .

178

Lexical analysis

195

4.1

Semantic relations

197

4.1.1

Translation equivalents

199

4.1.2

Semantic

4.2

4.3

fields

207

Semantic decomposition

219

4.2.1

Methodological considerations

219

4.2.2

Semantic tests for discourse particles . . .

223

4.2.3

English oh

226

4.2.4

Tests for the features of oh

232

4.2.5

Further English discourse particles

239

Consequences for lexical representation: Invariant meanings

258

Contents

5

Lexical representation 5.1

A unified model of the meanings and functions of discourse particles 5.1.1

5.3

6

261 261

The contextual meanings of discourse particles

262

From contextual meanings to discourse functions

270

5.1.3

The different word classes

275

5.1.4

The general function of discourse particles

277

5.1.2

5.2

ix

Aspects of the lexicon

284

5.2.1

General properties of linguistic léxica . . .

285

5.2.2

The structure of lexical entries

289

5.2.3

Types of lexical information

292

5.2.4

Linguistic generalisations in ILEX/DATR .

296

A frame- and construction-based lexicon for discourse particles

300

Conclusion and prospects

321

6.1

From cognitive semantics to lexical pragmatics . .

321

6.2

Automatic processing of discourse particles . . . .

323

References

327

Appendix A: Questionnaire

357

Appendix B: DATR Program

359

Index

367

Chapter 1 Introduction: The domain 1.1

Aims

This study concerns English and German discourse particles, small items such as German ja, also, ne, oh or ach and English yes, yeah, oh or well which predominantly occur in spontaneous spoken language. Discourse particles are "grammatically peripheral" (Fraser 1990: 391), that is, they do not enter any grammatical relationships with other parts of utterances, and they may fulfil such a broad range of functions that Hentschel and Weydt (1989) suggest the context-dependency of their meanings to be their most prominent feature, thus defining discourse particles as essentially syncategorematic. The current investigation addresses the problem of polysemy, "the occurrence of more-or-less discrete and more-or-less unitary bundles of semantic properties associated with particular word forms" (Cruse 1992: 2); since the most important contribution of discourse particles is in the pragmatic domain, particularly their functional polysemy, that is, the occurrence of more-or-less discrete and more-or-less unitary bundles of functional properties associated with particular word forms, will be investigated. In other words, this study attempts to account for the fact that a particular discourse particle lexeme may get different interpretations which are perceived as related in some way. Consider the following examples:1 The examples are from the Verbmobil corpus described in section 1.3.

2

Introduction: The domain

(1)

13BAR: what about the 18th of December? 14RIC: yeah, yeah, that work.

(2)

124ENG: so that won't work either. 125UMI: yeah, that's not good.

(3)

1UMI: yeah, we've got to get together and discuss Stufe A für die Studienordnung.2

(4)

3RIC : I'm Rie and I am what do I do? (whispering) software . yeah, I'm working for a software account.

The function of yeah in the first example is to accept the proposal the communication partner has made, it functions as a feedback signal in the second example. In example (3), the function of yeah is to introduce a new topic, occurring in the first turn in the dialogue which refers to the common task to schedule an appointment. In example (4), it functions as a repair marker, reorganising the speaker's utterance after he was reminded of the identity assigned to him for the purpose of the recording (see the description of the corpora used in section 1.3). As the four examples show, yeah may fulfil at least four different functions. Thus, the questions that need to be answered in this investigation are the following: - What is the relationship between a discourse particle lexeme such as yeah and its function as a feedback signal, an answer particle, a topic signal and a repair marker? - Are the different readings of such a lexeme somehow related, i.e. is there some general mechanism behind its functional spectrum, or are the possible interpretations completely independent of each other? - Is there an invariant component in all of the occurrences of a discourse particle lexeme? 2

The speakers in these dialogues are native speakers of English who live in Germany.

Aims

3

- Can each lexeme fulfil an endless range of functions or is there a systematic restriction to its functional spectrum? - What is the relationship between structural properties like the position in which a discourse particle token occurs and its interpretation? - Is there a general mechanism for the interpretation of all discourse particles? Based on such a general mechanism is it possible to find criteria for a definition of the class? - How is it possible that lexical items which function as discourse particles can often function in other word classes as well? The goal of this investigation is to find a general systematic model of the polysemy of discourse particles, providing answers to the above questions and explaining not only how particular lexemes get their functional interpretations in particular contexts but also what the essential properties of the word class of discourse particles are and how this word class is related to other word classes. So far such a model of the polysemy of discourse particles does not exist; Abraham (1991) criticises that all "descriptions given so far have, almost without exception, resulted in multiple meaning distinctions represented by one single phonetic form, without ever accounting for a common core meaning and the conditions under which the variant meanings come to hold" (Abraham 1991: 203). Hentschel and Weydt (1989) describe the current research situation as suffering from the so-called "particle paradox:" On the one hand there are approaches which provide detailed studies of the individual functions which discourse particles can fulfil, without being able to explain how a particular discourse particle gets its different interpretations, how these readings are related, and why it fulfils just exactly these pragmatic functions and not others. Most of these studies just list the different functions (for instance, Wolski 1986); this approach is also referred to as the maximalist approach (Mosegaard Hansen 1998: 239). On the other hand there are analyses which try to isolate what is common to the readings of a certain lexeme, thus identifying an invariant component for each discourse particle. This perspective has been called the minimalist ap-

4

Introduction: The domain

proach (Mosegaard Hansen 1998: 240). These approaches leave open how the abstract kernel meaning relates to the observable functional interpretations. Furthermore, most studies are restricted to a particular range of functions and thus their complete functional spectrum does not become apparent.3 Very few are concerned with a general mechanism by means of which the discourse particle lexemes are related to their complete range of functions. These studies, among them most influentially Schiffrin (1987), but also Östman (1983), Mosegaard Hansen (1998) Ehlich (1986), and Schourup (1983)4 are however not enirely satisfying. The former studis explain the functional polysemy of discourse particles by means of relations to different aspects of conversation, or "planes of talk" (Schiffrin 1987). Östman (1983) and Mosegaard Hansen (1998) both use only three such aspects and can therefore distinguish only three different functions of discourse particles, and in Schiffrin's model the relationship between the discourse particle lexemes and the "planes of talk" is unclear (see also section 5.1 and Redeker (1991) for a detailed analysis). The latter two approaches attempt to identify a general function of discourse particles from which their other functions can be inferred. However, the fact that they arrive Examples are, for instance, the analysis of interjections as expressions of emotions which neglects their textual functions (e.g. Angermeyer 1979); the very detailed study by Willkop (1988) which is restricted to functions with respect to the speaker-hearer-exchange system and the argumentation structure; or Maynard (1993) whose study focusses on aspects of subjectivity and emotionality. Jucker and Ziv (1998) write in their introduction to a collection of papers on discourse markers: "the first three papers ( . . . ) focus on text-structure signalling, the next set of papers ( . . . ) concentrate on cognitive aspects, and the remaining four papers ( . . . ) analyse contrastive markers, which display a range of attitudinal, cognitive and interactional properties, thus obviating the inherent problem of functional-domain specificity as criterial in the analysis" (Jucker and Ziv 1998: 5). Thus, even in this new book on the theory and description of discourse markers most studies are restricted to a particular functional domain. These approaches are discussed in detail in section 5.1 when the model proposed in this investigation has been presented.

Aims

5

at completely different basic functions casts doubt on the plausibility of the relationship proposed. Consequently, so far no unified account of the range of the meanings and functions of discourse particles has been presented; the aim is therefore to develop a lexical representation for discourse particles which shows that there is a single mechanism which explains their functional polysemy and therefore also the characteristics of the word class. The problem just identified for the description of the word class under consideration is however a general problem; what has been labelled the "particle paradox" holds for other word classes as well. The concept of polysemy, in contrast to homonymy, implies that the different senses of a single word form display a semantic relationship (Lyons 1977).5 The task is not just to match a number of word forms with a list of possible functions but to ask whether it is possible to get beyond simple enumeration, as Pustejovsky (1995) calls it. Therefore not only the meaning spectrum of each lexeme but also the conditioning factors which determine its variation must be analysed. The analysis thus needs to focus on the conditions under which a lexeme may get a certain interpretation, and on how these factors interact in order to provide a model of the interpretation of occurrences of the respective lexical item. For other word classes, a number of accounts of the relations between the meanings associated with a certain lexeme have been proposed. Lyons (1977), for instance, discusses two ways of characterizing the relatedness of different word senses: historical development and shared semantic properties. With respect to both criteria he argues that they do not allow a categorical evaluation of relatedness since either may apply to different degrees. With respect to historical relatedness the question is how far back the analysis may go while still 5

The terms homonymy and polysemy furthermore both suggest that there are form-related properties which are constant while the functional or semantic features vary. Discourse particles, however, are extremely variable in their phonological and prosodie realisation, and their interpretation depends on the structural contexts in which they occur. In how far discourse particles can therefore be regarded as being formally stable and which realisations constitute a single lemma has to be considered in the investigation.

6

Introduction: The domain

yielding plausible results, that is, results which are in accordance with the intuitive judgement of relatedness. Concerning the sharedness of semantic features the problem is likewise to identify the kind and number of shared properties necessary for the judgement of two meanings as being similar. Thus, the closeness of senses basically remains a matter of intuitive decisions. More recently, further concepts to account for polysemy have been developed. Pustejovsky and Anick (1988), Pustejovsky (1991, 1995) propose a systematic relationship between senses based on the different aspects of the qualia structure, the semantic properties of nouns. For instance, the qualia structure may account for the opposition between fast typist vs. fast driver, and for the event structure of verbs (for example, bake a cake vs. bake a potato). The different aspects of the semantic structure are incorporated into the semantic interpretation of larger structures by means of rules of composition, including cocomposition and type coercion (Pustejovsky 1991: 437). The features employed in the description are not necessarily meant to be cognitively relevant but are assumed if needed for semantic composition. The study of polysemy is also a central concern in cognitive linguistics, a number of different approaches to language which share "the cognitive commitment" (Taylor 1995: 4), the assumption that "language is a mental, i.e. cognitive phenomenon" (Taylor 1995: 4). This commitment does not imply a particular research strategy itself, and so different approaches can be subsumed under the term; for instance, in two-level semantics (e.g. Bierwisch 1983, Bierwisch and Schreuder 1992, Bierwisch and Lang 1989), which shares the cognitive commitment, an inventory of functions at the conceptual level is responsible for systematic polysemy. For example, words like school or university may mean the building, the institution, an ensemble of processes, or the institution as a principle (Bierwisch 1983: 81). Depending on the context, the polysemy is determined by a general conceptual function applied to the abstract semantic meaning of each lexeme, yielding the concrete reading, i.e. school as an institution or as a building. The respective lexical item is seen as under-specified and unambiguous. Polysemy in this approach, which distinguishes sharply between semantic and encyclopedic knowledge, the two levels, is thus a matter of world knowledge.

Aims

7

An alternative cognitive linguistic research direction is the wholistic, non-modular, content-oriented approach (cf. Gibbs 1996) advocated by, for instance, Lakoff (1987), Langacker (1991), Fillmore (1982). This approach will be referred to as cognitive semantics in the following. Initiated by findings from Rosch (1975), Rosch and Mervis (1975), Rosch et al. (1976), which indicate that natural language categories are not always based on necessary and sufficient criteria, the notion of prototype was developed to account for the relationship between word senses (e.g. Coleman and Kay 1981, Wierzbicka 1989, Geeraerts et al. 1984). This distinguishes between the core and the periphery of concepts and may result in word senses which do not share any essential properties at all (Lakoff 1987: 95). In network models (Langacker 1988, Norvig and Lakoff 1987), a central meaning for each word can be identified to which other senses are related; these relations belong to an inventory of cognitively relevant operations, such as metaphor, metonymy, profiling, etc. (Norvig and Lakoff 1987: 204). Lexical items thus exhibit radial structures (see also Lakoff 1987). In this variant of cognitive linguistics, word senses are thus not regarded to be similar because there are objective similarities between the objects denoted, but because of underlying conceptual structures, particularly métonymie and metaphorical relations (Lakoff and Johnson 1980), which provide the links between the different readings (e.g. Lakoff 1987, Sweetser 1990: 5). Thus, "word meaning is not necessarily a group of objectively "same" events or entities; it is a group of events or entities which our cognitive system links in appropriate ways" (Sweetser 1990: 9). In Sweetser (1990), metaphorical mapping was furthermore worked out to account for the polysemy and the historical development of English modals, conjunctions, and conditionals; by means of reference to three different conceptual domains, the polysemy of the three different kinds of linguistic phenomena can be explained. Another cognitive semantic approach is presented by Fillmore and Atkins (1992) according to whom polysemy is constituted by two different concepts and their interaction: "Frame semantics makes it possible to separate the notion of the conceptual underpinnings of a concept from the precise way in which the words anchored in them get

8

Introduction: The domain

used" (Fillmore and Atkins 1992: 101). They argue for a description against a structured experiential background which constitutes a kind of conceptual prerequisite for understanding (cf. also Fillmore 1975, 1982, 1994), and they consider the grammatical patterns of the particular item as a determining factor such that "the interrelations between two notions: semantic frame and syntax" (Fillmore and Atkins 1992: 101) must be specified to account for the relations between the senses of a lexical item. Frame semantics consequently allows the description of the interaction of the lexeme, syntax and a conceptual background frame. Which model of polysemy is suitable for the description of the multifunctionality of discourse particles depends on what informational resources are found to condition the interpretation of their occurrences and how they interact. In this investigation, it will be argued that a cognitive semantic viewpoint is the best starting point. While not drawing on a priori distinctions between semantic and encyclopedic knowledge, as two-level semantics demands, cognitive semantic concepts such as metaphorical mapping and the reference to a conceptual background frame can explain the relationship between those factors which condition the interpretation of discourse particle occurrences. Furthermore, the inclusion of syntax in a model of polysemy, as in frame semantics, allows to consider the structural properties of discourse particles. This accounts for the fact that different discourse particles may fulfil similar functions, on the one side, and restricts the generative component of the model to actual, lexicalized meanings on the other. Thus, the functional polysemy of discourse particles can be described by means of the interaction of their contextual properties and a conceptual background structure that is constituted by aspects of the communicative situation to which speakers attend regarding their communication partners. As a means of associating a discourse particle lexeme with the conceptual frame, the cognitive semantic concept of metaphorical mapping between domains, as developed in Sweetser (1990), can be used to explain the reference of discourse particles to the background structure. The development of such an approach to the functional polysemy of discourse particles demands not only that the conditioning factors which are involved in the interpretation of dis-

Aims

9

course particle occurrences and their interaction are determined, it furthermore requires a device, such as the invariant contribution of the respective lexeme, which allows to show why one lexical item may fulfil a certain function and not another. It also needs to show that speakers really attend to the conceptual background frame proposed, and it has to explain why just these particular domains are involved as opposed to some others, as well as to account for how these types of information, which the hearer may use to interpret an occurrence of a discourse particle, interact with the distributional patterning and the surface features of the lexical item under consideration. For a lexical pragmatic account of the functional polysemy of discourse particles it therefore has to be determined: - what the functional spectrum of a discourse particle is and what the structural contexts are in which it may occur (chapter 2); - which domains determine the functional polysemy of the class (chapter 3); - what the contribution of each lexeme is (chapter 4); - how the different types of information interact and how the interaction can be formalised in a lexical representation (chapter 5). An open question is thereby also the methodology for obtaining the information on the three interacting resources which condition the interpretation. Rather than adopting a particular linguistic methodology, this investigation will take a problem-oriented approach. It will propose methods to solve the problems occurring depending on the particular requirements for an explanatory account of the meanings and functions of discourse particles. So while the model to be developed will be based on concepts developed in cognitive semantics, a number of methodological questions will need to be addressed in the analysis of discourse particles and those factors which influence their interpretation. For instance, frame semantics has so far been predominantly employed for the description of nouns, verbs, and adjectives (Baker et al. 1998: 86). Thus far it is not clear:

10

Introduction: The domain

- how the different readings of discourse particles can be distinguished; - what may constitute a frame for the interpretation of discourse particles and how it can be identified; - how the invariant contribution of discourse particle lexemes can be analysed. In this investigation methods will be proposed that can provide solutions to the methodological problems occurring, however, without comparing the methods chosen to other methods in all possible detail. Thus, this investigation can exemplify a number of different methods as solutions to particular problems, however, it cannot discuss all possible alternatives. The aim is instead to develop a methodologically sound model of the functional polysemy of discourse particles which is based on cognitive semantics concepts. In particular, it builds on a cognitive background structure to which the meanings of discourse particles refer such that the interpretation of discourse particles in context is guided not only by their structural properties, but also by a group of entities which are linked by our conceptual system in an appropriate way (Sweetser 1990: 9). Since this investigation aims at the lexical representation of a number of linguistic items, the perspective is furthermore semasiological. This study involves the lexical representation of the pragmatic behaviour of these lexical items in so far as it accounts for their use in discourse on the basis of a number of partly new and partly well-tested methods. Consequently, in contrast to previous analyses in the area of lexical pragmatics (e.g. Mercer 1992, Blutner et al. 1996, Lascarides and Copestake 1995) which are primarily concerned with contextdependent and defeasible propositional information, the lexical pragmatic approach taken here concerns the distribution and the functions of the items under consideration with respect to pragmatic domains such as the structure of discourse, face work, or the management of speech (see also Levinson 1983: pp.47-53) in which the functions of discourse particles are located. The object domain is thus essentially pragmatics while the concepts and methods to be employed will be

Aims

11

largely drawn from studies in cognitive semantics. One of the main points of this investigation is thus that if the concepts developed in cognitive semantics are applied to questions of lexical pragmatics, that is, of the functional variation of certain lexical classes, a descriptively adequate model of the functional polysemy of discourse particles can be developed. Such a model is desirable as there is so far no unified mechanism to account for the broad range of functions discourse particles fulfil has been proposed. In spite of the many interesting properties of discourse particles which have been discovered, researchers have so far failed to provide a unified description of all of their functions. Thus there is no comprehensive definition of the word class. It is also desirable because discourse particles display a quantitatively prominent status in spoken language dialogues. For example, in the corpora from the toy-airplane construction domain (Sagerer et al. 1994, Brindöpke et al. 1995), the proportion of discourse particles ranges between 3.8% in simulated human-to-machine communication and 9.8% in informal human-tohuman communication. In particular the proportion of discourse particles of the 150 most common words is impressive (Fischer and Johanntokrax 1995: 6): Even in human-to-machine communication, the proportion of discourse particles amounts to 6.6% with respect to the 150 most frequent words. Rudolph finds particles to constitute even 23.8% of the total number of words in her corpora of German conversation, including however different types of particles, such as modal, scalar, focus, and discourse particles (Rudolph 1991: 208). As long as there is no explanation of the functions of such particles in spoken language dialogues, we are lacking insight into almost 10% of speakers' linguistic efforts. Cognitive semantic approaches have so far focussed mainly on the relationship between linguistic structures and their cognitive motivation (see also Fischer 1999); the linguistic units under consideration were thereby largely abstracted from particular usage events. The motivation for this focus has been that while language use has always been addressed under a functional perspective, linguistic structures were for a long time considered as independent of other cognitive processes. The achievement of cognitive semantic and cognitive grammar approaches to language is that they show that linguistic struc-

12

Introduction: The domain

ture is also deeply related to general cognition. Applying a cognitive semantic perspective on linguistic items such as discourse particles, which have their functions primarily in the pragmatic domain, bridges the gap between functional considerations developed in the analysis of talk-in-interaction and the perspective on linguistic structures developed in cognitive aproaches to language. The result will be a lexical model which represents the conventional aspects of discourse particles and which can explain how these motivate the functions discourse particles may fulfil and how lexical and functional aspects interact. Thus, the model accounts for the relationship between the structured inventory of conventionalised linguistic units and their usage events. The model to be developed will therefore provide a missing link between cognitive semantic and functional approaches to language. The unified model to be developed will have to serve not only to account for the broad range of functions discourse particles may fulfil; the functional variability of these items makes it also difficult to distinguish them from other word classes. The investigation of their functional polysemy therefore also concerns the definition of the class of discourse particles and its boundaries. Regarding a definition of the linguistic domain under consideration, in the following, proposals for defining discourse particles on the basis of semantic, functional, and formal criteria will be discussed. The first step taken is a discussion of the properties of discourse particles proposed in the literature in order to develop a preliminary definition of the domain under consideration (section 1.2). As a precondition to concrete analyses, the data and the methods employed in the study will then be considered. It will be argued that a variety of methods are necessary to account for the plurality of functions and the methodological problems involved in their analysis, as well as the use of a number of different large corpora of spontaneous spoken language dialogues. Section 1.3 thus concerns the corpora used in the investigation while in section 1.4 the methods employed in the analysis of discourse particles are considered. The discussion of the motivations for this study and the methods to be employed will allow a further, more detailed outline of the structure of this investigation.

Definition

1.2

13

Definition

It is notorious in the literature on discourse particles, discourse markers, interjections, hedges, connectors, segmentation markers, modal particles, feedback signals, cue phrases, filled pauses, etc. that the scope of every investigation has to be defined anew (cf. also Fraser 1999). The great number of different descriptive terms for this heterogeneous group indicates that firstly there is no single accepted word class definition, and that secondly the terms chosen depend very much on the perspective under which discourse particles are studied. Thus, the term connector focusses on the connecting function of discourse particles while cue phrase indicates that discourse particles function in the speaker-hearer interaction system. Table 1.16 shows the variability of the labels by different authors for the same English discourse particle lexemes, table 1.27 shows the same regarding German discourse particles. However, the fact that the same lexemes are labelled with different terms by different authors points to the fact that the items under consideration actually constitute a more homogeneous group than suggested by the number of different descriptive labels. Overviews of the domain, for instance, Keil (1994) and Willkop (1988); Schiffrin (1987) and Fraser (1988, 1990, 1999), describe the difficulties of drawing concrete lines between postulated word classes. To be able to discuss the different proposals in the literature on discourse particles for distinguishing the different word classes, the distinctions implied by the different descriptive terms will be provisionally disregarded and (basically following Willkop 1988: 59) a tentative terminology will be used to refer to the items under consideration, irrespective of the terminology employed by the authors to be discussed. Anticipating the results of this investigation, those lexemes will be re6

dm = discourse marker, dp = discourse particle, pause = pause marker, int = interjection, cue = cue phrase, parallel = parallel marker, * = does not apply.

7

DP = Diskurspartikel, GP = Gliederungspartikel, MP = Modalpartikel, Int = Interjektion, Segm = Segmentierungsmarker, Antw = Antwortpartikel, * = does not apply, jal and ja2 being the discourse and the modal particle respectively.

14

Introduction: The domain Table 1.1: Terms for English Discourse Particles by Different Authors

Schiffrin '87 Fraser '88 James '74 Hirschberg and Litman '93 Schourup '83 Wierzbicka '91

well dm dm int

oh dm int int

pause int

cue dp

*

*

dp int

dp int

*

ah *

now dm dm *

uh/um *

pause int

I mean dm parallel

but dm dm

*

*

cue dp

*

*

*

dp

dp

*

*

*

*

*

Table 1.2: Terms for German Discourse Particles by Different Authors

Ehlich '86 Willkop '88 Keil '94 Abraham '91 Rasoloson '94

jal

ja2

#

*

GP Antw

MP MP DP

* *

*

oh Int Int Int

ach Int GP Int

äh/ähm Int GP Int

also * GP Segm

mhm Int GP Antw

*

*

*

*

*

Int

*

Int

*

Int

garded as discourse particles that have under-specified meanings reporting on mental processes that are specified by means of reference to particular aspects of the communicative situation. Their general function is to mark an utterance as non-initial. Thus the items under consideration are indeed regarded as a homogeneous class with a single pragmatic function, discriminable by their semantic content and by the types of objects to which their indexical elements refer. Discourse particle will thus be regarded as the hyperonym of segmentation marker, interjection, and hesitation marker. Discourse particles will be considered as interjections when they signal the spontaneous expression of a cognitive state. As a starting point, some ostensive definitions are offered: English oh, ah and German oh, ach will be regarded as interjections, while lexemes that lack the spontaneous character of interjections, for example English well, yes, okay and German ja, also,

Definition

15

gut, hm will be counted as segmentation markers. Hesitation markers, for instance German ahm and äh or English uh and um, signalling ongoing thinking, will also be considered part of the class of discourse particles. The term hesitation marker is thus a co-hyponym of segmentation marker and interjection. The term modal particle will refer to those items which occur utterance-mediàlly and which are integrated in the phrasal intonation contour. The term discourse particle will be used because of its neutrality: Discourse particles occur in discourse and are small independent words (Zwicky 1985). Discourse marker, for instance, suggests that the respective items do not function in their own right but stand for something else, an assumption which will have to be investigated and which therefore should not be presupposed. The classification proposed and to be motivated during this investigation is shown in figure 1.1. particle

Figure 1.1: Taxonomy of Descriptive Terms

In the following an attempt will be made to find criteria for distinguishing these terms and for defining the nature of discourse particles on the basis of different proposals from the literature. In general, there are three different possibilities:8 the items involved may have different semantic features; they may be functionally different; or they may have different formal or distributional properties. In the following, these three options will be discussed. 8

A comprehensive overview on the literature on discourse particles provide Weydt and Ehlers (1987) and, more recently, Pons Borderia (1998).

16

Introduction: The domain

1.2.1

Semantic properties

Most discourse particles contribute only weakly if at all to the prepositional content of utterances. They are hardly needed in determining the information about the world the speaker conveys to the hearer. In particular computational linguistics and automatic speech processing, both based on formal theories of language, focus on the detection and analysis of the prepositional information of utterances. The contribution of discourse particles to this task is not obvious. For instance, the German discourse particle ja can function as an answer signal and may thus contribute prepositional information: cdl_g072a_TIS013: "ja, paßt mir wunderbar." However, we can also find ja in connection with the rejection of a proposal: cd3_m024n_EVB010: "ja, da kann ich nicht." If the description provides an entry for ja with agreement as its meaning, the analysis for the second example will result in a contradiction; the meaning seems to come from the context, not from the lexeme. Hentschel and Weydt therefore propose their syncategorematic content, i.e. a non-lexical, context-dependent meaning, as a semantic criterion to distinguish particles from other words, including interjections (Hentschel and Weydt 1989: 6).9 However, this is a negative criterion for the definition, dependent also on the respective semantic theory and the view of what constitutes lexical information according to which this definition is made. A content-related criterion proposed to distinguish interjections from other words is their emotionality: "words whose only function is to express emotion" (Leech and Svartvik 1975: 134), "somewhat primitive expressions of feeling, only loosely integrated into the lin9

Hentschel and Weydt (1989) hold the distinction between particles and interjections to be most problematic: "Zu den wichtigsten und ungelösten Problemen gehört die Abgrenzung der Partikeln gegenüber den Interjektionen" (Hentschel and Weydt 1989: 17). On the basis of the criterion proposed, interjections should be autosemantica since they denote aspects of reality, such as sounds or emotions (Hentschel and Weydt 1989: 7). However, the authors feel the need to postulate a continuous region between interjections and particles to account for items like na ja, ach so, tja or tschüß (Hentschel and Weydt 1989: 17) which, in their view and in contrast to 'normal' particles, display some kind of semantic content.

Definition

17

guistic system" (Leech et al. 1982). For German interjections, Angermeyer (1979) argues that an interjection is an "Ausrufe-, Ausdrucks-, Empfindungswort" [word of exclamation, expression, emotional involvement] (Angermeyer 1979: 40), "wichtiger Bestandteil unseres sprachlichen Umgangs mit Tieren" [important part of our way to deal with animals] (Angermeyer 1979: 41). The classification he proposes distinguishes disagreeing from agreeing inteijections; a more finegrained categorisation concerns the emotions they express (Angermeyer 1979: 46). Consequently, the emotional content is proposed to constitute the basic trait of inteijections. Gelhaus (1995) writes in the DUDEN grammar that the function of interjections is "Interesse beim Hörer für die Gefühlslage des Sprechers oder für die geschilderte Situation zu wecken" (Gelhaus 1995: 374). They are denied a lexical meaning ("jedenfalls bedeutungsarm" (Gelhaus 1995: 369)), and a grammatical status.10 The emotionality of inteijections is however not a reliable criterion to distinguish them from segmentation or hesitation markers; not all inteijections have an emotional component but can refer to other cognitive states as well (Wierzbicka 1992b), and many segmentation markers also refer to the speaker's mental states, for instance, now: "it is tied to the speaker's now of actual utterance and it indicates but does not specify current covert thinking on the part of the speaker" (Schourup 1983: 105). Wilkins (1992), basing his assumptions on the decomposition of the meanings of inteijections according to Wierzbicka (1986, 1991, 1992b), argues that the semantics of interjections includes components such as I, you, here and now. Interjections are therefore considered as deictic items which refer to aspects of the communicative situation. The main property of interjections in this view is thus to contain indexical elements in their semantic structure by means of which utterances are anchored in the speech situation. This view is also held by Schiffrin (1987) and Östman (1983) regarding interjections, segmentation markers, and pragmatic routines (here: you know). Thus, this In contrast to this, James (1972, 1973, 1974), for instance, argues for a competence-based use of interjections which, according to her, constitute an independent category in generative grammar.

18

Introduction: The domain

property is not a criterion by means of which interjections could be distinguished from segmentation markers or speech routines. Furthermore, an indexical meaning has also been proposed for other word classes, for instance, for conjunctions and modal particles (Diewald and Fischer 1998). It seems that there is no single semantic property by means of which discourse particles or one of their subclasses could be defined. 1.2.2

Functional

properties

Depending on the perspective of the investigation as well as on the background of the investigator, researchers have determined a number of different functions the items under consideration fulfil in spontaneous spoken language dialogues. However, these functions usually have been found to apply not only to a single word class but also to items from at least one of the other classes under consideration. Therefore, neither the subclasses postulated nor the class of discourse particles itself can be defined on the basis of the functional criteria, as the following overview of the functions proposed for discourse particles can show: Gülich (1970) was the first to propose several main functions of (French) discourse particles, including interjections, conjunctions, adjectives, adverbs, and certain verb forms, all of which share the distributional property of occurring at text segmental boundaries (Gülich 1970: 9). These functions are to support the construction of discourse structure, to provide an orientation concerning the content and the structure of the conversation, and to provide help in the formulation process. She distinguishes narrative from dialogical texts; in the former, discourse particles fulfil mainly segmentation functions, marking the narrative structure (Gülich 1970: 54); in the latter, discourse particles fulfil the following functions: as opening signals they support addressing the hearer (Gülich 1970: 89), taking the turn (Gülich 1970: 100),11 and introducing direct speech (Gülich 1970: 101); furthermore, 11

The term turn is taken here to refer to "the continuous period of time during which a person is talking" (Oreström 1983: 23). Yet, as Oreström notices, "[t]here is unanimous agreement, however, that certain brief, spontaneous

Definition

19

inside turns they help to bring up a new topic or to stress information (Gülich 1970: 197), as well as help to deal with repairs and interruptions (Giilich 1970: 164). Finally, as closing signals, they create a relation to the hearer as post-determining question particles (Giilich 1970: 228-229). Gülich (1970) concludes that discourse particles fulfil the same functions in spoken language as punctuation marks do in written language and that they are only identifiable as a class if the object of study is the dialogue instead of the sentence (Gülich 1970: 301). The idea that the main function of discourse particles is the structuring of the dialogue is supported by Stubbs (1983): There "are several distributional facts about [discourse particles, KF], which cannot be explained by the syntax and semantics of single sentences" (Stubbs 1983: 73). Discourse particles thus seem to operate on units which are larger than sentences. In addition to the large number of different items identified by Gülich (1970) (conjunctions, adverbs, interjections, etc.) which may fulfil structuring functions, Van Valin (1973) argues for a macrostructural function of mostly the modal uses of German doch, the function of this item predominantly being the structuring of the spoken language text for the hearer. He concludes his investigation by stating that "important natural language phenomena must of necessity remain beyond the linguist's grasp until a theory of discourse, a theory designed to deal with inter-sentential phenomena, is developed" (Van Valin 1973: 87). Likewise, Schiffrin (1987) as well as Maschler (1994, 1997), Lenk (1998), Fraser (1999), and Heritage and Sorjonen (1994) argue for the role of discourse particles with respect to larger informational units than adjacent utterances such as topics. This observation is supported by results from psycholinguistic experiments (Bestgen and Vonk 1995) that the use of particular discourse particles influences the availability of information from previous utterances. reactions from the listener (termed 'back-channel items' (...)) signalling continued attention, agreement, and various emotional reactions are not to be classified as turns" (Oreström 1983: 23). Thus he, and this investigation will follow his terminology here, distinguishes speaking-turns, or just "turns", from back-channel items.

20

Introduction: The domain

The structuring function of discourse particles also concerns the information structure of dialogues, the informational organisation of discourse (cf. also Willkop 1988). For German discourse particles, including the hesitation marker äh, Rasoloson (1994) shows that the items under investigation are used, among other functions, to mark important information and to segment utterances in such a way that relevant information is grouped together (see also Rudolph 1985). Yet, Krivonosov argues for an argumentation-structuring function of German modal particles, such that they divide the content of sentences into given and new information (Krivonosov 1989: 33-35). Likewise, discourse particles may be used to comment on certain units of speech, for example, as meta-languaging devices, in bilingual communication (Maschler 1994), but also in monolingual conversation (Gülich and Kotschi 1996). A related function of modal particles may be their support of the interpretation process by indicating the respective speech-act (Kawashima 1989: 281), as 'illocutionary indicators' (Helbig 1977: 34). For English, Fraser (1988, 1990), Schiffrin (1987), Blakemore (1987), Redeker (1990) consider discourse particles, including, for instance, conjunctions such as but, in their function to connect and segment utterances. That is, their role is "to bracket units of talk" (Schiffrin 1987: 34). Blakemore (1987) analyses discourse particles within the framework of relevance theory (Sperber and Wilson 1986) where they are looked at as meta-pragmatic instructions for the processing of utterances, i.e. discourse particles serve here to establish coherence between utterances. König and Requardt (1991) and König (1997) reach a similar conclusion regarding German modal particles. It seems that all of the items under consideration, such as interjections, segmentation markers, hesitation markers, and modal particles, but also conjunctions (and even punctuation marks in written language) can be described as fulfilling structuring functions with respect to local and global content and structure of discourse, so that these cannot be seen as a distinguishing criterion between the different classes of particles. With respect to the turn-exchange system, in ethnomethodological analyses the features of discourse particles with respect to the turn-

Definition

21

taking system have been analysed (Sacks et al. 1974, Schegloff 1982). Furthermore, Duncan and Fiske (1977) and Duncan (1972,1974), who argue for a signal-based turn-exchange, hold that discourse particles, as well as non-verbal cues, serve the function to take, hold, yield or support a turn. For German discourse particles, their role with respect to the speaker-hearer-exchange system is investigated by Willkop (1988). Also for English interjections, segmentation markers, and pragmatic idioms, there are analyses that focus on functions with respect to the turn-exchange system (e.g. Schiffrin 1987: 293). For hesitation markers, their contribution to turn-taking has been investigated by, for instance, Maclay and Osgood (1959). A function regarding the speaker-hearer-exchange system has not been assigned to modal particles so that this functional property may be peculiar to interjections, pragmatic idioms, hesitation and segmentation markers, as well as non-verbal signs. There is another domain with respect to which discourse particles may fulfil certain functions: Hockett argues that the phenomena of spontaneous spoken language do not reveal interesting facts about the language, but about the speaker's personality (Hockett 1958: 143). Thus, the function of spoken language items can be seen to provide a personal fingerprint in the conversation (cf. also Maynard 1993). The first to analyse the interpersonal functions of discourse particles systematically was Schourup (1983) (but see also Martin (1971)). As "evincives," discourse particles, including interjections, segmentation markers and pragmatic idioms, display a certain mental content for purposes concerning the speaker-hearer relationship, for instance, to make the speaker's mental processes transparent for the hearer. The effect may be that the speaker makes him- or herself trustworthy (Brown and Levinson 1987). Drescher (1997) writes for French interjections that "as manifestations of empathy they contribute once again to the emotional harmony essential to cooperative conversations" (Drescher 1997: 241). Brown and Levinson (1987) argue that discourse particles fulfil a role with respect to the management of face. Vismans (1991) describes the same function of Dutch modal particles. For German discourse particles, Ehlich (1986) provides a theoretical framework to treat them as signs in the so-called expeditive

22

Introduction: The domain

field, a functional domain which concerns the speaker-hearer relationship. For this purpose, he expands Bühler's sign model, adding the expeditive field to the deictic and the symbolic field. Similarly, especially modal particles have been found to express speaker attitude, "die sogenannte subjektive Modalität als eine subjektive Wahrnehmung des Geschehens" (Krivonosov 1989: 31). Discourse particles in reformulations can also be used to evaluate the part of discourse which is reformulated, and consequently express speaker attitude on the one hand and support the formulation process on the other (Giilich and Kotschi 1987, 1996). A function determining the relationship between speakers and hearers thus also does not seem to be peculiar to a particular word class. Concerning speech management functions,12 segmentation markers, interjections, and hesitation markers have been analysed in their role as repair markers (Jefferson 1974, Levelt 1983, Giilich and Kotschi 1996). In addition, hesitation markers have been discussed in particular with regard to in their function to provide time for speech planning (e.g. Goldman-Eisler 1958, Howell and Vetter 1969). While silence can serve speech planning purposes, too, this function has not been proposed for modal particles and may thus be a property which could distinguish them from the class of discourse particles. To sum up, it can be concluded that the research situation concerning the functions of segmentation markers, hesitation markers and interjections is very heterogeneous; functional criteria as they have been proposed in the literature do not provide a reliable basis to distinguish the different subcategories of discourse particles. Furthermore, not only has a large functional spectrum been attributed to the different subclasses of discourse particles, but the functions proposed also do not seem to be specific to discourse particles since non-verbal cues, modal particles, speech routines and even punctuation marks have been found to behave similarly in several respects. Thus, neither the subclasses postulated nor the class of discourse particles itself can be 12

The term speech management is taken here to refer to "Formulierungs-" and "Textherstellungsverfahren" (Giilich and Kotschi 1996: 37-41), that is, to the concrete effort of formulation and text production.

Definition

23

defined on the basis of the functional criteria arrived at in previous analyses. However, if it is correct that discourse particles are a functional, rather than a semantic or syntactic class, it may still be possible that it is not a single function which is characteristic of a certain word class but that the functional range of each class, i.e. the intersection of the different functions, may serve as a distinguishing trait of discourse particles. Combining the different properties proposed may thus be a method to distinguish discourse particles from other word classes; for instance, modal particles can be distinguished from discourse particles by their inability to fulfil functions regarding speech management and turn-exchange. However, defining a word class by means of negative criteria, such as functions it does not fulfil, or on the basis of a list of possible functions which largely intersect with those displayed by other word classes, makes it difficult to decide in concrete cases whether an item belongs to a particular class, and it remains unsatisfactory with respect to the essential properties of the word class. Regarding the subclasses of discourse particles, it has furthermore been shown that they are functionally very similar and that thus functional criteria do not serve as distinguishing criteria. 1.2.3

Form-related

properties

Formal criteria proposed to define the class of interjections include the following (cf. Ehlich 1986, Willkop 1988, Wierzbicka 1992b): - They are sentential: interjections constitute complete utterances; - they bear no clear grammatical relationships to other elements in the sentence; - they are not inflectable; - they may be stressed; - they may be phonologically ill-formed. Segmentation markers, which "segment units of talk" (Schiffrin 1987: 34), on the other hand, are characterised by the following formal attributes (cf. Willkop 1988, Schiffrin 1987, Schourup 1983):

24

Introduction: The domain

- They bear no clear grammatical relationships to other elements in the sentence, being only loosely integrated; - they are not inflectable; - they may be stressed; - they may be phonologically ill-formed; - they connect utterances as a kind of "discourse glue" (Fraser 1988). As can be seen from the two lists, there is a large overlap between the distinguishing formal properties of interjections and segmentation markers assumed so far. Those features which the descriptions do not share do not suffice to characterise the two classes as exclusive sets; thus, it was traditionally believed that segmentation markers and interjections can be distinguished by the fact that interjections always constitute entire utterances. However, segmentation markers may do so as well (cf. also Fraser 1988), for example: (5)

031054: also irgendwas hab(e) ich wahrscheinlich eben falsch gemacht aber [well I have probably just made a mistake but]

The frequency of this construction renders it unlikely that the speaker has intended to go on after the occurrence of aber (cf. Diewald and Fischer 1998) and that the utterance is just interrupted. A comparable example of English discourse particles in a sentential construction is (6):

(6)

5UMI: Uhhh this week is completely out of the question for me. 6UMI: I've got 7ENG: Well, you've got a very heavy cold, but well 8UMI: ((laughs)) 9ENG: Wh wh why is this week out of the question? (TP13, 7)

Definition

25

Other obvious examples of the sententiality of segmentation markers are those ones which may function as answer particles, such as German ja, nein or mhm and English yes, yeah, no, uh-huh or okay. Concerning their position between utterances which they connect, Schiffrin (1987) argues for such a function with respect to English oh, a prime candidate for an interjection, as well as the conjunction but and the speech routine y ' know. Regarding the distribution of the items under consideration, Zwicky (1985) argues that interjections and segmentation markers are similarly distributed, too. This is supported by Gülich's observation concerning the same locations at which interjections, segmentation markers, adverbs, adjectives, etc. can be found (Giilich 1970: 9). So it seems that it is not possible to find form- or distribution-related properties of interjections which could help to distinguish them from segmentation markers. In contrast to discourse particles, which are usually found outside utterance boundaries, German modal particles are taken to occur in utterance-medial positions, in particular, in the middle-field of a sentence (Abraham 1991), and to refer to the whole proposition (Helbig and Buscha 1993, Krivonosov 1989: 93). So discourse and modal particles can be distinguished by means of their syntactic positions. Furthermore, most modal particles cannot be stressed; others allow both stressed and unstressed variants, but the stressed variant has a meaning different from the unstressed form. Furthermore, discourse particles may be distinguished from sentence adverbs by the fact that the latter may also occur within clause boundaries. Willkop (1988) argues for a similar function of sentence adverbs if they occur in the same position as discourse particles; however, usually they are syntactically integrated, occurring in the 'Mittelbzw. Nachfeld' of sentences in German (Willkop 1988: 63-64). For English, for example, Traugott (1995) argues for a continuous relationship for items such as indeed, after all, and in fact between their function as adverbs and as discourse particles. Adverbs can occur integratedly and can therefore be distinguished from discourse particles, yet they may also occur in the same positions as discourse particles and fulfil similar functions there as well.

26

Introduction: The domain

To conclude, there are form-related criteria which allow to distinguish discourse particles from other word classes, for instance, the position with respect to the sentence they come along with and the integratedness into it, though adverbs can only be distinguished from discourse particles by the mere possibility to become integrated into the sentence structure. However, form-related properties do not suffice to distinguish the subcategories of discourse particles from each other.

A preliminary delimitation of the domain The discussion of the semantic, functional, and form-related properties proposed in the literature on discourse particles has shown that the features suggested do not suffice for a distinction between the different subclasses of discourse particles. Therefore, in this investigation, criteria for the distinction between interjections, segmentation and hesitation markers will have to be developed. Furthermore, the borderlines of the class of discourse particles are not clear; the semantic, functional and form-related properties proposed do not provide necessary and sufficient conditions for their definition. Useful semantic features to distinguish this word class from others have not been determined at all, while functional and form-related properties can only be used as parts of large lists, describing prototypical instances of discourse particles. These lists however do not provide a characterization of the respective word classes. On the basis of the previous discussion, as a preliminary delimitation of the object of study, discourse particles can be considered as those items which do not enter grammatical relationships with other elements of sentences; which usually occur outside sentence boundaries and constitute entire utterances themselves; which fulfil structuring, meta-languaging, turn-exchange, interpersonal and speech management functions and whose meanings contain pointers to the speaker and to other aspects of the utterance situation. Thus, the class of discourse particles can be looked at as an intersection of all of the properties proposed by previous authors. As the discussion has shown, these features are not essential characteristics of discourse particles,

Corpora

27

peculiar only to them and not to other classes. Thus a definition that consists of essential instead of only accidental properties needs to be developed and defended throughout this study. The problems with defining the different classes under consideration may point to the fact that the phenomena are categorically indistinguishable. Contrastive analyses support the assumption of a functional continuum for items from several different word classes, covering also conjunctions (cf. also Sweetser 1990) and speech routines (Stein 1995), since translators use items from all of these categories as translation equivalents (cf. section 4.1.1). Likewise, English tag questions have been found to be used as translation equivalents for German modal particles (Kohler 1978, Fillmore 1981, Nehls 1989).13 Yet, in spite of this functional continuum and the apparent impossibility to provide a definition of discourse particles based on the functional, form-related, and semantic aspects proposed in previous studies, it will be shown in section 5.1 that the classes under consideration can be distinguished by means of their semantic properties. The set of German discourse particles that will be considered here, all of which are in accordance with the criteria listed above and which occur often enough in the corpora to provide a reliable basis for a statistical analysis, consists of the following items: ach, ah, ähm, also, gut, hm, ja, ne, nein, and oh. The English discourse particles which will be discussed here in detail are ah, oh, well, and yes. However, in order not to restrict the scope of the investigation too much from the outset, in addition to the lexical items mentioned above, modal particles, tag questions and speech routines will be considered occasionally, not to provide a detailed description of their pragmatic functions, but so as not to lose the overall view of the phenomena. 1.3

Corpora

In section 1.2.2, the different functions discourse particles can fulfil were discussed as they have been proposed in the literature, among 13

Nehls (1989) also finds English auxiliary verbs to function as equivalents of German modal particles. This shows how fuzzy the functional borderlines even between particles and other linguistic classes are.

28

Introduction: The domain

them the structuring of discourse and the regulation of the speakerhearer relationship; these functions already indicate that the use of discourse particles must depend on the type of discourse and the relationship between speakers as imposed on them by the situation. That is, if discourse particles fulfil a particular function regarding the interaction with the communication partner, this function will differ depending on particular properties of the partner. For instance, the interpersonal functions of discourse particles will change in comparison with free natural conversation if the communication partner is an automatic speech processing system. Consequently, the choice of a corpus is an important question since it influences the obtainable results. Many linguistic theories regard face-to-face communication as the most basic form of communication (Clark 1996: 8-9); written language and monologic speech are regarded as being derived from the main communicative functions of language in face-to-face interaction (Fillmore 1981, Diewald 1991, Petriô 1995). Consequently, determining the role of discourse particles in these kinds of dialogues should be the proper starting point since other uses may be explainable on the basis of their functions in natural face-to-face conversation. In this investigation, however, another starting point will be taken. The German task-oriented dialogues which will be analysed first allow the controlled comparison of the influence of certain variables on the functions and distributions of discourse particles, and thus support finding those factors which determine their meanings and uses. After that, for the analysis of English discourse particles, natural conversational data will be used in order to show that the results obtained on the dialogues from task-oriented, experimental, situations are in accordance with those for unconstrained face-to-face interaction. All corpora under consideration consist of utterances by female and male speakers. In order not to confuse the reader, since the labels of each turn are not transparent with respect to the speaker's gender, in this investigation all speakers will be referred to as females, unless it is clear from the dialogue excerpts themselves that one of the speakers is male, or the speaker's gender is part of the investigation, as it will be the case in section 3.2.

Corpora

1.3.1

29

The German corpora

For German, three corpora are available that allow the study of the variable communication partner since they are comparable in being recorded in the same task-oriented situation; they differ only with respect to the communication partners to whom speakers are talking. Consequently, in the first part of this investigation, these corpora will be analysed in order to identify the role of the speakers' aims concerning their (different) communication partners in the use of a discourse particle. In order to show that the results of the analyses of certain discourse particles are not only true of a particular corpus, another corpus of German task-oriented dialogues, the Verbmobil corpus, will be consulted. Consequently, four German corpora are used in this investigation, all of which consist of task-oriented dialogues in which cooperativity is necessary to achieve the results required. The German examples used throughout this investigation will be glossed in English turn by turn in order to allow readers unfamiliar with German to follow the argumentation. The examples will be translated such that they are functionally equivalent to the source language turn and such that they preserve the structure of the original turn as much as possible. The Verbmobil corpus (Verbmobil-Database 1995, Bade et al. 1994, Jekat-Rommel et al. 1994) The Verbmobil corpus consists of a number of different corpora recorded in the same scenario but under different conditions. The biggest corpus consists of 226 dialogues and a total number of 67,801 words. The participants task is to agree upon a date, a time and sometimes on a place for an appointment. However, this corpus differs from natural conversation with respect to the setting in which the corpus was recorded and the recording modalities themselves. In particular, speakers were asked to press a button before they were allowed to speak. Therefore, the corpus contains very few instances of turn supporting feedback signals because very few speakers pressed the button just to communicate "I am listening" or "please continue" (Schegloff 1982); the turn-exchange system may therefore differ from natural communi-

30

Introduction: The domain

cation. However, the Verbmobil database provides a very rich source of data because of its impressive size. Furthermore, there are dialogues from the appointment scheduling domain, which were recorded for the development of the scenario (Bade et al. 1994, Jekat-Rommel et al. 1994); these dialogues are natural unconstrained interactions among native speakers of English and of German as well as interpreted dialogues and Wizard-of-Oz data in which automatic translation is simulated by a human interpreter mediated by a sound blaster. These various small test corpora recorded for Verbmobil scenario design will be occasionally used in this investigation for illustrative purposes. All dialogues in the Verbmobil corpus are transcribed according to the following conventions: TRANSCRIPTION CONVENTIONS (KOHLER ET AL. 1 9 9 4 )

< >

< o >

/-

/+ +/ =

0 (@ @)

marks a nonverbal feature marks a breathing event marks a pause marks hesitated speech marks a non-articulatory noise marks a technically-caused break marks a pronunciation variant marks an interruption marks the right border of an up-taking event marks the left border of an up-taking event marks a word fragment marks a non-understandable sequence mark overlapping speech mark overlapping speech with previous speech mark overlapping speech with following speech

The three corpora of the instruction domain The following three German corpora were recorded in the instruction domain and constitute a useful resource to study the influence of the

Corpora 31 variable communication partner. The task the participants had to perform was in all cases to instruct someone to build a toy-airplane. The three corpora differ however regarding the constructor; in the first setting she was another human being; in the other two corpora, the participants were seated in front of a computer. One time, the instructors were told that they were talking to an automatic speech processing system, another time they were informed that they were actually communicating to a human "wizard." The data therefore provide a basis to compare human-to-human communication with (simulated) verbal human-computer interaction and computer-mediated, indirect, humanto-human communication. The dialogues were recorded on DAT and transcribed according to Fink et al. (1995). Each turn is numbered and labelled concerning the dialogue in which it occurs; in the human-tohuman dialogues, the role of the speaker, i.e. whether she is an instructor (I), a constructor (K), or the conductor of the recording (V), is furthermore marked, while in the human-to-machine corpora the instructor is identified by means of two letters. Examples are 011070, meaning 'first dialogue, instructor speaking, turn number 70' in the human-to-human scenario, and 01MF070, referring to the first dialogue, turn number 70, MF being the letters for the identification of the speaker in the simulated human-to-machine scenario and the indirect, computer mediated, human-to-human scenario.

The Human-to-Human Scenario (Sagerer et al. 1994) The 22 instructors in this scenario had to perform two tasks: Firstly, they were asked to construct a toy-airplane themselves, following an illustration. Secondly, each of them had to instruct another person, one of the 22 human constructors, to build this toy-airplane. There were different conditions concerning how much of each other the participants could see; in one scenario they could see their partners and what they were doing, in another they were restricted in sight regarding the other's construction, whereas in a third scenario they could not see their communication partners at all. However, no systematic variation according to sight conditions could be found concerning the distribution of discourse particles; consequently, the dialogues will be treated as a uniform corpus here.

32

Introduction: The domain

To ensure comparability with the dialogues in the human-tomachine scenario in which the constructors' utterances consisted of prefabricated units uttered by the (simulated) artificial communicator, only the instructors' utterances are considered in the quantitative analyses in chapter 3. The 22 dialogues consist of altogether 25,914 words, with a mean length of 1,178 words per dialogue. There was an equal number of male and female participants who were all university students.

The (Simulated) Human-to-Computer Scenario (Brindöpke et al. 1995) The tasks the 40 participants had to solve in this experiment were almost the same as in the previous corpus but with the difference that they had to instruct a (supposedly) artificial system via microphone to build the toy-airplane. In fact, the behaviour of the artificial communicator was simulated by two people ("wizards") in another room. One person built the toy-airplane according to the participant's instructions. After every instruction, a photograph of the resulting state of the construction was taken and the picture was transferred to the monitor in front of the speaker. In the meantime, the other person could select verbal messages, simulating the speech processing system's verbal output, and send them to a text-to-speech synthesizer, which the human participant could hear via head phones. In order to make the simulated behaviour of the artificial system convincing, the wizards had to decrease their cooperativity according to the following restrictions (Brindöpke et al. 1995: 14): - reject instructions that contain words that one would assume an artificial system would not understand; - reject instructions that require memory; - ignore instructions concerning objects that are not precisely specified by the instructor;

Corpora

33

- reject instructions which are too global or under-specified; - randomly reject a certain number of instructions to simulate recognition errors. In an additional third task, the participants were asked to fill out questionnaires on their opinion about the characteristics of the artificial system. Only three of the forty speakers answered that they had doubted the existence of such an automatic speech processing system during the recording while the remaining 37 speakers believed that they had indeed communicated with an artificial system. The corpus consists of 40,268 words. The mean length is 1,007 words per dialogue. Although the number of male and female participants is not exactly equal, the corpus is big enough to ensure reliable results in the studies on gender variation in section 3.2. All of the speakers were enrolled as students at the University of Bielefeld.

The Indirect Human-to-Human Communication Corpus There was another, smaller corpus of ten speakers recorded in the same setting as the previous one with the only difference being that the speakers were told that there was a person simulating the speech processing system. With this corpus, the influence of the conceptualization of a communication partner as a human being can be directly investigated in comparison to the one in which the speakers believed they were talking to a machine. Unfortunately, the group of female speakers in this corpus was too small to yield statistically reliable results for the study of gender differences, so in the study of this variable (section 3.2), only the natural human-to-human dialogues and the simulated human-to-computer interaction will be analyzed.

34

Introduction: The domain TRANSCRIPTION CONVENTIONS (FINK ET AL. 1 9 9 5 )

{ }





< -

- >

0 / 1.3.2

The English

marks the beginning of a feature marks the end of a feature, here: quiet marks a simultaneous feature marks the beginning of human noise marks the end of the human noise breathing marks an isolated breathing noise marks the beginning of a noise marks the end of a noise of the microphone marks an isolated noise of the microphone marks a short pause marks a little longer pause marks a pause of two seconds mark parts of a word that are not realised marks a word fragment

corpora

After the influence of extra-linguistic variables on German discourse particles have been analysed, English discourse particles will be considered, firstly in order to determine whether results from the studies on German corpora can be transferred to English discourse particles, and secondly to find out whether the study of informal face-toface conversation, exhibiting many variables which can hardly be controlled, yields results comparable to those found for German discourse particles in task-oriented situations. Thus, English discourse particles in natural informal conversation will be considered. For this purpose, the prosodically labelled London/Lund Corpus is consulted which was mainly recorded in the relaxed natural situations of informal communication between educated British speakers. There are 87+13 texts of ca. 5,000 words each, the 87 texts alone make up ca. 435,000 words.

Methods

35

TRANSCRIPTION CONVENTIONS (SVARTVIK AND QUIRK 1 9 8 0 )

< 1234 A> Λ o *, ** **, + + \ / { } ' : ! !!

() (( ))

shows the utterance number ( 1234) of the respective dialogue and identifies speaker (A) marks the beginning of a tone unit marks the end of a tone unit mark overlapping speech marks falling intonation marks rising intonation mark a subordinate tone unit marks continuance marks a higher tone than the preceding syllable marks a higher tone than preceding pitchprominent syllable marks a much higher tone than preceding pitch-prominent syllable marks a brief pause marks a unit pause mark a contextual comment mark incomprehensible speech

Furthermore, for comparative reasons, the English counterpart to the German Verbmobil dialogues described above will be considered. This corpus was recorded by researchers at Carnegie-Melon University and thus consists of American English; additionally, data recorded by native speakers of English from Britain and the U.S. at the University of Hamburg (Jekat-Rommel et al. 1994) will be analysed. Again, the speakers' task is to schedule appointments. The transcription conventions are the same as in the German Verbmobil dialogues.

1.4 Methods The goal of this investigation is a description of discourse particles which does not only state the functional spectrum of a discourse particle but which describes the conditions under which it is assigned a

36

Introduction: The domain

certain functional interpretation. Regarding the heterogeneous results that are provided in the literature, this does not seem to be an easy enterprise. Wilkins summarises the methodological situation concerning interjections as follows: "Interjections are hard to handle in linguistic terms, not because they are peripheral to the concerns of linguistics, but because they embody, almost simultaneously, all of the concerns of linguistics" (Wilkins 1992: 155). The discussion of the similar functions segmentation markers, hesitation markers and interjections fulfil has shown that this is quite likely to apply to all subclasses of discourse particles. 1.4.1

Interpretative methods

The functions discourse particles fulfil as well as their meaning aspects have to be accessed by means of interpretative methods. A speaker has intuitions about what a word means, how an utterance can be interpreted, and why; the speaker is thus a valuable source of information. However, intuition does not only constitute a useful informational resource, it also involves an element of subjectivity (Taylor and Cameron 1987). Since all interpretative methods rely on judgements of these kinds, they all are subject to this criticism. The problem is thus to find methods which make use of speakers' intuitions as reliably as possible. A common way to access one's intuitions is introspection; the limits of using introspection in the analysis of discourse particles and of those factors which condition their functions will be discussed in the following. Discourse particles and introspection Regarding discourse particles, Angermeyer (1979), in his discussion of the treatment of interjections in schools, states that "Das Thema stellt keine Ansprüche von der Sache her an den Lehrer. Man wird sich auf sein Allgemeinwissen verlassen können!" (Angermeyer 1979: 40), i.e. he proposes introspection ('common knowledge') as a suitable method for the (unproblematic) analysis of the meanings and functions of interjections.

Methods

37

To test in how far the semantic and functional aspects of discourse particles are accessible to native speakers by means of introspection and whether speakers really display something like common knowledge about their meanings and uses, a questionnaire was designed in which speakers were asked to verbalize their knowledge about discourse particles. The questionnaire investigated the following aspects of discourse particles: - Are speakers able to identify and describe the meaning of a discourse particle? - Are they able to describe the use of a discourse particle? - Are they able to detect differences in meaning and distribution between different discourse particles? - What is their attitude towards discourse particles? These aspects were tested for both English and German (cf. the questionnaire in the appendix), that is, questions were designed in order to reveal the speakers' verbalisation of their introspectively available knowledge. In order to determine the speakers' attitudes towards discourse particles, they were asked whether they believe that discourse particles should be taught to children or foreign language learners. The examples employed in the tests were English oh and ah as well as er and urn, while for German the examples oh and ach and äh and ahm were used.14 To get a broad range of answers, only open questions were asked in the questionnaire. Participants were aged between 16 and 87. The number of female and male participants was approximately equal for German, and for English two-thirds were female and one-third was male. For English, since the group of male participants was too small to yield statistically reliable results, male and female speakers was considered as a single group. Social class, as a possible variable, was accounted for, using education as the operational measure of social class, 14

For the determination of the translation equivalents the results from the contrastive analyses described in section 4.1.1 were used.

38

Introduction: The domain

however, no significant differences could be found. The data concerning the 76 participants are shown in table 1.3. Table 1.3: Participants in the Questionnaire Study

group age range Mean Age Ν Educational Background high-school/ Lehre College/Abitur University degree involved in philologies

German-m 20-84 40.3 21 German-m 14.3% 14.3% 71.4% 35.0%

German-f 16-87 37.8 29 German-f 31.0% 13.8% 55.2% 52.8%

English 17-83 30.8 26 English -%

46.2% 53.8% 19.2%

The results are as follows: - Regarding the question of whether speakers are able to identify and describe the meanings of discourse particles by means of introspection, speakers were asked to report on their knowledge of the meaning of oh', more than half of the meaning descriptions provided by the participants are given as a list of emotion terms containing the word surprise (in German the terms Überraschung or Erstaunen). Several speakers opted for "no meaning." The exact data can be found in table 1.4.

- Concerning the uses of different discourse particles, less than half of the female German speakers distinguished between the uses of äh and ähm, whereas 66.7% of the German males and 61.5% of the English speakers acknowledged the difference, the latter with respect to the English counterparts er and urn. However, even if the difference was realised, 50-75% of the speakers were not able to justify their decision. As an explanation for the distributional difference between ach and oh (German), which was acknowledged by approximately 80% of the German

Methods

39

Table 1.4: The Meaning of Oh

surprise pity disappointment excitement emphasises has no meaning

German-m 57.1% 4.8% 9.5% 14.3% 19.0% 19.0%

German-f 72.4% 10.3% 17.2% 31.0% 13.8% 17.2%

English 53.8% 11.5% 8.0% 11.5%

speakers, partly positive and negative speaker attitude was mentioned (43.5% of the female speakers, 52.9% of the male speakers), but as many as a quarter of those who distinguished between these two lexemes provided explanations which are difficult to interpret (for example: "spontane vs. nachdenkliche Überraschung" ('spontaneous vs. thoughtful surprise') as the distinction between ach and oh). Only 3.8% of the English participants claimed that there is a difference between ah and oh (English). Regarding hesitation markers, several speakers answered that since the two items are different in form, they have to be different in meaning as well, thus using a sign concept to motivate the distinction. The quantitative results are represented in tables 1.5 and 1.6. Speakers' judgements appear to be quite random in their attempt to identify differences in the distribution of interjections and hesitation markers: Approximately 20% of the speakers state that they believe that the use of discourse particles is dependent on the speaker's sex. However, for each group, as many asserted that women use more discourse particles than men do as that men use more than women do. All three groups of speakers were split down the middle on the question whether the use of interjections and hesitation markers is generally topic dependent. Approximately fifty percent of the German speakers wrote that especially difficult topics made the

40

Introduction: The domain

number of discourse particles increase while difficulty of topic was only mentioned by ca. 20% of the English participants. Furthermore, highly emotional topics were reported to influence the number of discourse particles in a conversation. On the question of the dependence of the communication partner, most speakers responded affirmatively. Table 1.7 shows the respective data.

Table 1.5: Use of Oh and Ah (English) / Oh and Ach (German)

to express surprise to express other emotions for turn-taking to emphasise to provide feedback Difference between interjections if so, the argument is... ... uninterpretable ... emotional difference

German-m 85.7% 38.1% -

19.0% 4.8% 80.9% 23.5% 52.9%

German-f 89.6% 37.9% 10.3% 13.8%

English 50.0% 23.1% 3.8% 15.4% 42.3% 3.8%

-

79.3% 39.2% 43.5%

25.0% -

Table 1.6: Use of Um and Er (English) / Ähm and Äh (German)

uncertainty speech planning, word search turn-taking/holding disagreement Difference between hesitations if so, the argument is... ... inconsistent ... based on sign-concept

German-m 47.6% 66.7% 14.3%

German-f 55.1% 75.9% 6.9%

English 23.1% 88.3% -

66.7%

27.6%

8.0% 61.5%

50.0% 14.3%

75.0%

66.7%

-

-

-

-

Methods

41

Table 1.7: The Use of Discourse Particles

gender-specific if so, women use more if so, men use them more topic-specific if so, more if topic is difficult if so, more if topic is emotional dependent on the partner if so, more if situation is formal if so, more if situation is hierarchical if so, more if speakers acquainted

German-m 42.9% 33.3% 33.3 % 52.4% 54.5% 36.4% 85.7% 27.8% 33.3% 22.2%

German-f 24.1% 50.0% 50.0% 55.2% 50.0% 25.0% 75.9% 9.1% 36.4% 27.3%

English 19.2% 20.0% 20.0% 53.8% 21.4% -

65.4% 17.6% 17.6% 5.8%

- Concerning speakers' attitudes towards discourse particles, 71.4% of the male German speakers and about half of the female German and the English participants answered that discourse particles should not be taught to children or foreign language learners, although between ca. 25% and 60% made a distinction between interjections and hesitation markers. Reasons put forth were linguistic norms, social reasons such as the stigmatisation of hesitation markers, and also that they displayed low intelligence, low concentration, or lack of knowledge. Explicitly negative statements about hesitation markers were made by 38.5% of the English, and approximately 10% of the German speakers. The exact results can be found in table 1.8. Regarding the usefulness of asking native speakers for informal statements on the meanings of discourse particles, it seems that the investigation is well-suited in providing information on speakers' opinions about discourse particles. An important result of the questionnaire study is that speakers may obviously regard discourse particles, and especially hesitation markers, as superfluous and undesirable elements of speech. This negative attitude towards them may influence their uses in certain situations.

42

Introduction: The domain Table 1.8: Reasons to Teach Not to Use Discourse Particles

there are reasons if so ... ... distinction adults/children ... distinction hesitations/ interjections ... because of ling, norms . . . lack of concentration ... lack of knowledge/ intelligence

German-m 71.4%

German-f 51.7%

English 57.7%

33.3%

13.3%

20.0%

40.0% 33.3% 26.7%

60.0% 53.3% 13.3%

26.7% 40.0% 6.7%

23.8%

24.1%

13.3%

With respect to the meanings of interjections, ca. 70% of the female German participants and ca. 50% of the male German and the English speakers (42.3% of the English speakers also claimed that a reason to use oh is to give feedback) provide lists of emotion terms. Thus, when asked for the meaning of the interjection oh, speakers answered with a number of possible emotions which oh may express. This procedure is also followed, for instance, by Angermeyer (1979), by Leech and Svartvik (1975), and in dictionaries; however, it is not clear what the explanatory value of a description of an interjection is which consists of terms as varied as pain, terror, shame, astonishment, and disapprobation (OED 2nd ed. for oh). Wierzbicka (1992a) furthermore shows that emotion terms are highly complex and culture-specific; she argues that they therefore should not be used as a linguistic metalanguage. The results from the questionnaire study concerning the use of discourse particles seem to be that speakers do not provide uniform judgements about their functions and distributions. Participants disagree very much on the variables that influence the distribution of discourse particles. Thus the data speakers provide are inconclusive regarding the conditions of discourse particle use and distribution. In comparison, data from analyses with non-introspective methods in chapters 2 and 3 will show that the distribution of discourse particles is gender-, topic,- and partner dependent. Thus, assuming

Methods

43

that there is a shared competence regarding the use of discourse particles, the heterogeneous data provided by the speakers on the basis of introspection are not satisfying. Consequently, the method of asking speakers about their knowledge on discourse particles yields interesting results regarding their opinions on the evaluation of these items. However, (at least) linguistically untrained speakers do not provide conclusive data regarding discourse particle meanings and uses on the basis of questions like the ones used in the questionnaire study. Asking speakers for their opinions is not the only way to appeal to speakers' intuitions. For instance, Wierzbicka, in spite of her rejection of introspective analyses which result in lists of emotion terms, argues for introspection as a valid method. She shows that the description of illocutionary meanings can be accomplished by means of a limited descriptive inventory (Wierzbicka 1986, 1992b). In particular, she sees the difference, in comparison to introspection as employed in this questionnaire study, in the amount of time and effort spent on each definition and the carefully developed Natural Semantic Metalanguage (Wierzbicka 1986) used as a methodological tool for these analyses. Other authors working on interjections in this framework are, for instance, Ameka (1992), Hill (1992), and Wilkins (1992). However, regarding the contents of interjections, Ameka (1992) criticises the heterogenous results obtained by different authors using the same methods. Consequently, even a carefully carried out introspective analysis needs further justification. It is thus not clear how far linguistic training can improve the results of intuition-based methods (note also the large number of philologically trained participants in the questionnaire study). Regarding the distribution of different functions of discourse particles, similar problems can be found: In Helbig (1988) the function of ja as a turn-initial signal which indicates perception, understanding, and topic continuity to the hearer is not mentioned at all. In Wolski (1986), this function ranges very low among the 10 possible readings for ja (Wolski 1986: 504-505), whereas in the Verbmobil dialogues this was found to be by far its most common function (Fischer 1995).

44

Introduction: The domain

Likewise, while Wierzbicka argues for a semanticist viewpoint concerning the relationship between semantics and pragmatics15 (Wierzbicka 1991: 18-19), her analyses do not offer any explanation for the different functions interjections can fulfil as outlined in section 1.2.2, and their distributions in linguistic contexts. Consequently, regarding the use of discourse particles, introspection does not provide satisfying results compared to those obtainable by non-introspective methods even if employed by trained linguists. In the generative paradigm, the most important method is introspection since "the speaker-hearer's linguistic intuition is the ultimate standard that determines the accuracy of any proposed grammar" (Chomsky 1965: 21); however, Chomsky acknowledges that "this tacit knowledge may very well not be immediately available to the user of a language" (Chomsky 1965: 21). So it may not be straightforward to obtain the information wanted; in generative linguistics therefore judgements of the well-formedness of sentences are employed to access speakers' intuitions (cf. also Cruse 1986: 17). Unfortunately, there are restrictions to the use of intuitive wellformedness judgements in the analysis of discourse particles. One problem concerns the fact that they may constitute entire utterances, i.e. the well-formedness judgements would have to apply to sequences of utterances instead of to single sentences. Another problem is the context-dependence of the meanings and functions of discourse particles; thus, test-frames are problematic insofar as they constitute nonnatural contexts for discourse particles. Still, in chapter 4 of this investigation, this method will be discussed and a way to deal with these problems will be proposed. Another approach employing introspection as a method is to ask speakers after the conversation is over what they wanted to convey with a particular utterance. This position is taken for the study of In Wierzbicka's argument for a semanticist pragmatics, she allows for phenomena to belong to some (not defined) non-linguistic pragmatics; the distribution of discourse particles in certain structural contexts, as accessible to conversation analysis, seems to belong to this domain (Wierzbicka 1991 : 19).

Methods

45

discourse particles, for instance, by Maschler (1994). Yet the investigation of speakers' reports on their intuitions as answers to questions in the questionnaire shows that they are not necessarily aware of what influences the choice of a particular discourse particle. There may be problems if topics are involved which speakers do not normally talk about and which constrain utterances very subtly and subconsciously. It seems that with respect to the distribution and functional aspects of discourse particles other methods are necessary. However, in spite of the problems with introspection as a method for the analysis of the meanings and particularly the functions of spontaneous spoken language phenomena, it is not possible to abandon it completely. It has been shown to be useful, for instance, to get information on the stigmatisation of certain items. Furthermore, introspective decisions are a valuable heuristic method for gaining initial information, for hypothesising, and generalising. However, intuition cannot be the only source of information since, as the results from the questionnaire study and the discussion have shown, the aspects involved in the use of spoken language items may not be exhaustively accessible to intuition. Corpus-based interpretative methods Conversation analysis (CA),16 like introspective methods, takes a pretheoretical, data-driven approach to conversational phenomena, but it is methodologically more constrained and may thus be a suitable option for the analysis of the functional range of discourse particles. In contrast to the common attitude towards discourse particles as 'noise to eliminate' (e.g. O'Shaugnessy 1993), CA seems to be a good starting point into analyses of their functions since it holds that "detailed study of small phenomena may give an enormous understanding of the way humans do things and the kinds of objects they use to construct and order their affairs" (Sacks 1984: 24); that is, in CA "nothing in talk-in-interaction should be dismissed as trivial or uninteresting before it has been subjected to analysis" (Hutchby and Wooffitt 1998: 16

Cf. Sacks (1984, 1996); for an overview see, for instance, Heritage (1995), Psathas (1995), Hutchby and Wooffitt (1998).

46

Introduction: The domain

23, emphasis original). Furthermore, as discourse particles have been proposed to fulfil several kinds of structuring functions, CA with its interest in the structural organisation of dialogues as an interactional achievement seems well suited for an analysis of the domain because it allows for a description of the role of discourse particles in the structural organisation of dialogues. In conversation analysis, corpora of talk-in-interaction provide the basis for the analyses (Heritage 1995: 395). One of its important features is its insistence on naturally occurring as opposed to experimentally produced data or elicited dialogues obtained under laboratory conditions. In this investigation, what is shared with this view is the reliance on spoken language corpora as a starting point since it concurs with Sacks (1984) that we should "be using observation as a basis for theorizing. Thus we can start with things that are not currently imaginable, by showing that they happened. We can then come to see that a base for using close looking at the world for theorizing about it is that from close looking at the world we can find things that we could not, by imagination, assert were there" (Sacks 1984: 25).

However, as already discussed in section 1.3, the analyses in this investigation will not be entirely based on naturally occurring data since the use of task-oriented corpora has a number of advantages: Certain variables can be studied systematically since the common task makes all dialogues comparable. As the objective of this investigation is to identify the influences on the interpretation of discourse particles, comparable corpora allow a controlled analysis of the role of the communication partner and other extra-linguistic variables on their use. Taking into consideration the methodological principles of CA, the use of task-oriented corpora however is not unproblematic. First of all, their major 'principles', such as the turn-taking system, preference organisation and sequential structure (Hutchby and Wooffitt 1998), have been developed on "mundane conversation", i.e. on language in which "what people say, how they say it and the length of the turn in which they say it - in other words, turn form, turn content and turn length are free to vary" (Hutchby and Wooffitt 1998: 47). In the German taskoriented human-to-human dialogues to be considered in this study, the

Methods

Al

form and length of the turns speakers may utter are not restricted, yet participants may not choose their topics freely. The dialogue structure mirrors the structure of the task to perform, i.e. it corresponds to the construction phases the instructor chooses in the instruction dialogues and the kind of information the speakers have to negotiate in the appointment scheduling dialogues of the Verbmobil corpus. However, although the topics are principally restricted, speakers still may structure the task in the way they prefer, and thus turn contents may also be varied freely. The restriction to a particular task thus does not seem to be a methodological drawback regarding the principles of CA. Another central property of CA is that it is concerned with the description of conversational organisation as it is produced by the participants themselves in actual conversations. It is assumed that the interpretation of a particular utterance manifests itself in the sequential structure of the dialogue. In the following turn it becomes apparent how the communication partner interprets the previous turn (Heritage 1995). Sacks et al. (1974) refer to this as a 'proof-procedure': "But while understandings of other turns' talk are displayed to coparticipants, they are available as well to professional analysts, who are thereby afforded a proof criterion (and a search procedure) for the analysis of what a turn's talk is occupied with. Since it is the parties' understanding of prior turns' talk that is relevant to their construction of next turns, it is THEIR understandings that are wanted for analysis. The display of those understandings in the talk of subsequent turns affords both a resource for the analysis of prior turns and a proof procedure for professional analyses of prior turns - resources intrinsic to the data themselves." (Sacks et al. 1974: 729, emphasis original)

Thus, by displaying their understanding of their partners' utterances, the conversationalists themselves create the guidelines for the interpretation of their utterances by the analyst. Methodologically the description exploits, and is at the same time restricted to, those aspects whose the speakers' own accounting for leaves traces in the sequential organisation of the dialogue. Regarding the corpora chosen for this investigation, for the simulated human-computer interaction and the indirect human-to-human dialogues the reliance on the communication partner's utterance in

48

Introduction: The domain

the interpretation constitutes a methodological problem: While human speakers constantly attend to their communication partners' utterances and consequently provide data for the next-turn-proof-procedure, the simulated system's output consists of stereotypical, prefabricated phrases. Nevertheless, the (simulated) system interprets the instructor's utterances, and consequently the speakers, and thus the analysts, do get feedback on their utterances by the system's verbal and nonverbal actions in the following turn. Whether this interpretation was satisfactory may become clear from the speakers' reactions to the simulated system's behaviour. For instance, in example (7) the speaker comments on the interpretation of her previous utterance as insufficient: (7)

43KB004: okay. fein, {braves Kind.} so. jetzt nehmen wir den [ okay. fine. {good boy} so. now we take the ] K: bitte wiederholen Sie Ihre Anweisung. [ please repeat your instruction] 43KB005: ich hab(e) noch nichts gesagt, okay. . [ I haven't said anything yet. okay .]

It can be concluded that even for the task-oriented human-computer dialogues the CA methods, including the procedure discussed, seem to be well suited (see also Douglas 1995). The comparison between different task-oriented corpora is not only a methodological obstacle but can also support the CA studies on discourse particles; Nothdurft (1984), for instance, criticises that "[d]ie conversational analysis vergißt, daß auch das Bemühen um Ordnung und Erkennbarkeit nur "for all practical purposes" erfolgt und nur Randbedingung der Realisierung individueller Handlungsziele darstellt" (Nothdurft 1984: 124-125), that is, speakers may only account for those properties by means of ordered sequential organisation

Methods

49

to which they want the hearer to react in a particular way. Thus it is possible that there will be aspects involved in the interpretation of discourse particles which cannot be identified as traces in the conversational structure. For instance, speakers use German ja frequently in situations when they are in danger of losing face (see section 2.1). It is thus plausible to assume that the speakers' aims regarding the interpersonal relationship to their communication partners may influence the employment of discourse particles in such situations. However, the regulation of the speakers' interpersonal relationships is not necessarily obvious in the sequential dialogue structure since speakers may not account for these aspects in the same way they indicate, for instance, the structure of an argumentation. Consequently, in addition to CA analyses of uses of discourse particles in informal task-oriented communication, comparable corpora will be analysed which differ only in one respect, namely the communication partners to whom speakers talk. By means of functional and distributional analyses of discourse particles in the different corpora, the influence of the speakers' conceptualisation of their communication partners on their speech particularly in the interpersonal domain can be detected. Comparative analyses can thus show what the domains are to which speakers attend depending on their conceptualisation of their communication partners and which tasks they regard to be relevant regarding their hearers. The choice of task-oriented corpora can also be useful in another way: CA restricts itself to evidence from the sequential structure of the dialogue. However, conversation is always situated, that is, there are situational factors which influence the conversation. In the instruction dialogues, for instance, speakers are supposed to perform a practical task; in the natural conversations collected in the London/Lund corpus, speakers eat biscuits, pour milk into their coffee and open doors while they are speaking. Consequently, situational factors such as the accompanying actions have to be accounted for as well since they may influence the use of discourse particles (see section 2.1). From the point of view of the ethnography of communication, CA has been criticised for its rejection of background knowledge as a valuable source of information supporting the interpretation of utterances (cf. Auer 1995: pp.427-428). The structure of conversation is not the only source of in-

50

Introduction: The domain

formation that is available to the speakers. By using corpora in which the scenario is predefined, we can also control for the situational aspects of the context which may influence the use and interpretation of discourse particles and may yield results for which CA alone could not account. Although the principles of CA so far seem to be well suited for the study of the conditions on the interpretation of discourse particles, there are also limits to its usefulness regarding the current goals. The aim of this investigation being a systematic cognitive semantic account of the functions of particular lexical items, the focus of the study is on the general structures and mechanisms by means of which a discourse particle gets its interpretation in a conversational context. Conversation analysis however is primarily oriented at the "accumulation of empirical findings about the multiplex practices organising social action" (Heritage 1995: 397). In contrast, a lexical approach demands the generalisation about the contextual occurrences with respect to a certain lexeme. A question the CA analysis leaves open is how the different contexts identified are related to the particular lexeme for each lexical item (cf. Heritage's (1984) analysis of oh). Thus it is questionable whether an analysis of the various contexts in which a discourse particle may occur corresponds to a representation of this lexeme in the mental lexicon. In the analysis of talk-in-interaction, category-based approaches present themselves as an alternative to CA (cf. Levinson 1983). These approaches assume that conversation can be analysed exhaustively into units of different size on different analytical levels, that each unit can be assigned a functional interpretation, and that these constitute structured sequences which display certain patterns. In contrast to conversation analysis in which only those categories are assumed of which can be shown in the sequential organisation of the conversation that speakers orient to them, category-based approaches usually work with a more general inventory of descriptive categories, their origin often being in speech act theory (Austin 1962, Searle 1969, 1975). In such a framework, discourse particles are analysed by assigning each occurrence a function from a restricted inventory; thus each instance is disambiguated regarding its function on the basis of a classification of communicative acts (see, for example, Stenström 1994).

Methods

51

From a CA perspective, category-based approaches can be criticised for their use of a prefabricated inventory of descriptive categories (Psathas 1995). Discourse analysis (DA) in the sense of Levinson (1983), a type of category-based approach, thereby presupposes the objective of the present investigation, namely that discourse particles can be assigned a certain speech act or interactional move if they constitute independent utterances (Coulthard 1985, Stenström 1994). Likewise, the inventory used to describe the functions of discourse particles cannot be presupposed; it is not only more desirable to start with those categories to which the participants attend themselves, in addition the heterogeneity of the functions proposed for discourse particles and the fact that previous studies have rarely accounted for all of them indicate that a methodology needs to be employed which goes beyond judgements of intuition. For this reason, the inductive approach of CA with its "strict and parsimonious structuralism and a theoretical asceticism" (Levinson 1983: 295) seems to be a more suitable starting point, while DA provides us with the distinctions needed for a lexical account of the functional polysemy of discourse particles, namely by identifying their different readings. If the two approaches, CA and DA, are therefore not seen as alternatives but as different means to different ends which may support each other, they are both suitable for the purposes of this investigation, namely to identify speaker categories in the use of discourse particles and to generalise about them. In particular, on the basis of the results from a CA analysis, functional correspondents may be grouped together and an inventory of interpretative categories can be proposed. If generalisations about different contexts can be found, a category-based description may represent these generalisations as particular descriptive categories. However, in designing a descriptive inventory, several aspects have to be considered: - The descriptive inventory has to build on speaker categories identifiable in a CA analysis. There have to be means of showing that the categories really match with the data; if necessary, the categories have to be adapted in a cyclic revision process. - The criteria for the assignment of a particular category have to be clear. Preferably, these criteria are those to which speakers

52

Introduction: The domain

attend. While it is difficult to get evidence of which information conversationalists rely on in a particular decision (Taylor and Cameron 1987), it makes sense to determine the upper boundary of the extent to which speakers may employ certain features in the interpretation process. To look for an empirical basis for interpretative categories with surface features as indicators of these classes can, for instance, be seen as a heuristic method for the formation of certain categories. If the descriptive inventory can be shown to have a basis in surface features, classes could be determined intersubjectively (and automatically, which is of relevance for automatic speech processing). This is independent of the question of whether interpretation is actually based on surface cues: Firstly it is an empirical matter whether it will be possible to find surface correspondents for each interpretative category, secondly it is questionable whether it can be assumed that speakers indeed infer functional properties on the basis of formal features. - The categories proposed have to be sufficiently contrastive, exhaustive and expressive. In the methodological approach proposed here, a categorical description of the functions of discourse particles will be developed, based on the detailed description of discourse particles in their different structural contexts. The former is necessary since in a lexical approach different readings have to be distinguished so as to develop a model of the polysemy of the items under consideration. The latter is needed as a sound methodological basis in so far as the categories proposed have to match those aspects of conversation to which speakers attend, i.e. which prove themselves as participant categories. How exactly the mapping between these two types of information is to be carried out will be discussed in the section on quantitative methods (section 1.4.2). The methods discussed so far allow us to analyse the functional spectrum of each discourse particle and to categorize the different occurences into functional classes. An open question is still the contribution of each discourse particle lexeme. By means of CA methods

Methods

53

only different contexts can be identified and hypotheses about the general value of each discourse particle can be generated. The inductive approach with its "healthy suspicion of premature theorising and ad hoc analytical categories" (Levinson 1983: 295) has to leave discourse particles ambiguous with unrelated functions in different types of conversational structures, for example, well in closings and well in dispreferred responses (Levinson 1983: 366). The functional categorization also leaves us with a number of different, unrelated readings. This is unsatisfactory from a lexical point of view; for a lexical description of polysemy, in contrast to homonymy, a way to relate the different senses of a lexeme has to be found. Here, methods from lexical semantics are more useful. In lexical semantics, two basic directions of study can be distinguished for the analysis of the contribution of lexemes: relational approaches (e.g. Cruse 1986) and decomposition, i.e. feature-based analyses (e.g. Wierzbicka 1985). These two viewpoints constitute different perspectives on how meanings are constituted; both will be discussed regarding their applicability to the analysis of the lexical meanings of discourse particles. The relational approach, developed in the structuralist paradigm, assumes that each word gets its meaning in opposition to other words; an important concept in this paradigm is therefore the semantic field (Trier 1973a,b, Lehrer and Kittay 1992, Lutzeier 1993) which represents the relations between related items. On the basis of the independent evaluation of closeness between the items under consideration and another variable, it is possible to create semantic fields automatically by means of computational methods (see section 1.4.2). The tertium comparationis can be provided by the translation equivalents of the respective discourse particles. These data are available from contrastive analyses. There are two main possibilities to relate the two languages in order to arrive at correspondence relations between the respective items: either both languages are discussed separately and the results are then compared on the basis of a common function; or, the contrastive study is based on already established equivalence relations as provided by translations. In this investigation, which aims at the description of

54

Introduction: The domain

both German and English discourse particles, both methods are employed. On the one hand, independent analyses of both languages will be carried out: The functional spectrum of German discourse particles will be investigated and then it will be tested whether the results found for German can be applied to English discourse particles. On the other hand, modern plays and their translations will be analysed in order to identify equivalence pairs between the discourse particles of the two languages. These can be used to determine semantic and functional distances between the items of each language and between languages. Consequently, the relational approach to the meanings of lexical items taken will also be based on the contrastive study of corpora, in this case, of plays and their translations, rather than on intuition or on substitution tests, which are common methods in the relational paradigm (Dupuy-Engelhardt 1993). In decompositional approaches it is assumed that speakers have a certain knowledge associated with a word which can be described in a feature-based manner. The three methods traditionally associated with this approach are semantic decomposition into features either on the basis of intuition (e.g. Wierzbicka 1985), contrast relations to similar items (e.g. Coseriu 1970), or test-frames (e.g. Cruse 1986). The features which will be used for the description of the expected properties of a discourse particle in this investigation will be derived from the interpretative analyses of corpus material. Thus the functional properties resulting from CA analyses constitute a valuable heuristic basis for semantic decomposition. Nevertheless, in order to arrive at a kernel meaning of each discourse particle, hypotheses have to be made about the invariant features of a discourse particle by generalising about its different contextual contributions. These can be modelled in Natural Semantic Metalanguage (Wierzbicka 1986). Methods for validation which go beyond plausibility judgements are therefore necessary. In lexical semantics, this is usually done by means of semantic tests. Specially adapted semantic test-frames will be employed to support the semantic analyses of both invariant and contextual features of discourse particle occurrences in their original contexts (Fischer 1998). To conclude, even the semantic analysis of the lexical contribution of each discourse particle lexeme may be based on evidence from corpora.

Methods 55 To sum up, conversation analysis will play the main role in the interpretative analyses of discourse particles since is a suitable approach for determining both the functional range of a discourse particle and the speaker categories involved. Its main strength regarding the purposes of this investigation is the interpretation of discourse particles in particular contexts on the basis of those aspects to which speakers attend. These analyses may serve as a basis not only for the development of a categorical descriptive inventory but also for feature-based semantic analyses while other methods by means of which the conditioning factors and their interaction can be studied directly and effectively support the analysis. Consequently, in this investigation a number of different methods will be combined. Only by regarding the different aspects involved in the interpretation of discourse particles, which are accessible by different methods, can a unified lexical pragmatic account of the relationship between discourse particle lexemes and their functional interpretations be developed.

1.4.2

Quantitative and computational methods

Since the situation influences the interpretation of utterances much more strongly in spoken than in written language, results from the study of spontaneous spoken language dialogues are partly corpus dependent (Biber et al. 1998); data from one corpus can only be transferred to another if the different conditions under which this corpus has been recorded are analysed and treated as constraints on the interpretation of the items under consideration. Quantitative analyses of the distribution of discourse particles can not only reveal the effects of differences between corpora, which then can be used for identifying the determinants of the interpretation and distribution of discourse particles, they can also point to possibly different functions of discourse particles in the same corpus, as in the case of English äh and ähm (for a full discussion see section 3.2). So quantitative results can be a useful heuristic basis for further functional analyses, which is even accepted among conversation analysts (cf. Heritage 1995: 404). Another aspect in which quantitative analyses may support functional investigations is the fact that concerning the use of discourse

56

Introduction: The domain

particles a high variability between individual speakers can be found. For instance, in the instruction dialogues, in which the speakers all fulfil the same tasks, the variance of the total number of discourse particles for each speaker is 4.13 with a mean of 4.69 per 100 words (Fischer et al. 1996). The interpersonal variation of their occurrences may therefore be quite high. This implies that the complete range of the functions of a particular lexeme can only be detected if the linguistic behaviour of a number of speakers is observed. Quantitative analyses of large corpora recorded by different speakers may support obtaining a general picture of the meanings and uses of discourse particles. Yet, statistical analyses have to be preceded by or combined with functional analyses; even analyses which only rely on word counts need further interpretation if they are to be of any explanatory value (Fischer et al. 1996, Heritage 1995: 402). However, supported by interpretative methods, quantitative analyses can provide further evidence and may lead to new hypotheses. Quantitative methods are also suitable firstly for the formation of interpretative categories of occurrences of discourse particles on the basis of more elementary features, and secondly for the identification of the relationship between certain structural properties and pragmatic interpretation. For these tasks, an artificial neural network classifier can be used which classifies discourse particle occurrences automatically on the basis of an analysis of their surface and functional features. It uses the paradigm of supervised learning, which is especially useful for phenomena for which an algorithmic description is difficult, but for which examples are available. Thus, the gap between the data obtained from a sequential analysis in the CA framework and the DA-style functional categories postulated will be bridged by means of a quantitative method, namely automatic classification. This method allows the automatic mapping of many individual instances of discourse particles to a limited number of general categories. Thereby it can be furthermore discovered which role a particular property of discourse particles, such as its position in the turn or the intonation contour, may play in the interpretation process. By modelling the classification process with an automatic classifier, it becomes clear whether the classes proposed are sufficiently contrastive and exhaustive: If they were not, they could

Methods

57

not be learnt by the simulated neural network. Thus on the basis of the accumulated facts about different structural properties of different occurrences of discourse particles, classes can be automatically assigned and the descriptive inventory can be verified. This provides the link between CA- and DA-style analyses and validates the descriptive inventory, the functional readings of discourse particles proposed. In section 4.1.2, computational methods will be furthermore employed to construct a semantic field as a topographical map on the basis of English-German equivalence pairs taken from from the analyses of modern plays and their translations. The two methods, classification by means of simulated neural networks and automatic similarity detection in a Kohonen map, belong to the connectionist paradigm of modelling human cognitive activities which is based on the idea that the brain works by means of interconnected nodes and activation patterns while knowledge is located in the differently weighted connections. Although it is a reductionistic model of the processes in the brain (cf. the notion of coherence, for instance, Weiss (1996), Weiss and Rappelsberger (1998)), the neural network approach displays a greater probability of being a suitable computational model of human activity than symbolic representations (cf. Dreyfus 1992) which are based on the functionalist view of artificial intelligence such that "the physical properties onto which the structure of the symbols is mapped are the very properties that cause the system to behave as it does. In other words the physical counterparts of the symbols, and their structural properties, cause the systems behavior" (Fodor and Pylyshyn 1988: 14, emphasis original).

This is because, first of all, artificial neural networks are designed after the model of neurons in the brain. Second and more importantly, the properties of connectionist models are similar to those observable in human behaviour: rule implicitness, graceful degradation and resistance against damage, soft constraints, parallel computation, and stochastic mechanisms (e.g. Smolensky 1988, Fodor and Pylyshyn 1988: 54-64). For these reasons, simulations by means of connectionist models have been employed in linguistic research as a method for evaluating experimentally the assumptions of linguistic theories, especially where no direct access to the relations under consideration,

58

Introduction: The domain

for instance, the relations between phonological features, is possible (e.g. Dell 1986, Eikmeyer and Schade 1991, Laubenstein et al. 1995, Schade 1992, Berg and Schade 1992, Scheler 1996a). To sum up, methods from the connectionist paradigm will be used to model the relationship between individual structural and interpretative features as they are identified in CA analyses as speaker categories on the one side and more general interpretative categories on the other. By employing both linguistic and computational methods on the issue, the generalisations proposed about the different linguistic contexts identified for the respective discourse particle will be validated by means of non-introspective methods. The same holds for the automatic calculation of the similarity relations between different lexical items: Similarity maps will be constituted by means of a computational procedure on the basis of translation equivalents proposed by professional translators. This method also supports intersubjective analyses of interpretative aspects of language and therefore helps to provide a sound methodological framework for the analyses. Finally, regarding simple statistics with respect to the differences between different corpora, the use of both quantitative and functional analyses may point to interdependencies for which either method alone could not account.

1.4.3

Linguistic

models

Linguistic modelling is a method different from both interpretative and quantitative methods since it is not empirical: It involves the development of hypotheses about the interaction of different factors which condition the interpretation of discourse particles. Furthermore, a linguistic model not only accounts for actual data but is also meant to allow predictions about potential phenomena. Linguistic modelling usually does not start from scratch, however; instead, the researcher can usually build on principles and concepts that have been introduced and described for other linguistic material. This procedure ensures not only that the model developed is not restricted to the phenomena under consideration, it also anchors a particular model in a research paradigm and builds on previous achievements without having to re-

Methods

59

peat the demonstration of the validity of the concepts employed (Kuhn 1970: 24). Such a model can be tested, for instance, in a computational formalisation. Linguistic modelling will be used at several points in this investigation. Firstly, categorical interpretation will be modelled by artificial neural networks. By means of this technique, it can be determined whether the classes chosen are learnable and therefore constitute a classification with sufficiently contrastive categories. This means that a number of determinants are proposed on the basis of CA analyses which condition the interpretation of discourse particles; general classes will be constituted by means of automatic classification. The classifier thus constitutes a model of the linguistic categorisation of discourse particle occurrences. Secondly, a semantic field also constitutes a model of the relations between different discourse particle lexemes because predictions can be made about the similarity of certain items of the same language based on the analysis of some hundred instances of translation equivalence pairs. Furthermore, proposing a certain invariant semantic component of each discourse particle also constitutes a model of its semantic structure, and it predicts the meaning of potential occurrences by claiming that they contain the abstract semantic property proposed. Most importantly, the interaction of the different types of information involved in the interpretation of discourse particles will be modelled. Since the aim of this investigation is the development of a unified model of the relations between a discourse particle lexeme and its functions in order to account for its functional polysemy, the last step in this investigation has to be the development of a lexical model of the meanings and functions of discourse particles and the formalisation of the dependencies involved. Thus, on the basis of the findings accumulated in this study, concerning the determinants of the interpretation of occurrences of discourse particles, a network of different informational resources will be built up whose interactions have to account for the mechanisms involved in the interpretation of discourse particles. Here, concepts from cognitive semantics can be useful. Thus, as has been shown for verbs, nouns, and adjectives (Baker et al. 1998), reference to a conceptual background frame can be used to explain the

60

Introduction: The domain

functional spectrum discourse particles display. Reference to particular domains (Sweetser 1990) can account for the relation between lexical content and pragmatic function, and the influence of positional and other surface information will be accounted for by means of constructions, as developed in construction grammar, an attempt to account for the syntactic information in frame semanics on the one hand and to formalise Langacker's cognitive grammar on the other (Kay 1995). Finally, the hierarchical construction- and frame-based lexicon to be developed will be implemented in DATR and thus will be computationally interpretable. This formalisation of the model allows to test the interaction of the different conditioning factors in the linguistic model. 1.5

The structure of the following

The main line of this investigation is summarised in table 1.9; it is oriented at the requirements of a systematic account of the functional polysemy of discourse particles. The first analytical step consists in identifying the functional spectrum of discourse particles and of those factors which condition their interpretation. In section 1.2.2, previous proposals regarding the different functions members of the class discourse particle fulfil have been considered: The functional spectrum assigned to them is indeed impressive. In order to determine the functions of a particularly multi-functional discourse particle, the interpretations of German ja in its different structural contexts will be analysed by means of CA methods. By means of the interpretative analyses in section 2.1, the different contextual positions in which ja can be found can be distinguished; the result is a list of different structural contexts in which ja may occur. In section 2.2, on the basis of these structural properties of discourse particles and their contexts, an investigation of how their different functional meanings associated with these structural contexts can be justifiedly summarized in a restricted descriptive inventory can be carried out. That is, by classifying the individual features assigned to a discourse particle occurrence in the structural analysis, classes will emerge which can be verified by learning experiments. Thus, the pos-

The structure of the following

61

sibility of assigning categorical interpretations such as those proposed in DA-style analyses will be considered by means of automatic classification in artificial neural networks. As a result of this investigation, the different readings of the functionally polysemous discourse particle ja can be distinguished. Furthermore, the contribution to the automatic classification of each type of feature involved can be analysed, focusing on the contribution that surface properties add to the interpretation. In particular, the relation between surface features and certain pragmatic functions that speakers attend to will be analysed. These structural properties can be associated with the respective functional properties to form linguistic constructions, form-meaning pairs. The elements of the conceptual background frame are the subject of chapter 3. Human-to-human communication is compared with (simulated) human-to-machine interaction. Usingthe systematic functional changes of discourse particles in the two settings, it can be determined to which communicative domains speakers attend by using discourse particles, and that discourse particles are interpreted on the basis of the communicative tasks to which speakers orient regarding their hearers. These domains may be interpreted as the constituent parts of the conceptual background model of a cognitive semantic model of functional polysemy. Chapter 4 is concerned with the possibility of applying concepts from lexical semantics to the study of discourse particles in order to arrive at the lexical content of discourse particles. In particular, two main approaches are taken: the study of lexemes by means of their relations to other lexemes and semantic decomposition. The analysis of the semantic contribution of each discourse particle may motivate the different functions identified. By means of a semantic feature analysis, accompanied by semantic tests, their invariant meaning aspects, as well as their contextual meanings, can be determined. In section 5.1, a cognitive semantic model of the interaction of the meanings and functions of discourse particles will be developed on the basis of the findings from the previous chapters. Thus, the analysis of those factors which constrain the interpretation of a discourse particle occurrence in the previous chapters will lead to an integrative model.

62

Introduction: The domain

This model will account for the relationship between discourse particle lexemes and their functions and will build on a conceptual background frame and metaphorical mapping between its domains. In section 5.2 certain aspects of linguistic lexicons will be discussed which are relevant to the lexical representation of discourse particles. Section 5.3 will then provide a lexical representation of the properties of discourse particles that also accounts for the compositionality of discourse particles in combinations. The basic domains that are germane to this investigation constitute the functional spectrum; the three conditioning factors structural context, speaker aims regarding their hearers, and lexical meaning. In addition, the interaction of these domains will be considered. Table 1.9 shows the different methodological steps involved in the study of these domains and the results obtained in this investigation.

The structure of the following

63

Table 1.9: The Main Domains, Methods, and Results

chapter 2

domain functional spectrum of discourse particles

method conversation analysis

results functional trum of ja

classification in artificial neural networks

functional readings; characterisation of the role of individual structural features identification of communicative domains (the conceptual background model) translation equivalents

3

domains to which speakers attend regarding their hearers

comparative distributional and functional corpus analyses

4

relational meaning

contrastive analysis

invariant ing

5

mean-

interaction these factors

of

spec-

computational similarity detection

semantic fields / similarity maps

decomposition and semantic tests on the basis of functional analyses modelling of interdependencies by means of cognitive semantic concepts

identification of invariant and contextual meanings

formalisation

computational lexicon

linguistic model

Chapter 2 Contexts and categories: Functional interpretation In section 1.2.2, the functional spectrum of discourse particles in spontaneous spoken language dialogues was described on the basis of the functions identified in the literature; thus, the class of discourse particles presents itself as extremely multi-functional. The first step to be taken here is to determine which functions an individual discourse particle can fulfil, that is, what the functional spectrum of a particular word form is in the dialogues under consideration. Furthermore, the circumstances under which the discourse particle lexeme fulfils these functions must be analysed, i.e. what factors condition its interpretation. Therefore, the analysis of a particular discourse particle will be carried out, focussing on those properties in the analysis which become manifest in the sequential organisation of the dialogue. However, since the aim of this investigation is a lexical account, the structural contexts thus identified will then be grouped together as functional categories in order to provide a small inventory of functional readings. This descriptive inventory has to be validated as distinct readings of the discourse particle lexeme. Therefore, in section 2.2, the contribution of different contextual properties to a general functional categorization will be analysed by means of quantitative methods. It will be determined to what extent it is possible to assign a categorical interpretation to each occurrence of a discourse particle so that different functions, which can be identified by means of methods of conversation analysis (CA),

66

Contexts and categories: Functional

interpretation

may be regarded as occurring in stable feature bundles, constituting a general descriptive inventory for the contextual interpretations of discourse particles. The objectives of this section are therefore the following: - firstly, the whole range of functions which a given discourse particle may fulfil in particular contexts will be analysed; - secondly, the different contexts in which it may occur will be determined and it will be proposed how these structural contexts can be combined to form a more general functional inventory; - thirdly, the possible role of individual contextual features in the interpretation of discourse particles as they become clear as speaker categories will be analysed; - finally, the possibility of assigning a categorical interpretation to each occurrence of a discourse particle will be discussed. Before these questions can be addressed, what a discourse particle lexeme actually is must be considered. That is, it is necessary to discuss the extremely variable phonological realisation of discourse particles, and which variants of a discourse particle should be combined in order to talk about a discourse particle lexeme. Rasoloson (1994), for instance, combines /(§>/, /Ε/, /@:/, /E:/ as well as /@m/, /Em/, /@:m/, /E:m/ to form the lexeme ÄH.1 However, Fischer and Johanntokrax (1995) report that äh, involving l@l, /Ε/, /@:/, /E:/, /?@/, /?E/, /?:/, /?E:/, and ähm, involving l@ml, /Em/, /@:m/, /E:m/, /?@m/, /?Em/, /?@:m/, /?E:m/, display different distributions. For example, in human-to-machine communication, ähm occurs more often in the context of pauses on both sides (5% äh vs. 24% ähm), äh occurs more often inside constituents such as NPs and VPs (10% äh vs. 1% ähm), and äh can be found more often turn-initially than ähm (71% äh vs. 54% ähm).2 Thus, there are distributional differences between the 1

The phonological representation used here is the SAMPA notation developed by Wells et al. (1992).

2

Further distributional differences between the two lexemes will be discussed in section 3.2.

Contexts and categories: Functional interpretation

67

two forms. To what extend these have to be taken into account in the lemmatization process, that is, in the process of deciding which realisations belong to the same lemma, is not yet clear. If the difference manifests itself only quantitatively, it is of no consequence to a lexical approach; yet, at the beginning of an investigation there is no way of knowing whether the difference is only a quantitative one. While it is practically impossible to consider every realisation in its own terms, some classification is necessary. The procedure proposed here is to start with as many descriptive categories as is practicably manageable while still considering the classification as preliminary. Whether, for example, the German discourse particles öh and öhm can be assigned to the lemmata äh and ahm respectively, is an empirical question depending on functional and distributional similarities; a cyclic revision of the lemmatization process may be necessary. Another type of property of discourse particles which may directly influence lemmatization are the different prosodie realisations in which a discourse particle lexeme can be found. For instance, mhm, a back-channelling signal with positive or neutral speaker attitude, is usually realised with fall-rise intonation and can be distinguished by means of its intonation contour and its function from hm which displays beginning divergence and which is usually accompanied either by falling or by rising intonation. Thus, these differences may be taken to be decisive for distinguishing the two almost contrary functions, thereby regarding mhm and hm as different lexemes. Ehlich (1986), however, in spite of acknowledging the different functions and the differing prosodie contours, does not distinguish these two types as distinct lexical items but treats the lemma HM as polysemous, taking the phonological basis as the criterion for lemmatization, attributing the functional difference to the differences in the prosodie realisation of the same lemma. Consequently, there are no a priori reasons for a classification of different instances of a discourse particle to one or more lemmata. The example to be discussed now is German ja, and as a starting point, all of its phonological and prosodie variants will be considered.

68

Contexts and categories: Functional

interpretation

2.1 The functional spectrum of German ja In this section, the multi-functional German discourse particle ja in the human-to-human instruction dialogues (Sagerer et al. 1994) will be analysed. In section 4.1.2 the choice of this particular example will become clearer: Ja is the only German discourse particle which fulfils all of the functions the class of discourse particles may fulfil in spontaneous spoken language dialogues and can thus be seen as a central example of the class. A context in which German ja may occur is after an explicit, positively formulated question, for example, as a complete turn, in turn final, and in turn initial position: (8)

071078: hast du? [got it?] 07K078 ja. [PTC] 071079: auf beiden Seiten festgeschraubt, gut. [fixed on both sides, fine.]

(9)

071082: da ist doch jetzt noch ein Loch frei nach vorne oder? [there is still a free hole in the front, isn't there?] 07K082: nach vorne, ja [in the front, PTC], 071083: ja, gut. [yes, okay.]

( 10)

04K021 : sollte ich den mit fünf nehmen? [was I supposed to take the one with the five (holes)?] 041022: ja < - > einen mit fünf und einen mit dreien.

[ PTC one with five and one with three. ] 04K022: den mit fünf chum: stöhnen >

The functional spectrum of German ja

69

[ the one with the five ] Occurrences of this context of ja can also be found in the Verbmobildialogues: (11)

cd2_gl04a_NAR006: und dann bis Sonntag, den zehnten bleiben, meinen Sie? [and then stay till Sunday the tenth you mean?] cd2_gl04a_UTB007: ja, genau, das denke ich so [PTC, exactly, that's what I think.]

All four instances of ja occur after yes/no-questions, and in both (9) and (10), the utterances in the turns in which they occur are agreements with what the previous speaker has suggested; the relevant information is even repeated: den mitfünf, and nach vorne. In (11), ja is accompanied by another particle and a statement that what her partner has suggested is indeed what she thinks, so it also constitutes an act of agreement. As the examples show, ja in this function can occur at different positions in the turn. However, it always constitutes an independent utterance; in (9), ja constitutes a separate intonational unit from gut as marked by the comma; in (10), it is separated by means of a pause from the rest of the turn. The turns following the utterances of ja indicate that the communication partners consider their questions to be answered: In (8), the speaker summarises the situation agreed upon, in (9) and (10), items from the previous turn are repeated as well. Considering that the preceding turn is a question, that a question makes an answer conditionally relevant, that ja can be used with other signals of agreement in the same turn, that it may also constitute an entire turn, and finally that the following turn indicates that the question has been answered, it can be concluded that ja is an answer signal, indicating agreement on some propositional information, in these examples irrespective of whether it occurs with other utterances in the same turn. Consequently, although the contexts in which the instances

70

Contexts and categories: Functional

interpretation

of ja occur may vary, the function ja fulfils is in all cases the same, depending mainly on the previous speech act. In addition to the prepositional function of ja in these three examples, to express agreement between the speakers, it can be argued that its two occurrences in (9) and (10) also fulfil functions with respect to the turn-taking system. In particular, in (9) ja functions as a turnyielding signal (Duncan 1972,1974) while in (10) it is used to take the turn (Duncan and Fiske 1977). Clearly, after a question, an answer is conditionally relevant, and if this answer begins with ja, it could very well function as a turn-taking signal. However, at this point we can only say that there is a coincidence between the occurrences of ja and changes with respect to the speaker role. Whether ja functions as a turn-taking or yielding "signal" or whether its occurrence at the turn beginning or end is merely accidental cannot be answered on the basis of these examples. We will return to this point later in this investigation. The generalised structural context in which ja constitutes an answer is thus the following: 1. Preceding the use ofja, the communication partner utters a question; 2. ja, constituting a complete utterance, occurs as an entire turn or is combined with information which is in accordance with the information suggested by the communication partner's question; 3. after the occurrence of ja, the communication partner treats the question as answered. Ja in this interpretation will be referred to as an answer signal. Thus the different structural contexts in which ja may occur, that is, turninitially, turn-finally and as an entire turn, can be summarized as a single interpretation of ja because of its same function in all those contexts. Ja can also be found as a complete turn as in example (8), however, it does not occur after a question. In the following examples, it occurs

The functional spectrum of German ja

71

after requesting (example (12)), informing (example (13)) and evaluating acts (example (14)): (12)

131058: du brauchst äh eine runde orange Schraube mit Kerbe [you need uh a round orange screw with a slot] 13K057: ja [PTC] 131059: und einen < - - > Siebenerträger [anda seven-hole bar]

Ja in (12) constitutes a complete utterance; often, however, transcribers do not interpret ja as an entire turn: (13)

131070: das ist das Fahrwerk mit den Rädern vorne der Motor [this is the undercarriage with the wheels the engine in the front]

In these occurrences, the previous turn is not a question but some kind of assertive act. In examples (12) and (13),ja occurs at a transition relevance place (Sacks et al. 1974), i.e. at the possible completion point of a complete proposition and a grammatical sentence (Selting 1995, 1996). The following turn (i.e. the rest of the instructor's turn in (13)), does not refer in any way to the utterance of ja; in both cases, the instructor goes on with the instruction or information, adding something on the same topic. These examples illustrate a sequence of an asserting or requesting act, an occurrence of ja at a transition relevance place (TRP), and the continuation by the first speaker after this. A hypothesis about the contribution of ja may therefore be that it is a signal that the communication partner may continue. Furthermore, since ja is displayed at a TRP, the speaker must have followed her partner at least in so far as to identify this TRP, that is, she has either parsed her partner's utterance correctly, or she has followed the contents of her utterance, or she got at least some acoustic cue that

72

Contexts and categories: Functional

interpretation

the unit is complete, or all of this. To identify the TRP, the speaker must at least perceive her partner, that is, hear what she is saying. The reaction by her partner after the occurrence of the back-channelling signal in examples (12) and (13), however, is to add some information which is based syntactically and semantically on her previous utterance. Consequently, on the basis of this use of ja the instructor seems to assume that her partner is not only perceiving but also understanding what she is saying. This is not to say that this is really the case, i.e. that the hearer is really understanding; however, in proceeding, the instructor builds on the understanding of her previous utterance. Consequently, for these occurrences ofja, the assumption that the speakers are signalling to their partners that they are perceiving and following them is plausible. Schegloff (1982), however, holds that "understanding" regarding feedback signals such as ja in these utterances means "I understand the state of the talk" and that "extended talk by another is going on by declining to produce a fuller turn in that position. It does not claim understanding in general, but displays a particular understanding through production of an action fitted to that understanding" (Schegloff 1982: 81). Curiously, the examples Schegloff quotes as evidence occur after instances of "re-cognitional reference" (Schegloff 1982: 80), i.e. speakers use the feedback signals after they have identified the person the current speaker is talking about, that is, when they have understood the information provided by their communication partner. However, on the basis of these same examples, Schegloff reduces the function of feedback signals to those with respect to the turn-taking system, therefore labelling them "continuers" (Schegloff 1982: 87). Yet, on the basis of an analysis of other-initiated repair, which usually occurs at transition relevance places, i.e. in the same places in which the feedback signals occur, he assumes that feedback signals may have an understanding-of-the-contents reading since they "pass the opportunity to initiate repair" (Schegloff 1982: 88). This would mean, however, that silence could perform the same task, which is obviously not the case; instead, turns are interactively constructed by means of feedback signals (Levinson 1983: 337).

The functional spectrum of German ja.

73

In contrast to Schegloff's analysis of feedback signals as mere "continues," it is assumed here that understanding that someone is still not finished producing a larger unit of talk implies perceiving what the other is saying and usually also understanding the contents of what is being said. At least the instructors in the previous examples seem to assume this. In fact, Schegloff's analysis, after his consulting of otherinitiated repair as a second body of evidence, amounts to the same interpretation of the function of feedback signals. In accordance with Schegloff, the two examples of ja discussed above support the communication partners in their role as speakers; however, in the following example, the speaker takes up the turn again after a short pause: (14)

051105: so und ich hoffe, das sieht jetzt aus wie (ei)n Flugzeug. [so and I hope it looks like an airplane now.] 05K105: ja. < - > doch. [PTC. it does. ]

In example (14), the constructor reacts to the instructor's concluding evaluation of the situation with ja. However, in this example the speaker does not necessarily support the other's right to speak; after a short pause which both speakers may use to take the turn, the current speaker goes on, adding something to her own assessment, this being also in agreement with what her partner has suggested. The assignment of the turn therefore seems to be open at this point. Consequently, a feedback signal such as ja in these constructions does not invariantly mean that the other speaker may continue. In the Verbmobil dialogues, in which the recording modalities of the corpus are such that the speakers had to press a button before they were allowed to speak, this use of ja cannot be found. Yet, in the dialogues without this restriction (Jekat-Rommel et al. 1994) there are also instances of ja in this function: (15)

62FRI: wir könn' das ja erstmal vorläufig festhalten und [we can stick to that for the moment and]

74

Contexts and categories: Functional

interpretation

63CRI: ja [PTC] 64FRI: wenn wir dann genau wissen, uns genau im klaren darüber sind und genauer beurteilen können, wieviel Zeit wir dann brauchen, könn' wir ja vielleicht nochmal < P > eh zur Not noch anders umlegen, umstellen. (TP13, 6) [when we know exactly, it is completely clear and we can estimate more precisely how much time we ΊI need then, we can perhaps again

reschedule if necessary.] The linguistic context for this use of ja, which will be referred to as a back-channelling signal, therefore seems to be the following: 1. Preceding the use of ja, the communication partner utters an asserting, requesting, or evaluating act; 2. ja occurs at a TRP; 3. the communication partner may go on after this without referring to the utterrance of ja and building on the information uttered previously. In the instruction dialogues, another variant of the use of ja as an entire utterance can be found, in which it is, however, preceded by another discourse particle and a pause: (16)

031028: und damit das festbleibt, drehst du da so einen < > diesen grünen Klotz rein. oder ist egal, welche Farbe. [and for this to hold you screw this green cube into it or doesn't matter what colour.] 03K028: mhm ja. [uhuh PTC] 031029: so und jetzt noch hinter den Propeller < - > äh hinter die Flügel. [ so and now another one behind the propeller uh behind the wings]

The functional spectrum of German ja

75

In this example, the previous turn consists of the instructor's request to combine some parts of the toy-airplane. After the occurrence of ja, the instructor goes on with her instructions. The following turn by the instructor furthermore signals that she regards the first instruction to be successfully carried out (so and und jetzt noch indicate that the previous construction phase is regarded as completed). This implies that the turn containing the instance of ja must have been some sign of successful completion to her. This indicates that ja may actually refer to the action which the instructor has requested the constructor to perform. For the interpretation of this utterance of ja, one therefore has to know about the extra-linguistic situation and the action the speakers are carrying out; mhm directly follows the request and may be a signal that the instruction has been understood; the constructor's pause may occur while the action requested is performed; and with ja she signals that the action is completed and that she awaits the next instruction. Thus ja can also be regarded as a feedback signal here, however, not with respect to understanding what is being said but concerning the action requested. The following example shows that for the participants it is not always clear whether ja refers to the understanding of the instruction only or also to the successful completion of the construction. In example (17), the instructor asks immediately after the occurrence of ja in 07K007 whether ja means here that the instruction has been carried out successfully, which results in parallel speech. So even though ja is no sign of completed action here, it shows that the communication partner believes this to be possible; the understanding of ja is negotiated here between the participants. The last ja in 07K008 then refers to the completed action as well as being an answer to the instructors question: (17)

071006: nee, eine eckige also steckst du durch und drehst einen roten Würfel drauf [no, one with corners. well, you stick it through and screw a red red cube onto it.] 07K006: einen roten Würfel [a red cube]

76

Contexts and categories: Functional

interpretation

071007: einen roten Würfel [a red cube] 07K007: ja Moment [PTC just a moment ] 071008: hast du? [ got it? ] 07K008: ja [PTC] Further examples of the action reading of ja are the following: ( 18)

091069: schraubst es oben drauf ( . . . ) [you screw it onto the top] 09K068: mhm ja. [uhuh PTC] 091070: so die weiteren. [so the next ones]

(19)

101043: da legst du jetzt eins dieser ganz langen äh Stäbchen. [you put there one of these very long uh sticks wie? [uhuh how? ] 101044: und zwar äh senkrecht < - > zueinander. [that is uh vertically to one another] 10K044: ja, okay, in der Mitte? < - > die Mittlere? < - >

[yes, okay, in the middle? the one in the middle? ] 101045: das soll sozusagen der ei/ erste Teil des Flügels werden, ne [this is supposed to be the one/ first part of the wing so to say, isn't it ]

The funcñonal spectrum of German ja

77

10K045: mhm < - -> ja [uhuh PTC] 101046: und befestigst das mit einer orangenen Schraube, die eine Einkerbung < - > einer runden, orangenen Schraube, die eine Einkerbung hat < - > befestigst du das von unten mit einer orangenen Flügelschraube [and this you fix with an orange screw which has a slot a round orange srew which has a slot you fix that from the bottom with an orange thumb screw] In turn 10K043 in example (19), the constructor employs mhm in reaction to the instruction. After this, a pause follows, but while in the previous examples the instance of a feedback signal preceeded the occurrence of ja, the following question 'how' in this example shows that mhm is only a signal of successful perception here. Yet, after a clarification sequence, the instructor's reaction to ja in turn 10K045 shows that she regards the instruction as having been carried out since she goes on with her instruction, building on the previous construction results. In (20), the speaker supports the action reading of the second ja by means of a statement that she has completed the action: (20)

071096: genau, so(_ei)n Kreuz [exactly, such a cross] 07K096: ja ja habe ich. Iyes PTC got it.]

In the Verbmobil dialogues in which the speakers' task is to schedule appointments, this reading of ja cannot be found. To sum up, the context for ja as an action signal is as follows: 1. Preceding the use of ja, the communication partner utters a request to carry out some action; 2. the speaker acknowledges the successful perception and understanding of the request;

78

Contexts and categories: Functional

interpretation

3. the speaker remains silent while the requested action is carried out; 4. the communication partner regards the action to be complete after the use of ja. Another context in which ja can be found is after assertions, however, unlike the back-channeling signals, it does not constitute an independent utterance or a complete turn in this case. For instance, in example (21), the instructor asserts that the two airplanes are identical, i.e. that the task has been successfully performed, and she evaluates this situation positively. After syntactic and semantic completion of this turn, the constructor utters ja and then adds a question which expresses a positive evaluation as well. Unlike the occurrences of ja in examples (12) to (14), ja does not signal that the other may continue here since the constructor continues using the turn for herself to add more information, taking up the same topic. Ja may therefore be interpreted here as a turn-taking signal, while in the previous examples, it mainly supported the other's speaker-role:3 (21)

141062: jetzt sind sie identisch < - > sehr schön. [now they are identical very nice.] 14K061: ja, ist das nicht toll? [PTC, isn't that great?]

In example (22), ja occurs after a question, but not after that kind of questions after which ja as an answer would be conditionally relevant: (22)

05K005: Flügelschrauben, was ist das? [thumb screws, what's that? ] 051006: ja diese Schrauben zum Festdrehen. [ PTC these screws for fixing]

3

Example (21), consisting of evaluatory statements about the construction results, also shows how the participants accept the task imposed onto them by the experimental situation as their own.

The functional spectrum of German ja

79

In the following example from the Verbmobil-corpus, the speaker uses ja before uttering the second part of the greeting, thus signalling that she is perceiving and understanding her partner and that she is relating her utterance to the previous one. This use has been found to be by far the most frequent function of ja in the Verbmobil corpus (Fischer 1995). (23)

cd3_g 191 aJCAEOOO: ja, hallo, hier ist Ina Weißpflug, ich wollte fragen, wann Sie vielleicht Zeit hätten, die beiden zweitägigen Treffen zu machen. \yes, hello, this is Ina Weißpflug. I would like to ask when you have time for the two two-day meetings.] cd3 _g 191 a_SAROO1 : ja, Tag, Frau Weißpflug [PTC, hello Mrs. Weißpflug.]

Thus, ja in these contexts is not necessarily an agreement signal. The following example shows that agreement need not be involved at all: (24)

05K093: zwei < - > im ganzen. [two altogether ] 051094: ja < - > aber jetzt kannst du nämlich *däh den Würfel nehmen und die rote Schraube und jetzt paßt das weil [ PTC but now you can take the cube and the red screw and now it fits since ]

What all of these turn-initial occurrences of ja have in common is that the speakers refer to the same topic that is brought up by their communication partners. Thus, the function of ja in this kind of context seems to be to show that one is following the partner's utterances on the one hand and to relate the current utterance to the previous on the other. The structural conditions for the use of ja in this function are therefore taken to be the following: 1. Preceding the use of ja, the communication partner utters, for instance, an asserting, evaluating, questioning act, yet no yesno-question;

80

Contexts and categories: Functional

interpretation

2. the use of ja precedes an utterance which is thematically related to the previous utterance by the communication partner. This use of ja will be referred to as a take-up signal. In contrast to the use of initial ja just discussed, ja can also be found at the very beginning of conversations in the Verbmobil scenario, for instance, the first occurrence of ja in example (23). There is no previous utterance to which ja could refer. In the instruction dialogues, there are also instances of utterance-initial ja which do not refer to the previous topic4 but which introduce a new phase in the conversation: (25)

051096: von oben die rote Schraube von unten den < - > Würfel \from the top the red screw from the bottom the cube] 05K096: ja \yes\ 051097: ja das ähm jetzt kommt der nächs/ die letzte Abteilung [PTC this uhm now comes the next/ the last part]

In 051096, the instructor asks the constructor to combine some parts which the constructor acknowledges after a break, the instructor then announces in her next turn that now another construction phase begins, this turn being introduced by ja. Thus, it can be assumed that the instructor regards the previous construction phase to be complete. Another example is (26), in which ja is used not turn- but utteranceinitially: (26)

4

091001: ja, alles klar < - > ja, also Ziel ist es, < - > ein Flugzeug zu bauen. \yes, all right, PTC well, the goal is to build an airplane] Topic will be defined here as an informational unit in discourse concerned with, with respect to the corpora under consideration, individual aspects of an appointment, such as date, time, or place, or a certain construction phase in the instruction dialogues.

The functional spectrum of German ja

81

09K001: mhm. [uhuh] 091002: m < - > und zwar < - > ja dann besteht es aus einer Längsachse < - > und hinten dann < - > kann ich auch so mit Händen alles? [m in particular PTC then it consists of a longitudinal axis and at the back then can I also use my hands?] 09V001 ja, < - > ja mach(e) mal wie du willst. ¡yes, yes, do as you like.] 091003: eine Längsachse und drauf ist im Prinzip alles drauf montiert. [ a longitudinal axis and in principle everything is fixed on it] 09K002: mhm. [uhuh] 091004: und denn, ya würde sagen, wir fangen am besten mit der Längsachse an. [and then, PTC I would say the best is to start with the longitudinal axis] 09K003: ja. [yes] The sequence selected in example (26) constitutes the first orientation in the dialogue in which the instructor refers to the object to be built for the first time in order to provide the constructor with the necessary background information. Turn 091001, being the first turn in the transcript, consists of a statement that everything is settled. This is followed by a pause and a statement about the aim of the communication, which is introduced by ja. Since the participants in these dialogues are instructed about their tasks before they actually begin constructing, it can be inferred that the instructor's assertion that everything is okay concerns the conditions of the recording situation, and that the assertion about the goal constitutes the first utterance regarding the instruction task. Ja is consequently placed at the beginning of the very first

82

Contexts and categories: Functional

interpretation

orientation concerning the aim of the enterprise and thus marks a thematic break, while the speakers now turn to the task they are supposed to carry out. In the instructor's next turn, she begins with a specification, then remains silent, and then starts again naming an important part of the airplane, this utterance again beginning with ja. She then asks a question which digresses from the scenario and the conductor of the recording (V) answers her question. Turn 091004, containing an instance of ja, serves the purpose of focussing attention on the main axis of the airplane as a starting point for the construction. Furthermore, the metalinguistic statement "würde sagen" [I'd say] indicates that the speaker is re-orienting herself. The instance of ja in this turn occurs turn-medially but distinguishes the first part of the turn, which is not carried any further, from the second, which is a grammatical (though elliptical) sentence (which however does not depend on the previous part of the turn). Thus, in these examples ja does not refer to what has been previously said; in contrast, something new is introduced. This new topic is something that can be expected, namely the next construction phase in the instruction domain and a greeting in the Yerbmobil dialogues. Further examples for topic signals are the following (the latter one being from the Verbmobil corpus): (27)

041071: genau wie so(_ei)n Kreuz [exactly like a cross] 04K071 : mhm [ uhuh] 041072: ja < - > und dann nimmst du < > [ PTC and then you take ]

(28)

cd2_gl 17aJAK003_2: und danach essen, ne? [and after that dinner, okay?] cd2_gl 17aJAK003_3: ja, ganz genau, Sie sagen es. [yes, exactly, that's what I meant]

The functional spectrum of German ja

83

cd2_gl 17aJAK003_4: ja, wenn wir dann noch einen zweiten Termin vereinbaren, ich denke mal so wir brauchen auch noch für andere geschäftliche Probleme ein bißchen Zeit. [PTC maybe we are going to schedule another appointment, I think we shall need some time for other business problems, too.] Thus, the contexts for ja in which it may introduce a new construction or an appointment scheduling phase fulfil the following conditions: 1. Preceding the use of ja, a previous topic has been concluded; alternatively, ja constitutes the first utterance in the conversation; 2. followingya, the speaker introduces a new topic which concerns the predefined common task in the dialogues. This use of ja will be labelled framer. In turns 04K024 and 04K026 of example (29) also a new topic is introduced by means of ja:5 (29)

041024: und dann müßtest du noch den mit fünf übrighaben. [and then you should have the one with the five holes left] 04K024: ja genau ja, ich hab(e) eben [PTC exactly PTC, before I have] 041025: ja macht ja nichts haben es ja gemerkt [PTC doesn't matter, we have realised it, you know 6]

5

The history of this excerpt is the following: The instructors usually ask their partners first to select some parts for the construction before they begin telling them how to combine these parts. Accordingly, the instructor in this dialogue has asked the constructor to identify a number of parts (041001: du mußt erstmal die Geräte zusammensuchen). Up to turn 041007, the exchange concerns the selection of parts only. Up to turn 04K019, these selected parts are combined when the constructor realises that something has gone wrong ('ich hab(e) aber noch zwei Stäbe'). The following turns concern the construction as it should have been, and in the sequence selected here, the constructor identifies the mistake and tries to explain it.

6

The occurrences of utterance-medial ja were translated on the basis of suggestions made by Fillmore (1984).

84

Contexts and categories: Functional

interpretation

04K025: einfach < - > den noch rausgegriffen den ich mir noch nicht zur Seite gelegt hatte. [ simply taken one of those which I had not laid aside.] 041026: ja wir haben es ja

04K026: so < - > ja jetzt hab(e) ich jetzt hab(e) ich noch die eine rote Schraube mit der Kerbe. [ so PTC now I have the red screw with the slot.] Turn 041024 is a request to select a part for the construction, which the constructor acknowledges with ja genau. The first occurrence of ja in 04K024 follows a statement which needs confirmation and thus can be interpreted as a signal of successful perception and understanding, as well as agreement with the information proposed. Then the constructor begins a new utterance with ja which refers to something she has previously done and which has turned out to be a mistake: She has taken a part from a set of parts which was not previously selected and laid aside. The instructor reacts to this confession two times with utterances stating that this does not matter; both times her turns are prefaced by occurrences of ja, and both contain medial uses of ja. The instructor could use the initial occurrences of ja to signal that she has understood the constructor's problem since the sequential organisation for these initial occurrences of ja in example (29) shows the same properties as ja does in the examples (21) to (24). Consequently, these occurrences of ja in the turns 04K025 and 041026 may indicate perception and understanding to the hearer, as well as a reaction to the same topic.7 In 04K026, the previous topic is settled as basically agreed upon by means of so; this agreement is used as a common starting point for the next construction phase, which is introduced with ja. What is still unclear about this example is why the speakers apply this large number of instances of ja in this exchange; the total number 7

There are three further utterance-medial occurrences of ja the functions of which will be discussed later.

The functional spectrum of German ja

85

of occurrences of ja in these six turns is eight. If the functions of ja in these occurrences are exhaustively described as giving feedback and introducing new topics, why do so many of these signals occur, compared to other parts of the dialogue? In other CA analyses (e.g. Levinson 1983, Smith and Clark 1993) it has been found that speakers spend much linguistic effort on dispreferred utterances; for instance, after proposals, accepting acts are preferred, refusals dispreferred. The different amounts of linguistic labour involved provide evidence of this; while accepting acts are quick and straightforward, refusals occur with delay, hesitation markers, and further acts of justification and explanation. An explanatory concept which may account for this kind of preference organisation is the concept of face (Goffman 1972, Brown and Levinson 1987). According to this, it is supposed that a situation in which one of the participants has made a mistake is possibly face-threatening (one indicator being the instructor's insistence that the mistake does not matter), and thus that the occurrences of ja have a function with respect to the speakers' faces in this exchange. A further relevant source of information for the interpretation of the role of these occurrences of ja may be its other functions analysed thus far. For instance, ja has been argued to function as an answer signal, indicating agreement on propositional information, or as a back-channelling signal, indicating perception and understanding. The large number of occurrences of ja in example (29) may indicate that speakers regard signals of agreement and understanding to be particularly necessary in a situation in which understanding between the communication partners has turned out not to be self-evident. The argument would be thus that ja means what it means in other situations, namely agreement and understanding, but that it is used here more often in these functions because of the face-threatening situation. Alternatively, ja could have a particular function here which differs from signalling perception, understanding, and topic continuity or topic change. A hypothesis from this point of view could be that there is a particular contribution of ja, namely to indicate basic agreement, and that it consequently may serve directly in maintaining a harmonious interpersonal relationship. If we assume, on the basis of

86

Contexts and. categories: Functional

interpretation

the functions it already has been found to fulfil, a common invariant meaning of ja such as basic agreement, it can be hypothesised that the use of ja under consideration also is to signal basic agreement, yet not necessarily on what is said or its perception and understanding, but on the relationship between the speakers themselves. This interpretation is very plausible if the rest of the utterance is in contradiction with the information the communication partner has suggested or presupposed in her utterance, as, for instance, in a rejection of a proposal, as in the following examples from the Verbmobil corpus: (30)

cd3_m024n_MPI009: gut. dann in der folgenden Woche der Mittwoch, der dreizehnte? [okay, then in the following week the Wednesday, the thirteenth?] cd3_m024n_EVB010: ja, da kann ich nicht. [PTC, I don't have time then]

(31 )

cd2_m016n_NCW012: einunddreißigster Mai dritter Termin, okay? [the third appointment on the thirty first of May, okay?] cd2_m016n_KHS013: ja, leider gibt es bei uns da Probleme. [PTC, unfortunately there are problems from our side.]

(32)

cd3_gl43a_SOK006: das wäre vom vierzehnten bis sechzehnten Oktober. [this would be from the fourteenth to the sixteenth of October.] cd3_gl43aJBAC007: ja, das paßt mir nicht so gut. [PTC, that does not suit me well.]

In these occurrences, the previous turns are suggestions concerning appointments, and the information of the turn containing the occurrence of ja is in conflict with the information presupposed or suggested by the communication partner. These signals cannot indicate agreement on the propositional information suggested by the communication partner. They can only signal, as back-channelling signals

The Junctional spectrum of German ja

87

do, that the speaker has understood what her partner was saying and that she is taking up the same topic. This is already cooperative action; however, if this is the function of ja here, why does it not occur before every turn in the dialogues, but why can it be found in rejections, which constitute "dispreferred seconds" (Levinson 1983: 338)? On structural grounds, it can be observed that in many cases in which problems occur, discourse particles, and especially instances of ja, are used. In the current corpus, for instance, these situations are seen when one of the communication partners has made a mistake or in which the constructor does not understand the other's instructions. However, in order to explain this observation, further explanatory concepts, such as the concept of face (Goffman 1972), that the speakers' "actions be un-imposed by others" (negative face) and that their "wants be desirable to at least some others" (positive face) (Brown and Levinson 1987: 62) are necessary. As an alternative explanatory concept, Allwood et al. (1992) argue for such an interpersonally related function of back-channelling items as in examples (12) to (14) on the basis of the requirements of communication: "First, communication requires that at least two agents are willing and able to communicate. Second, communication requires that the receiving agent is willing and able to perceive the behavioral or other means whereby the sending agent is displaying or signalling information. Third, communication requires that the receiving agent is willing and able to understand the content that the sender is displaying or signalling. ( . . . ) Finally, communication requires that the receiving agent is willing and able to react attitudinally and behaviourally to various aspects of the content that the sender is displaying or signalling" (Allwood et al. 1992: 2-3).

These authors draw upon a model of communication to justify their analysis of the four functions of linguistic feedback, including the feature of interpersonal contact. On the basis of the sequential organisation of the dialogues under consideration, the question regarding the function of ja in these contexts cannot be conclusively answered. There are however other ways of studying the contributions of these conditioning factors more effectively; in order to identify the functional role of ja in face-threatening

88

Contexts and categories: Functional

interpretation

situations, other methods to investigate the speaker-hearer relationship and the invariant contribution of the discourse particle lexeme will be used. Chapter 3 will be concerned with these types of information which may contribute to the interpretation of occurrences of discourse particles. Ja can furthermore be found utterance-medially, in connection with pauses and hesitations, for example: (33)

041009: nimmst du < - > ahm < - > ja < - > den ja nimmst du < - > die gelbe Schraube mit der Kerbe. [ you take uhm PTC the PTC you take the yellow screw with the slot.] 04K009: mhm. [ uhuh]

(34)

011092: und dann hast du diese zylinderförmigen mit äh < - > ja ja diese Zylinder, diese Plastikdinger, ne. [and then you have these cylinder-shaped with uh PTC PTC these cylinders, these plastic things, okay?] 01K092: ja ja. [okay]

Note that ja is enclosed by pauses and realised hesitatingly. This shows that the first occurrences of ja in examples (33) and (34) are used while the speakers are still planning their utterances. Ja and the pauses are placed somewhere in a syntactically well-formed utterance, that is, the speaker continues after the occurrence of ja in the same way she has has begun her utterance. If ja and the pauses and hesitation markers were edited out, a well-formed sequence would be the result. This type of ja, which will be referred to as a filler, can be described as follows:

The functional spectrum of German ja

89

1. Ja occurs in the middle of an utterance; 2. it occurs in the context of pauses and hesitated speech; 3. no information is cancelled, and ja occurs inside an otherwise syntactically well-formed utterance. Instances of ja in such a context are quite rare in the corpora under consideration. More often ja occurs at syntactic restarts. In such a context, editing the respective utterances would result in two sentences, the first one of which, at least, will be syntactically incomplete. Examples are the latter two occurrences of ja in examples (33) and (34) and the following: (35)

011004: äh rote < - > ja auch Kreise mit {(ei)nem} < - > Loch in der Mitte und die steckst du da jetzt erstmal da rein, [uh red PTC also circles with {a} *really* A oh it's :l\ovelyo you ((can)) Apour it \out Agreat ¡creamy h\eado Ayou kn/owo Ajust 'like the :{r\eal} !th\/ingo (1.7)

In both examples, the function of oh seems to be to introduce an evaluation that refers to what the speaker has talked about before. As its German counterpart, oh can also be found at thematic breaks. These occurrences may be a reaction to what is going on in one's own mind, as in (243), or they may be caused by the actions performed while speaking, as in (244), (245) and (246): (243)

if you ((in"Aherit a)) uni'versity from b\ureaucratso . Awhat do you exp\ecto ( - - laughs) < 1459 B> Ay=eso. < 1460 B> A oh w=ello < 1461 B> [@] Athank you very m\ucho (1.2)

(244)

the Atrouble /iso . Aoh for 'God's s\akeso the Akey won't go in the l\ock (1.2)

(245)

AI 'don't think I've A\oho there A\is milko (1.8)

(246)

oh it's awful isn't it yes - yeah - oh. sorry I'm pouring that out the wrong way (1.7)

In example (243), the speaker concludes her topic on universities, using oh well, and then turns directly to her partner. In examples (244) to (246), the speakers produce a statement, and after the utterance of oh they refer to some action carried out while speaking. In (244) and (245), this even interrupts their previous utterances. In the next example, oh is used turn- and furthermore utterancemedially; in the utterance following oh, the information stated before is cancelled:

Semantic decomposition

(247)

231

that - ASunday br\eakfasto - - Ais s\ausages o - - Aoh n\onsense { AArab\ellao } o - ASunday breakfast is boiled \eggso * ( - - -laughs)* (1.3)

This use of oh can also be found in the Verbmobil dialogues: (248)

HH_TA_SIM_040/28/JONO 13: Umm < P > is early in/- Oh early in May actually isn't very good for me either.

The following two instances of oh occur when the speaker obviously has a speech management problem, although no information is cancelled in these examples. The first one occurs in combination with yeah, the second one is integrated in the following tone unit: (249)

*he was A[eib] he was A[ei]* oh Ay\eaho . Ahe was 'able to t/alk to them ando . (1.7)

(250)

A=and [@:m]o - [i] . oh ((there's no)) there's Ano com!p/\arisono it Atasted !so - [u: @: lai?] !w\ateryo ((you know)) Al\ifelesso

Finally, English oh can be found in a context not identified for German oh: It can be employed turn-medially before quoted speech, for instance: (251)

((I mean)) do you Aguaran"!t\ee themo he said A oh n/\oo I Adon't guaran"!t\/ee themo (1.1)

(252)

ADavid !Pr\endergasto . Aand they said \\oh_yeso ADavid !Pr\endergasto the Aman from over!s\easo (1.2)

While instances of oh (and, as will be shown soon, of other English discourse particles) are only found at the beginning of quoted speech, it is unlikely that they belong to the speech quoted, i.e. speakers do

232

Lexical analysis

not repeat exactly what the speaker quoted has uttered. Instead, oh is taken here to mark that speech is quoted from someone else. The reason that German oh was not found to be used in this function in our corpora may be either that it cannot be used in this function in German conversation, or that speakers do not quote any speech in the task-oriented dialogues investigated. Willkop also does not report such a function of oh in her corpora of informal spontaneous conversation, recorded in a cafe and a Greek restaurant among people who know each other well (Willkop 1988: 229-238). She does however report an example of German na ja which serves this purpose. So in principle it seems to be possible in German to mark quoted speech in the same way as in English. However, it is first of all not clear whether German oh can be used in this way, and secondly, although marking of quoted speech may in principle be possible in German, it may be far less common that in English. This is thus one of the 'larger' pragmatic facts where the languages differ, as discussed by Fillmore (1984). Otherwise, it can be concluded that although German and English oh differ with respect to their function in quoted speech, they are extremely similar with regard to the rest of their functional spectrum. They are also similar concerning the co-occurrence of features in the discourse functions they fulfil, such as to signal perception and understanding in the function as a back-channel.

4.2.4

Tests for the features of oh

The analysis of the English interjection oh by means of interpretative methods has provided us with a number of different functions it may fulfil that may now serve as a basis to propose invariant and contextual features for this lexeme. These will now be validated using semantic tests. For instance, as shown in examples (236) to (240), oh may function as a feedback signal. It should now be possible to formulate the functions of a discourse particle in a description based on NSM and to evaluate these supposed features of a back-channelling signal: "I hear what you are saying" and "I want you to say more about it" in an occurrence of such a signal as, for instance, the following example:

Semantic decomposition

(253)

233

a. withAout. doing too much h/arm I sup_poseo on the Acompre!h\ension - {Aqu\estiono}o < 1149 B > Aoh !y\eso, and indeed I hear what you are saying. (normal)

The test sentence results in a normal sentence. The but-test yields an odd sentence: (253)

b. withAout. doing too much h/arm I sup_poseo on the Acompre!h\ension - {Aqu\estiono}o < 1149 B > Aoh !y\eso, but actually I hear what you are saying. (odd)

In this test, redundancy is created between the meaning aspect of oh present in this occurrence and "I hear what you are saying." Thus, the sentence is evaluated as odd because the expectation of a contradiction created by but is not fulfilled. In example (253c.) the feedback signal is contradicted with the negated feature. In Cruse's test-frames, these tests render strange objects but normal sentences. In the domain of discourse particles, there are conflicting statements but normal sentences and strange speakers, for instance: (253)

c. < 1147 A > withAout. doing too much h/arm I sup_poseo on the Acompre!h\ension - {Aqu\estiono}o Aoh !y\eso, but actually I don't hear what you are saying, (contradiction)

The feature "I hear what you are saying" is therefore a feature of the utterance of oh in (). The other feature of a back-channel "I want you to say more about it" can also be shown to be present: (254) a. withAout . doing too much h/arm I sup_poseo on the Acompre!h\ension - {Aqu\estiono}o Aoh !y\eso, and indeed I want you to say more about it. (normal)

234

Lexical analysis

b. < 1147 A > withAout. doing too much h/arm I sup_poseo on the Acompre!h\ension - {Aqu\estiono}o < 1149 B > Aoh !y\eso, but actually I want you to say more about it. (odd) c. withAout . doing too much h/arm I sup_poseo on the Acompre!h\ension - {Aqu\estiono}o < 1149 B > Aoh !y\eso, but actually I don't want you to say more about it. (contradiction) Next, it will be examined whether the supposed invariant meaning of oh can be shown to be present in the examples discussed so far. On the basis of the analysis in the previous section, it is proposed here that oh invariantly signals "I now know something I didn't think of before and I feel something because of it." If this is really an invariant aspect of its meaning, it has to be present in all of the examples of oh. That this is the case will be illustrated with three examples of the different readings identified in paragraph 4.2.3, which all show the kernel meaning of oh in the tests: (255) a. in Aspite of these 1/adieso *( laughs)* *( laughs)* oh dear, and indeed I now know something I didn't think of before and I feel something because of it. (normal) b. in Aspite of these 1/adieso *( laughs)* < 1100 b > *( — laughs)* oh dear, but actually I now know something I didn't think of before and I feel something because of it. (odd) c. in Aspite of these 1/adieso *( laughs)* < 1100 b > *( — laughs)* oh dear, but actually I don't know anything now I didn't think of before and I don't feel anything because of it. (contradiction)

Semantic decomposition

235

(256) a. the Atrouble /iso . Aoh for 'God's s\akeso the Akey won't go in the l\ock, and indeed I now know something I didn't think of before and I feel something because of it. (normal) b. the Atrouble /iso . Aoh for 'God's s\akeso the Akey won't go in the l\ock, but actually I now know something I didn't think of before and I feel something because of it. (odd) c. the Atrouble /iso . Aoh for 'God's s\akeso the Akey won't go in the l\ock, but actually I don't know anything now I didn't think of before and I don't feel anything because of it. (contradiction) (257) a. that - ASunday br\eakfasto - - Ais s\ausages o - - Aoh n\onsense { AArab\ellao } o - ASunday breakfast is boiled \eggso, and indeed I now know something I didn't think of before and I feel something because of it. (normal) b. that - ASunday br\eakfasto - - Ais s\ausages o - - Aoh n\onsense { AArab\ellao } o - ASunday breakfast is boiled \eggso, but actually I know something now I didn't think of before and I feel something because of it. (odd) c. that - ASunday br\eakfasto - - Ais s\ausages o - - Aoh n\onsense { AArab\ellao } 0 - ASunday breakfast is boiled \eggso, but actually I don't know anything now I didn't think of before and 1 don't feel anything because of it. (contradiction) The previous analysis of English oh has shown that it may fulfil a number of different functions; as the function of oh as a back-channel signal could be validated, it should be possible to evaluate its other functions by means of the tests proposed. In the following examples, oh functions as a take-up and should therefore display the features "I have heard what you said" (test example (258)), "I want to say something now" (test example (259)) and "I want to say more about it" (test

236

Lexical

analysis

example (260)), corresponding to the features perception, understanding and topic continuity postulated in section 2.2 as typical features of a take-up signal. In (261), the feature of a framer ("I want to say something new now") should be identifiable, and in (262) to (263), oh functions as a repair marker, supposedly with the features "I want to say something else" and "I know now what I want to say":9 (258) a. I Alike the e:qu\ipment M/ikeo . ((it's very)) oh it's very good quality equipment, and indeed I have heard what you said, (normal) b. I Alike the e:qu\ipment M/ikeo . ((it's very)) oh it's very good quality equipment, but actually I have heard what you said, (odd) c. I Alike the e:qu\ipment M/ikeo . ((it's very)) oh it's very good quality equipment, but actually I didn't hear what you said, (contradiction)

(259) a. I Alike the e:qu\ipment M/ikeo . ((it's very)) oh it's very good quality equipment, and indeed I wanted to say something now. (normal) b. I Alike the e:qu\ipment M/ikeo . ((it's very)) oh it's very good quality equipment, but actually I wanted to say something now. (odd) 9

These features have to be transferred into past tense or present perfect in the tests if, for instance, regarding a repair marker, the utterance is already repaired or, concerning a take-up, the information has already been added by the end of the turn.

Semantic decomposition

237

c. I Alike the e:qu\ipment M/ikeo . ((it's very)) oh it's very good quality equipment, but actually I didn't want to say anything now. (contradiction)

(260) a. I Alike the e:qu\ipment M/ikeo . ((it's very)) oh it's very good quality equipment, and indeed I wanted to say more about it. (normal) b. I Alike the e:qu\ipment M/ikeo . ((it's very)) oh it's very good quality equipment, but actually I wanted to say more about it. (odd) c. I Alike the e:qu\ipment M/ikeo . ((it's very)) oh it's very good quality equipment, but actually I didn't want to say more about it. (contradiction)

(261) a. the Atrouble /iso . Aoh for 'God's s\akeso the Akey won't go in the l\ock, and indeed I am saying something new now. (normal) b. the Atrouble /iso . Aoh for 'God's s\akeso the Akey won't go in the l\ock, but actually I am saying something new now. (odd) c. the Atrouble /iso . Aoh for 'God's s\akeso the Akey won't go in the l\ock, but actually I am not saying anything new now. (contradiction)

(262) a. that - ASunday br\eakfasto - - Ais s\ausages o - - Aoh n\onsense { AArab\ellao } o, and indeed I want to say something else, (normal)

238

Lexical

analysis

b. that - ASunday br\eakfasto - - Ais s\ausages o - - Aoh n\onsense { AArab\ellao } o, but actually I want to say something else, (odd) c. that - ASunday br\eakfasto - - Ais s\ausages o - - Aoh n\onsense { AArab\ellao } o, but actually I don't want to say anything else, (contradiction)

(263) a. that - ASunday br\eakfasto - - Ais s\ausageso - - Aoh n\onsense { AArab\ellao } o, and indeed I know now what I wanted to say. (normal) b. that - ASunday br\eakfasto - - Ais s\ausageso - - Aoh n\onsense { AArab\ellao } o, but actually I know now what I wanted to say. (odd) c. that - ASunday br\eakfasto - - Ais s\ausageso - - Aoh n\onsense { AArab\ellao } o, but actually I don't know now what I wanted to say. (contradiction) Thus, all functional properties proposed for oh as a take-up or framing signal or as a repair marker were validated with the tests.10 To sum up, it can be concluded that - for oh an invariant meaning could be identified; - each occurrence furthermore displays certain contextual features depending on the discourse function it fulfils; - these contextual features are the same as those identified for German discourse particles; - the combination of the contextual features is the same as for German discourse particles; 10

In section 5.3, its function as a marker of quoted speech will be discussed.

Semantic decomposition

239

- the analyses of both English oh in natural informal conversation and German spoken language items in task-oriented dialogues yield comparable results with respect to the discourse functions of discourse particles. 4.2.5

Further English discourse particles

The findings of the study of oh will now be supported by means of analyses of three further English discourse particles: ah, well and yes. The contexts in which English ah occurs are in most cases similar to those found for English oh; there are some, however, which differ slightly from those found for oh. Ah can be found, in many cases, turn-initially after question-answer pairs: (264)

< 1396 B> and Awhat are !y\ou _theno Al'm on the :academic :c\ouncilo A\aho *((Av\ery nice po_sitiono))* (1.2)

(265)

Awho's th\ato AGeorge !F\ornbyo A{\ah} ¡George !F\ornby {Ay\eaho} o (2.4)

In example (264), speaker Β asks a question and acknowledges the answer by means of ah and an assessment. The speaker uses ah here to take the turn in order to display the receipt of the information provided by the communication partner. Speaker A in (265) furthermore repeats the answer after the occurrence of ah, thus this example supports the interpretation that ah is acknowledging the partner's answer here, taking up the information provided in the previous utterance. In other words, these occurrences of ah are typical take-up signals. Example (266) is another turn-initial signal through which the speaker reacts to the previous utterance, however, in this case he is going to answer the question himself:

240

(266)

Lexical analysis

but Archy how did you know you wanted to do research on plastics when there weren't any A\aho beAc\/auseo A well there was a ¡little :bit of !B\ake'liteo beAfore the *:w\/aro (1.12)

Thus, ah acknowledges the understanding of the previous utterance here in so far as the speaker begins to speak about the same topic, trying to answer the question posed. Ah can also make up an entire turn. In the following two examples, the communication partner immediately continues after the occurrence of ah, connecting the following utterance by means of and to the previous turn: (267)

< 1221 A > you Aknow ¡that's the !s\econd 'time that's 'happened to m/eo ah and the Afirst 'time I just rdon't 'think it was 'my f/aulto (2.5)

(268)

but I've Agot an !\eye dis_easeo . ah A/and [@:m]o it's Aall 'part of 'general ex"h\austiono and Aover!w\ork {for Aall those :y\earso}o (1.10)

Here ah occurs at TRPs, and the previous speaker immediately continues after the occurrence of ah. It thus seems to be a signal to her that she may continue, i.e. that the partner has perceived and understood her utterance and that she is following. Ah can also be used in repair contexts, for instance in connection with other lexemes: (269)

the Amoon's 'further aw/ayo from the A/earth than the s\uno A { { \ a h } "g\arbage} it's :n\oto you Akn/owo . [m] ( - *-* laughs) (1.10)

Semantic decomposition

241

The noun garbage and the following statement explicitly mark the previous information as wrong. In the next example, however, although it constitutes a restart, no information is cancelled; ah here fulfils two functions at the same time, occuring on the one hand after a false start, on the other at a thematic break. The speaker thus changes the topic after the occurrence of ah: (270)

Ayou have to [t] you Ahave to 'tune it !\ino with a Apair of "!pl\ierso the [b] - ah I Ath\ought to myselfo I'll An\ever see 'one of 'those ag/aino . "AB\/ob ((will)) 'fetch me 'one from !L\ondono (1.7)

The next excerpt is an example in which ah is used in combination with well to introduce a meta-communicative statement that concludes the previous topic: (271)

and they all Adied be'fore they :f'inished ito and so Al 'started to 'feel rather !m\/orbid a'*bout 'this*o (laughs - - -) * (laughs)* A\ah w/ello that's eAn/ough ((of 'mine))o (1.9)

Thus a framing function can be assigned to ah in combination with well in this example. In examples (272) and (273) ah occurs in the middle of utterances, however, without information being cancelled in the following utterance. It thus serves as a filler. (272)

< 165 A > Abut [=@m]o < 166 A > the the Ad\immer 'ones of c/ourseo A\are {r\eally "aho} o . Anot very 'good ma!t\erialo A [ \ m ] o - . (1.6)

242

(273)

Lexical

analysis

< 1065 A> Ajust be'low Fl\eeto beAl\ow Fl/eeto - . A\aho - ABr\ookwoodo . (1.11)

A quite common function of ah, as well as of oh, is furthermore to indicate that the speaker is now reporting someone else's speech, here in combination with well: (274)

oh it's the guests who say ah well can we help you with the washing-up now and we say *my God no* (1.12)

(275)

and Apr\esident _saido - A{/\afc} w/\ello - the Asimple :truth \/iso that. Aif you're going to boil eggs . (1.3)

Examples (269), (270), and (271) as well as the markers of quoted speech in (274) and (275) show that surprise is not necessarily a feature of ah as suggested by participants in the questionnaire study reported in section 1.4. In example (266), surprise would also be a highly unlikely reaction to the partner's question. For a repair marker such an interpretation is even more implausible. The same holds for the English interjection oh for which similar functions were found. Since the current aim is the identification of the oppositions between discourse particle lexemes so as to include these items in the lexicon, it will now be determined what the lexeme ah invariantly contributes to its contextual interpretations. The invariant meaning of ah, as can be seen from the similar functional spectrum it fulfils, is, like that one of oh, the receipt of new information. However, unlike oh, it does not display any emotional content (cf. also Wierzbicka 1992b). This is illustrated by examples (265), (267), and (271 ), as well as (268): (276) a. Awho's th\ato AGeorge !F\ornbyo A{\ah} ¡George !F\ornby {Ay\eaho} o, and indeed I know something now I didn't think of before, (normal)

Semantic decomposition

243

b. Awho's th\ato AGeorge !F\ornbyo A{\ah} ¡George !F\ornby {Ay\eaho} o, but actually I know something now I didn't think of before, (odd) c. Awho's th\ato AGeorge !F\ornbyo A{\ah} ¡George !F\ornby {Ay\eaho} o, but actually I don't know anything now I didn't think of before, (contradiction)

(277) a. < 1221 A> you Aknow ¡that's the !s\econd 'time that's 'happened to m/eo ah, and indeed I know something now I didn't think of before, (normal) b. you Aknow ¡that's the !s\econd 'time that's 'happened to m/eo ah, but actually I know something now I didn't think of before, (odd) c. you Aknow ¡that's the !s\econd 'time that's 'happened to m/eo ah, but actually I don't know anything now I didn't think of before, (contradiction) (278) a. but I've Agot an !\eye dis_easeo . ah, and indeed I know something now I didn't think of before, (normal) b. but I've Agot an !\eye dis_easeo . ah, but actually I know something now I didn't think of before, (odd) c. but I've Agot an !\eye dis_easeo .

244

Lexical analysis

ah, but actually I don't know anything now I didn't think of before, (contradiction) (279) a. (laughs - - - ) A\ah w/ello that's eAn/ough ((of 'mine))o, and indeed I know something now I didn't think of before, (normal) b. (laughs - - - ) A\ah w/ello that's eAn/ough ((of 'mine))o, but actually I know something now I didn't think of before, (odd) (laughs ) A\ah w/ello that's eAn/ough ((of 'mine))o, but actually I don't know anything now I didn't think of before, (contradiction) The utterance containing ah from example (271), for instance, also demonstrates the function of ah as a framer as shown for occurrences of oh: (280) a. (laughs - - - ) A\ah w/ello that's eAn/ough ((of 'mine))o, and indeed I am saying something new now. (normal) b. (laughs ) A\ah w/ello that's eAn/ough ((of 'mine))o, but actually I am saying something new now. (odd) c. (laughs - - - ) A\ah w/ello that's eAn/ough ((of 'mine))o, but actually I am not saying anything new now. (contradiction) Since this function has not been found for English oh, let us finally consider an analysis of ah as a filler that is supposed to carry the meaning of a turn-holding signal "I want to say some more:" (281) a. Abut [=@m]o the the Ad\immer 'ones of c/ourseo A\are {r\eally "aho} o . Anot very 'good ma!t\erialo, and indeed I wanted to say some more, (normal)

Semantic decomposition

245

b. < 165 A > Abut [=@m]o < 166 A> the the Ad\immer 'ones of c/ourseo A\are {r\eally "aho} o . Anot very 'good ma!t\erialo, but actually I wanted to say some more, (odd) c. Abut [=@m]o the the Ad\immer 'ones of c/ourseo A\are {r\eally "aho} o . Anot very 'good ma!t\erialo, but actually I didn't want to say anything any more, (contradiction) Thus, even the features by means of which discourse particle occurrences contribute to the turn-exchange system can be shown to be present in the tests. Well, besides its use as an adverb, can be found in certain idiomatic constructions, e.g. as well as and may well VERB. Furthermore, it has been suggested to mark dispreferred seconds (cf. Levinson 1983). For instance, in the following example, well prefaces a rejecting statement, it is even combined with no: (282)

unAtil No"!v\/embero - < 175 A> *. ((so it's))* Aup to th\emo *Awell ((no))* . Alet's have a look ((at [dai])) Al'm . Al'm !back on the twenty-n\inth {AR\eynardo}o (1.1)

In example (283), the speaker first disagrees with what the partner has suggested, but then reformulates this disagreement to smooth the force of this utterance. This statement is introduced with well: (283)

you Asound a bit !d\ubiouso An\oo ((Awell)) !I I get rather fed :up of :some of these/ y/oungsterso and the Aclaptrap they {t/alk} s/ometimeso - -

In examples (284) and (285) speaker Β reacts to a compliment, this utterance being introduced by means of well:

246

(284)

Lexical

analysis

Athis Athis is !\awfully k/indo *beAcause* *((well))* ΛΙΊ1 Al'll get them through *.* qu\/icklyo (1.1)

(285)

you Amust have coined ((this)) your!s\elfo Aw\ello [@] I Aput I Aput the linguistic :j\argon(l.l)

In the latter two examples, the two instances of well occur turninitially. The utterances they introduce can be said to at least highlight different aspects of the information provided by the communication partner. Thus they cannot be interpreted as instances of agreement since the speakers modify the compliments by their communication partners. Another example is (286) where the second well also introduces an answer in which Β modestly reduces her contribution by stating that the work has been done in cooperation. The first and third well in (286) occur at the beginning of questions, which also constitute potentially face-threatening acts (cf. Brown and Levinson 1987): (286)

< 1 B> ((of ASpanish)). graph\ologyo Aw=ello . < 3 A> ((if)) did Ay/ou_set_thato < 4 B> A well !J\oe andJo Aset it betw\een_uso Aactually !Joe 'set the :p\apero and *((3 to 4 sylls))* *Aw=ello . "Am/\ay* I _asko Awhat goes !\into that paper n/owo < 11 A> beAcause I ¡have to adv=iseo (1.1)

However, there are also cases in which well occurs turn-initially and in which the utterance introduced by it just adds something to what the communication partner has said. Neither case constitutes a dispreferred second or a potentially face-threatening act. In the following three examples, in which the speakers agree to their partners' assessments, well functions to take up what the other has said, to add something to the topic, and to introduce statements of agreement:

Semantic decomposition

(287)

247

*((it))* Adoesn't pattern with :\anythingo *Ay\eso* well Athat's the whole :tr\ouble (({you As=eeo}o)) (1.1)

(288)

to An\on-head {of ((a)) deAp\artmento } o well Athat's ((a)) p/\ointo as Aw/\ello \Ay/\eso (1.1)

There are a number of examples, however, in which well introduces a statement which constitutes a modification or reformulation of the partner's statement: (289)

I Athink they've got quite a good o!p\inion of himo A well [@] :I !I have t\ooo A[/m]o (1.1)

(290)

[@:m] - - you've Anot heard !P\eel_mentioned {in Athis conn=exiono}o well AN\ightingale_mentioned it {c\asually to_meo}o that APeel !m\ight try f/\or ito do you know anything Adefinite ((ab/out himo)) (1.1)

In example (289), the speaker modifies the suggested "they" to include herself. In example (290), the speaker provides the information requested, yet the source for this information was different from what the communication partner had suggested. In example (291), an answer is introduced by well. With this answer, Β modifies the partner's proposition slightly, first agreeing on one test question then adding that there were possibly two: (291)

((it's Ajust 'one !qu\estion that they have to do /isn't ito)) A well there were [@] !\/oneo or Atw\o we've *got on th/ereo* Ayou s/eeo (1.1)

248

Lexical analysis

This use of well can also be found in the Verbmobil dialogues (HH_SIM_TA_039/25): (292)

GRA024: < P > Daytime or evening? HAM025: < P > Umm < P > well, < P > I have < P > +/Tuesday/+ on Tuesday, Wednesday < P > and Friday I have the whole day free.

In this example, the speaker does not react to the partner's question directly but basically states that it does not matter to her when to meet. These examples seem to support Lakoff's analysis of well as a marker for the violation of the maxim of quantity, i.e. the speaker introduces an utterance of which she knows that she cannot provide the full or the expected answer (Lakoff 1973). To sum up, however, well could be found to introduce not only dispreferred seconds and not fully satisfying answers, but also statements of agreement and acknowledgement. Thus, a description of well as "marker of dispreferred seconds" (Levinson 1983) or of statements that violate the maxim of quantity (Lakoff 1973) does not seem to be justified. The same holds if well concludes the old topic and introduces a new one, as the following examples show: (293)

I Amean a Iman coming from well say from . :Batley to :Br\/ightono ((well then)) there'd Abe a !r\eason probablyo ((for)) . A wanting to make !th\at ch/angeo *Aw\ouldn't thereo* (1.1)

(294)

[@:m] you're Avery kind old S/amo - - "Abl\ess_youo well Athat finishes th\ato . [@:m]. ((now)) Awhat was the \other thing {I Awanted to :=ask_youo}o. (1.1)

In these examples, there is no sense of insufficiency or dispreference (cf. also Svartvik 1980). The readings of well discussed so far are thus taken to be take-up and framing signals, depending on whether they continue on the same topic or introduce a new one.

Semantic decomposition

249

Furthermore, well may be found utterance-medially as a marker of reformulation. In the latter example, the previous information is indeed cancelled, thus marked as insufficient, however, in the former, well in combination with scry just introduces a specification: (295)

I Amean a !man coming from well say from. :Batley to :Br\/ightono (1.1)

(296)

the A[ouw] the Aonly . the Aonly !\other . possib/ilityo . [@:m] Aw\ello . not the A\only other possib/ilityo . but it Aseems to me that you've got :two . _two th=ingso (1.1)

As English oh and ah, well can also be found as a marker of quoted speech: (297)

Ahow much !\are theyo he said Athey're thirty for half a cr/\owno - - ((Aw\ell she _saido)) that Adoesn't make "!s\enseo that's the Acheapest of the !l\oto A((he said well you can)) do what you !l\ike a_bout/ ito but . if you Ause these according to di:r\/ectionso/ I can Aguaranîtee they're !!s\afeo (1.1)

It can be concluded that in contrast to German ja and English yes, well is not a clear signal of agreement. However, it is not a MARKER of problematic information either. It was shown that it may also introduce preferred seconds with agreement readings. In addition it is not clear why anybody should mark that she is going to say something the partner will not like. The functional spectrum of well can only be accounted for if it is understood as a signal of conclusive evaluation, and in a semantic continuum with its 'homonym' the adverb. The meaning suggested for well is therefore one of conclusive evaluation, signalling furthermore that several different (possibly even contradictory) premises are taken into account: "after I have thought about all I know about it I say this."

250

Lexical

analysis

(298) a. Athis Athis is !\awfully k/indo *beAcause* *{{well))* ΛΓ11 Al'll get them through *.* qu\/icklyo, and indeed I say this after I have thought about all I know about it. (normal) b. Athis Athis is !\awfully k/indo *beAcause* *((well))* Al'll Al'll get them through *.* qu\/icklyo, but actually I say this after I have thought about all I know about it. (odd) c. Athis Athis is !\awfully k/indo *beAcause* *((wel[))* Al'll Al'll get them through *.* qu\/icklyo, but actually I say this after I have not thought about all I know about it. (contradiction)

(299) a. [@:m] you're Avery kind old S/amo - - "Abl\ess_youo well Athat finishes th\ato and indeed I say this after I have thought about all I know about it. (normal) b. [@:m] you're Avery kind old S/amo - - "Abl\ess_youo well Athat finishes th\ato but actually I say this after I have thought about all I know about it. (odd) c. [@:m] you're Avery kind old S/amo - - "Abl\ess_youo well Athat finishes th\ato but actually I say this after I have not thought about all I know about it. (contradiction) (300) a. the A[ouw] the Aonly . the Aonly !\other . possib/ilityo . [@:m] Aw\ello . not the A\only other possib/ilityo, and indeed I say this after I have thought about all I know about it. (normal)

Semantic decomposition

251

b. the A[ouw] the Aonly . the Aonly !\other . possib/ilityo . [@:m] Λw\ello . not the A\only other possib/ilityo, but actually I say this after I have thought about all I know about it. (odd) c. the Afouw] the Aonly . the Aonly !\other . possib/ilityo . [@:m] Aw\ello . not the A\only other possib/ilityo, but actually I say this after I have not thought about all I know about it. (contradiction) While the tests yield the results expected, it may be objected that they have not reliably shown the respective feature to be present. After all, this feature is so general that it may not be wrong of any utterance. However, on the one hand even if the information expressed by well was redundant, this would not be uncommon in natural languages; for instance, the pragmatic idiom I think can be added to any utterance and yet speakers use it only sometimes. On the other hand, the tests can show that using the negated suspected meaning of well in a test does not produce a contradiction in connection with any utterance: (301)

This is a nice house, but actually I say this after I have not thought about all I know about it. (normal)

This statement is normal; the connective but actually may be paraphrased best by something like "but to be honest." Thus it is not true that the property proposed for well is potentially present in any utterance. Besides its invariant semantic aspects, in the following example well as a take-up shows the same functional features as, for instance, oh did as a take-up in examples (258-260): (302) a. ((it's Ajust 'one !qu\estion that they have to do /isn't ito)) A well there were [@] !\/oneo or Atw\o we've *got on th/ereo*, and indeed I have heard what you said, (normal) b. ((it's Ajust 'one !qu\estion that they have to do /isn't ito))

252

Lexical

analysis

Awell there were [@] !\/oneo or Atw\o we've *got on th/ereo*, but actually I have heard what you said, (odd) c. ((it's Ajust 'one !qu\estion that they have to do /isn't ito)) Awell there were [@] !\/oneo or Atw\o we've *got on th/ereo*, but actually I didn't hear what you said, (contradiction)

(303) a. ((it's Ajust 'one !qu\estion that they have to do /isn't ito)) Awell there were [@] !\/oneo or Atw\o we've *got on th/ereo*, and indeed I wanted to say more about it. (normal) b. ((it's Ajust 'one !qu\estion that they have to do /isn't ito)) Awell there were [@] !\/oneo or Atw\o we've *got on th/ereo*, but actually I wanted to say more about it. (odd) c. ((it's Ajust 'one !qu\estion that they have to do /isn't ito)) Awell there were [@] !\/oneo or Atw\o we've *got on th/ereo*, but actually I didn't want to say more about it. (contradiction) (304) a. ((it's Ajust 'one !qu\estion that they have to do /isn't ito)) Awell there were [@] !\/oneo or Atw\o we've *got on th/ereo*, and indeed I wanted to say something now. (normal) b. ((it's Ajust 'one !qu\estion that they have to do /isn't ito))

Semantic decomposition

253

Awell there were [@] !\/oneo or Atw\o we've *got on th/ereo*, but actually I wanted to say something now. (odd) c. ((it's Ajust 'one !qu\estion that they have to do /isn't ito)) Awell there were [@] !\/oneo or Atw\o we've *got on th/ereo*, but actually I didn't want to say anything now. (contradiction) Thus, well can be concluded to display "I say this after I have thought about all I know about it" invariantly in all of its occurrences. As an example for one of the functional readings of well it was shown that as a take-up it indicates the same context-dependent meanings as the other discourse particles in this function. Yes can often be found after the discourse particle right with rising intonation as a complete turn, as the following example shows. After speaker A's utterance, which is concluded with a signal that makes some kind of feedback relevant, speaker Β reacts with an instance of yes with rising intonation. Speaker A seems to interpret this occurrence of yes as a feedback signal since she continues her assessment, connecting her utterance to the following using and. After a short pause, speaker Β acknowledges the information given by means of a second occurrence of yes, in this instance combined with an explicit statement of agreement, this time with falling intonation contour and in second position after a hesitation marker: (305)

at the Aexaminers' m\eetingo - - ((Ar/ighto)) ((Ay/wo)) ((and)) I Athink that went very :w\ello — [@:h] - /\y\eso Al Al ¡think it d\id on the wh/oleo (1.1)

254

Lexical

analysis

The first use of yes that is interpreted by the communication partner as a signal to continue can also be found in the Verbmobil corpus; in the example (HH_TA_SIM_041/27), it is followed by another signal of agreement and a hesitation marker. After a short pause, the communication partner continues: (306)

DOU025: So it would probably be better for me

in June, the beginning of June, perhaps. LUU026: < P > Yes. Of course. Umm/DOU027: < P > What do you suggest?

The following example shows yes turn-initially after the checking signal you see but with falling intonation contour. This use of yes is followed by another occurrence of yes: (307)

well ANightingale :said that he . he :might want to/ :get a: way from . Lower N\etherhall you s/eeo Ay\eso - - Λy=eso I [@] I Awouldn't be sur!pr\ised at th/ato - - I Areally !w\ouldn'to (1.1)

After the occurrence of you see, speaker A uses yes to signal that she is still following, strengthening this by means of another use of yes. That neither instance of yes necessarily means agreement to a proposition can be seen from the utterance following yes, with which speaker A asserts that she does not know about it but would not be surprised if it was the case. Thus, yes serves here to signal that the speaker has heard and understood her communication pamer's utterance and that she is continuing on the same topic. In the following example, speaker A acknowledges with yes I see the partner's utterance that there were one or two questions. With two further instances of yes speaker A reacts to her partner's utterance that ends in you see with rising intonation. These occurrences of yes obviously constitute a concluding evaluation of the situation and therefore complete the topic, since the speaker explicitly announces another topic after these multiple occurences of yes:

Semantic decomposition

(308)

255

Λ well there were [@] !\/oneo or Atw\o we've *got on th/ereo* Ayou s/eeo *Ayes !I s\eeo* . Λy=eso . Λy=eso - [@:m]. Aone \other thing S/amo - (1.1)

The following examples of yes from the Verbmobil dialogues (HH_TA_SIM_041/27 and HH_SIM_042/26) occur in reaction to a clarifying question. They can thus be considered as answer signals: (309)

LUU047: Uh would it be < P > good for you to meet in the center < P > of Hamburg? DOU048: Yes. Hm LUU049: So umm < P > maybe we could meet < P > at umm < P > the Alsterpavillon?

The fact that the communication partner continues after the occurrence of yes hm with a concluding so, building on the speaker's agreement in the following utterance, implies that the occurrence of yes expresses agreement with the propositional information suggested by the communication partner here. In the following example, yes seems to fulfil the same function since the speaker repeats her partner's utterance. This is acknowledged with okay: (310)

JON039: That's good? WEL040: Yes, that's good. JON041: Okay.

In of the following example, speaker A does not continue with her utterance, an attempt to continue her partner's statement. After a short pause and yes with falling intonation, she yields the floor to her partner. Her communication partner completes the utterance, which A again acknowledges with yes: (311)

you'll Afind that it !is 'this qu\/estiono . *((which))*

256

Lexical analysis

*on A which* they are . Ay\eso they're "Ap\assingo on *A((that))* !qu\estiono *Ay=eso* < 1065 (B> **-** and it's Anot a question of :fifteen **exA\actlyo** marks out of a :h\undredo (1.1) The first instance of yes thus may be interpreted as a filler, occurring in the context of pauses and within an utterance, however, it also signals to her partner that she may continue. The second instance of yes is a feedback signal, acknowledging the expression proposed by her partner as the expression she was looking for. Example (312) also shows many different uses of yes\ the first two are feedback signals by speaker A the agreement of which is then diminuished by quite. However, speaker A then continues with two further instances of yes, the latter in combination with the interjection oh. After this, A explicitly states agreement, concluding this with another instance of yes. However, what follows is a contradition to what Β has previously said. Thus, yes is used here several times in order to express perception and understanding; the speaker seems to worry about the degree to which her feedback signals may also be taken as signals of agreement since she keeps qualifying this agreement reading. In the end it becomes apparent that the previous occurrences of yes do not express clear prepositional agreement: (312)

((Athen)) !this immediately :brings \ino the Afunds **.** of another di!v\isiono *Ay=eso* **/\y=eso** Ay=eso . Ay=eso ((A/\oh -.yeso)) Al [kwa:] I Aquite a!gr=ee with _youo Athis [@] Athis was made !v\ery _clearo ((A;y\eso)) [@:m] - Abut ( . clears throat) !what 'Oscar [@ dh@V @] this iAd\ea of Oscaro Ag\oing {to the ASt=ateso}o (1.2)

Semantic decomposition

257

As German ja and English well, English yes can occur in connection with dispreferred utterances. In (313), Β reacts with yes as a backchannelling signal, however, also trying to take the turn to state a restriction to her agreement: (313)

is sort of Abegging for the !m\oono - - ((you As/eeo)) Ay\eso [@:m] *Abut [i i]* *Athis is* Athis is :one of the :things that - [@] (1.1)

As it has been argued in connection with well, the fact that these discourse particles occur preceding dispreferred seconds does not necesssarily mean that they 'mark' dispreferred seconds. On the contrary, in order to maintain a positive speaker-hearer relationship, speakers employ signals indicating agreement in these positions especially when this relationship is endangered. The meaning proposed f o r j e s is therefore: "I think that we both think the same." This feature is shown to be present in the following examples: (314) a. ((and)) I Athink that went very :w\ello - - [@:h] - Ay\eso, and indeed I believe that we both think the same, (normal) b. ((and)) I Athink that went very :w\ello [@:h] - Ay\eso, but actually I believe that we both think the same, (odd) c. ((and)) I Athink that went very :w\ello — [@:h] - Ay\eso, but actually I don't believe that we both think the same, (contradictory) (315) a. well ANightingale :said that he . he :might want to/ :get a: way from . Lower N\etherhall you s/eeo

258

Lexical analysis

Λy\eso, and indeed I believe that we both think the same, (normal) b. well ANightingale :said that he . he :might want to/ :get a: way from . Lower N\etherhall you s/eeo Ay\eso, but actually I believe that we both think the same, (odd) c. well ANightingale :said that he . he :might want to/ :get a:way from . Lower N\etherhall you s/eeo Ay\eso, but actually I don't believe that we both think the same, (contradictory) (316) a. they're "Ap\assingo on *A((that))* !qu\estiono *Ay=eso*, and indeed I believe that we both think the same, (normal) b. they're "Ap\assingo on *A((that))* !qu\estiono *Ay=eso*, but actually I believe that we both think the same, (odd) c. they're "Ap\assingo on *A((that))* !qu\estiono *Ay=eso*, but actually I don't believe that we both think the same, (contradictory) 4.3

Consequences for lexical representation: Invariant meanings

To sum up, for all four discourse particles discussed, different invariant meanings were proposed. These were evaluated using semantic tests. The contextual features discourse particles display when they fulfil a certain function were also shown to be present in the particular occurrences, no matter which lexeme it was. Therefore, the tests developed make it possible not only to validate analyses of the semantic

Consequences for lexical representation

259

content of a discourse particle, and therefore to compare their meanings, but they also allow to test the context dependent features. These are not only related to propositional information, but also to domains such as the flow of information, discourse structure, or speech management. Even in the case of a filler, the turn-holding function was demonstrated via the tests. For all English discourse particles the same combinations of features in their structural contexts could be identified, as for German discourse particles. Basically, the same descriptive inventory seems to be relevant for both English and German items with the exception of some intonation contours. In particular, English and German differ here in that English back-channels were occasionally found with rising intonation. The inventories differ, however, concerning the extensive use of English oh, ah, and well as markers of quoted speech. This use could not be found in the German corpora under consideration (and on which was only rarely reported with respect to different German corpora by other scholars). Thus, oh, ah, well, and yes have been found to function as take-ups, oh, ah, and yes as back-channels, oh and well as markers of reformulation, ah and well as fillers, and finally oh, ah, and well as framers, repair markers, and as quotation markers. With the exception of the back-channelling signals, all of these fit in the same general structural contexts as proposed as a descriptive inventory for German discourse particles in section 2.3. It can be concluded, on the one hand, that for English and German discourse particles the same co-occurrence relations can be assumed between the different functions that constitute their contextual interpretations. On the other hand, the sets of discourse functions are different for the two languages; for English discourse particles a further construction quotation marker is necessary. Which discourse particle can fulfil which discourse function has to be regarded as conventionalised and therefore language-specific. The same is true of the structural properties of the constructions. The differences between the two languages are, however, surprisingly minor. The following chapter shows how the invariant aspects of discourse particles identified here interact with the other factors that influence their interpretation.

Chapter 5 Lexical representation In this chapter, the findings of this investigation with respect to the interpretation of discourse particles will be tied together in a unified model. Since the aim of this study is a lexical representation of discourse particles, a linguistic model that accounts for their functional polysemy will be developed. This model show that the different lexical items considered can indeed be regarded as belonging to a single class since they can be described by means of a unified model. Next, different aspects of the lexicon will be discussed that influence the representation of discourse particles. Finally, their lexical representation in a computational frame- and construction-based lexicon will be presented.

5.1 A unified model of the meanings and functions of discourse particles In the last three chapters, three concepts were introduced that were found to be of relevance for the description of discourse particle lexemes: constructions that specify the structural contexts in which discourse particles may occur, a communicative background frame that specifies those aspects of the communicative situation to which speakers attend by means of discourse particles, and thirdly, the lexical, invariant, meaning aspects of each lexeme. It now has to be discussed how the different aspects of this model interact and how exactly the information involved is distributed be-

262

Lexical

representation

tween these concepts. A model of the interaction of these conditioning factors has to explain the relation between the invariant meanings of discourse particles, the different domains to which they may refer, and the constructions in which they occur. Such a model, which locates the discourse particle lexeme in a space between the linguistic frame and possible structural contexts (Fillmore and Atkins 1992: 101), will now be developed.

5.1.1

The contextual meanings of discourse particles

The first aspect that needs to be explained is how a given discourse particle with its invariant meaning may get its contextual interpretations. It will be proposed here that the context-dependent meanings of discourse particles are created by means of the reference of their invariant meanings to aspects of particular communicative domains. The mechanism assumed is similar to the one developed in Sweetser (1990); she argues that conjunctions like and, or and i/have very abstract meanings that are so general that they apply to three different metaphorically related domains, the sociophysical, the epistemic and the speech-act domain. In her approach, the conjunctions get their pragmatic interpretation in reference to one of these domains. Likewise it is assumed here that discourse particles carry an abstract, schematic meaning which is contextually specified by means of reference to a particular communicative domain. Sweetser (1990) supports her identification of the three domains that she assumes to be metaphorically related by evidence from lexical semantic change as well as from the study of polysemy and pragmatic ambiguity. That the three domains that she assumes are indeed cognitively relevant is supported by the fact that she can successfully apply her analysis to a number of different linguistic phenomena, such as English perception verbs, modality, or conjunction. In contrast, the domains assumed here as a background against which discourse particles are interpreted have been determined as speaker categories in chapter 3. That is, their cognitive relevance was demonstrated by showing that speakers attend to them, depending on their communication partners. Although the method of determining the domains of reference differ, the mechanism

A unified model of discourse particles

263

through which the schematic meanings of discourse particles become contextually specified in reference to these communicative domains is however comparable to Sweetser (1990). This will now be illustrated regarding the English interjection oh, the German hesitation markers äh and ähm, as well as the German segmentation marker ja. It has been argued in chapter 4 that the meaning of English oh is "I now know something I didn't think of before and I feel something because of it." Furthermore, oh has been found to function, for example, as a feedback signal with the features "I have heard what you have said" and "I want you to say more about it." The relationship between these propositions can now be explained as follows. Since the meaning of oh consists in the receipt of new information, oh may function as a feedback signal if the information received originates in the communication partner's utterance. The invariant meaning aspects of oh may therefore refer to the perceptual domain such that the under-specified meaning of oh "I now know something... " is specified by the perception of what the communication partner has just uttered. This may fulfil the function of giving feedback to the partner. The relationship between the invariant meaning of oh and its contextual features is thus explained by reference to an aspect of the communicative background frame, in particular to the perceptual aspects of the information flow, which speakers have been found to orient to in chapter 3. The contextual meaning of oh is thus "I have heard what you just said and I feel something because of it." The strategic function that results from signalling this kind of information is maintaining the flow of information, if this is a task to which the speaker attends. The back-channelling function of oh, signalling "I have heard what you have said," is thus licenced by the task to ensure successful perception in the flow of information in the dialogue. The polysemy of discourse particles in general can be accounted for by their reference to communicative domains, such as perception, propositional information or the interpersonal relationship between the speakers. These domains have been identified in section 3.3.2 as relevant to the speakers depending on their aims regarding their communication partners. With respect to English oh, its invariant meaning to signal the recognition of something new may also refer to the under-

264

Lexical

representation

standing of the information provided by the communication partner. Thus, it may signal "I understand what you are saying." Furthermore, besides referring to what the communication partner has uttered, as demonstrated in the analysis of oh in section 4.2.3, its invariant meaning "I now know something... " may also refer to something that does not originate in the communication partner's utterances. In particular, the reference elements may be three different types of information: the action context (e.g. the key not going into the lock, or pouring out milk the wrong way), the discovery of an insufficiency in the speaker's own speech, or remembering a topic one still would like to talk about. In each case, the something in "I now know something... " is filled in with reference to a certain communicational domain identified in section 3.3.2 as one of the domains relevant to the language users. With this mechanism, the functions of oh as an up-taking and as a feedback signal, a framer, a repair marker, and its reference to the extra-linguistic situation can be accounted for. By signalling emotional involvement by means of the latter part of the meaning paraphrase: "I feel something because of it," speakers may furthermore use oh to maintain a harmonious interpersonal relationship, as in the following example from the Verbmobil corpus: (317)

mkps_7_02: well, it looks like, um, mm, hm, next Monday and Tuesday, um in the morning would be fine, or even Wednesday in the morning. < P > um after that I'll be out of town for a bit though. flmb_7_03: oh goodness this is going to be interesting to schedule, because, I have class, oh, Monday in the morning, from nine until twelve, and, Tuesday in the morning, I've got a meeting from ten to twelve.

Here, oh refers to what the partner has just said; however, in signalling emotional involvement, the speaker shows that she does not reject her partner's proposals without some feeling of, for example, regret. What kind of feeling this is is however neither specified by the invariant meaning of oh nor by the communicative domain to which it refers. Instead, it depends on the speaker's aims, and these may become apparent through the prosodie realisation.1 1

A complete, more formal, analysis of oh will be presented in section 5.2.

A unified model of discourse particles

265

Let us now regard the hesitation markers äh and ähm. Both indicate that the speaker is currently thinking. In the communicative situation of conversational interaction, this signal can refer to different communicative domains about which the speaker is currently thinking. It can therefore be strategically employed in certain constructions, depending on what tasks the speaker regards to be relevant. For instance, if äh or ähm are used at the beginning of the turn, "I am thinking" may refer to what the communication partner has just uttered, thus signalling "I'm thinking about what you just said," and indicating at least perception of what has been said. This may serve different purposes: For example, showing that one is thinking about what the communication partner has uttered may be useful in face-threatening acts to demonstrate that one is not going to hurt the other's feelings by thoughtlessness (Brown and Levinson 1987). It may also indicate that one is not certain about how to answer, and is withholding, for instance, a dispreferred reaction. Finally, it may demonstrate that the contents of what is going to be said are negotiable (Smith and Clark 1993). Initial äh and ähm may also refer to a new topic. Similar to English oh, the German hesitation markers display mental processes that may lead to a change in the thematic structure. The thinking process then refers to something that came into one's mind that, for instance, still needs to be discussed. Unlike English oh, the hesitation markers do not display sudden recognition but a current thinking process. As repair markers, German äh and ähm refer to the formulation process, for instance, the search for the right construction. Between informational units such as "possible sentences" (Selting 1995), hesitation markers are used to mark informational boundaries by displaying planning breaks with respect to the next informational unit. These planning breaks also indicate linguistic structure to the hearer and facilitate her understanding (Fischer 1999). The function hesitation markers fulfil before nouns has already been discussed in section 3.2: If the maintainance of the interpersonal relationship is a task to which the speaker orients, marking a noun or a noun phrase can reduce the force of the utterance. This is achieved by indicating that what is said is at disposition. By displaying uncertainty ("I am still thinking about it"), the speaker presents the content

266

Lexical

representation

of the utterance as negotiable. So the different functions the hesitation markers fulfil are motivated by their invariant meaning, while their polysemy, and therefore their poly-functionality, arises from their reference to different communicative domains such as the understanding of the previous utterance, a new topic, or the management of one's own speech. Now let us finally return to German ja. Like English yes, it can be assigned a simple invariant meaning that expresses a relation between the speaker and the hearer: "I think that you and I think the same." The polysemy of ja results from the reference of this meaning to different communicative domains. First, ja may be the answer to a question. In a question-answer pair, a reaction by the hearer with respect to the information asked for is made relevant by the speaker's question. Ja then signals that the hearer believes the same proposition to apply to a certain state of affairs as the speaker suggests in her question. Here the agreement relation expressed by ja refers to the propositional information proposed by the communication partner. In contrast, when the speaker understands the flow of information with regard to the perception and understanding of the transferred message as a domain she attends to, ja may be used as a back-channelling signal. In this case, it does not necessarily refer to the proposition uttered by the partner; instead, the agreement relation is interpreted with respect to the flow of information. What is expressed by ja is the identity of the informational status of the communication partners on the basis of what has been said. This signal may serve to indicate "I am following" or "I understand what you are saying." The relation of accordance expressed by ja therefore refers to the successful transfer of information; if securing the informational flow is a task to which the speaker attends, the occurrence of ja will thus function as a feedback signal. In the same way, the semantic content ofja may refer to the speakerhearer relationship. Here, 'thinking the same' concerns the basic relationship between the communication partners. Thus, the different reference regarding the interpersonal and the propositional level ac-

A unified model of discourse particles

267

counts, for example, for the different readings of ja after the proposal for a date in the utterances, as in the following example:2 (318)

Wie wäre es am Donnerstag? a. ja da kann ich. b. ja da kann ich nicht.

In (318a), ja signals perception and understanding of the partner's question and additionally refers to the proposed information itself. In (318b), it also signals agreement, however, not concerning the proposition since the proposal is rejected, but regarding the speaker-hearerrelationship which is signalled to be basically harmonious (Holly 1979: 210). That is, ja signals here agreement regarding the interpersonal relationship BECAUSE OF the disagreement on the propositional level. Here, the model can explain why signals of positive speaker attitude occur in "dispreferred seconds." Instead of having to classify them as "markers of dispreferred seconds" and therefore to introduce two distinct readings of ja, the model proposed here allows a unified picture of its meanings and functions. In these examples, ja provides feedback on the partner's utterance, signalling that the information is successfully transferred. At the same time it directly relates the current utterance to the previous, signalling that both speakers are talking about the same thing. And finally, it may signal a relationship of accordance between the communication partners directly. In these readings, the reference elements are the domains perception and understanding, the topic structure, and the interpersonal relationship. For ja as a signal of completed action, as found after requests to combine parts in the human-to-human toy-airplane construction scenario, its invariant meaning concerns the action which is supposed to be carried out. Thus, the agreement relation refers to the execution of an action requested such that there is identity between the action requested and the action performed. 2

This example was constructed in order to compare the different contributions of ja in the same context; however, comparable, attested, cases were discussed in section 2.1.

268

Lexical

representation

In turn-final ja with rising intonation, it is the agreement between speaker and hearer that needs affirmation: with respect to perception and understanding, some propositional information, or the interpersonal relationship. Thus, the checking signal functions as a question of whether the agreement relation expressed by ja holds in the case under consideration. As shown in section 2.1, this agreement relation may refer to different types of information; it may concern propositional information, the flow of information, or the speaker-hearer relationship. In section 2.1 it was furthermore argued that ja may refer to the speech management domain. It was shown that ja as a repair marker usually occurs directly before the repaired item, rather than immediately after the reparandum. On the basis of the model proposed here it can now be argued that the function of ja in these occurrences is not to mark a speech management problem, as, for instance, oh does, but first to indicate that the speaker knows now what to say, i.e. that there is a relation of identity of what is being planned and what is going to be said, and second that there is a relation of accordance between what has been said before and what is going to come. Thus ja does not mark a reparandum as wrong, even though information may indeed be cancelled. The relationship of accordance invariantly expressed by ja therefore does not refer here to an aspect of the relation between speaker and hearer, but to an accordance of speech planning and speech production processes. As Ducrot (1996), for instance, shows, to construe different aspects of the same speaker as different 'voices' is a common phenomenon in language. The German modal particle ja can also be characterised with the model proposed. Unlike discourse particles, however, modal particles do not refer to aspects of the communicative situation, including the structure of the conversation itself. Instead, they refer to the pragmatic pretext, i.e. a proposition which is not explicitly mentioned but which is "at hand" because it is perceptionally salient to both speakers or because it can be presupposed to be generally available background knowledge; it makes up a common ground for the speakers (for details cf. Diewald and Fischer 1998). Ja signals here an accordance relation with respect to the proposition at hand and the current utterance

A unified model of discourse particles

269

(the German modal particle also, in contrast to this, signals a consequential relationship between the pragmatic pretext and the current proposition). The contextual meanings of discourse particles can thus be explained by reference of their invariant meanings to certain communicative domains. Their polysemy is therefore determined by the number of domains to which they can refer. Consequently, in order to be restrictive enough, the model proposed has to constrain the number of communicative domains involved. It is therefore important that the domains relevant here have been identified on independent grounds in chapter 3 and that they constitute not just a random selection of descriptive levels but a coherent whole, a structured conceptual background frame that reflects what speakers consider to be relevant tasks in the communicative situation. Thus, discourse particle meanings refer to a conceptual model of the aspects relevant to speakers in a communicative situation, depending on which tasks they want to perform regarding their hearers. Thus, the communicative domains identified are part of the structured frame on the background of which discourse particles are interpreted.3 To conclude, after having discussed these three examples of the ways that the invariant meaning aspects of discourse particles create certain context-specific interpretations by referring to different communicative domains, a general schema for the contextual readings of discourse particles can now be proposed: invariant meaning (DOMAIN) = > contextual meaning

This model, which can be paraphrased as "in reference to a communicative domain, the invariant meaning of a discourse particle constitutes a certain contextual meaning," allows the creation of a set of possible interpretations for each discourse particle, and its polysemy depends on the number of domains identified as relevant. 3

The domains involved in this model are basically those identified in the comparative analyses of chapter 3, however, the turn-exchange features cannot be related to the invariant properties of a discourse particle, and furthermore the expressive function is also located on a different descriptive level, as will be shown in the next paragraph.

270

5.1.2

Lexical

representation

From contextual meanings to discourse functions

The contextual meanings of discourse particles are not identical with the functions they fulfil; that is, the relationship is not direct. On the one hand, whether discourse particle occurrences may fulfil certain functions depends on whether the speaker attends to the respective domain regarding a certain task. On the other hand, several signals may be used to fulfil the same function, i.e. to perform the same task. Consequently, not every occurrence of a discourse particle with a particular contextual meaning creates a particular pragmatic function, nor is a particular communicative function necessarily caused by a particular contextual reading of a discourse particle. When the speaker regards maintaining an interpersonal relationship as an important task in a certain situation, for example, if she is going to reject a proposal by her partner, she may use several different discourse particles with reference to different communicative domains to achieve her goal. Thus, if she wants to signal that the rejecting act she is going to perform does not endanger the harmonious relationship between herself and her communication partner, she may signal that she takes the proposal seriously, that is, she will show that she has heard and understood her partner's utterance and that she takes up the same idea to add something to it. She may do so by means of several different discourse particles which refer to the domains perception, understanding, and informational structure so that their contextual meanings are signals of successful perception, understanding, and continuing of the previous topic. Yet, there are alternatives to reach the same end: The speaker may also signal emotional involvement, using, for instance, oh to show that she regrets that she has to reject her partner's proposal. She may also signal that she has been thinking about the proposal, using a hesitation marker, and thus is not taking the decision easily. She may also refer to the interpersonal relationship directly, employing a discourse particle whose contextual meaning signals basic agreement regarding the speaker-hearer relation. Consequently, the speaker's intentions can be realised by a number of different discourse particles in reference to different communicative domains. Likewise, the relationship between discourse particles and their functions depends on the tasks the speaker wants to fulfil. In chapter 3

A unified model of discourse particles

271

it was found that the use of a discourse particle may, but need not, have a strategic, i.e. partner-oriented purpose. In monologic speech, for instance, speakers use discourse particles only in their expressive function, i.e. without aiming at attaining a partner-oriented goal. However, the reference of the meaning of a discourse particle lexeme to a certain communicative domain is independent of the question whether the speaker considers a signal in the respective domain as strategically suitable. While discourse particle meanings are contextually specified by reference to particular communicative domains in all of their occurrences, they do not necessarily fulfil strategic functions with respect to these domains. Only if a certain domain constitutes an area to which the speaker attends such that it constitutes a task she wants to perform, does she use a certain discourse particle with a strategic function. Thus, a discourse particle that in one occurrence segments information into the units that are easiest to grasp for the hearer need not do so in another one, depending on what the speaker aims at with regard to her hearer. The conditioning factor that determines the function of a contextual meaning of a discourse particle is therefore the speaker's aims regarding the hearer. These aims can be specified as partner-oriented tasks directed to the respective domain. A task is consequently defined here as a speaker aim directed towards a certain communicative domain. These tasks regarding certain aspects of the communicative situation condition the use of a discourse particle; for instance, if speakers regard managing their speech as an important task, they will use discourse particles as repair markers and devices for gaining time for planning their speech. If they want their partners to understand, they may use discourse particles to structure and package the information in a way for their partners easiest to grasp. If they view the maintainance of the interpersonal relationship as an important task, they will try to show their partners by means of discourse particles such as yes, yeah in English and ja in German that they basically agree with them, especially when this interpersonal understanding is endangered by refusing, requesting, or proposing acts. The different functions discourse particles fulfil therefore have to be interpreted against a background of these tasks.

272

Lexical

representation

In task-oriented human-to-human dialogues, such as those from the instruction and the appointment scheduling domain, it can be assumed that speakers are oriented to all of the communicative domains concerning the fulfilment of specific hearer-oriented purposes since the common task requires a certain amount of cooperation between the speakers. The same seems to hold for English discourse particles in informal face-to-face conversation, as has been shown in section 4.2. In these types of dialogues, it can be taken for granted that speakers attend to certain goals with respect to the communicative domains and thus mean discourse particles not just to express their mental processes, as in monologic speech, but also to fulfil a number of partner-oriented purposes. However, as was shown in chapter 3, in the communication with a (simulated) automatic speech processing system this does not apply. Thus while discourse particles in particular constructions fulfil certain pragmatic functions, one must also consider that they can be used without a strategic purpose. In section 5.2, a way to account for the different functions that a contextual meaning of a discourse particle can fulfil will be proposed. So far it has been shown that the invariant meaning of a discourse particle in reference to a certain communicative domain can fulfil a certain function if the speaker attends to this domain while performing a certain task. In chapter 2, several of these functions were found to cooccur in more-or-less stable feature-bundles, and furthermore paired with certain surface properties. Thus, conventionalised co-occurrence relations could be identified for both functional and surface aspects of occurrences of discourse particles. It was suggested that these pairs of general surface and interpretative features are representable as constructions. These constructions were called discourse functions. The contextual meanings of a discourse particle combined in a discourse function are not the only interpretative features. In order to identify the contribution of the discourse functions themselves, two discourse particles, both of which may be typically used as take-ups, ja and oh, will be considered. The functions combined in a take-up were proposed as the following: - a signal of successful perception; - a signal of successful understanding;

A unified model of discourse particles

273

- a signal that one is going to take up the information provided by the communication partner; - a turn-taking function. Ja and oh fulfil these functions in different ways: Ja signals successful perception and understanding by displaying a relation of accordance between the information uttered by the communication partner and the information processed by the hearer, while oh displays the receipt of new information. Furthermore, in referring to what the partner has said, speakers of both ja and oh display that they take up the same topic. However, the relevance of what is going to come to what has been previously said is demonstrated for ja and oh in different ways. For ja, it signals that one is basically saying the same thing, just adding something to the previous information. For oh, it displays an immediate cognitive reaction. These two features may furthermore make ja and oh become turn-taking signals: One may be justified to take the turn if one either does not interrupt the speaker because one is basically saying the same thing, or if one shows a sudden, spontaneous reaction that refers to what the previous speaker has said. The proposal for the role of discourse particles regarding the turn-taking system is thus that they are accounts of the uses speakers make of transition relevance places (see Fischer ip). Unlike the other functions of discourse particles discussed, their turn-taking function is not created by contextual specification of the schematic meaning to a particular communicative domain; there is no direct relationship between the invariant meanings of discourse particles and signals such as "I want to say something now." Instead, the turn-taking interpretation evolves from other contextual meanings of an occurrence of a discourse particle. The discourse functions proposed in section 2.2 thus also involve functional features that are not directly related to the signalling of a contextual meaning of a discourse particle. Consequently, on the one hand, the interpretative information of a discourse function consists of the functions created by a selection from the contextual meanings of a discourse particle. On the other hand, interpretative information such as turn-taking features, emerge from the construction itself. Consequently, "I want to say something now," i.e. a turn-taking signal, is

274

Lexical

representation

not a contextual meaning of ja or oh, but an emergent feature of the combination of features in the discourse function take-up. The nature of a construction such as a discourse function can be illustrated further using the example of the checking function. Here the contribution of the construction becomes clearest since in a checking signal the contextual meanings of a discourse particle are not statements but can be thought of as questions that make the communication partner's feedback relevant. Thus, a discourse function not only combines different form-related aspects with the functions based on the contextual readings of a discourse particle, it may also provide a contribution to the interpretation by carrying further emergent functional components such as turn-exchange features and by providing a structure for the interpretation of the contextual meanings of discourse particles. To conclude, the discourse functions, conventional form-meaning pairs, which apply if the speaker attends to the respective communicative domains as relevant tasks in the conversation, define cooccurrence relations between different functional, and between functional and surface features. Although the turn-exchange features are not directly signalled by the discourse particle, i.e. they are not contextual meanings of discourse particles, these features emerge from the combination of functional and surface features in the respective discourse function. A discourse function, carrying formal and functional aspects, can thus be looked at as a construction as defined in construction grammar (Fillmore and Kay 1995, Fischer 1996c). To sum up, what a discourse particle means in a certain occurrence depends on the following conditioning factors: - the meaning of the lexeme; - reference to a certain communicative domain; the result is the contextual meaning of the discourse particle; - whether the speaker attends to the domain with the goal to perform certain tasks; this results in the contextual function; - finally, the discourse function in which the respective discourse particle occurs.

A unified model of discourse particles

275

The interaction of these different sources of information in their contribution to the interpretation of a discourse particle occurrence is summarised in figure 5.1: The lexical meaning of a discourse particle refers to a communicative domain, thereby licensing a particular contextual meaning. Several of these contextual meanings may constitute the interpretative side of a discourse function. These are paired with particular typical surface properties, if, parallel to the relationship between lexical meaning, communicative domain and contextual meaning, the speaker aims refer to the respective communicative domains such that she attends to a particular communicative task. If speakers orient to these domains, the contextual meanings get a particular functional interpretation with respect to this communicative domain.

5.1.3

The different word classes

On the basis of the general mechanism through which a discourse particle lexeme gets its context-dependent interpretation in a particular utterance situation developed here, a definition for the class discourse particle can be proposed, and different subclasses can be distinguished. Common to all are not only their under-specified invariant meanings, which are contextually specified by reference to a certain communicative domain on the basis of a model of communicative tasks to which speakers attend, but also the nature of their lexical meanings, which all report on a speaker's mental state. This does not mean that they really express the speaker's cognitive processes but that they are signals of the mental aspects that can be employed strategically (Fischer 1999). The different subclasses can be distinguished with regard to the kinds of processes involved: Interjections, such as English oh, ah, oops and German oh, ah, ach, aha, all display the sudden recognition of some kind of information, i.e. in NSM terms, they all contain an "I now"-component. Hesitation markers however do not display a sudden change in the speakers' mental states but instead indicate a current process: "I am thinking." Segmentation markers, which were shown in this investigation to fulfil many more functions than just segmenting information, can be divided into two different kinds: those whose meanings directly involve the communication partner (e.g. for

276

Lexical representation

eω n Έ υ o•ι θ! υ «ts Μ Ρ

C o 'Ü O

&

e _o υ c points to the node-path pairs at which the respective discourse functions are represented. Thus, each discourse particle is linked to those template nodes that specify the cooccurrence relations of their functional and surface properties. Since it is the lemma that determines which discourse functions the respective discourse particle is going to fulfil, at the lemma node a pointer can be found that lists all of the respective discourse functions. The path thus contains the list of all discourse functions that the German interjection oh may fulfil. All together, the lexical entry for German oh looks like the following: Oh:



Interjection ('orthographic representation oh') ('phonological variants loi, /ο:/, Hol, Πο:Γ)





< action >



( Ί know now sth I did not think of before and I feel sth because of it') (Backch:, or Framer: < fune >, or Take-up:, or Action:, or Repair:) ( Ί have heard now what you said') ( Ί have understood now what you said') ( Ί now know something I want to say about what you have said') ( Ί know now what is happening') ( Ί know now that I have said something wrong') ( Ί feel something') ( Ί know now sth new I want to say')

302

Lexical

representation

==

==

("" , or "" 'and I fear sth because of it') ('phonological representation /? o ? o:/').

Since DATR aims at the minimisation of redundancy, those features that are shared by all discourse particles or subclasses of discourse particles, such as interjections, can be inherited from more general nodes than the lemmata themselves. Thus, the lemma node for each discourse particle needs to represent only idiosyncratic information. An example of such a more general node is INTERJECTION: Interjection:



== == ==

Discourse_particle interjection ('mental state: "I now" - component').

This node is only part of an inheritance hierarchy, pointing to another, one level higher, node itself. Besides the information that more properties are to be inherited from the root node DiSCOURSE_PARTICLE, the node represents the word class specific properties of interjections, as discussed in section 5.1.3. In addition to lemma nodes that represent the information that belongs to each lexical item and the more general nodes such as INTERJECTION that describe the properties of classes of items, the description to be developed also has to account for the constructions in which discourse particles may occur. As argued in section 5.1, the linguistic constructions identified, the so-called discourse functions, combine different contextual meanings of the respective discourse particle with particular structural properties and furthermore carry functional and surface features. Thus, a discourse function specifies the actual function a discourse particle occurrence displays, as well as its typical position in turn and utterance and the intonation contour. These nodes can be described as templates, i.e. as those nodes at which linear precedence and co-occurrence relations are defined. Consider, for instance, the discourse function TAKE-UP:

A frame- and construction-based

Take-up:







lexicon

303

Discourse_particle ( , , , , , < turn-taking > , typically ) ("" ' = > maintains channel') (" " ' = > ensures flow of information') ("" ' = > saves speakers faces') ("" ' = > takes up previous topic') (takes the turn) ( , , < inton_contour> ) (fall or fall-rise intonation) (turn-initial position).

At nodes representing discourse functions, the respective surface features are listed at the path < s u r f > . These features were identified in the CA analysis and then verified as typical of instances of discourse functions by the artificial neural network classifier. For take-ups these surface properties are position with respect to turn and utterance as well as a particular intonation contour. Yet for answer particles the relevant structural feature is the previous speech act. The values for the paths in the list are evaluated at the same node; if information is not found, it is evaluated at the more general node DISCOURSE.PARTICLE to which the empty path points. For instance, < s u r f > points to another path . Since discourse particles typically occur utteranceinitially, this information is represented at that more general node. The respective functions connected to the construction can be found as values of the path . The paths evaluated there point to the contextual meanings of the respective lexical item that are inherited from the lemma nodes by global inheritance, i.e. that are contributed by the respective discourse particle and that motivate the functional

304

Lexical representation

meaning conveyed by the discourse particle in the respective construction. Those functions that are peculiar to the construction itself are not inherited from the respective lemma node but are specified at the discourse function node itself. This is particularly true of the turn-taking properties of discourse particle occurrences, which are dependent on the contruction in which they occur. As with the surface features, the respective discourse function may also point to types of information that are neither specified at the individual discourse particle node nor by the construction itself, but that are so general that they apply to all or most discourse particles. An example is the segmentation function. This information can also be found at the node DISCOURSE_PARTICLE, which is accessed by means of the empty path of both the lemma nodes and the template nodes, such as the discourse functions: Discourse_particle :





< intonation >

==

0 Functions: ("") Comm_domains: ('mental state: "I" - component'), particle ('segments chunks of information') Redup.particle (utterance-initial position) (intonationally not integrated).

The root node DISCOURSE-PARTICLE does not point to anything else, i.e. the empty path leads to an empty list. At this node all the properties are coded which are true of all or most discourse particles. It would be redundant to state them for each individual discourse particle or at each discourse function node. These properties are, besides the word class specific semantic property to contain an indexical component pointing to the speaker, the syntactic category, the segmentation function, and the typical syntactic position. There are also a number of path-value equations that provide the mechanisms for the automatic

A frame- and construction-based

lexicon

305

interpretation. These devices, in particular the paths , and , will be discussed in more detail during the discussion of the other nodes that are involved in the inferential processes. There is an inheritance hierarchy based on generalisation, i.e. INTERJECTION is more general than OH, and DISCOURSE_PARTICLE is more general than INTERJECTION. These are thus hierarchically ordered in a simple inheritance hierarchy. However, the nodes discussed so far make up, and are embedded in, a complex inheritance network. The nodes involved are connected by links of generalisation, the inheritance hierarchy, and by links of specification. For instance, the information evaluated at a quoted path is passed on to the respective lemma node by means of global inheritance. An example are the quoted paths in the discourse functions, such as TAKE_UP. TAKE.UP is represented in ILEX as a template node, as such it makes use of global inheritance, i.e. a specification relation where those kinds of information are passed on to the template node that are specified at the lemma node where the query started, i.e. the node for which the information was requested. This mechanism now allows to specify the contextual meaning of a particle occurrence if it refers to the communicative domain perception, but this domain is not attended to by the speaker. Such a query would make use of, for example, the path "". This path is not specified at the node OH and is thus passed on to the node INTERJECTION and from there to DISC0URSE_PARTICLE via the respective empty paths. Here, the first part of the path can be evaluated, the information that is found there points to the node COMM.DOMAINS, and the rest of the path queried is carried on to that node: Comm_domains:

==

==

< express action >

==

0 ("" '(perception) = > ' "") ("" '(understanding) = > ' "") ("" '(action) = > ' "")

306

Lexical

representation

< express macrostr>

==

< express speechmanage >

==

< express proposition >

==

==

==

==

("" '(macro structure) = > ' "") ("" '(speech management) = > ' "") ("" '(proposition) = > ' " < proposition > ") ("" '(information structure) = > ' "") ("" '(interpersonal relation) = > ' "") ( , , , , typically Take-up: , or < express action >, or < express speechmanage>, typically Repair: < surf >).

The node COMM.DOMAINS provides a list of the different communicative domains that have been identified as attended to by the speakers. This provides the frame of reference to which the meanings of discourse particles relate in order to create contextual meanings. Here the invariant meaning of the discourse particle queried is inherited into a structure that expresses the function developed in section 5.1, namely: invariant meaning (DOMAIN)

= > contextual meaning

At the template node COMM_DOMAlNS this function is specified by means of the invariant meaning of the respective lexical item, the respective communicative domain that is coded locally at the node itself and the respective contextual meaning that results from reference of the invariant meaning to the communicative domain under consideration. The theory may be queried, for instance, for all of the contextual meanings a discourse particle may express. In this case, the invariant meaning and the contextual meanings of each discourse particle are retrieved. A query to the representation may however also

A frame- and construction-based lexicon

307

be OH:, for instance, if only a particluar discourse function is of relevance. For such a query, the node FUNCTIONS is necessary through which the individual discourse functions can be accessed directly. If the query is OH:, is passed on to DISCOURSE.PARTICLE via the empty

path where it is evaluated as a pointer to this node. Here, the respective discourse function is evaluated. The discourse particle queried is the global environment and the values of OH are instantiated at the node TAKE-UP.

Functions:







== == == == == == == == == ==

0 Take-up: Framer: Action :< fune > Ans wer : < fune > Modal: < func> Repair: Check: Β ackch : < fune > Quotation : < fune >.

To sum up, the model described in section 5.1 is formalized here in such a way that for each discourse particle what it expresses if it occurs without any partner-oriented function in monologic discourse can be retrieved separately. In addition, the functions it fulfils in cooperative partner-oriented dialogues in both the instruction and the appointment scheduling domain can also be retrieved. While the invariant meaning aspects, as well as those contextual properties that are local features of each lexeme, are represented at the respective lemma node, the relevant contextual meanings and the resulting functions are identified by means of inheritance relations, The relevant information is thus collected at the template nodes representing the communicative frame, and at the constructions specifying the structural contexts in which discourse particles may occur. The template nodes thus provide the final set of contextual meanings of each discourse particle, their typical structural properties and their partner-oriented functions.

308

Lexical

representation

The following theorem is the result from accessing the theory by the query O H : < F U N C T I O N T A K E - U P > : Oh: = as a take-up: I know now sth I did not think of before and I feel sth because of it (perception) => I have heard now what you said => maintains channel, I know now sth I did not think of before and I feel sth because of it (understanding) => I have understood now what you said => ensures flow of information , I know now sth I did not think of before and I feel sth because of it (information structure) => I now know something I want to say about what you have said => takes up previous topic , I know now sth I did not think of before and I feel sth because of it (interpersonal relation) => saves speakers faces , segments chunks of information , takes the turn , typically utterance-initial position , turn-initial position fall or fall-rise intonation In accordance with the model developed in section 5.1, the query result is structured in the following way: invariant meaning

(DOMAIN)

= > contextual meaning = > function

The theorem derived from the representation has the following structure: First of all, the invariant meaning of the lexeme is stated, followed, in brackets, by the communicative domain to which it refers, and the resulting contextual meaning, indicated by an arrow. After this, the typical partner-oriented function is described which the contextual meaning may fulfil, also introduced by an arrow. At the end of the query, the typical surface properties are listed.

A frame- and construction-based lexicon

309

Another example for a discourse function is CHECK, in which the contextual meanings of the respective discourse particle are formulated in the form of a question. The construction thus contributes to the functional interpretation: Check:

< perception > < understanding > < proposition >





Discourse_particle (< perception > , < understanding > , < proposition > , , , 'typically' ) ('Is it true that' " , ) (utterance-final position) (turn-final position) (rising intonation).

Like the discourse function node TAKE-UP, the template node representing the discourse function CHECK combines properties inherited from the respective lemma nodes with functional and surface properties that are peculiar to the construction itself. At this node, the principle of default and override can furthermore be illustrated since surface features that are specified at the more general node DlSCOURSE_PARTICLE are 'overridden' with an entry at this more specific node. In particular, the more general information is that discourse particles typically occur utterance-initially while checking signals occur utterance-finally. Although both nodes specify conflicting information on the same path, no conflict arises since the information at a more specific node, the discourse function, which points to the root node

310

Lexical

representation

DISCOURSE.PARTICLE, is taken to be more relevant than the default information specified at the general node. So far it has been shown how the model of the functional polysemy of discourse particles developed in section 5.1 can be formalised in an ILEX/DATR lexicon. The representation still has to account for the fact that discourse particles may occur in certain combinations. For instance, they may occur in reduplicated form. Since the information on the interpretation of reduplication can be similarly described for all discourse particles, the general node DISCOURSE-PARTICLE contains a pointer to the node at which this information is specified: the node REDUP_PARTICLE. This node states linear precedence relations, in particular regarding the orthographic representation of the combination, and idiosyncratic properties such as the combinatorial intonation contour. In contrast, the function and the phonological representation have to be specified at the particular lemma nodes since lexemespecific, conventionalised, information is involved in the reduplication of discourse particles: Redup_particle:

== ==



== == ==

Discourse_particle ("" , , , ) ("" and "") ("") (rise-fall or falling intonation).

If a query is made with respect to features of a particular reduplicated discourse particle, such as OH:, the query is passed on over the empty path via INTERJECTION to the root node DlSCOURSE_PARTICLE. Here the pointer to the REDUP_PARTICLE node

is evaluated. At that node, the functional and the formal, i.e. the orthographic and phonological, information about the respective discourse particle lexeme is inherited globally from the node at which the query started, i.e. OH, which constitutes the global environment in such a query. This information is built together at the node REDUP-PARTICLE. Since this node defines linear precedence relations between items, it is considered as a template node.

A frame- and construction-based

lexicon

311

Oh: = I know now sth I did not think of before and I feel sth because of it (understanding) => I have understood now what you said , or I know now sth I did not think of before and I feel sth because of it (action) => I know now what is happening and I fear sth because of it orthographic representation oh and orthographic representation oh , phonological representation /ο ? ο:/, rise-fall or falling intonation Reduplication is, however, not the only form of combination of discourse particles. As argued in the previous section, a common combination is between an interjection and a segmentation marker, such as oh and ja. Again, the node lNTERj_SEGM_COMB defines the relation between discourse particles, in particular, combinations between interjections and segmentation markers, by referring to the features of the constituent parts of the combination using global inheritance: Interj _segm_comb :

==

==

Discourse_particle ("" "" , "

< speechmanage>

lexicon

313

( Ί want to add something to what you are saying') ('we both think that we should talk about sth new') ('we think the same about what to do') ('we think the same') ( Ί now know what to say') ('phonological representation /jA ja:/') ( and 'but it is not new to me').

If oh and ja now occur as a first and second part in an interjectionsegmentation marker construction respectively, the contextual meanings of each of them are retrieved from the respective lemma nodes and combined at the template node LNTERJ_SEGM_COMB. The surface features are typically those of a take-up signal. The information on the combination of oh and ja can be retrieved by means of the query OHJA:

:

Ohja: = I know now sth I did not think of before and I feel sth because of it (perception) = > I ve heard now what you said I think that you and I think the same (perception) = > I hear what you are saying , I know now sth I did not think of before and I feel sth because of it (understanding) => I ve understood now what you said I think that you and I think the same (understanding) => I understand what you are saying , or I know now sth I did not think of before and I feel sth because of it (action) => I know now what is happening I think that you and I think the same (action) => we think the same about what to do or I know now sth I did not think of before and I feel sth because of it (understanding)

314

Lexical

representation

=> I ve understood now what you said I think that you and I think the same (proposition) => we think the same , typically utterance-initial position , turn-initial position, fall or fall-rise intonation The functioning of the reduplication template can now also be illustrated on ja, thus showing that the representation proposed here accounts for the combinatorial properties of different discourse particles in a general way. The model proposed here is therefore capable of providing the descriptive basis for the actual and potential properties of particular instances of the word class discourse particle, that is, it accounts for the properties modelled in section 5.1. Consequently, lexical information on the structural and interpretative aspects of discourse particles, as well as their combinatorial behaviour, can be provided by the formalisation proposed. The theorem that is derived as an answer to the query Ja: < R E D U P > thus uses idiosyncratic as well as general information from the template nodes: Ja: = I hear what you are saying and I understand what you are saying but it is not new to me , orthographic representation ja and orthographic representation j a , phonological representation /jA ja:/ , rise-fall or falling intonation So far, different possible functional interpretations of discourse particle occurrences have been identified when they occur in particular discourse functions or in certain combinations. Yet, discourse particles may also be used without a strategic purpose when they are not directed at the communication partner, as in monologic discourse. On the basis of the findings of chapter 3 it was argued in section 5.1 that whether a contextual meaning of a discourse particle fulfils a par-

A frame- and construction-based lexicon

315

ticular function depends on the speaker's aims regarding her communication partner. Thus if the speaker attends to a certain domain concerning a particular task, the respective discourse particle can be assigned a strategic purpose, as specified at the discourse function nodes. However, discourse particles can also be used without a strategic function, for example, in monologic discourse where the speaker does not direct her speech towards the communication partner but speaks to herself. It is proposed here that in this case discourse particles only signal their contextual meanings, without a particular strategic function regarding the hearer. Still, in these cases the invariant meaning of the discourse particle lexeme refers to the different communicative domains. In the representation proposed here, the contextual meanings of discourse particles can be retrieved without a functional interpretation if in the queried path the information expressed is requested. In order to be able to access all of the possible contextual meanings of a respective discourse particle if they do not occur in particular discourse functions, the path at the template node COMM_DOMAiNS combines pointers to the local paths of the lemma nodes with their typical surface form. The example provided here will be the properties expressed by German ja : Ja: = I think that you and I think the same (perception) => I hear what you are saying , I think that you and I think the same (understanding) => I understand what you are saying , I think that you and I think the same (information structure) => I want to add something to what you are saying , I think that you and I think the same (macro structure) => we both think that we should talk about sth new , typically utterance-initial position , turn-initial position , fall or fall-rise intonation , or I think that you and I think the same (action) => we think the same about what to do , I think that you and I think the same (speech management) => I now know what to say , typically utterance-medial position, falling or integrated intonation

316

Lexical representation

As a take-up, for instance ja fulfils the same functions regarding the hearer as the interjection oh does. Yet, they both achieve these goals differently; for instance, while oh takes up the information provided by the communication partner by signalling that the speaker is suddenly aware of some information, ja does so by signalling that the speaker basically thinks the same and only wants to add something to what has been said. Thus, while the functional interpretation in both cases is the same, the contextual meanings that cause a particular function are not identical. Compare the results from the query JA: with those for OH::

Ja: < function take-up > = as a take-up: I think that you and I think the same (perception) = > I hear what you are saying => maintains channel I think that you and I think the same (understanding) => I understand what you are saying => ensures flow of information I think that you and I think the same (information structure) => I want to add something to what you are saying => takes up previous topic , I think that you and I think the same (interpersonal relation) => we think the same => saves speakers faces , segments chunks of information , takes the turn , typically utterance-initial position , turn-initial position , fall or fall-rise intonation The different contributions of oh and ja can also be compared when they both function as repair markers. Oh: = as a repair marker: I know now sth I did not think of before and I feel sth because of it (speech management) => I know now that I have said something wrong typically utterance-medial position, falling or integrated intonation

A frame- and construction-based

lexicon

317

In the function as repair markers, both signal different information due to their different invariant meaning aspects. That is, by indicating the recognition of something new, oh marks a previous utterance as insufficient or the information transferred as wrong, while ja as a repair marker signals that the accordance relation between speech planning and production is reconstituted. Ja: < function repair> = as a repair marker: I think that you and I think the same (speech management) => I now know what to say , typically utterance-medial position, falling or integrated intonation The examples discussed so far have been about two German discourse particles. Ja was chosen for discussion here because it had been investigated thoroughly in chapter 2. German oh was chosen because of its combinability with ja and its comparable properties regarding the functions as take-up and repair marker, as well as its occurrence in reduplicated form. Now, an English discourse particle will be discussed. It was found in chapter 4 that English discourse particles fulfil the same discourse functions as their German counterparts, with the exception that many of them may also be used to introduce quoted speech. Thus, the functions of English oh are taken to differ from those of German oh here only in one respect: English oh may also function as a marker of quoted speech and may therefore inherit from German oh everything but the range of functions and certain idiosyncratic aspects, such as its phonological realisation. Oh_e:

==



== == ==

Oh (Oh: , Quotation : < fune > ) ('phonological variants /ou/, loi, Hol, /?ou/') ('phonological representation /?o ?ou/').

318

Lexical

representation

Another discourse function has to be introduced to account for the functions of English oh, namely that it can occur before quoted speech. This is also a function of English ah and well. Otherwise, English oh inherits all of the other functions from its German counterpart.10 The relation between German and English oh is however not one of dependency, it is just an economical way of treating the phenomena under consideration. DATR allows here to constitute an ontologically neutral inheritance hierarchy just on the basis of different levels of specificity. The inheritance relation regarding English and German oh could thus just as well have been the other way around, the choice being a matter of simplicity only. In principle, however, on the basis of the structuralist assumption that each language is a self-contained system, defined by its internal oppositions, it is of course undesirable to mix the constructions of two languages. The following node represents the discourse function quotation marker: Quotation:



Discourse_particle ('states quoted speaker is going to continue the topic', , '=> signals other speaker is speaking', 'typically' ) Framer:.

Querying the representation regarding English oh in the function as a quotation marker yields the following result: Oh_e: = states quoted speaker is going to continue the topic segments chunks of information , => signals other speaker is speaking , As argued in chapter 4, there is a minor difference between the English and the German sets of constructions such that English back-channelling items can also occur with rising intonation while German back-channels can only be found with falling and falling-rising intonation contours.

A frame- and construction-based

lexicon

319

typically utterance-initial position, fall, fall-rise, or rise-fall intonation To sum up, it was shown that the functional polysemy of discourse particles can be accounted for in the lexical representation proposed. This applies to the occurrences of discourse particles in monologic discourse as well as in discourse functions when the speaker attends to certain tasks regarding her communication partner. This was achieved by formalising the interaction between the three conditioning factors: structural context (chapter 2), communicative domains (chapter 3), and invariant meaning (chapter 4). Furthermore, it was shown that the representation also holds for combinations of discourse particles, thus accounting for compositional relations in the lexicon. It can also be concluded that DATR, interpreted by the ILEX concept, provides a useful representation language to account for the information and the relations specified in section 5.1. Thus, the representation is a formalisation of the interdependencies described in the unified model of the meanings and functions of discourse particles developed throughout this investigation.

Chapter 6 Conclusion and prospects 6.1 From cognitive semantics to lexical pragmatics In this investigation a model was developed that accounts for a number of complex observations regarding the distribution and interpretation of discourse particles, and, in particular, for their functional polysemy. This model integrates different informational resources: the contribution of each lexeme, the contextual properties of each discourse particle occurrence, and a background model of the domains to which speakers attend regarding their hearers by using discourse particles. In accordance with Fillmore and Atkins (1992), each discourse particle occurrence is defined relative to "syntax," i.e. the structural position in which it may occur, and to a conceptual frame. This frame consists of the communicative domains to which speakers attend regarding their communication partners. The frame proved to be suitable not only for the description of the polysemy of discourse particles (in analogy with Sweetser ( 1990)), but also for the definition of the word class discourse particle. In the analyses of the contextual properties that condition the interpretation of a discourse particle occurrence, conversation analysis was used as a starting point. That is, an approach to conversational data was used that restricts itself to those aspects that become interactively manifest as those aspects to which speakers orient in the sequential organisation of the dialogue. The properties determined are thus identified as speaker categories, relevant in the interaction between speakers and

322

Conclusion and prospects

hearers. These categories are not only domains of social interaction but can also each be understood as a conceptual domain of particular cognitive relevance. The conceptual framework with respect to which the functional polysemy of discourse particles is explained is therefore identified through an empirical analysis of the interaction between speakers in discourse. Methodologically, the procedure taken here has been to develop the conceptual background frame structure on the basis of results from empirical analyses of what speakers want to achieve regarding their communication partners. The approach taken is to start from cognitive semantic concepts, to extend it by a further methodological dimension by identifying cognitive categories with speaker categories as evident in talk-in-interaction, and thus to arrive at a lexical pragmatic approach to the functional polysemy of discourse particles. Consequently, the current investigation has shown how cognitive semantic approaches to language, mostly oriented to lexical semantic knowledge, can be used to shed light on the functional variability of items whose functions are mainly located in the pragmatic domain, as well as the other way around: It was shown that methods from the analysis of talk-in-interaction can contribute to establishing cognitive semantic concepts such as a cognitive background frame. The methodology employed in this study is therefore part of the point made. Besides providing a method for developing cognitive frames, further innovative steps in methodology were made regarding the inherent restrictions that apply to conversation analysis as a method to analyse lexical items. These restrictions can be overcome using a combination of conversation analysis and categorical approaches to discourse. Before, no method had been proposed that connected these two perspectives. In this study, a computational method was proposed that bridges the gap between the data and results from conversation analysis and the more general categories of discourse analysis (see section 1.4). In particular, while CA analyses provided the speaker categories regarding the functional spectrum of each discourse particle, computational methods were used to bridge the gap between the many individual observations in the CA analysis and a DA-style descriptive inventory. These also provided insights into the role of individual contextual features in the categorical interpretation of instances of discourse particles.

Automatic processing of discourse particles

323

The results from contrastive analyses of modern plays and their translations, equivalence pairs of the two languages under consideration, could be used to create semantic fields by means of computational methods. Semantic decomposition and testing with test-frames were used to identify the contribution of each discourse particle lexeme. The lexical model developed finally was formalised in a computational lexicon. The study therefore combines a range of different methods, and thus is a methodologically interdisciplinary analysis of the domain under consideration. To sum up, the cognitive semantic model of the interaction between structural aspects and the conceptual structure outlined by Fillmore and Atkins (1992) could be applied to a lexical model for the interpretation of the spontaneous spoken language items under consideration. The description includes information about the surface properties of discourse particles such as the positions in which they occur with respect to the turn and the utterance and their prosodie realisations. It also shows in how far these contextual properties influence the interpretation of discourse particle occurrences. The model contains a complex network of micro- and macro-structural aspects of the lexicon, involving conventionalised, idiosyncratic features on the one hand and aspects of the communicative situation on the other. It thus presents a first attempt in spoken language lexicology. 6.2

Automatic processing of discourse particles

The study may contribute to increasing attention to discourse particles in automatic speech processing. Interest in spontaneous spoken language dialogues, particularly from the perspective of automatic speech processing, has been growing during the last decade. Speech processing systems have to deal with spontaneous spoken language as it is really produced, and not as it (perhaps) should be. Interest in the behaviour of the frequently occurring discourse particles has therefore increased as well, since they constitute a very high proportion of the words uttered. Whether discourse particles will soon play a more important role in automatic speech processing depends largely on two questions: It has

324

Conclusion and prospects

to be determined to what extent the many different contextual meanings that actual occurrences of discourse particles display in spontaneous spoken language dialogues can be accounted for on the basis of surface features accessible to automatic speech processing. This has been part of the goals of this investigation. Secondly, even if the different readings of discourse particles can be automatically distinguished, automatic speech processing systems will only process discourse particles if their analyses support the aims of the systems. Since discourse particles only rarely contribute to the propositional information in an utterance, they are usually regarded to be superfluous. Since, however, in section 2.2 correlations between discourse particles in certain functions and domain-specific dialogue acts were found, there is information that could be used for the automatic identification of the macrostructure of a dialogue. This in turn can be used for the disambiguation of utterances or for their pragmatic-based translation (Schmitz and Quantz 1995). Consequently, for automatic speech processing, the results of section 2.2 are particularly conclusive. Even in a relatively unrestricted scenario such as the human-to-human construction dialogues, up to 60% of the occurrences of ja can be automatically determined on the basis of information that is easily available to such systems. This recognition rate may be even better for other discourse particles that do not display such a broad range of functions. If the domain is more restricted than the one used in the classification experiments in section 2.2, the correct assignment rate can also be much higher. For the instruction dialogues in the toy-airplane construction domain, only a slight increase in correct classification was found when information on the previous and current dialogue acts was added; however, in a human-to-computer scenario for appointment scheduling a much higher assignment rate of correct interpretation could be reached since the domain is more restricted (Fischer and Brandt-Pook 1998). As spoken language human-to-machine interfaces become more and more important, information other than the propositional content of an utterance may get more weight. For example, in human-tocomputer communication, turn-taking has to be modelled in order for a system to know when the speaker has finished her contribution. For language generation, natural modelling may also require the correct

Automatic processing of discourse particles

325

employment of pragmatically determined items such as discourse particles. Correlations between syntactic positions and certain discourse particles can also be exploited in speech processing systems for the automatic segmentation of turns into utterances. Even if discourse particles do not contribute to the propositional content of utterances, they can still be useful for automatic speech processing systems. Consequently, there are many potential uses for the results from analyses of discourse particles, and the results from this investigation can provide a linguistic basis for the analysis and automatic disambiguation of discourse particles in automatic speech processing systems. To conclude, this investigation has provided a cognitive lexical pragmatic account of the functional polysemy of discourse particles. Factors were determined that influence the interpretation of discourse particles and a model of the interaction of these factors was developed. By showing statistical correlations between certain types of information, such as the position in turn and utterance and the prosodie contour or the preceding dialogue act, the results of this analysis point to areas of application. Thus the investigation not only provides insights into a seemingly chaotic domain, as an exercise in the methodological foundation of lexical pragmatics on the basis of a cognitive semantic conception of polysemy, it also provides a first step towards the integration of discourse particles in automatic speech processing systems.

References Abraham, Werner 1991 Discourse particles in German: How does their illocutive force come about? In Werner Abraham (ed.), Discourse Particles: Descriptive and Theoretical Investigations on the Logical, Syntactic and Pragmatic Properties of Discourse Particles in German, (Pragmatics & Beyond 12), 203-252. Amsterdam: John Benjamins. Aitchison, Jean 1987 Words in the Mind. An Introduction to the Mental Lexicon. Oxford/New York: Blackwell. Allwood, Jens, Joachim Nivre, and Elisabeth Ahlsén 1992 On the Semantics and Pragmatics of Lingistic Feedback. Journal of Semantics 9: 1-26. Ameka, Felix 1992 The meaning of phatic and conative interjections. Journal of Pragmatics 18: 245-271. Angermeyer, Alfred 1979 Die Interjektion. Linguistik und Didaktik 37: 39-50. Auer, Peter 1995 Ethnographie Methods in the Analysis of Oral Communication. Some Suggestions for Linguists. In Uta Quasthoff (ed.), Aspects of Oral Communication, (Research in Text Theory 21), 419-440. Berlin/New York: Mouton de Gruyter.

328

References

Austin, John L. 1962 How to Do Things with Words (2nd ed.)· Cambridge, Massachusetts: Harvard University Press. Bade, Ute, Susanne Heizmann, Susanne Jekat-Rommel, Shinichi Kameyama, Detlev Krause, Ilona Maleck, Birte Prahl, and Wiebke Preuß 1994 Wizard-of-Oz-Experimente mit dem Verbmobil-Simulator. Memo 24, Verbmobil. Baker, Collin F., Charles J. Fillmore, and John B. Lowe 1998 The Berkeley FrameNet Project. In Proceedings of Coling/ACL '98, 86-90. Barwise, Jon and John Perry 1984 Situations and Attitudes (2nd ed.). Cambridge, Mass.: ΜΓΓ Press. Bauer, H.-U and Klaus Pawelzik 1992 Quantifying the neighborhood preservation of self-organizing feature maps. In Transactions on Neural Networks, Volume 3, 570579. IEEE. Beattie, Geoffrey W. 1979 Planning Units in Spontaneous Speech: Some Evidence from Hesitation in Speech and Speaker Gaze Direction in Conversation. Linguistics (17): 61-78. Berg, Thomas and Ulrich Schade 1992 The Role of Inhibition in a Spreading-Activation Model of Language Production. Part 1 : The Psycholinguistic Perspective. Journal of Psycholinguistic Research 22: 405-434. Bestgen, Yves and Wietske Vonk 1995 The Role of Temporal Segmentation Markers in Discourse Processing. Discourse Processes 19: 385-406. Biber, Douglas, Susan Conrad, and Randi Reppen 1998 Corpus Linguistics. Investigating Language Structure and Use. Cambridge Approaches to Linguistics. Cambridge: Cambridge University Press.

References

329

Bierwisch, Manfred 1983 Semantische und konzeptuelle Repräsentation lexikalischer Einheiten. In R. Ruzicka and W. Mötsch (eds.), Untersuchungen zur Semantik, 61-99. Berlin: Akademie-Verlag. Bierwisch, Manfred and Ewald Lang 1989 Dimensional Adjectives: Grammatical Structure and Conceptual Interpretation. Berlin: Springer. Bierwisch, Manfred and Robert Schreuder 1992 From Concepts to Lexical Items. Cognition 42\ 23-60. Blakemore, Diane 1987 Semantic Constraints on Relevance. Oxford: Blackwell. Bleiching, Doris 1990 Das Wortfeld 'family' als semantisches Netz. Thesis for Staatsexamen, University of Bielefeld. Bleiching, Doris 1992 Prosodisches Wissen im Lexikon. In G. Görz (ed.), Konvens '92, 59-68. Berlin: Springer. Blutner, Reinhard, Annette Leßmöllmann, and Rob van der Sandt 1996 Conversational Implicature and Lexical Pragmatics. In Proceedings of the AAAI Spring Symposion on Conversational Implicature, Stanford, 1-9. Boomer, Donald S. 1965 Hesitation and Grammatical Encoding. Language and Speech 8: 148-158. Boomer, Donald S. and Allen T. Dittmann 1962 Hesitation Pauses and Juncture Pauses in Speech. Language and Speech 5: 215-220. Brindöpke, Christel, Michaela Johanntokrax, Arno Pahde, and Britta Wrede 1995 Darf ich Dich Marvin nennen? Instruktionsdialoge in einem Wizard-of-Oz Szenario: Materialband. SFB-Report "Situierte künstliche Kommunikatoren" 95/7, University of Bielefeld.

330

References

Brown, Penelope and Stephen Levinson 1987 Politeness. Some Universals in Language Usage (2nd (original 1978) ed.). Cambridge University Press. Chafe, Wallace 1980 Some Reasons for Hesitating. In Hans W. Dechert and Manfred Raupach (eds.), Temporal Variables in Speech. Studies in Honour of Frieda Goldman-Eisler, 169-180. The Hague, Paris, New York: Mouton. Chomsky, Noam 1957 Syntactic Structures. The Hague: Mouton. Chomsky, Noam 1965 Aspects of the Theory of Syntax. Cambridge, Massachusettes: The M.I.T. Press. Christenfeld, Nicholas 1994 Options and Ums. Journal of Language and Social Psychology 13(2): 192-199. Clark, Herbert H. 1971 The Importance of Linguistics for the Study of Speech Hesitations. In D. Horton and J. Jenkins (eds.), Perception of Language, 69-78. Columbus, OH: Charles E. Merrill. Clark, Herbert H. 1996 Using Language. Cambridge University Press. Clark, Herbert H. 1999 Speaking in Time. In Proceedings of the ESCA Workshop on Dialogue and Prosody, De Koningshof, Veldhoven, The Netherlands, Sept.lst-3rd, 1999, 1-6. Clark, Herbert H. and Edward F. Schaefer 1989 Contributing to Discourse. Cognitive Science 13: 259-294. Coleman, Linda and Paul Kay 1981 Prototype Semantics: The English Verb Lie. Language 57(1): 2644.

References

331

Coseriu, Eugenio 1970 Sprache - Strukturen und Funktionen. Tübingen: Niemeyer. Coulthard, Malcolm 1985 An Introduction to Discourse Analysis. London/ New York: Longman. Cruse, D. Allen 1986 Lexical Semantics. Cambridge University Press. Cruse, D. Allen 1992 Polysemy and Related Phenomena from a Cognitive Linguistic Viewpoint. In Patrick Saint-Dizier and Evelyne Viegas (eds.), Proceedings of the 2nd Seminar on Computational Lexical Semantics, January 23-25, 1992. IRIT, Toulouse, France. Daly-Kelly, Nancy A. 1995 Linguistic and Acoustic Characteristics of Pause Intervals in Spontaneous Speech. In Proceedings of the 4th European Conference on Speech Communication and Technology, 1023-1026. Dell, Gary 1986 A Spreading Activation Theory of Retrieval in Sentence Production. Psychological review 93: 283-321. Diewald, Gabriele 1991 Deixis und Textsorten im Deutschen. (Reihe Germanistische Linguistik 118). Tübingen: Niemeyer. Diewald, Gabriele and Kerstin Fischer 1998 Zur diskursiven und modalen Funktion von aber, auch, doch und ja in Instruktionsdialogen. Linguistica 38(1): 75-99. Douglas, Sarah A. 1995 Conversation Analysis and Human-Computer Interaction Design. Cambridge Series on Human-Computer Interaction 10: 184-203. Drescher, Martina 1997 French Interjections and their Use in Discourse. In Susanne Niemeyer and René Dirven (eds.), The Language of Emotions. Amsterdam/Philadelphia: John Benjamins.

332

References

Dreyfus, Hubert L. 1992 What Computers Still Can't Do. A Critique of Artificial Reason. Cambridge, Massachusettes/London: The M.I.T. Press. Ducrot, Oswald 1996 Slowenian Lectures/Conférences Slovènes, ed. by Igor Zagar, Ljubljana: ISH. Duncan, Starkey Jr. 1972 Some signals and rules for taking speaking turns in conversations. Journal of Personality and Social Psychology 23(2): 283-292. Duncan, Starkey Jr. 1974 On the structure of speaker-auditor interaction during speaking turns. Language in Society 3: 161-180. Duncan, Starkey Jr. and Donald W. Fiske 1977 Face-to-face interaction: Research, methods, and theory. Hillsdale, N.J.:Erlbaum. Dupuy-Engelhardt, Hiltraud 1993 Wortfeldpraxis nach den Prinzipien der Lexematik. In Peter Rolf Lutzeier (ed.), Studien zur Wortfeldtheorie/Studies in Lexical Field Theory, (Linguistische Arbeiten 288), 23-34. Tübingen: Niemeyer. Edmondson, Willis 1981 Spoken Discourse. A model for analysis. (Longman Linguistics Library 27). London/New York: Longman. Ehlich, Konrad 1986 Interjektionen. meyer.

(Linguistische Arbeiten 111). Tiibingen:

Nie-

Eikmeyer, Hans-Jürgen and Ulrich Schade 1991 Sequentialization in Connectionist Language-Production Models. Cognitive Systems 3: 128-138. Evans, Roger and Gerald Gazdar 1990 The DATR Papers. Cognitive science research paper, University of Sussex.

References

333

Evans, Roger and Gerald Gazdar 1996 DATR: A Language for Lexical Knowledge Representation. Computational Linguistics 22(2): 167-216. Fillmore, Charles and Paul Kay 1995 Construction Grammar. CSLI Lecture Notes, Leland Stanford Junior University. Fillmore, Charles J. 1972 Types of Lexical Information. In D.D. Steinberg and L.A. Jakobovits (eds.), Semantics. An Interdisciplinary Reader in Philosophy, Linguistics and Psychology, 370-392. Cambridge University Press. Fillmore, Charles J. 1975 An Alternative to Checklist Theories of Meaning. In Proceedings of ΒLS I, 123-131. Berkeley Linguistic Society. Fillmore, Charles J. 1981 Pragmatics and the Description of Discourse. In Peter Cole (ed.), Radical Pragmatics, 143-166. New York etc.: Academic Press. Fillmore, Charles J. 1982 Frame Semantics. In Linguistics in the Morning Calm, 111-137. The Linguistic Society of Korea: Seoul: Hanshing Publishing Co. Fillmore, Charles J. 1984 Remarks on contrastive pragmatics. In Jacek Fisiak (ed.), Contrastive Linguistics. Prospects and Problems, Trends in Linguistics: Studies and Monographs 22, 119-141. Berlin/New York/Amsterdam: Mouton. Fillmore, Charles J. 1994 Ethnographic Lexicography. In Plenary Lectures delivered at the 6th Euralex International Congress, Amsterdam, August 30 September 3, 1994, 27-48. Fillmore, Charles J. and Beryl T. Atkins 1992 Toward a Frame-Based Lexicon: The Semantics of risk and its Neighbors. In Adrienne Lehrer and Eva F. Kittay (eds.), Frames, Fields and Contrasts, 75-102. Hillsdale N.J.: Lawrence Erlbaum.

334

References

Fink, Gemot, Michaela Johanntokrax, and Brigitte Schaffranietz 1995 A Flexible Formal Language for the Orthographic Transcription of Spontaneous Spoken Dialogues. In Proceedings of the Fourth European Conference on Speech Communication and Technology, 871-874. Fischer, Kerstin 1995 Lexikonaufbau: Diskurspartikeldatenbank. Memo 85, Verbmobil. Fischer, Kerstin 1996a Compositionality and Lexical Semantics for Compounds. In Franz Hundsnurscher and Edda Weigand (eds.), Lexical Structures and Language Use, 193-203. Tübingen: Niemeyer. Fischer, Kerstin 1996b A Construction-based Approach to the Lexicalization of Interjections. In M. Gellerstam, J. Järborg, S. Malmgren, Κ. Nören, L. Rogström, and C. Rojder Papmehl (eds.), Euralex '96: PROCEEDINGS, 85-91. University of Gothenburg. Fischer, Kerstin 1996c Distributed Representation Formalisms for Discourse Particles. In Dafydd Gibbon (ed.), Natural Language Processing and Speech Technology. Results of the 3rd KONVENS Conference, Bielefeld, October 1996, 212-224. Berlin/New York: Mouton de Gruyter. Fischer, Kerstin 1998 Validating semantic analyses of discourse particles. Journal of Pragmatics 29: 111- 127. Fischer, Kerstin 1999 Die Ikonizität der Pause: Zwischen kognitiver Last und kommunikativer Funktion. In Ipke Wachsmuth and Bernhard Jung (eds.), Proceedings der KogWis99, Bielefeld, September 1999, 250-255. Gesellschaft für Kognitionswissenschaft: Infix. Fischer, Kerstin i.p. Discourse Particles, Turn-Taking, and the Semantics-Pragmatics Interface. Revue de Sémantique et Pragmatique (8).

References

335

Fischer, Kerstin and Hans Brandt-Pook 1998 Automatic Disambiguation of Discourse Particles. In Proceedings of Coling/ACL '98 Workshop on Discourse Realtions and Discourse Markers, 107-113. Fischer, Kerstin and Michaela Johanntokrax 1995 Ein linguistisches Merkmalsmodell für die Lexikalisierung von diskurssteuernden Einheiten. Report 18, SFB 360 "Situierte künstliche Kommunikatoren". University of Bielefeld. Fischer, Kerstin and Britta Wrede 1997 Discourse Particles in Female and Male Human-ComputerInteraction. In Milton Keynes De Montford University (ed.), Women into Computing, 36-49. Fischer, Kerstin, Britta Wrede, Christel Brindöpke, and Michaela Johanntokrax 1996 Quantitative und funktionale Analysen von Diskurspartikeln im Computer Talk. Sprache und Datenverarbeitung 27(1-2): 85-100. Flowerdey, John and Steve Tauroza 1995 The effect of discourse markers on second language lecture comprehension. Studies in Second Language Aquisition 17: 435-458. Fodor, Jerry and Zenon Pylyshyn 1988 Connectionism and Cognitive Architecture. Cognition 28: 3-71. Francis, Gill and Susan Hunston 1992 Analysing Everyday Conversation. In Malcolm Coulthard (ed.), Advances in Spoken Discourse Analysis, 123-165. London: Routledge. Fraser, Bruce 1988 Types of English Discourse Markers. Acta Linguistica Hungarica 38(1-4): 19-33. Fraser, Bruce 1990 An approach to discourse markers. Journal of Pragmatics 14: 383-395.

336

References

Fraser, Bruce 1999 What are Discourse Markers? Journal of Pragmatics 31: 931— 952. Geeraerts, Dirk, Stefan Grondelaers, and Peter Bakema 1984 The Structure of Lexical Variation: Meaning, Naming, and Context. (Cognitive Linguistics Research 5). Berlin/New York: Mouton de Gruyter. Gelhaus, Hermann 1995 Die Wortarten. In Duden: Die Grammatik (5th ed.), Volume 4 of Der Duden in 12 Bänden. Mannheim: Dudenverlag. Gibbon, Dafydd 1990 Prosodie Association by Template Inheritance. In G. Gazdar and W. Daelemans (eds.), Inheritance in Natural Language Processing. Proceedings of First International Workshop on Inheritance in Natural Language Processing, Tilburg. Gibbon, Dafydd 1992 ILEX: A Linguistic Approach to Computational Léxica. Computano Linguae. Zeitschrift für Dialektologie und Linguistik 73: 3253. Gibbon, Dafydd 1997 Compositionality in the Inheritance Lexicon: English nouns. University of Bielefeld Ms. Gibbon, Dafydd and Firmin Ahoua 1991 DDATR: un logiciel de traitement d'héritage par défaut pour la modélisation lexicale. Cahiers Ivoiriens de Recherches Linguistiques 27: 5-59. Gibbon, Dafydd and Claudia Sassen 1997 Prosody-Particle Pairs as Discourse Control Signs. In Proceedings of the Fifth European Conference on Speech Communication and Technology.

References

337

Gibbs, Raymond 1996 Cognitive Semantics. In Eugene H. Casad (ed.), Cognitive Linguistics in the Redwoods, 821-836. Berlin/New York: Mouton de Gruyter. Goffman, Erving 1972 Encounters. Two Studies in the Sociology of Interaction. Penguin University Books. Harmondsworth: Penguin. Goffman, Erving 1978 Response Cries. Language 54: 787-815. Goldberg, Adele 1995 Constructions. The University of Chicago Press. Goldman-Eisler, Frieda 1958 Speech Production and the Predictability of Words in Context. Quarterly Journal of Experimental Psychology 10: 96-106. Good, David A. and Brian L. Butterworth 1980 Hesitancy as a Conversational Resource: Some Methodological Implications. In Hans W. Dechert and Manfred Raupach (eds.), Temporal Variables in Speech. Studies in Honour of Frieda Goldman-Eisler, 145-152. The Hague, Paris, New York: Mouton. Gülich, Elisabeth 1970 Makrosyntax der Gliederungssignale im gesprochenen Französisch. München: Wilhelm Fink Verlag. Gülich, Elisabeth and Thomas Kotschi 1987 Reformulierungsverfahren als Mittel der Textkonstitution. Untersuchungen zu französischen Texten aus mündlicher Kommunikation. In Wolfgang Mötsch (ed.), Satz, Text, sprachliche Handlung, (Studia grammatica 25), 199-261. Berlin: Akademie Verlag. Gülich, Elisabeth and Thomas Kotschi 1996 Textherstellungsverfahren in mündlicher Kommunikation. Ein Beitrag am Beispiel des Französischen. In Wolfgang Mötsch (ed.), Ebenen der Textstruktur. Sprachliche und kommunikative Prinzipien, (Reihe Germanistische Linguistik 164), 37-80. Tübingen: Niemeyer.

338

References

Gumperz, John 1982 Discourse Strategies. (Studies in Interactional Sociolinguistics 1). Cambridge University Press. Halliday, M.A.K. 1992 Language as System and Language as Instance: The Corpus as a Theoretical Construct. In Jan Svartvik (ed.), Directions of Corpus Linguistics. Proceedings of Nobel Symposium 82, Stockholm, 4-8 August, 1991, (Trends in Linguistics, Studies and Monographs 65), 61-77. Berlin/New York: Mouton de Gruyter. Handke, Jürgen 1995 Mental and machine-readable lexicons (What can machines learn from the mind?). In René Dirven and Johan Vanparys (eds.), Current Approaches to the Lexicon, (Duisburg Papers on Research in Language and Culture 24), 273-299. Frankfurt a.M.: Lang. Harras, Gisela 1993 Lexikalische Feldstruktur und kommunikatives Hintergrundwissen. Am Beispiel deutscher Sprechaktverben. In Peter Rolf Lutzeier (ed.), Studien zur Wortfeldtheorie/Studies in Lexical Field Theory, (Linguistische Arbeiten 288), 75-86. Tübingen: Niemeyer. Heibig, Gerhard 1977 Partikeln als illokutive Indikatoren im Dialog. Deutsch als Fremdsprache 14: 30-44. Heibig, Gerhard 1988 Lexikon deutscher Partikeln. Leipzig: Verlag Enzyklopädie. Heibig, Gerhard and Joachim Buscha 1993 Deutsche Grammatik (15th ed.). Leipzig/Berlin/München: Langenscheidt - Verlag Enzyklopädie. Hentschel, Elke and Harald Weydt 1989 Wortartenprobleme bei Partikeln. In Harald Weydt (ed.), Sprechen mit Partikeln, 3-18. Berlin/New York: de Gruyter.

References

339

Heritage, John 1984 A change-of-state token and aspects of its sequential placement. In J. Atkinson and John Heritage (eds.), Structure of Social Action: Studies in Conversation Analysis, 299-345. Cambridge University Press. Heritage, John 1995 Conversation Analysis: Methodological Aspects. In Uta Quasthoff (ed.), Aspects of Oral Communication., (Research in Text Theory 21), 391-418. Berlin/New York: de Gruyter. Heritage, John and Maija-Leena Soijonen 1994 Constituting and Maintaining Activities across Sequences: AndPrefacing as a Feature of Question Design. Language in Society 23: 1-29. Herman, Vimala 1995 Dramatic Discourse. Dialogue as Interaction in Plays. Routledge: London, New York. Hill, Deborah 1992 Imprecatory inteijectional expressions: Examples from Australian English. Journal of Pragmatics 18: 209-223. Hinrichs, Erhard and Paola Monachesi 1996 Tutorial on Syntax and Semantics in HPSG. Tutorial Notes at Konvens '96. Hirschberg, Julia and Diane Litman 1993 Empirical Studies on the Disambiguation of Cue Phrases. Computational Linguistics 19(3): 501-530. Hitzenberger, L. and C. Womser-Hacker 1995 Experimentelle Untersuchungen zu multimodalen natürlichsprachigen Dialogen in der Mensch-Computer-Interaktion. Sprache und Datenverarbeitung 79(1): 51-61. Hockett, Charles 1958 A Course in Modern Linguistics. New York: Macmillan.

340

References

Holly, Werner 1979 Imagearbeit in Gesprächen. Zur linguistischen Beschreibung des Beziehungsapekts. (Reihe Germanistische Linguistik 18). Tübingen: Niemeyer. Howell, Richard W. and Harold J. Vetter 1969 Hesitation in the Production of Speech. The Journal of General Psychology (81): 261-276. Hutchby, Ian and Robin Wooffitt 1998 Conversation Analysis. Cambridge: Polity. James, Deborah 1972 Some aspects of the syntax and semantics of interjections. In Papers from the 8th regional meeting ofCLS, 162-172. Chicago Linguistic Society. James, Deborah 1973 Another look at, say, some grammatical constraints, on, oh, interjections and hesitations. In Papers from the 9th regional meeting ofCLS, 242-251. Chicago Linguistic Society. James, Deborah 1974 The Syntax and Semantics of Some English Interjections. (Papers in Linguistics 1,3). Ann Arbor, Mich.: University of Michigan. Jefferson, Gail 1974 Error correction as an interactional resource. Language in Society 3: 181-199. Jekat-Rommel, Susanne, Ilona Maleck, and Birte Prahl 1994 TP-13 Datenerhebung. Memo 18, Verbmobil. Jucker, Andreas H. and Yael Ziv 1998 Discourse Markers: Inroduction. In Andreas H. Jucker and Yael Ziv (eds.), Discourse Markers. Descriptions and Theory, (Pragmatics and Beyond - New Series 57), 1-12. John Benjamins: Amsterdam, Philadelphia.

References

341

Kamp, Hans and Uwe Reyle 1993 From Discourse to Logic. Introduction to Modeltheoretic Semantics of Natural Language, Formal Logic and Discourse Representation Theory. (Studies in Linguistics and Philosophy 42). Dordrecht/lì oston/London : Kluwer. Kamp, Hans and Antje Roßdeutscher 1992 Remarks on Lexical Structure, DRS-Construction and Lexically Driven Inferences. Arbeitspapiere des Sonderforschungsbereichs 340 21, Universität Stuttgart. Sprachtheoretische Grundlagen für die Computerlinguistik. Kawashima, Atsuo 1989 Textsorte und Partikeln im Japanischen und Deutschen. In Harald Weydt (ed.), Sprechen mit Partikeln, 276-281. Berlin/New York: de Gruyter. Kay, Paul 1995 Construction Grammar. In J. Verschueren, J.-O. Östman, and J. Blomaert (eds.), Handbook of Pragmatics: Manual, 171-177. Amsterdam: Benjamins. Kay, Paul and Chad Κ. McDaniel 1979 On the Logic of Variable Rules. Language in Society 8: 151-187. Kay, Paul and Chad K. McDaniel 1981 On the Meaning of Variable Rules: Discussion. Language in Society 10: 251-258. Keil, Martina 1994 Analyse von Partikeln für ein sprachverstehendes System - am Beispiel telefonischer Zugauskunftsdialoge. Master's thesis, Universität Erlangen-Nürnberg, Phil. Fak. II. Kilgariff, Adam 1992 Inheriting Polysemy. In Patrick Saint-Dizier and Evelyne Viegas (eds.), Proceedings of the 2nd Seminar on Computational Lexical Semantics, January 23-25, 1992, 198-211. IRIT, Toulouse, France.

342

References

Kohler, Klaus 1978 Englische "Question Tags" und ihre deutschen Entsprechungen. Arbeitsberichte des Instituts für Phonetik der Universität Kiel (10): 61-77. Kohler, Klaus, Gloria Lex, Matthias Pätzold, Michael Scheffers, Adrian Simpson, and Werner Thon 1994 Handbuch zur Datenaufhahme und Transliteration in TP 14 von VERBMOBIL - 3.0. Technisches Dokument Nr. 11, Institut für Phonetik und digitale Sprachverarbeitung der ChristianAlbrechts-Universität zu Kiel. König, Ekkehard 1997 Zur Bedutung von Modalpartikeln im Deutschen: Ein Neuansatz im Rahmane der Relevanztheorie. Germanistische Linguistik 136: 57-75. König, Ekkehard and S. Requardt 1991 A relevance-theoretic approach to the analysis of modal particles in German. Multilingua 70(1/2): 63-77. Krause, Jürgen 1992 Fazit und Ausblick: Registermodell versus metaphorischer Gebrauch von Sprache in der Mensch-Computer- Interaktion. In Jürgen Krause and Ludwig Hitzenberger (eds.), Computertalk, (Sprache und Computer 12), 157-170. Hildesheim: Olms. Krivonosov, Aleksej 1989 Zum Problem der Klassifizierung der deutschen Partikeln. In Harald Weydt (ed.), Sprechen mit Partikeln, 30-38. Berlin/New York: de Gruyter. Krzeszowski, Tomasz P. 1990 Contrasting Languages. The Scope of Contrastive Linguistics. (Trends in Linguistics, Studies and Monographs 51). Berlin/New York: Mouton de Gruyter. Kuhn, Thomas S. 1970 The Structure of Scientific Revolutions (second enlarged ed.). Chicago: University of Chicago Press.

References

343

Labov, William 1969 Contraction, Deletion, and Inherent Variability of the English Copula. Language 45: 715-762. LakofF, George 1987 Women, Fire and Dangerous Things. What Categories Reveal about the Mind. The University of Chicago Press. Lakoff, George and Mark Johnson 1980 Metaphors We Live by. The University of Chicago Press. Lakoff, Robin 1973 Questionable answers and answerable questions. In B. Kachru (ed.), Issues in Linguistics, 453-467. Urbana: University of Illinois Press. Lalljee, Mansur and Mark Cook 1975 Filled Pauses and Floor Holding: the Final Test? Semiotica 12: 219-225. Langacker, Ronald 1988 A Usage-based Model. In B. Rudzka-Ostyn (ed.), Topics in Cognitive Linguistics, 127-161. Amsterdam: Benjamins. Langacker, Ronald W. 1991 Concept, Image and Symbol. The Cognitve Basis of Grammar. (Cognitive Linguistics Research 1). Berlin/New York: Mouton de Gruyter. Langer, Hagen 1992 DELASOUL: Eine constraintbasierte Beschreibungssprache für lexikalische Repräsentationen. Technical Report ASL-TR-2692/UBI, University of Bielefeld. Lascarides, Alex and Ann Copestake 1995 Pragmatics of Word Meaning. In Proceedings of Semantics and Linguistic Theory (SALT 5), Austin, Texas. Laubenstein, Uwe, U. Schade, H. Hendrichs, and V. Korz 1995 Artificial Neural networks in the Analysis of Behavioral Topology. Naturwissenschaften 82(10): 479 ff.

344

References

Le Ny, Jean-Francois 1992 Mental Lexicon and Lexicon in a Machine. In Patrick Saint-Dizier and Evelyne Viegas (eds.), Proceedings of the 2nd Seminar on Computational Lexical Semantics, January 23-25, 1992, 1-12. IRIT, Toulouse, France. Leech, Geoffrey, Margaret Deuchar, and Robert Hoogenraad 1982 English Grammar for Today: A New Introduction. London et al.: Macmillan Press. Leech, Geoffrey and Jan Svartvik 1975 A Communicative Grammar of English. London: Longman. Lehrer, Adrienne 1993 Semantic Fields and Frames: Are They Alternatives? In Peter Rolf Lutzeier (ed.), Studien zur Wortfeldtheorie/Studies in Lexical Field Theory, (Linguistische Arbeiten 288), 149-162. Tübingen: Niemeyer. Lehrer, Adrienne and Eva F. Kittay 1992 Frames, Fields and Contrasts. Hillsdale N.J.: Lawrence Erlbaum. Lenk, Uta 1998 Discourse Markers and Global Coherence in Conversation. Journal of Pragmatics 30: 245-257. Lenke, Nils, Stefan Dreckmann, and Rolf Heinze 1995 GOLD: An object-oriented lexical tool for German. In René Dirven and Johan Vanparys (eds.), Current Approaches to the Lexicon, (Duisburg Papers on Research in Language and Culture 24), 319-336. Frankfurt a.M.: Lang. Levelt, Willem 1983 Monitoring and Self-repair in Speech. Cognition 14: 41-104. Levin, Beth 1993 English Verb Classes and Alternations. A Preliminary Investigation. The University of Chicago Press. Levinson, Stephen 1983 Pragmatics. Cambridge University Press.

References

345

Lindner, Katrin 1991 'Wir sind ja doch alte Bekannte.' The use of German ja and doch as modal particles. In Werner Abraham (ed.), Discourse Particles: Descriptive and Theoretical Investigations on the Logical, Syntactic and Pragmatic Properties of Discourse Particles in German, (Pragmatics & Beyond N.S. 12), 163-201. Amsterdam: John Benjamins. Lindsey, A. Elisabeth, John O. Greene, Rhonda G. Parker, and Marianne Sassi 1995 Effects of Advance Message Formulation on Message Encoding: Evidence of Cognitively Based Hesitation in the Production of Multiple-Goal Messages. Communication Quarterly 43(3): 320331. Lutzeier, Peter Rolf 1981 Wort und Feld. Wortsemantische Fragestellungen mit besonderer Berücksichtigung des Wortfeldbegriffes. (Linguistische Arbeiten 103). Tübingen: Niemeyer. Lutzeier, Peter Rolf 1993 Wortfelder als kognitive Orientierungspunkte. In Peter Rolf Lutzeier (ed.), Studien zur Wortfeldtheorie/Studies in Lexical Field Theory, (Linguistische Arbeiten 288), 203-214. Tübingen: Niemeyer. Lutzeier, Peter Rolf 1999 Das Gerüst des Lexikons. Überlegungen zu den organisierenden Prinzipien des Lexikons. In Hans Otto Sillmann and Ingo Warnke (eds.), Internationale Tendenzen der Syntaktik, Semantik und Pragmatik: Akten des 32. Linguistischen Kolloquiums in Kassel 1997, (Linguistik International 1). Frankfurt a.M. et al.: Lang. Lyons, John 1977 Semantics, Volume 1 and 2. Cambridge University Press. Maclay, Howard and Charles E. Osgood 1959 Hesitation Phenomena in Spontaneous English Speech. Word I: 19-44.

346

References

Martin, James G. 1971 Some Acoustic and Grammatical Features of Spontaneous Speech. In D. Horton and J. Jenkins (eds.), Perception of Language. Columbus, OH: Charles E. Merrill. Maschler, Yael 1994 Metalanguaging and Discourse Markers in Bilingual Conversation. Language in Society 23: 325-366. Maschler, Yael 1997 Discourse Markers at Frame Shifts in Israeli Hebrew Talk-inInteraction. Pragmatics 7(2): 183-211. Maynard, Senko K. 1993 Discourse Modality. Subjectivity, Emotion and Voice in the Japanese Language. (Pragmatics & Beyond 24). Amsterdam/ Philadelphia: John Benjamins. Mercer, Robert 1992 Presuppositions and Default Reasoning: A Study in Lexical Pragmatics. In James Pustejovsky and Sabine Bergler (eds.), Lexical Semantics and Knowledge Representation. Proceedings First SIGLEX Workshop, Berkeley, CA, June 1991. Berlin/Heidelberg: Springer. Mosegaard Hansen, Maj-Britt 1998 The Semantic Status of Discourse Markers. Lingua 104: 235-260. Müller, Robert 1993 Wortfeldtheorie und Kognitive Psychologie. In Peter Rolf Lutzeier (ed.), Studien zur Wortfeldtheorie/Studies in Lexical Field Theory, (Linguistische Arbeiten 288), 215-228. Tübingen: Niemeyer. Nehls, Dietmar 1989 German Modal Particles Rendered by English Auxiliary Verbs. In Harald Weydt (ed.), Sprechen mit Partikeln, 282-292. Berlin/New York: de Gruyter. Nerbonne, J., K. Netter, and C. Pollard 1994 German in Head-Driven Phrase Structure Grammar. CSLI Publications.

References

347

Norvig, Peter and George Lakoff 1987 Taking: A Study in Lexical Network Theory. In Proceedings of the Berkeley Linguistic Society Meeting, Volume 13. Nothdurft, Werner 1984 "...ähfolgendes Problem äh...". Die interaktive Ausarbeitung "des Problems" in Beratungsgesprächen. (Forschungsberichte des Instituts für Deutsche Sprache Mannheim 57). Tübingen: Narr. Oreström, Bengt 1983 Turn-taking in English Conversation. Lund Studies in English. Lund: CWK Gleerup. O'Shaugnessy, Douglas 1993 Locating Disfluencies in Spontaneous Speech: An Acoustic Analysis. In Proceedings of the 3rd European Conference on Speech Communication and Technology, 2187-2190. Östman, Jan-Ola 1983 You know: A discourse functional approach. (Pragmatics & Beyond Π:7). Amsterdam: Benjamins. O'Sullivan, Emer and Dietmar Rosier 1989 Wie kommen Abtönungspartikeln in deutsche Übersetzungen von Texten, deren Ausgangssprachen für diese keine direkten Äquivalente haben? In Harald Weydt (ed.), Sprechen mit Partikeln, 204216. Berlin/New York: de Gruyter. Oviatt, Sharon 1995 Predicting Spoken Disfluencies during Human-Computer Interaction. Computer Speech and Language (9): 19-35. Petric, Teodor 1995 Indexikalische Leistungen der Modalpartikeln und ihre natürlichkeitstheoretische Bewertung. Linguistica 35: 245259. Pollard, Carl and Ivan Sag 1987 Information-based Syntax and Semantics, Volume 1 of Stanford University CSL1 Lecture Notes. University of Chicago Press.

348

References

Pollard, Carl and Ivan Sag 1994 Head-driven Phrase Structure Grammar. University of Chicago Press. Pons Borderia, Salvador 1998 Conexión y Conectores. Estudio de su relación en el registro informal de la lengua. Facultat de filologia, Universität de València. Prince, Ellen F. 1981 Towards a Taxonomy of Given-New Information. In Peter Cole (ed.), Radical Pragmatics. New York etc.: Academic Press. Psathas, George 1995 Conversation Analysis. The Study of Talk-in- Interaction. (Quality Research Methods 35). Thousand Oaks/London/New Delhi: Sage. Pustejovsky, James 1991 The Generative Lexicon. Computational Linguistics 17(4): 409441. Pustejovsky, James 1995 The Generative Lexicon. Cambridge, Massachusetts: The MIT Press. Pustejovsky, James and Peter G. Anick 1988 On the Semantic Interpretation of Nomináis. In Proc. of the 12th International Conference on Computational Linguistics, Budapest, 518-523. Pustejovsky, James and Branimir Boguraev 1993a Introduction: Lexical Semantics in Context. Journal of Semantics 12: 1-14. Pustejovsky, James and Branimir Boguraev 1993b Lexical knowledge representation and natural language processing. Artificial Intelligence 63: 193-223.

References 349 Rasoloson, Janie Noëlle 1994 Interjektionen im Kontrast. Am Beispiel der deutschen, madagassischen, englischen und französischen Sprache. (Arbeiten zur Sprachanalyse 22). Frankfurt am Main, New York, Bern: Peter Lang. Rathmayr, Renate 1989 Zur Übersetzung gesprochener Sprache: Probleme der Lexikoneintragung von Partikeln und gestischen bzw. mimischen Zeichen. In Mary Snell-Hornby (ed.), Translation and Lexicography, 149-160. Amsterdam: Benjamins. Redeker, Gisela 1990 Ideational and Pragmatic Markers of Discourse Structure. Journal of Pragmatics 14: 367-381. Redeker, Gisela 1991 Linguistic Markers of Discourse Structure. Linguistics 29: 1139— 1172. Reinhard, Sabine 1990 Adäquatheitsprobleme automatenbasierter Morphologiemodelle am Beispiel der deutschen Umlautung. Master's thesis, Universität Trier. Reiter, Norbert 1980 Die Perfidie des deutschen ja. Deutsche Sprache 4: 342-355. Riehemann, Susanne 1994 Morphology and the Hierarchical Lexicon, http://hpsg.stanford.edu/hpsg/papers.html. Ritter, Helge and Teuvo Kohonen 1989 Self-organizing semantic maps. Biological Cybernetics 61: 241254. Ritter, Helge, Thomas Martinetz, and Klaus Schulten 1991 Neuronale Netze: Eine Einführung in die Neuroinformatik selbstorganisierender Netzwerke (2nd extd. ed.). Addison-Wesley: Bonn.

350

References

Rosch, Eleanor 1975 Cognitive Representations of Semantic Categories. Journal of Experimental Psychology: General 104: 192-233. Rosch, Eleanor and Caroline Mervis 1975 Family Resemblances: Studies in the Internal Structure of Categories. Cognitive Psychology 7: 573-605. Rosch, Eleanor, Wayne Mervis Caroline ad Gray, O. Johnson, and R BoyesBraem 1976 Basic Objects in Natural Categories. Cognitive Psychology 8: 382-439. Rudolph, Elisabeth 1985 Zur Rolle argumentativer Partikeln und Konjunktionen in der Text-Kohärenz. In W. Kürschner and R. Vogt (eds.), Grammatik, Semantik, Textlinguistik. Akten des 19. linguistischen Kolloquiums, Vechta 1984, Volume I. Niemeyer: Tübingen. Rudolph, Elisabeth 1991 Relationships between Particle Occurrence and Text Type. Multilingua 70(1-2): 203-223. Sacks, Harvey 1984 Notes on Methodology. In J. Atkinson and John Heritage (eds.), Structure of Social Action: Studies in Conversation Analysis, 2127. Cambridge University Press. Sacks, Harvey 1996 Lectures on Conversation, Volume I & II. Oxford: Blackwell. Sacks, Harvey, Emanuel A. Schegloff, and Gail Jefferson 1974 A simplest systematics for the organization of turn-taking for conversation. Language 50(4): 696-735. Sagerer, Gerhard, Hans-Jürgen Eikmeyer, and Gert Rickheit 1994 "Wir bauen jetzt ein Flugzeug": Konstruieren im Dialog. Arbeitsmaterialien. Report sfb 360, University of Bielefeld.

References

351

Sajavaara, Kari and Jaako Lehtonen 1980 Prisoners of Code-Centered Privacy: Reflections on Contrastive Analysis and Related Disciplines. In Papers in Discourse Analysis. Jyväskylä Contrastive Studies. Sajavaara, Kari, Jaako Lehtonen, and Liisa Korpimies 1980 The Methodology and Practice of Contrastive Discourse Analysis. In Papers in Discourse Analysis. Jyväskylä Contrastive Studies. Schade, Ulrich 1992 Konnektionismus - Zur Modellierung der Opladen: Westdeutscher Verlag.

Sprachproduktion.

Schegloff, Emanuel Α. 1982 Discourse as an Interactional Achievement: Some Uses of 'uh huh' and Other Things that Come between Sentences. In Deborah Tannen (ed.), Analysing Discourse. Text and Talk. Washington: Georgetown University Press. Scheler, Gabriele 1996a Learning the Semantics of Aspect. In Harold Somers and Daniel Jones (eds.), New Methods in Language Processing. University College London Press. Scheler, Gabriele 1996b VGEN. Technical report, ΑΙ/Cognition group, Department of Computer Science Munich University of Technology Research Unit VII. Scheler, Gabriele and Kerstin Fischer 1997 The Many Functions of Discourse Particles: A Computational Model of Pragmatic Interpretation. In Proceedings of Cogsci 1997. Schiffrin, Deborah 1987 Discourse markers. (Studies in Interactional Sociolinguistics 5). Cambridge University Press.

352

References

Schmid, Hans-Jörg 1993 Cottage and Co: Can the theory of word-fields do the job? In Peter Rolf Lutzeier (ed.), Studien zur Wortfeldtheorie/Studies in Lexical Field Theory, (Linguistische Arbeiten 288), 107-120. Tübingen: Niemeyer. Schmitz, Birte 1998 Pragmatikbasiertes Groos.

maschinelles

Dolmetschen.

Heidelberg:

Schmitz, Birte and J. Joachim Quantz 1995 Dialogue Acts in Automatic Dialogue Processing. In Proceedings of the Sixth Conference on Theoretical and Methodological Issues in Machine Translation, TMI-95, Leuven, 33-47. Schourup, Lawrence C. 1983 Common Discourse Particles in English Conversation. Ph. D. thesis, Ohio State University. Searle, John 1969 Speech Acts. An Essay in the Philosophy of Language. Cambridge University Press. Searle, John 1975 Indirect Speech Acts. In P. Cole and J.L. Morgan (eds.), Syntax and Semantics - Speech Acts, Volume 3, 59-82. New York etc.: Academic Press. Selting, Margret 1995 Der 'mögliche Satz' als interaktiv relevante syntaktische Kategorie. Linguistische Berichte 158: 298-325. Selting, Margret 1996 On the Interplay of Syntax and Prosody in the Constitution of Turn-constructional Units and Turns in Conversation. Pragmatics 6(3): 371-388. Smith, Vicki L. and Herbert H. Clark 1993 On the Course of Answering Questions. Journal of Memory and Language 32: 25-38.

References

353

Smolensky, Paul 1988 On the Proper Treatment of Connectionism. Behavioral and Brain Sciences 11: 1-74. Sperber, Dan and Deirdre Wilson 1986 Relevance: Communication and Cognition. Oxford: Blackwell. Stein, Stephan 1995 Formelhafte Sprache: Untersuchungen zu ihrer pragmatischen und kognitiven Funktion im gegenwärtigen Deutsch. (Sprache und Gesellschaft 22). Frankfurt a.M.: Lang. Stellmacher, Dieter 1972 Gliederungssignale in der gesprochenen Sprache. Germanistische Linguistik 4: 518-530. Stenström, Anna-Brita 1994 An Introduction to Spoken Interaction. Learning about Language. London/ New York: Longman. Stubbs, Michael 1983 Discourse Analysis. The Sociolinguistic Analysis of Natural Language. (Language in Society 4). The University of Chicago Press. Svartvik, Jan 1980 Well in Conversation. In Sidney Greenbaum, Geoffrey Leech, and Jan Svartvik (eds.), Studies in English Linguistics for Randolph Quirk, 167-177. London: Longman. Svartvik, Jan and Randolph Quirk 1980 A Corpus of English Conversation. Lund: Gleerup. Sweetser, Eve 1990 From Etymology to pragmatics. Metaphorical and cultural aspects of semantic structure. (Cambridge Studies in Linguistics 54). Cambridge University Press. Swerts, Marc, Anne Wichmann, and Robert-Jan Beun 1996 Filled Pauses as Markers of Discourse Structure. In Proceedings ofthelCSLP '96, 1033-1036.

354

References

Tannen, Deborah and Robin Tolmach Lakoff 1994 Gender and Discourse, Chapter 4: Conversational Strategy and Metastrategy in a Pragmatic Theory: The Example of Scenes from a Marriage, 137-173. Oxford University Press. Taylor, John 1995 Models of Meaning in Comparison: The Two-Level Model (Manfred Bierwisch) and the Network Model (Ronald Langacker). In René Dirven and Johan Vanparys (eds.), Current Approaches to the Lexicon, (Duisburg Papers on Research in Language and Culture 24), 3-26. Frankfurt a.M.: Lang. Taylor, Talbot J. and Deborah Cameron 1987 Analysing Conversation: Rules and Units in the Structure of Talk. Oxford et al: Pergamon Press. Todt, Dietmar 1981 Zum Auftreten von Fiillauten in spontan gesprochenen Berichten. Nova acta Leopoldina N. F. 54(245): 597-611. Traugott, Elizabeth 1995 The Role of the Development of Discourse Markers in a Theory of Grammaticalization. Paper presented at ICHL XII, Manchster. Trier, Jost 1973a Aufsätze und Vorträge zur Wortfeldtheorie. The Hague: Mouton. Trier, Jost 1973b Der deutsche Wortschatz im Sinnbezirk des Verstandes: von den Anfängen bis zum Beginn des 13. Jahrhunderts (2nd (original 1931) ed.). Germanische Bibliothek, Reihe 3: Untersuchungen und Einzeldarstellungen. Heidelberg: Winter. Van Valin, Robert Detrick Jr. 1973 A Pragmatic Analysis of German doch. Master's thesis, University of California at Berkeley.

References

355

Verbmobil-Database 1995 Data Collection for a Speech to Speech Translation System Scheduling Domain. Institut für Phonetik, München, Teilprojekt 14 and Carnegie Mellon University. Verbmobil CDROM 1-3 Transliteration Database. Vismans, Roel 1991 Dutch Modal Particles in Directive Sentences and Modalized Directives in English. Multilingua 70(1-2): 111-122. Weber, Heinrich 1993 Zur Feldstruktur der Seinsverben. In Peter Rolf Lutzeier (ed.), Studien zur Wortfeldtheorie/Studies in Lexical Field Theory, (Linguistische Arbeiten 288), 35-53. Tübingen: Niemeyer. Weiss, Sabine 1996 EEG-Kohärenz und Sprachverarbeitung: Die funktionelle Verkopplung von Gehirnregionen während der Verarbeitung unterschiedlicher Nomina. In Gert Rickheit (ed.), Studien zur klinischen Linguistik: Modelle, Methoden, Interventionen. Opladen: Westdeutscher Verlag. Weiss, Sabine and Peter Rappelsberger 1998 Left Frontal EEG Coherence Reflects Modality Independent Language Processes. Brain Topography 7/(1): 33-42. Wells, John, William Barry, Martine Grice, Adrain J. Fourcin, and Dafydd Gibbon 1992 Standard Computercompatible Transcription. Technical report, Phonetics and Linguistics Department, UCL. Weydt, Harald and Klaas-Hinrich Ehlers 1987 Partikel-Bibliographie: internationale Sprachforschung zu Partikeln und Interjektionen. Frankfurt a.M.: Lang. Wiemer, Rudolph Otto 1971 Beispiele zur deutschen Grammatik. (Schritte 19). Berlin: Wolfgang Fietkau Verlag. Wierzbicka, Anna 1985 Lexicography and Conceptual Analysis. Ann Arbor: Karona.

356

References

Wierzbicka, Anna 1986 A Semantic Metalanguage for the Description and Comparison of Illocutionary Meanings. Journal of Pragmatics 10: 67-107. Wierzbicka, Anna 1989 Prototypes in Semantics and Pragmatics: Explicating Attitudinal Meanings in Terms of Prototypes. Linguistics 27: 731-769. Wierzbicka, Anna 1991 Cross-Cultural Pragmatics. The Semantics of Human Interaction. (Trends in Linguistics. Studies and Monographs 53). Berlin/New York: Mouton de Gruyter. Wierzbicka, Anna 1992a Defining Emotion Concepts. Cognitive Science 16: 539-581. Wierzbicka, Anna 1992b The semantics of inteijection. Journal of Pragmatics 18: 159192. Wierzbicka, Anna 1995 Lexicon as a key to history, culture and society. In René Dirven and Johan Vanparys (eds.), Current Approaches to the Lexicon, (Duisburg Papers on Research in Language and Culture 24), 103155. Frankfurt a.M.: Lang. Wilkins, David P. 1992 Inteijections as deictics. Journal of Pragmatics 18: 119-158. Willkop, Eva-Maria 1988 Gliederungssignale im Dialog. München: Iudicium. Wolski, Werner 1986 Partikellexikographie. Tübingen: Niemeyer. Zwicky, Arnold M. 1985 Clitics and Particles. Language 67(2): 283-305.

Appendix A: Questionnaire This questionaire is about discourse particles and inteqections, those small words like ah, oh, hm, um, er. It serves scientific purposes only, and the data will be handled absolutely confidentially. Thank you very much for your help. 0. Questions about you: a) your age b) your sex c) your educational background d) your subject area 1. Why do you think people use words like a) ah and ohi b) urn and eri 2. Do you think that oh has a meaning, and if so, what could that be? 3. Do you think there is a difference between urn and eri 4. Do you think there is a reason to teach NOT to use words like ah and oh or urn and er a) to children? b) to foreigners, learning English? 5. Would you think the use of a) oh and ah, and b) of urn and er is gender specific, depends on the topic, depends on the partner in the conversation?

358

Appendix A: Questionnaire

Fragebogen Dieser Fragebogen betrifft die Verwendung von Diskurspartikeln und Interjektionen wie beispielsweise ach, ach so, also, naja, oh, äh. Er dient ausschließlich wissenschaftlichen Zwecken, und die erhobenen Daten werden selbstverständlich absolut vertraulich behandelt. Vielen Dank für Ihre Hilfe. 0. Fragen zur Person: a) Ihr Alter b) Ihr Geschlecht c) Ihre Ausbildung d) Ihr Arbeitsbereich 1. Welche Gründe könnte es Ihrer Meinung nach geben, die Worte a) ach undähm oh, zu verwenden? b) äh und 2. Glauben Sie, daß oh eine Bedeutung hat, und wenn ja, was könnte das sein? 3. Glauben Sie, daß es einen Unterschied in der Verwendung gibt zwischen a) ach und oh, b) äh und ähml 4. Gibt es Ihrer Meinung nach einen Grund, jemandem beizubringen, Worte wie ach und oh, oder äh und ähm NICHT zu benutzen, und wenn ja, welchen? a) Kindern? b) Erwachsenen, die Deutsch lernen? 5. Glauben Sie, daß die Verwendung von oh und ach, bzw. äh und ähm geschlechtsspezifisch, themenabhängig, abhängig vom Gesprächspartner ist? Bitte begründen Sie Ihre Antworten.

Appendix Β: DATR Program Lemma Nodes Oh:



< perception > < understanding >

< speechmanage >



Interjection ('orthographic representation oh') ('phonological variants loi, /ο:/, Hol, Πο\Γ) ( Ί know now sth I did not think of before and I feel sth because of it') (Backch:, or Framer:, or Take-up: , or Action:, or Repair: ) ( Ί have heard now what you said') ( Ί have understood now what you said') ( Ί now know something I want to say about what you have said') ( Ί know now what is happening') ( Ί know now that I have said something wrong') ( Ί feel something') ( Ί know now sth new I want to say') ("" , or "" 'and I fear sth because of it') ('phonological representation /? o ? o:/').

360

Appendix Β: DATR Program

Oh_e:



= = = ==

Ja:





< perception > < understanding >

< action >

< speechmanage >

Ohja:



Oh (Oh: , Quotation : < fune > ) ('phonological variants /ou/, loi, Hol, /?ou/') ('phonological representation /?o ?ou/').

Discourse .particle ('orthographic representation ja') ('phonological variants /jA/, /ja/, /ja:/') ( Ί think that you and I think the same') (Answer: , Backch: , Take-up: , Framer: , Repair: , Action: , Modal: ) ('we think the same') ( Ί hear what you are saying') ( Ί understand what you are saying') ( Ί want to add something to what you are saying') ('we both think that we should talk about sth new') ('we think the same about what to do') ('we think the same') ( Ί now know what to say') ('phonological representation /jA ja:/') ( and 'but it is not new to me').

Interj _segm_comb "Oh:" "Ja:".

Appendix Β: DATR Program

361

Object Nodes Representing Generalisations Discourse_particle:





< intonation >

== 0 == Functions : < fune > == ("") == Comm_domains : < express > == ('mental state: "I" - component'). =:= particle == ( ' segments chunks of information ' ) == Redup_particle == (utterance-initial position) == (intonationally not integrated).

Inteijection:



== == ==

Discourse_particle interjection ('mental state: "I now" - component

Template Nodes Interj _segm_comb :

Di scourse_particle ("" " " , "" " " , or "" "" or "" " " , 'typically' Take-up:).

362

Appendix Β: DATR Program

Redup-particle:



Take-up:





< inton_contour>

Repair:





Discourse.particle ("" , , , ) ("" and "") ("") (rise-fall or falling intonation).

Discourse.particle ( , , , , , < turn-taking > , typically ) ("" '=> maintains channel') ("" '=> ensures flow of information') ("" '=> saves speakers faces') ("" '=> takes up previous topic') (takes the turn) ( , , < inton _contour > ) (fall or fall-rise intonation) (turn-initial position).

Discourse.particle ('as a repair marker:' "", typically ) ( , ) (utterance-medial position) (falling or integrated intonation).

Appendix Β: DATR Program

Backch:

< perception > < understanding >

Check:

363

Discourse .particle ('as a backchannel:' , < understanding > , , typically ) ("" '=> maintains channel') (" " '=> ensures flow of information') (entire turn) ('supports the others turn').

< perception >

== Discourse_particle == ( , , < proposition > , , , 'typically' ) == ('Is it true that' ""

< understanding >

== ('Is it true that' "'

== ('Is it true that' ""

== ( Ί wish that' "" 'or I wish that' "") == ('yields the turn') == ( , , ) == (utterance-final position) == (turn-final position) == (rising intonation).



< turn-position >

364

Appendix Β: DATR Program

Framer:



Action:





Answer:



Quotation:



Discourse.particle ('as a framer:' , 'typically' ) ("" '=> marks thematic breaks') ( ) (fall, fall-rise, or rise-fall intonation).

Discourse.particle ('as an action signal:' "" , typically ) (, , ) (fall or fall-rise intonation) (turn-final position) (utterance-final position).

Discourse_particle ('as an answer signal:' "", typically ) ('entire utterance' , , ) (after clarifying questions) (fall or fall-rise intonation).

Discourse_particle ('states quoted speaker is going to continue the topic', , '=> signals other speaker is speaking', 'typically' ) Framer: < surf >.

Appendix Β: DATR Program

Modal:



365

0 (modal particle) ('refers to pragmatic pretext') ('as a modal particle:' "" '(pragmatic pretext) = > ' "", typically ) ( , ) (integrated intonation) (utterance-medial position).



Comm_domains:

==

< express perception >

==

< express understanding >

==

==

< express macrostr>

==

==

==

==

< express interpers>

==

==

0 ("" '(perception) = > ' "") ("" '(understanding) = > ' "") ("" '(action) = > ' "< action >") ("" '(macro structure) = > ' "") ("" '(speech management) = > ' "") ("" '(proposition) = > ' "") ("" '(information structure) = > ' "") ("" '(interpersonal relation) = > ' "") ( , , , , typically Take-up: , or < express action>, or < express speechmanage>, typically Repair: ).

366

Appendix Β: DATR Program

Functions:







0 Take-up: Framer: Action :< fune > Answer: Modal: < fune > Repair: Check: < fune > Backch: Quotation: .

Index äh, 15, 37, 38, 55, 66, 67, 127, 138, 157-165, 168-170, 173, 174, 183, 185, 203, 204, 206, 207, 211, 263, 265 ahm, 15, 55,66,67,138,158-168, 180, 203, 204, 211, 265 aber, 24, 206, 214, 217 ach, 37-39, 140, 151-154, 157, 178, 185, 198, 206, 207, 211,214,226,295 action signal, 76, 77, 107, 117, 118, 123, 142, 177, 267, 301 adverb, 18, 19, 25, 26, 140, 214, 245, 249 ah, 37, 39, 206, 213, 214, 226, 239-245, 249, 259, 295 aha, 198, 206 alright, 206, 215 also, 127, 134-137, 206, 214, 277 anchorage, 280, 281, 283 and, 206, 215 and-test, 220, 225 answer signal, 2, 16, 25, 69, 70, 85, 101, 107, 109, 110, 114, 117, 118, 123, 140, 147, 179, 180, 196, 206, 211, 214, 215, 217, 218,

255, 266, 277, 291, 296, 303 appointment scheduling, 2, 29, 30, 35,47, 80, 177,272, 307, 324 attribute-value description, 99, 109, 110, 296 automatic classification, 56, 59, 61,63, 98, 114, 121 automatic similarity detection, 57, 58, 63 automatic speech processing, 52, 99, 115, 176, 285, 288, 323-325 autosemantica, 16 back-channel, 2, 19, 67, 74, 78, 85-87, 100, 106, 117, 118, 122, 129, 137, 153155, 165, 177, 181, 211, 221, 227, 232, 233, 235, 257, 259, 263, 266, 301, 318 but, 20, 25, 233 but-test, 220, 224, 233 categories contrastivity of, 52, 56, 59, 98, 110-114, 121, 124 exhaustivity of, 52, 56, 98 expressivity of, 52

368

Index

checking signal, 93, 97, 100, 108, 118, 123, 180, 211, 212, 254, 268, 274, 291, 309 cognitive commitment, 6 cognitive grammar, 11, 60, 121 cognitive semantics, 7, 8, 10-12, 59,61,63, 321-323, 325 coherence discourse, 20, 278, 281 neuro-phyiological, 57 common ground, 94-96, 132, 190, 268, 277 communicative background frame, 8, 9, 61, 62, 125, 175, 178, 179, 192, 195, 261, 263, 269, 283, 284, 296, 306, 307, 321, 322 communicative situation, 8, 14, 17, 26, 28, 50, 94-96, 179, 184, 261, 265, 268, 269, 277, 280-284, 295, 323 compositionality, 6, 62, 294, 297 computer talk, 126, 174, 175 conjunction, 7, 18, 20, 25, 27, 206, 214,215,217,262 connectionism, 57, 58 connector, 13, 220 connotation, 199 construction, 60, 61, 120-122, 124, 195, 259, 261, 262, 265, 272, 274, 292, 293, 296, 302, 309,313,318 construction grammar, 60, 121, 274, 286, 287, 297 context-dependency, 1, 10, 16, 44 contextual meaning, 8, 63, 66,

122, 232, 238, 242, 258, 262, 263, 269, 270, 272275, 292, 302, 306-309, 312,313, 321 continuer, 72, 73 contrastive analysis, 27, 37, 53, 54, 63, 199, 200, 212, 226, 323 conversation analysis, 44-51, 55, 56, 58-60,63,66, 85,96, 97, 197, 322 cooperativity, 29, 32, 87, 139 cue phrase, 13, 14 DATR, 60, 286, 287, 296-300, 302, 310, 312, 318, 319 default, 287, 288, 297, 298, 309, 310 defeasibility, 10, 297 dialogue act, 101, 104, 105, 119, 294, 324 dictionary, 42, 285, 289 bilingual, 285 disclosure, 280, 283 discourse analysis, 51, 98, 322 discourse function, 196, 211, 232, 238, 259 discourse marker, 13-15, 278 discourse particle attitude towards, 37, 41, 45 definition, 284 prototype of, 211 discourse representation theory, 286 discourse structure, 4, 10, 18-20, 26, 28, 46-50, 101, 130, 132, 133, 137, 138, 149, 158, 160, 166, 175, 182,

Index

369

183, 186, 188, 270, 271, 287 dispreference, 53, 85, 87, 161, 245, 246, 248, 257, 265, 267 divergence, 67,139 do(n't) you, 206, 207 doch, 19, 206, 215 dramatic dialogue, 54, 57, 200204, 212, 213, 323

flow of information, 96,105,180182, 221, 259, 263, 266, 268, 295 frame semantics, 7-9, 60, 209, 210 framer, 2, 83, 98, 106, 113, 118, 119, 122, 133, 136, 137, 142, 147, 150, 153, 154, 167, 211, 214, 236, 244, 259, 264, 301 functionalism, 57

eh, 206 emotion, 4, 16, 17, 19, 21, 38, 40,42,43,199, 242, 264, 270 encyclopedic knowledge, 6, 8, 199,209 er, 37, 38, 203, 206 ethnography of communication, 49 evincive, 21, 280, 283 exception, 288, 289, 297 expeditive field, 22, 277, 278 expressive function, 191, 271, 278, 283 expressive paradox, 220 extra-linguistic variables, 34, 46, 125

generative grammar, 17, 44, 286 gewiß, 206 good, 206 gut, 69, 131, 140-143, 157, 183, 206, 214, 277

face, 10, 21, 49, 85, 87, 88, 162, 175, 189,246, 265, 281 face-to-face communication, 28, 34, 204, 226, 272 family resemblance, 209 filler, 88, 91, 96, 109, 123, 166, 168, 170, 211, 241, 244, 256, 259

ha, 206 Hamming distance, 113, 120, 121 he, 206, 214 hedge, 13 hesitation marker, 14, 15, 17, 2022, 26, 36, 39-42, 85, 138, 157-174, 176, 180, 182, 183, 188, 189, 192, 265, 266, 270, 275, 277 hey, 206 hm, 67, 127, 138, 139, 157, 198 homography, 207 homonymy, 5, 53 HPSG, 286, 287, 297 huh, 206 hä, 206 I mean, 279 idiosyncrasy, 286-288, 292, 296, 297, 299, 301, 302, 310, 312, 314, 317, 323

370

Index

immediate dominance, 297 in Ordnung, 206 indexicality, 14, 17, 278-280, 304 information management, 20, 105 information marker, 109, 124, 211,219 inheritance, 286, 287, 296-299, 302, 303, 305, 307, 311, 318 instruction dialogues, 47, 49, 56, 68, 80, 100, 101, 166, 204, 324 interjection, 4, 13-26, 36, 39-43, 199, 206, 207, 211, 214, 215, 217-219, 221, 226, 232, 242, 256, 275, 277, 278, 296, 300, 302,311 interpersonal function, 21, 26, 28, 49, 85, 87,105,133, 158, 174, 185-189, 264, 265, 267, 268, 270, 271, 283 interruption, 19, 158 invariant meaning, 2, 3, 9, 10, 54, 59, 61, 63, 86, 96, 97, 212, 221, 226, 232, 234, 238, 242, 251-253, 258, 259, 261-267, 269, 272275, 279, 293, 296, 306 is(n't) it, 206 ja, 16, 25, 43, 49, 60, 61, 68-80, 82-101, 105-114, 116120, 122, 125, 127-133, 144, 152, 177, 180, 182184, 189, 190, 196, 198, 205-207, 211-215, 226, 249, 257, 266-268, 271274, 277, 290, 291, 312-

317, 324 jawohl, 206 klar, 206 lemmatization, 67 lexical access, 285, 292, 315 lexical architecture, 285, 287, 289 lexical entry, 289, 292, 296, 298, 301 lexical pragmatics, 9-11, 55, 321, 322, 325 lexicon linguistic, 284, 285, 289, 296 mental, 50, 284 linear precedence, 297, 299, 310 linguistic opposition, 53, 210, 318 London/Lund corpus, 34, 49, 223, 239 macro-structure, 19, 96, 104, 135, 183,219, 221,267, 324 maxim of quantity, 248 meta-language, 20, 26, 136 metaphor, 7, 8, 62, 262, 282, 283 metonymy, 7 mhm, 25, 67, 77, 206 modal particle, 13-15, 18-23, 25, 27, 94, 95, 100, 108, 117, 123, 131, 132, 135, 190, 199, 200, 205, 206, 211, 268, 269, 277, 290, 291 monologic speech, 28, 104, 128, 130, 133, 134, 147, 149, 154, 175, 191, 271, 280, 307, 314 motivation, 11, 12, 61, 266, 288, 300

Index na, 130, 206, 207, 214 nanu, 198 natural semantic metalanguage, 42, 43, 54, 221-223, 232 natürlich, 206 ne, 118, 133, 134, 158, 180, 189, 225, 277 nein, 25, 143-146, 157, 196, 198, 206, 211-215 neural networks, 56-59, 61, 63, 98, 99, 110-113, 119, 303 nicht, 206, 214 no, 25, 206, 215 non-linguistic action, 49, 75, 77, 105, 146, 156, 184, 185, 198, 230, 264 non-verbal cue, 21, 22, 204 non-verbal cues, 48 now, 206, 215 nun, 207 object-orientation, 298 oh (E), 25, 37, 39, 42, 201, 206, 207, 213, 219, 221, 224239, 242, 259, 263, 264, 270,279,295,317,318 oh (G), 37, 42, 119, 120, 122, 147-151, 157, 174, 185, 196, 201, 206, 211, 213, 226, 232, 268, 273, 295, 300, 301, 311, 313, 316318 oho, 198, 206, 207, 214 okay, 25, 206, 228 override, 287, 288, 297, 309 particle paradox, 3, 5

371

partner manipulation, 277, 278 partner-orientation, 132, 139, 149, 174, 175, 271, 272, 278, 307, 308 pause, 14, 88, 188, 256 perception, 43, 77, 84, 85, 96, 105-107, 117, 122, 129, 133, 137, 180, 181, 227, 232, 236, 256, 263, 265268, 270, 272, 291, 296 perceptron, 112, 114, 115 phonological realisation, 5,23,24, 66, 67, 285, 293, 296 planes of talk, 4, 278-280 politeness, 281 polysemy, 1, 3-9, 261-263, 266, 269, 279, 282, 321, 322 pragmatic idiom, 17, 21, 22, 25, 27, 251 pragmatic pretext, 105, 108, 123, 190, 268, 269, 277 primitive pragmatic, 98 semantic, 222 propositional information, 10, 16, 69, 85, 86,104,105, 107, 123, 141-144, 179, 180, 255, 259, 263, 266-268, 277, 282, 287, 295, 296 prosody, 223, 226 intonation, 15, 67, 91, 93, 97, 100, 101, 115-118, 120, 122-124, 133, 135, 138, 180, 195, 226, 253-255, 259, 268, 291, 293, 302 stress, 23-25, 290 prototype, 7, 109, 120, 121, 208,

372

Index

211,288 punctuation mark, 19, 20, 22 qualia structure, 6 questionnaire, 37-45, 152, 154 questionnaire study, 242 quotation marker, 259, 312, 318 quoted speech, 231,232,238, 242, 249, 259 recording modalities, 29, 73 reduplication, 301, 310-312, 314, 317 reformulation, 22 reformulation marker, 91, 109, 124, 135, 137, 188, 211 register, 126, 128, 174, 289 repair, 72, 73, 152, 188 repair marker, 2, 22, 89-91, 96, 105, 107, 109, 118, 119, 123, 131, 146, 150, 154, 155, 162, 163, 168-171, 174, 178, 185, 192, 196, 211, 236, 238, 259, 264, 265, 268, 271, 301, 316, 317 rhythm, 163 right, 206, 215, 253 schön, 206, 214 segmentation marker, 13-15, 17, 20-26, 36, 206, 207,214, 215, 217-219, 275, 277, 296,311,313 self-organising (Kohonen) map, 57, 58, 63, 199,212,213, 215-218

semantic decomposition, 17, 53, 54, 63, 218, 219, 223, 323 semantic field, 53, 63 semantic normality, 197, 198, 223 semantic oddity, 220, 225 semantic relation, 53, 54, 197, 210, 219 semantic test, 54, 63, 221, 223, 226, 232 sententiality, 23, 25 sicher, 206 silence, 22, 72 similarity, 59, 113, 196, 199, 206, 210 situation semantics, 286 so, 75, 84, 151, 152, 184, 206, 215, 255 speaker aims, 62, 219, 271, 275 speaker categories, 55, 58, 120, 262, 322 speaker personality, 21 speech management, 10, 22, 23, 26, 96, 104, 137, 147, 167, 174, 175, 178, 185, 186, 192, 231, 259, 266, 268, 278, 284, 287, 295 speech planning, 22, 40, 105, 124, 158, 160, 162, 166, 192, 265,271,317 structural context, 5, 44, 52, 60, 65,66,70, 105-108, 121, 122, 261,262, 307,319 substitution test, 54, 208 sure, 206 surface features, 9, 52, 60, 61, 96, 97, 99, 101, 112, 115-

Index

119, 121, 122, 205, 272, 274, 275, 292, 293, 296, 302, 303, 309, 323, 324 surprise, 38-40, 152, 242 syntactic middle-field, 25 tag question, 27, 205, 206, 215, 218 take-up signal, 80, 106, 109, 114, 118, 122, 133, 183, 189, 211, 235, 236, 238, 239, 248, 251, 253, 259, 272, 274, 295, 301-303, 307, 308, 316 task-orientation, 28, 46, 48, 49, 226, 232, 239 test-frame, 220, 221, 223-225, 233 then, 206, 215

time gaining, 166 tja, 206, 214 topic, 80, 183 change, 2, 19, 80, 82, 83, 85, 96, 106, 122, 131, 143, 230, 236, 241, 248, 254, 265,266 continuity, 43, 71, 78, 79, 84, 85, 87, 96,106,122, 149, 152, 229, 236, 240, 248, 254, 270, 273, 290, 291, 296 dependency, 39, 41, 42, 177 transcription conventions, 30, 34, 35 transition relevance place, 71, 72, 129, 148, 273 translation, 53, 54, 200, 202, 205, 206, 212, 213, 215, 218,

373

324 translation equivalence, 27, 37, 53, 57-59, 63, 199, 200, 204, 205, 323 turn, 18 tum function, 20, 21, 23, 26, 29, 46,70,72,78,92,93,98, 104, 105, 108, 133, 158, 175, 178, 189, 190, 244, 245, 259, 269, 273, 274, 281, 304 two-level semantics, 6, 8 uh, 15, 203, 206 uh-huh, 25, 206 um, 37, 38

uncertainty, 40, 158, 174, 266 und, 206, 214 under-specification, 6, 14, 263, 275, 283, 284 understanding, 43, 72, 73, 75, 77, 84-86, 96, 105-107, 117, 122, 129, 133, 137, 140, 148, 151, 158, 180182, 227, 232, 256, 265268, 270, 273, 291, 296 validation, 54, 57, 58, 65, 232, 235, 238, 258 variable rule, 177 Verbmobil corpus, 1, 29, 30, 35, 43, 47, 73, 77, 166, 177, 204, 294 was, 206, 214 well, 53, 206, 213, 215, 226, 241,

242, 245-251, 253, 257, 259

374

Index

well-formedness, 44, 198 Wizard-of-Oz, 30-32, 139, 190 word class, 3, 5 , 1 2 , 1 3 , 1 8 , 23,26, 206, 215, 217, 218, 275, 277, 278, 304, 321

yeah, 2, 25, 206, 213, 215, 228, 231, 271 yes, 25, 206, 213, 215, 225, 226, 228, 239, 249, 253-259, 271 you know, 17, 25, 280