Cognitive Semantics: Meaning and Cognition (Pragmatics & Beyond New Series)

COGNITIVE SEMANTICS

Pragmatics & Beyond New Series Editor: Andreas H. Jucker (Justus Liebig University, Giessen) Associate Editors: Jacob L. Mey (Odense University) Herman Parret (Belgian National Science Foundation, Universities of Louvain and Antwerp) Jef Verschueren (Belgian National Science Foundation, University of Antwerp) Editorial Address: Justus Liebig University Giessen, English Department Otto-Behaghel-Strasse 10, D-35394 Giessen, Germany e-mail: [email protected] Editorial Board: Shoshana Blum-Kulka (Hebrew University of Jerusalem) Chris Butler (University College of Ripon and York) Jean Caron (Université de Poitiers); Robyn Carston (University College London) Bruce Fraser (Boston University); John Heritage (University of California at Los Angeles) David Holdcroft (University of Leeds); Sachiko Ide (Japan Women’s University) Catherine Kerbrat-Orecchioni (University of Lyon 2) Claudia de Lemos (University of Campinas, Brasil); Marina Sbisà (University of Trieste) Emanuel Schegloff (University of California at Los Angeles) Paul O. Takahara (Kobe City University of Foreign Studies) Sandra Thompson (University of California at Santa Barbara) Teun A. Van Dijk (University of Amsterdam); Richard Watts (University of Bern)

55 Jens Allwood and Peter Gärdenfors (eds) Cognitive Semantics. Meaning and Cognition.

COGNITIVE SEMANTICS MEANING AND COGNITION

Edited by

JENS ALLWOOD University of Gothenburg

PETER GÄRDENFORS Lund University

JOHN BENJAMINS PUBLISHING COMPANY AMSTERDAM/PHILADELPHIA

8

TM

The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences — Permanence of Paper for Printed Library Materials, ANSI Z39.48-1984.

Library of Congress Cataloging-in-Publication Data Cognitive semantics : meaning and cognition / edited by Jens Allwood, Peter Gärdenfors. p. cm. -- (Pragmatics & beyond, ISSN 0922-842X ; new ser. 55) Includes bibliographical references and index. 1. Semantics--Psychological aspects. 2. Cognition. I. Allwood, Jens S., 1947- . II. Gärdenfors, Peter. III. Series. P325.5.P78C64 1998 401’.43--dc21 98-44717 ISBN 90 272 5068 5 (Eur.) / 1 55619 817 5 (US) (Hb; alk. paper) CIP ISBN 90 272 5069 3 (Eur.) / 1 55619 818 3 (US) (Pb; alk. paper) © 1999 - John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. • P.O.Box 75577 • 1070 AN Amsterdam • The Netherlands John Benjamins North America • P.O.Box 27519 • Philadelphia PA 19118-0519 • USA

“I KNOW, WE WON’T REVOLUTIONIZE THE WORLD WITH IT, BUT...”

v

Contents PREFACE Semantics as Meaning Determination with Semantic-Epistemic Operations Jens Allwood

vii 1

Some Tenets of Cognitive Semantics Peter Gärdenfors

19

Function, Cognition, and Layered Clause Structure Peter Harder

37

From Vision to Cognition. A study of metaphor and polysemy in Swedish Sören Sjöström Polysemy and Differentiation in the Lexicon Åke Viberg

67 87

Space and Time Elisabeth Engberg-Pedersen

131

Conceptual Engineering Kenneth Holmqvist

153

Situated Embodied Semantics and Connectionist Modeling Jordan Zlatev

173

NAME INDEX

195

SUBJECT INDEX

197

PREFACE

vii

Preface Interest in semantics during the 20th century has displayed a succession of different foci. The century started with a strong interest in historical semantics, often combined with an interest in the cognitive processes connected with historical change. The main focus was nonetheless on historical change rather than on cognition. Some of the researchers associated with this trend were the German Christian Reisig, the Frenchman Michel Breal and the Swede Gustaf Stern. The impact of Ferdinand de Saussure’s linguistic structuralism meant that the focus shifted from diachronic to synchronic studies in semantics as well, but the strong interest in the cognitive processes underlying linguistic structures often survived. In Saussure’s own work, the interest in the psychological processes underlying language was explicit while, in later structuralist writers, it was often more implicit, perhaps in response to the strong behaviorist tendencies of American linguistic structuralism. The third and perhaps strongest influence on semantics in this century does not come from historical or structural linguistics but from ideas originating in the philosophy of logic and mathematics. Ever since the proposals made by Alfred Tarski in the 1940s, an increasing number of successful attempts have been made to treat semantics with formal methods. Unfortunately, this interest in formalizing semantics has often been connected with a lack of interest in, or an unwillingness to, investigate meaning as a cognitive phenomenon over and above the models investigated in formal semantics. For this reason, we can now see that toward the end of the century, there is a dissatisfaction with the semantics offered by existing formal linguistic theories and a growing interest in explicitly focusing on meaning as a cognitive phenomenon. Cognitive linguistics and, more specifically, cognitive semantics have appeared as labels for a number of slightly different approaches to linguistics and semantics. One thing they have in common, however, is their desire to focus on the relation between language, meaning and cognition. Some do so with the aid of concepts developed in connection with information technology, others do so entirely without any such links.

viii

PREFACE

In this book, which represents Scandinavian work in cognitive semantics, we shall look at some of these approaches in eight articles, all of which are concerned with semantics and cognition. They present extensions of, and critical commentaries on, existing approaches, as well as attempts to develop new approaches integrating semantics and cognition. Jens Allwood presents an operational approach to semantics which is characterized by being cognitive, dynamic and context-sensitive. The meanings of utterances in context are produced through a combination of the meaning potentials of the words in the utterances constrained by semantic operations and various types of available contextual information. The approach is an attempt to create a cognitive semantics which is relevant to understanding how meaning is determined in linguistic interaction between several interlocutors. It thus represents a more pragmatic and social approach than has often been the case in the dominant school of cognitive semantics. Peter Gärdenfors examines the relations between cognitive semantics and recent approaches in the philosophy of language. He contrasts cognitive semantics with standard formal extensional and intensional semantics. He then examines six basic tenets of cognitive semantics and outlines a first step in developing a cognitive semantics based on conceptual spaces. He argues that semantics is a relation between language and cognitive structure, and that the appropriate framework for cognitive structures is a conceptual space. He also argues that cognitive semantics must take in social aspects of language, in particular power relations, and that this invalidates Putnam’s argument that meanings must refer to something non-cognitive. Peter Harder discusses the relation between functionalism and cognitive semantics. He relates functional meaning to communicative interaction, which he claims is evolutionarily older than cognition. In his view, ability to carry functional meanings still characterizes human language, but functional meanings now form a basis for conceptual meanings. He develops a “layered model of the clause”, where this relation is made explicit and provides a model for how cognition is embedded in interaction. Sören Sjöström and Åke Viberg provide examples of how cognitive semantics can be developed by being extended to new empirical material. Sören Sjöström describes and discusses the polysemy of lexical expressions (verbs, nouns and adjectives) connected with vision in Swedish. He uses his analysis to explore the relation between vision and cognition. For example, he claims that light metaphorically represents knowledge and that, accordingly,

PREFACE

ix

perception of light represents understanding, non-perception of light lack of understanding, illumination, explanation etc. Åke Viberg is interested in investigating the semantic structure of verbs in Swedish from a crosslinguistic perspective. In his paper, he investigates the semantic field of “physical contact verbs” e.g. stryka (stroke), smeka (caress), kittla (tickle), skava (abrade) and slå (strike/hit/beat). According to Viberg, verbal semantic fields are usually organized around one (sometimes several) “nuclear verb/s”. The verb slå is such a verb for physical contact verbs, and Viberg claims that other verbs of the field can in fact be seen as elaborations or specializations of some aspects of slå. In this way, the analysis of the nuclear verb slå can be used to impose a structure on the whole field of physical contact verbs. However, some verbs in the field, Viberg claims, belong to other fields as well, especially those verbs which also pattern as “soundsource verbs”. Furthermore, Viberg shows that physical contact verbs are related to action verbs and motion verbs, a fact which is then revealed in the meaning extensions of physical contact verbs. Elisabeth Engberg Pedersen extends the realm of cognitive semantics even further by discussing how it can be used to analyze deaf sign language. Her topic is the use of metaphor in spoken language and sign language. She first discusses spatio-temporal expressions in spoken languages and then gives a description of uses of space to express time in Danish Sign Language. She shows that linguistic means used to express temporal relations and means to express spatial relations are related in systematic ways. Using Gibson’s psychology as a point of departure, she argues that although time and space are not distinct perceptual domains, it is possible to distinguish conceptualizations of time and space at some cognitive levels. Neither time nor space can, however, necessarily be seen as a metaphorical extension of the other. What is seen as a metaphorical extension depends on the language user’s sense of the basic meaning of individual expressions. Kenneth Holmqvist and Jordan Zlatev discuss two different possibilities of implementing a cognitive semantics approach using computer programs. Kenneth Holmqvist first presents a computer model of Langacker’s cognitive grammar. The first part of the model involves a model of “image schemata”. The second part of the model consists of a semantic composition process modeling composition as image superimposition. This process is seen as incremental and as involving “semantic expectations”. The point of the model is that it allows for an experimental investigation of the processing order of

x

PREFACE

different mechanisms of superimposition, which in turn allows for a gradual refinement of the model as a whole. Jordan Zlatev then uses computer modeling in order to give an account of how linguistic expressions are grounded in experience. He presents an approach which he calls “situated embodied semantica”, in which meaning emerges from a pairing of linguistic expressions with situations. Connectionist modeling is used to test the feasibility of the approach and for gaining insights into such issues as learning categories without necessary and sufficient conditions for membership, the context dependence of meaning and the ability to utter and comprehend novel expressions. All in all, we believe this book well represents some of the basic lines of work for extending cognitive semantics — theory construction, new empirical domains and formal modeling on a computer. Jens Allwood and Peter Gärdenfors Göteborg and Lund

SEMANTICS

AS

MEANING DETERMINATION

1

Semantics as Meaning Determination with Semantic-Epistemic Operations Jens Allwood Department of Linguistics, Göteborg University

1.

Introduction

This paper develops and summarizes an approach to semantics which has so far only been available in Swedish, cf. Allwood (1989). The approach is characterized by the fact that it is cognitive, dynamic and context-sensitive. Meaning and concepts are primarily taken to be cognitive phenomena and are studied in terms of operations on information rather than as static entities. The operations are context-sensitive, so that meaning is seen as determined by operations which are sensitive to and make use of linguistic and extralinguistic context. The 1989 paper also gives an analysis of the nature of meaning and of concepts and of the relation between that analysis and classical theories of meaning. Ways of determining concepts and meanings are discussed, and a number of conceptual or cognitive operations for doing this are proposed. There is also a discussion of the linguistic counterparts of these operations and of how they can be used to determine the meaning of linguistic expressions in context. Finally, the paper presents a number of examples of how different linguistic constructions can be analyzed.

2.

Background

The approach makes the following assumptions: i.

All conventionalized linguistic expressions (morphemes, words, idioms, phrases etc.) are connected with “meaning potentials”, cf. Rommetveit

2

JENS ALLWOOD

(1974). A meaning potential is basically a person’s memory of the previous uses of a particular expression and can be seen as the union of all the information the person can associate with the expression. The semantic part of this information will include both what is sometimes called “encyclopedic” and “lexical” information concerning the phenomenon the expression refers to or is otherwise associated with. Philosophical arguments for this position can be found in Quine (1953), and more linguistically flavored arguments can be found in Haiman (1980) and Langacker (1987). ii. When used, a linguistic expression activates its meaning potential. The context-free meaning of a linguistic expression is seen as an activation potential, i.e. as a potential to activate (parts of) the meaning potential associated with a particular expression. iii. The actual meaning of the expression is determined through cognitive operations, the function of which is to achieve compatibility between the meaning potential of a particular expression, the meaning potential of other expressions, and the extralinguistic context. Actual determinate meanings of linguistic expressions thus result from partial activations of the meaning potentials of the expressions guided by cognitive operations. iv. A subset of the operations can be characterized as semantic-epistemic operations, i.e. as operations which have both a linguistic expression and a conceptual-epistemic effect. It is these operations which motivate the use of the term “semantic-epistemic operation”. The basis for these operations are cognitive operations such as discrimination, similarity abstraction, typification and reification, which exist independently of language but are expanded and elaborated by being connected with language. v. The linguistic expressions of the semantic-epistemic operations are mostly what are known as “syncategorematic” expressions, e.g. conjunctions, prepositions, pronouns, quantifiers, some adverbs, some interjections, inflectional and derivational affixes. vi. Another part of the vocabulary is made up of “categorematic” roots and stems (the roots and stems of nouns, verbs, adjectives, and some interjections and adverbs). The meaning potentials of such roles are claimed to serve as arguments for various semantic-epistemic operations. The interplay between the meaning potentials of categorematic roots and stems and syncategorematic operations is further claimed to be a major facet of

SEMANTICS

AS

MEANING DETERMINATION

3

linguistic competence, playing an important role in the production, comprehension and acquisition of language. Below I will illustrate this by first discussing two of the eight types of semantic-epistemic operations proposed in Allwood (1989). I will then illustrate these two types of operations by examining their role in the determination of the concepts of conflict, war and peace. In Allwood (1989), eight types of semantic-epistemic operations were proposed. Each type includes a set of operations, making up a total of about 90 operations. The operations are linguistic and/or cognitive regularizations of underlying spontaneous cognitive processes. Each operation can be seen in two modes, as a process and as an end state, with a category resulting from the operation. We might say that all the operations can be seen both as processes and as products resulting from these processes. The types are the following: i. ii. iii. iv. v. vi. vii. viii.

Basic semantic-epistemic categories Basic conceptual structure Anchoring in time and space Relations Processes Roles derived from relations and processes Properties Quantity, modality and evaluation

Below I will now try to illustrate the approach by discussing two types of operations: i. ii.

Basic semantic-epistemic categories and Roles derived from relations and processes

The operations can be jointly or successively applied to meaning potentials in a way which sometimes involves reiteration or recursion. To avoid confusion, however, let me first briefly comment on the types which are not illustrated in this paper. Under the heading of “basic conceptual structure”, one can find, for example, operations of typification, part-whole structure and instantiation.These can be used together with basic categorization operations to elaborate and give a concept further structure. Under the headings “relations”, “processes” and properties”, one can find operations which can be used together with the basic category operations of “relations”, “processes” and “properties” to further specify these categories. Operational

4

JENS ALLWOOD

types (iii), (vi) and (viii) are, in a similar way, used to further elaborate and structure concepts in the contexts where this is relevant.

3.

Basic semantic-epistemic categories

Language provides support for the conceptual structuring of the world in many ways. One of them is by providing support for a fundamental classification of real phenomena. Below, I will present six categories which have turned out to be useful in conceptual-semantic analysis of many languages. They have also often been pointed out by philosophers engaged in conceptual analysis, cf. Aristotle (1938), Kant (1975), Husserl (1913), Barwise and Perry (1983). The six categories are not, in the present approach, primarily assumed to be ontological. Instead, they are assumed to be semantic-epistemic, i.e. conceptual categories supported by linguistic mechanisms. Whether they also have an ontological status is left open. The categories are the following: i.

Entity e.g. substances like water, concrete objects like trees, abstract objects/substances like freedom, collective objects like police, holistic objects like nature ii. Property e.g. blue or strong iii. Relation e.g. between or and iv. Process e.g. run or give v. State e.g. the state of being strong or the state that X is between Y and Z vi. (Course of) events e.g. single events like X closed the door, and courses of events like building a house The relations between the categories can be depicted as in the following figure: states

properties

(courses of) events

relations

processes

entities Figure 1. Relations between semantic-epistemic categories.

SEMANTICS

AS

MEANING DETERMINATION

5

The different types of entities are regarded as the base of the categories. They serve as bearers (or arguments) of properties, relations and processes. When entities are combined with properties or relations, the result is a state. When they are combined with processes or dynamic relations, the result is an event or a course of events. The arrows going directly from entities to states or courses of events are there, since, as we shall see below, it is possible, by a process of repeated abstraction and reification, to linguistically create entities which encapsulate states or courses of events. From a linguistic point of view, it is convenient to divide the categories into basic and derived, simple and complex in the following manner: 1.

Basic categories

A: Simple Entity:

Property: Relation:

Process: B: Complex State: (course of events) 2.

objects substances collections holistic static dynamic

e.g. tree e.g. water e.g. police e.g. nature e.g. blue, strong e.g. in, and e.g. give, hug (Dynamic relations are also classified as processes.) e.g. run, give

e.g. X is blue e.g. X closed the door, X built a house

Derived categories Entity: Property: Relation: Process:

blueness, strength, inclusion, running watery, natural, included, running bluer than, being in love with to water, strengthen, include

The derived categories are linguistically derived from the basic categories by iterative (recursive) morphological or syntactic processes. Structurally this means that the semantic-epistemic categories themselves should be seen as

6

JENS ALLWOOD

operations which can be applied to sensory (cognitive) input to support a basic linguistic level of categorization of cognitive and sensory data. It is not assumed that such categorization must always be supported by language, or that it must take place at all. Perception can involve nonlinguistically organized experience and even non-conceptual experience. The term “basic” here primarily refers to the fact that, linguistically, we are dealing with morphologically simple roots rather than derived or compounded stems. It is not assumed that such roots always correspond to cognitively basic structures. They can be associated with one or more different processes and products of typification, e.g. prototypes, stereotypes or ideal types (cf. Allwood 1989), but do not need to be. Thus, rather than seeing prototype formation (cf. Berlin and Kay 1969, Rosch 1977 or Lakoff 1987) as the fundamental mode of cognitive organization, it is seen as one of the important types of cognitive operations which are compatible with language. Semantic-epistemic category operations can be applied either to categories on the basic level or to categories on a derived level to form new derived categories. From a conceptual-semantic point of view, this is achieved by a combination of the general cognitive operations of instantiation, abstraction and reification with the operations which correspond to the semanticepistemic categories and with additional semantic-epistemic operations based on similarity or causality. Compare, for example,, watery, which has been formed by similarity-based property extraction from the substance water, or blacken, which has been formed to allow an association of a causative or inchoative relational process with “being black” as a resulting state. The nature of the linguistic processes is not the same in all languages. In Swedish, for example,, blacken would correspond, on the one hand, to svärta (cause to become black) and, on the other hand, to svartna (to become black). In English, to maintain this distinction would require the use of syntactic rather than morphological means, e.g. “cause to become black” and “become black”, respectively. In Chinese, most derivations that in English or Swedish are done with the help of morphology would be done by compounding or syntax.

4.

Roles derived from relations and processes

Language does not only support the formation of basic semantic-epistemic categories, it also supports the act of relating entities through static or dynamic

SEMANTICS

AS

MEANING DETERMINATION

7

relations (processes). In principle, each expression of a static or dynamic relationship between entities, when it is used, highlights properties of the entities which are required for the relation to be applicable and make sense. Simultaneously, compatible properties of the relation and the entities involved are highlighted. The properties which by a particular process or relation are required for a specific argument can be called the role constituting properties. Since there is no a priori limit to how fine-grained these properties may be, there are, in principle, as many roles as there are different relational expressions. It is, however, possible to generalize and to create a list of role types. (This list can be short or long.) In Fillmore (1970), a fairly short list was proposed, while Allwood (1989) suggested a slightly longer list, which is presented below. The roles are grouped together if they have a common component. Since they also have differentiating components, they can be separated whenever there is a need for this. A. B. C. D. E. F.

Cause - motive - reason - origin Result - function - product - effect Direction - purpose - goal Need (of agent, instrument, process, patient) Object - material Agent (agent types in e.g. perception, cognition, emotion or different types of movement) G. Potential (of e.g. agent, instrument or process, such as dispositions) H. Resource (resource agent, resource source) I. Patient - other participants (who potentially can become agents) J. Instrument K. Manner - organization L. Surrounding (except time and space), e.g. physical, social, generic and unspecified The list is not exhaustive but includes some of the most important role types. Since the role designations are somewhat general and vague, explicit definitions are needed to make the roles mutually exclusive. The roles can be used to distinguish different kinds of processes from each other, e.g. the criterion of intentional control (i.e. possible agency) can be used to distinguish activities that require intentional control from other processes. We can picture the roles as in Figure 2. Each relation/process highlights a specific set of roles. For example,, in

8

JENS ALLWOOD

the sentence John loves Mary, the relation love puts John in the agent role (having the agent relation to love) and Mary in the patient role. The arguments (mostly entities) which the relation is applied to occupy these roles with respect to the relation/process. Since the same real course of events can be the source of several different linguistic relational descriptions, one and the same entity, depending on the choice of relational description, can occupy several conceptual roles. Compare the following examples: (1) (2)

John (agent) hired a car (object) from Bill (source) Bill (agent) rented a car (object) to John (patient/goal)

As we can see, the role designations of John and Bill vary depending on which linguistic relational description we choose.

goal agent source cause/origin

Relation/ Process

result

patient instrument location

object

Figure 2. Roles derived from Relations and Processes

SEMANTICS

AS

9

MEANING DETERMINATION

5.

An illustration

I.

Roles of conflict, war and peace

Let me now try to illustrate how the basic semantic-epistemic categories and role analysis can be used in analyzing how we structure concepts through linguistic means. My illustrations will be based on an analysis of the use of the concepts conflict, war and peace, as they appear in the minutes of the Swedish Parliament 1978-79 (cf. Riksdagens snabbprotokoll, riksmötet 1978/79). The analysis was based on a concordance of the material and resulted in tables of the following kind. The table is only part of a more comprehensive table and only shows relational expressions requiring conflict, war, and peace to be seen as container-like entities. Table 1. Conflict, war and peace as containers providing space for other phenomena (in the minutes of the Swedish Parliament 1978-79) Konflikt (conflict) användas i avspänning i befinna sig i delta i dra sig ur dras in i dödas i fatta beslut i fungera i få vara i gå med i gälla i hamna i hålla utanför i inbegripen i inblandad i indragna i invecklad i klara sig i komma i komma i X med komma ur

(used in) X (detente in) X (be in) X/(be at) X (participate in) X (pull out of) X (*pulled into) X (killed in) X (make decisions in) X (function in) X (be left in) X (*join in) X (hold in) X (land in) X (keep out of) X (in) X (involved in) X (involved in) X (*pulled into) X (*embroiled in) X (survive in) X (*come in) X (come in X with) (*come out of) X

2 1 9 6 1 1

krig (war)

fred (peace)

5

1 1 1 1 1 1 2 1 5 1 1 6 2 23

1 20 1 1 1 1 1 1

3

10

JENS ALLWOOD

leva i ligga i botten på liv i lämna i rädda ur råka i skapa i ske i skede i stå i svåranvändbar i söka i tillverka i tvingas ut i uppgift i utveckling i är i ∑

(live in) X (lie at the bottom of X (life in) X (leave in) X (save from) X (*fall into) X (create in) X (happen in) X (phase of) X (*stand in) X (hard to use in) X (search for in) X (manufacture in) X (forced into) X (task in) X (development in) X (is at) X

process identity possession legal restriction attitude speech

1 1 1

1 3 1 1

1 1 1 1

1 1 1 3 1 71

objects of (i) (ii) (iii) (iv) (v) (vi)

1 1

5 49

localization (i) spatial container (ii) temporal (iii) circumstantial

conflict war peace

source instrument

goal abstract agent of

consequence result

(i) inchoative course of events (ii) stative relation (iii) dynamic course of events (iv) causal force (v) premise, content

Figure 3. Role types of “conflict”, “war” and “peace”

9

SEMANTICS

AS

MEANING DETERMINATION

11

When all the usages in the material were generalized, the three concepts were found to occupy the following roles (Figure 3). These role types can be illustrated by the following examples, which are translations of the Swedish originals: 1.

Abstract agent of: (i) Inchoative process: Krig bryter ut (War breaks out) (ii) Stative relation: Konflikten varar (The conflict lasts) (iii) Course of events: Krig rasar (War rages) (iv) Causal force: Krig är en orsak till armod (War is a cause of misery) (v) Premise, content: Fred betyder frihet (Peace means freedom)

2.

Instrument: Användandet av krig som ett politiskt argument (The use of war as a political argument)

3.

Source: Fly från krig (Escape from war)

4.

Localization: (i) Spatial container: Vara i konflikt med (Be in conflict with) (ii) Temporal: Före konflikten (Before the conflict) (iii) Circumstantial: Konfliktsituation (Conflict situation)

5.

Objects of (i) Process: Studera konflikten (Study the conflict) (ii) Identity: Detta är krig (This is war) (iii) Possession: Guds fred (God’s peace) (iv) Legal restriction: Reglera konflikten (Regulate conflict) (iv) Attitude: Älska fred (Love peace) (iv) Speech: Diskutera fred (Discuss peace)

6.

Goal: Arbeta för fred (Work for peace)

7.

Result/consequence: Detta är en orsak till konflikt (This is a cause of conflict)

What the examples illustrate is how a particular relation or process, as expressed by the linguistic context (mostly a predication), determines the role the three concepts can assume. In a particular context, language leads us to structure the concepts in such a way that they become compatible elements of a larger unit.

12 II.

JENS ALLWOOD Semantic-epistemic categorization of conflict, war and peace

What kind of semantic-epistemic categories could be reasonably assigned to the concepts of conflict, war and peace? Table 2 presents some relevant data. Table 2. Semantic-epistemic status of Conflict, War and Peace

State Event Course of Events

konflikt (conflict)

krig (war)

fred (peace)

konflikt råder (conflict exists) konflikt bröt ut (conflict broke out) konflikt pågår (conflict is going on)

krig råder (war exists) krig bröt ut (war broke out) krig pågår (war is going on)

fred råder (peace exists) fred inträdde (peace came into existence) ?fred pågår ?(peace is going on)

All three concepts can be viewed as states or events. Compare the expressions (which all have Swedish translations equivalent to the English expressions) a state of conflict, war or peace and in the event of conflict, war or peace. When it comes to “courses of events”, this category seems more natural for “conflict” and “war” than for “peace”. “Conflict” and “war” are focused on various processes connected with war and conflict, while “peace” seems to be focusing on the end state resulting from processes leading to peace. Compare the following Swedish participle forms, where fredad ((appeased) (resultative)) and krigande (warring-(dynamic)) are possible but not *krigad (warred), *konfliktad (conflicted) or *fredande (appeasing). The three concepts also exhibit a process/product ambiguity. Compare (A) process with (B) product. A. Process (1) ?Freden pågick i 5 år (The peace went on for 5 years) (2) Kriget (konflikten) pågick i 5 år (The war (conflict) went on for 5 years) B. Product (1) Det uppnådda kriget (konflikten) studerades av alla parter (The war (conflict) reached was studied by all parties) (2) Den uppnådda freden studerades av alla parter (The peace reached/achieved was studied by all parties) It is slightly easier to imagine a situation where B(i), rather than B(ii), makes

SEMANTICS

AS

MEANING DETERMINATION

13

sense. Conversely, it is easier to linguistically support a conceptualization as courses of events of war and conflict than of peace, and this makes A(i) less acceptable than A(ii). In short, peace can be most easily conceived of as a state or an event. Conflict and war can most easily be conceived of as events or as courses of events, which implies that the process/product ambiguity for the three concepts is not symmetrical. The concept of “meaning potential” is illustrated by the possibility of viewing concepts like conflict, war and peace in three distinct ways, viz. as a state, as an event and as a course of events, and at another level as an entity or even as a substance, cf. below. All five possibilities are, so to speak, potential determinations of the meanings of the words in different contexts. Thus, the term “meaning potential” also signifies a way of reconceptualizing what traditionally has been called the “polysemic structure” of the meaning of a certain word. (1)

Conflict, war and peace as countable entities (peace ( *peaces/peace treaties One ( war many ( wars (conflict ( conflicts

(2)

(Conflict, war and peace as substances There is more war, conflict, peace to come A little war, conflict, peace might not hurt you

Given the appropriate linguistic contextual means, the meaning of “conflict” (or “war” or “peace”) can be determined in one or another of the directions indicated. The reason for the use of the term “meaning potential” can now be more fully grasped. Meaning is viewed as having potentialities which can be drawn upon by extra- and intralinguistic context. In a given context, the “meaning potential” of a term is determined in a way which is relevant and appropriate for that context. III. Unpacking abstraction All three concepts (conflict, war and peace) can be seen as reified states or courses of events, i.e. abstract objects derived from underlying conceptualizations of states or courses of events which, in turn, consist of relations, processes and entities occupying certain roles in these relations or processes. The idea is illustrated in Figure 4.

14

JENS ALLWOOD

abstract objects conflict, war, peace

course of (events)

State

property

relation

process

Entities in process/relation determined roles

Figure 4. The internal structure of reified states and (courses of) events

On the basis of examples like blue -> blueness, strong —> strength, we might think that language, above all, provides support for processes of abstraction and reification. However, diagram 4 raises the question of whether linguistic processes also allow us to move in the other direction, using linguistic means to unpack and recover more concrete conceptual material. I believe the latter to be the case and will try to illustrate this by considering various linguistic ways in which the term conflict can be turned into a relation. This can be done either directly or indirectly. Let us first consider the direct cases. The examples are given in Swedish (taken from the minutes of the Swedish Parliament 1978-79) with English translations. In most cases they seem to work equally well in Swedish and English. A. Directly 1. Via implicit relation focused by preposition mellan (between) Conflict( state) → relation Construction: Det är X mellan A och B There is X between A and B

SEMANTICS

AS

MEANING DETERMINATION

15

Det är konflikt mellan A och B There is conflict between A and B 2.

Joint venture construction Construction: A har X med B A has X with B A har konflikt med B A has conflict with B

3.

Adversative construction Construction: A har X mot B A has X against B A har konflikt mot B A has conflict against B

B. Indirectly (via metaphorical extension), 1. Conducting vehicle Construction: A för X mot B A conducts X against/toward B A för krig mot B A conducts war against B C. Container constructions (i) A är i X med B A is in X with B (ii) A kommer i X med B A come into X with B (iii) A drar sig ur X med B A pulls out of X with B

konflikt krig konflikt krig konflikt krig fred

All three cases are possible with conflict and war, and number (iii) is also possible with peace. The examples clearly show that we not only have linguistic means for abstraction and reification, but also for unpacking and recovering conceptual material underlying abstractions and reifications. In both cases, the linguistic means can be both morphological and syntactic, and they can involve conventionalized metaphorical extension, as in the above cases of linguistic expressions based on ideas of “vehicle conducting” and “containers”. Semantically speaking, we can, for example, unpack the reified course of events krig (war)

16

JENS ALLWOOD

by forming krig-are (warrior) and krig-a (to make war). In fact, we seem to have a store of linguistic constructions (cf. Fillmore 1988) which can be used for various purposes of semantic-conceptual structuring.

6.

Concluding Remarks

In this paper, I have tried to demonstrate an approach to semantics which is characterized by the assumption that language provides a set of tools and mechanisms for structuring information which is maximally useful in human action and interaction. One of the ways in which this is achieved seems to be by providing regular linguistic support for a number of conceptual (semanticepistemic) operations. These allow information to be flexibly structured in a regular and predictable way, probably in harmony with certain innate predispositions, to meet requirements of context such as those given by the currently relevant linguistic and extra-linguistic activity and purpose, the perceptual environment and the stored background information of the discourse. The approach has been illustrated by discussing operations pertaining to certain basic semantic-epistemic categories and role relations. The primary goal of the approach is to gain a deeper understanding of human conceptual and linguistic capacities, but I believe some of the features of the approach can also be used for purposes such as: -

conceptual analysis historical conceptual-semantic studies comparative socio-cultural analysis studies of the relation between grammar and lexicon modeling linguistic/semantic processing

Acknowledgments I would like to thank Elisabeth Ahlsén and Joakim Nivre for comments and discussion.

SEMANTICS

AS

MEANING DETERMINATION

17

References Allwood, J. 1989 Aristotle. 1938

Om begrepp - deras bestämning, analys och kommunikation (ms). Göteborgs universitet, Institutionen för lingvistik.

Categories. On Interpretation and Prior Analytics. Translated by H.P. Cooke and H. Tredennick. London and Cambridge, Mass. Loeb Classical Library. Harvard University Press and W. Heineman Ltd. Barwise, J., & Perry, J. 1983 Situations and Attitudes, Cambridge, Mass., MIT Press. Berlin, B. and Kay, P. 1969 Basic Color Terms. Berkeley, University of California Press. Fillmore, C. 1970 Subjects, Speakers and Roles. Ohio Working Papers in Linguistics, No. 4. The Ohio State University, Columbus, Ohio. Fillmore, C. 1988 The mechanisms of construction grammar. In Papers for the Fourteenth Annual Meeting of the Berkeley Linguistics Society, pp 163 -182 Haiman, J. 1980 “Dictionaries and Encyclopedias”. Lingua 50:329-57. Husserl, E. 1913 Logische Untersuchungen II. Tübingen, Max Niemeyer Verlag. 1968. Reprinted from the 2nd revised edition, Halle, 1913. Kant, I. 1975 Die Drei Kritiken. Stuttgart, Alfred Kroner Verlag. Konkordans: Riksdagens snabbprotokoll, riksmötet 1978/79. (Concordance of the quick minutes of the Swedish Parliament 1978/79). Dept of Swedish language, Språkbanken, University of Göteborg. Lakoff, G. 1987 Women, Fire and Dangerous Things. Chicago, The University of Chicago Press. Langacker, R.W. 1987 Foundations of Cognitive Grammar, Vol. 1, Theoretical Prerequisites. Stanford, Stanford University Press. Quine, W.V.O. 1953 “Two dogmas of empiricism”. In From a Logical Point of View. Harvard, Harvard University Press. Rommetveit, R. 1974 On Message Structure. London, John Wiley & Sons. Rosch, E. 1977 “Human categorization.” In Warren (ed.) Studies in Cross-Cultural Psychology. London, Academic Press.

SOME TENETS OF COGNITIVE SEMANTICS

19

Some Tenets of Cognitive Semantics Peter Gärdenfors Department of Cognitive Science, Lund University

1.

Realistic versus cognitive semantics

As an introduction, I want to contrast two approaches to formal semantics: one realistic and one cognitive. The fundamental difference concerns what kinds of entities are the meanings of words. According to the realistic approach to semantics, the meaning of an expression is something out there in the world. Cognitive semantics, on the other hand, identifies meanings of expressions with mental entities. Realistic semantics comes in two flavors: extensional and intensional. In the extensional type of semantics, one starts out from a language L, which may or may not be defined in formal terms, and maps the constituents of L onto a “world.” Names are mapped onto objects, predicates are mapped onto sets of objects or relations between objects, etc. By compositions of these mappings, sentences end up being mapped onto truth values. The main objective of this kind of semantics is to determine truth conditions for the sentences in L. A consequence of this approach is that the meaning of an expression is independent of how individual users understand it. As an account of natural language, the extensional theory of reference implicit in this kind of semantics was soon found to be wanting. In order to handle some of the problems, so-called intensional semantics was developed by philosophers, logicians and linguists. In this kind of semantics, the language L is mapped onto a set of possible worlds instead of only a single world. Still, the goal of the semantics is to provide truth conditions for the sentences in L. The meaning of a sentence is taken to be a proposition, which is identified with a set of possible worlds — the set of worlds where the sentence is true. The classic form of this semantics is Kripke’s (1959) semantics for

20

PETER GÄRDENFORS

modal logics. With respect to natural language, Montague’s (1974) work has provided a lot of inspiration for intensional semantics. The second paradigm of semantics is cognitivistic. The core idea of this approach is that meanings of expressions are mental. A semantics is seen as a mapping from the linguistic expressions to cognitive structures. Language itself is seen as part of the cognitive structure, and not as an entity of independent standing. Within cognitive semantics, the emphasis is on lexical meaning rather than on the meaning of sentences. This kind of semantics will be presented further in the following section. It is interesting to note that if de Saussure is read properly, he proposes a cognitive analysis of the signification relation. The following excerpt from the first paragraphs of the first chapter illustrates this (de Saussure 1966: 65-66): [F]or some people a language, reduced to its essentials, is a nomenclature: a list of terms corresponding to a list of things. […] This conception is open to a number of objections. It assumes that ideas already exist independently of words […]. It does not clarify whether the name is a vocal or psychological entity […]. Furthermore, it leads one to assume that the link between a name and a thing is something quite unproblematic, which is far from being the case. None the less, this naive view contains one element of truth, which is that linguistic units are dual in nature, comprising two elements. […] the two elements involved in the linguistic sign are both psychological and are connected in the brain by an associative link. This is a point of major importance. A linguistic sign is not a link between a thing and a name, but between a concept and a sound pattern.

However, it should be emphasized that even if de Saussure presents concepts as major elements in his definition of a sign, it does not follow that he would have endorsed the cognitive approaches to semantics and grammar that have been developed in later years. My first aim in this paper is to present some of the main tenets of cognitive semantics. I will contrast these tenets with traditional types of semantics, but my goal is not primarily to criticize these kinds of semantics. In the third section, I will present the bare bones of a formal cognitive semantics.1 One limitation of cognitive semantics as it has developed is that it focuses on individuals’ meanings of words. However, there are several social aspects of language that should also be accounted for within a cognitivistic program, and in the fourth section, I will take an example of Putnam’s as a point of

SOME TENETS OF COGNITIVE SEMANTICS

21

departure for a discussion of how a social meaning can be determined on the basis of individual meanings.

2.

Six tenets of cognitive semantics

I shall give a programmatic presentation of cognitive semantics in the form of six tenets together with some comments.2 The approach of a cognitively oriented semantics will be contrasted with the “realistic” view. Prime examples of works in the cognitive tradition are Lakoff’s (1987) and Langacker’s (1987). Related versions of cognitive semantics can be found in the writings of Jackendoff (1983, 1990), Fauconnier (1985), Talmy (1988), Sweetser (1990) and many others. There is also a French semiotic tradition, exemplified by Desclés (1985) and Petitot-Cocorda (1985), which shares many features with the American (mainly Californian) group. I.

Meaning is conceptualization in a cognitive model (not truth conditions in possible worlds).

The prime slogan for cognitive semantics is: Meanings are in the head. More precisely, a semantics for a language is seen as a mapping from the expressions of the language to some mental entities. A consequence of the cognitivist position that puts it in conflict with many philosophical semantic theories is that no form of truth conditions of an expression is necessary to determine its meaning. The truth of expressions is considered to be secondary, since truth concerns the relation between the mental structure and the world. To put it tersely: Meaning comes before truth.3 II. Cognitive models are mainly perceptually determined (meaning is not independent of perception). Since the cognitive structures in our heads are connected to our perceptual mechanisms, directly or indirectly, it follows that meanings are, at least partly, perceptually grounded. This, again, is in contrast to traditional realistic versions of semantics which claim that, since meaning is a mapping between the language and the external world (or several worlds), meaning has nothing to do with perception.

22

PETER GÄRDENFORS

We can talk about what we see and hear. Conversely, we can create pictures, mental or real, of what we read or listen to. This means that we can translate between the visual form of representation and the linguistic code.4 A central hypothesis of cognitive semantics is that the way we store perceptions in our memories has the same form as the meanings of words. III. Semantic elements are based on spatial or topological objects (not symbols that can be composed according to some system of rules). In contrast to the Mentalese of Fodor and others, the mental structures applied in cognitive semantics are the meanings of the linguistic expressions; there is no further step of translating conceptual structure to something outside the mind. Furthermore, instead of being a symbolic system having a syntactic structure like “Mentalese”, the conceptual schemes that are used to represent meanings are often based on geometric or spatial constructions. As a framework for a geometric structure used in describing a cognitive semantics, I have proposed (Gärdenfors 1988, 1991, 1996a, 1996b, 1997) the notion of a conceptual space. A conceptual space consists of a number of quality dimensions. Examples of quality dimensions are: color, pitch, temperature, weight, and the three ordinary spatial dimensions. Some of the dimensions are closely related to what is produced by our sensory receptors, but there are also quality dimensions that are of an abstract non-sensory character. The notion of a dimension should be understood literally. It is assumed that each of the quality dimensions is endowed with certain topological or metric structures. For example,, “time” is a one-dimensional structure which we conceive of as being isomorphic to the line of real numbers. Similarly, “weight” is one-dimensional with a zero point, isomorphic to the half-line of non-negative numbers. Some quality dimensions have a discrete structure, i.e., they merely divide objects into classes, e.g., the sex of an individual. Some of the quality dimensions seem to be innate and to some extent hardwired in our nervous system, as for example, color, pitch, and probably also ordinary space. Other dimensions are presumably learned. Learning new concepts often involves expanding one’s conceptual space with new quality dimensions. Functional properties used for describing artifacts may be an example here. Still other dimensions may be culturally dependent. “Time” is a good example — in contrast to our linear conception of time, some cultures

SOME TENETS OF COGNITIVE SEMANTICS

23

conceive of time as circular, so that the world keeps returning to the same point in time, and in other cultures it is hardly meaningful at all to speak of time as a dimension. Finally, some quality dimensions are introduced by science (see Gärdenfors 1993b). There is a strong similarity between the notion of a conceptual space and the domains as used in Langacker’s (1987) semantic theory. The following quotation from Langacker (1987: 5) concerning his notion of “domains” strongly supports this thesis: “What occupies the lowest level in conceptual hierarchies? I am neutral in regard to the possible existence of conceptual primitives. It is however necessary to posit a number of ‘basic domains,’ that is, cognitively irreducible representational spaces or fields of conceptual potential. Among these basic domains are the experience of time and our capacity for dealing with two- and three-dimensional spatial configurations. There are basic domains associated with various senses: color space (an array of possible color sensations), coordinated with the extension of the visual field; the pitch scale; a range of possible temperature sensations (coordinated with positions on the body); and so on. Emotive domains must also be assumed. It is possible that certain linguistic predications are characterized solely in relation to one or more basic domains, for example, time for (BEFORE), color space for (RED), or time and the pitch scale for (BEEP). However, most expressions pertain to higher levels of conceptual organization and presuppose nonbasic domains for their semantic characterization.”

IV. Cognitive models are primarily image-schematic (not propositional). Image-schemas are transformed by metaphoric and metonymic operations (which are treated as exceptional features in the traditional view). The most important semantic structure in cognitive semantics is that of an image schema. Image schemas have an inherent spatial structure. Lakoff (1987) and Johnson (1987) argue that schemas such as “container,” “sourcepath-goal” and “link” are among the most fundamental carriers of meaning. They also claim that most image schemas are closely connected to kinesthetic experiences. Metaphors and metonymies have been notoriously difficult to handle within realist semantic theories. In these theories, such linguistic figures have been treated as a deviant phenomenon that has been ignored or incorporated via special stylistic rules. In contrast, they are given key positions within cognitive semantics.5

24 V.

PETER GÄRDENFORS Semantics is primary to syntax and partly determines it (syntax cannot be described independently of semantics).

This thesis is anathema to the Chomskyan tradition within linguistics. Within Chomsky’s school, grammar is a formal calculus, which can be described via a system of rules, where the rules are formulated independently of the meaning of the linguistic expressions. Semantics is something that is added, as a secondary independent feature, to the grammatical rule system. Similar claims are made for the pragmatic aspects of language. Within cognitive linguistics, semantics is the primary component (which, in the form of perceptual representations, existed before language was fully developed). The structure of the semantic schemas puts constraints on the possible grammars that can be used to represent those schemas. To give a trivial example of how semantics constrains syntax, consider the role of tenses. In a Western culture where time is conceived of as a line, it is meaningful to talk about three basic kinds of time: past, present and future. This is reflected in the grammar of tenses in most languages. However, in cultures where time has a circular structure, or where time cannot be given any spatial structure at all, it is not meaningful to make a distinction between, say, past and future. And there are languages which have radically different tense structures, which reflect a different underlying conceptual structuring of time. The Chomskyan tradition within linguistics has been dominated by syntactic studies. Since grammars are represented by formal rules, they are suitable for computer implementations. This kind of work has indeed been the main focus of computational linguistics. Within cognitive semantics, computer-friendly representations are much more rare. One notable exception is Holmqvist (1993, 1994, this volume), who develops implementable representations of image schemas and other concepts from the cognitive linguists. To some extent, he is inspired by Langacker’s compositional image schemas and Lang’s spatial models (see Lang, Carstensen and Simmons 1991), but he extends their formalisms to a much richer computational structure. In his model (1994), he also utilizes an old idea of Behaghel to generate grammatical structure solely from the valence expectations of different lexical items. The result is something that looks like a rule-governed syntax, albeit there is no single explicit syntactic rule in the system.

SOME TENETS OF COGNITIVE SEMANTICS

25

VI. Concepts show prototype effects (instead of following the Aristotelian paradigm based on necessary and sufficient conditions). The classical account of concepts within philosophy is Aristotle’s theory of necessary and sufficient conditions. However, one very often encounters problems when trying to apply the Aristotelian theory to concepts represented in natural language. As a result of a growing dissatisfaction with the classical theory of concepts, prototype theory was developed within cognitive psychology.6 Within cognitive semantics, one attempts to account for prototype effects of concepts. A concept is often represented in the form of an image schema and such schemas can show variations just like concepts normally do. This kind of phenomenon is much more difficult to model using traditional symbolic structures.

3.

Fundamentals of a formal cognitive semantics

Here, I will only outline the first steps in developing a cognitive semantics based on conceptual spaces. According to the cognitive view, semantics is a relation between language and a cognitive structure, and I submit here that the appropriate framework for the cognitive structure is a conceptual space. On this assumption, formulating a semantics for a specific language can be broken down into two major steps: (1) To specify the mapping between the lexicon of the language and the appropriate conceptual spaces. (2) To describe the operations on the image schemas (which are defined with the aid of the conceptual space) that correspond to different syntactic formation rules. In this section, I will discuss only the first step. Following the technical style of philosophical semantics, one can define an interpretation for a language L as a mapping of the components of L onto a conceptual space. As a first element of such a mapping, individual names are assigned vectors (i.e., points in the conceptual space) or partial vectors (i.e., points with some arguments undetermined). In this way, each name (referring to an individual) is allocated a specific color, spatial position, weight, temperature, etc. Following Stalnaker (1981: 347), a function which maps the individuals onto a conceptual space will be called a location function. The fundamental lexical hypothesis is then the following: Predicates in

26

PETER GÄRDENFORS

natural languages generally denote connected regions in some conceptual space.7 For instance, I conjecture that all color terms in natural languages express connected regions with respect to the psychological representation of the three color dimensions. It is well-known that different languages carve up the color circle in different ways, but all carving seems to be done in terms of connected sets. Technically, as a second element of the interpretation mapping, the basic predicates of the language are assigned regions in the conceptual space.8 Such a predicate is satisfied by an individual only when the location function locates the individual at one of the points included in the region assigned to a predicate.9 Some of the so-called intensional predicates, like “tall,” “former” or “alleged,” are not basic, but “secondary” in the sense that their regions cannot be described independently of other predicates. There is no class of all “tall” objects. Rather, “tall” requires a contrast class, like “tall woman” or “a tall tower” in order to be well defined. In Gärdenfors 1997, I provide an analysis in terms of conceptual spaces of the role of contrast classes in determining the reference of certain such secondary predicates. If we assume that an individual is completely determined by his/her/its set of properties, then all points in the conceptual space can be taken to represent possible individuals. On this account, a possible individual is a cognitive notion that need not have any form of reference in the external world. This construction will avoid many of the problems that have plagued other accounts of possible individuals. A point in a conceptual space will always have an internally consistent set of properties — since e.g., “blue” and “yellow” are disjoint predicates in the color space, it is not possible that any individual will be both blue and yellow (all over). There is no need for meaning postulates in order to exclude such contradictory predicates. One important contrast to the traditional intensional semantics is that the cognitive semantics outlined here does not presume the concept of a possible world. However, different location functions describe alternative ways that individuals may be located in a conceptual space. Consequently, these location functions can be given the same role as possible worlds in the traditional semantics. This means that we can define the notion of a possible world as a possible location function, and this can be done without introducing any new semantic primitives to the theory.

SOME TENETS OF COGNITIVE SEMANTICS 4.

Some social aspects of meaning

I.

Putnam’s counter-arguments

27

A heavy attack against the very possibility of cognitivistic semantics has been launched by Putnam (1975, 1988). He contends that meanings can’t be in the head. His argument starts from the following assumptions about meaning and mental representations, all of which seem to be accepted by the cognitive semanticists (Putnam 1988: 19): 1. 2.

3. 4.

Every word the speaker uses is associated in his mind with a certain mental representation. Two words are synonymous (have the same meaning) only when they are associated with the same mental representation by the speakers who use those words. The mental representation determines what the word refers to, if anything. Putnam claims that these three conditions cannot be simultaneously satisfied. The reason is that we “cannot individuate concepts and beliefs without reference to the environment” (1988: 73).

A central part of his argument can be illustrated by the following example (Putnam 1975: 226-227): “Suppose you are like me and cannot tell an elm from a beech tree. We still say that the extension of ‘elm’ in my idiolect is the same as the extension of ‘elm’ in anyone else’s, viz., the set of all elm trees, and that the set of all beech trees is the extension of ‘beech’ in both of our idiolects. Thus ‘elm’ in my idiolect has a different extension from ‘beech’ in your idiolect (as it should). Is it really credible that this difference in extension is brought about by some difference in our concepts? My concept of an elm tree is exactly the same as my concept of a beech tree (I blush to confess). (This shows that the identification of meaning ‘in the sense of intension’ with concept cannot be correct, by the way). […] Cut the pie any way you like, meanings just ain’t in the head!”

The upshot seems to be that meanings must refer to something non-cognitive. A related argument has been presented by Burge (1979). In my opinion, the lesson to be learned from Putnam’s argument is not that cognitive semantics is impossible, but that it has generally forgotten about the social structure of language. In Gärdenfors (1993a), I argue that the social meanings of the expressions of a language are indeed determined from their

28

PETER GÄRDENFORS

individual meanings, i.e., the meanings the expressions have for the individuals, together with the structure of linguistic power that exists in the community. In contrast to Putnam and others, I claim that no reference to the external world is needed to handle the problem he presents. The question of linguistic power concerns who decides on what is the “correct meaning” of an expression in a society. In other words, who are the masters of meaning? Given some rather weak assumptions concerning how the meaning of a word is determined, it can be shown that there are two basic types of power structures: oligarchic and democratic.10 An oligarchic (or dictatorial) power structure would arise when the social meanings of words are determined by a group of linguistic experts writing dictionaries, encyclopedias, handbooks, etc. When a language user is in doubt about the meaning of a locution that falls under the realm of the oligarchy, he would rely on the judgments of these experts. In contrast, a democratic power structure would control those parts of the language where linguistic meaning is identified with “common usage.” For such parts of the language, a dictator or a small group of speakers cannot, by themselves, change the meaning of an expression; for this, the consent of almost all language users is required. This is analogous to prices in a free market — a single agent cannot decide to change the price of a product. I do not claim that all parts of the semantics of a language are governed by the same power structure. A more realistic description is to say that a language is a conglomerate of several sublanguages, each with its own conditions of linguistic power. The semantics of the language of lawyers is determined by criteria that are different from those of the language of entomologists; which in turn are different from those used for slang expressions. For lawyers’ and entomologists’ expressions, the power structures may very well be oligarchic, while the use of slang is a more democratic business. In support of this, it seems as if hedge words like “technically” can be used for expressions that are governed by an oligarchic power structure, but not for those the meaning of which are determined democratically: “Technically, a spider is not an insect” is correct, but “Technically, a hooker is a prostitute” sounds odd.11 Putnam (1975: 227-229) describes something very much like an oligarchic power structure in his hypothesis about the “division of linguistic labor.” This hypothesis maintains that every linguistic community “[…] possesses at least some terms whose associated “criteria” are known only to a subset of the speakers who acquire the terms, and whose use by the other speakers depends upon a structured cooperation between them and the speakers in the relevant

SOME TENETS OF COGNITIVE SEMANTICS

29

subsets” (1975: 228). He claims that the hypothesis accounts for the failures of the assumptions that knowing the meaning of a locution is just a matter of being in a certain psychological state, and that the meaning of a term determines its extension. Putnam’s argument for this is that [w]henever a term is subject to the division of linguistic labor, the “average” speaker who acquires it does not acquire anything that fixes its extension. In particular his individual psychological state certainly does not fix its meaning; it is only the sociolinguistic state of the collective linguistic body to which the speaker belongs that fixes the extension. (1975: 229).

The last remark indicates that Putnam thinks of the fixation of social meaning in much the same way as in my analysis. However, it seems as if he misses the possibility of democratic power structures, which is a different way of determining social meaning.12 In a more recent book, Putnam (1988) also discusses “conceptual role” semantics, in particular in relation to natural-kind terms. He argues that the meaning of such terms cannot be given in terms of their conceptual roles only, but “once we have identified a word as a natural-kind term, then we determine whether it is synonymous with another natural-kind term primarily on the basis of the extensions of the two words” (1988: 50). Here, extension is, of course the set of things in the world that the word applies to. So natural-kind terms presume a realistic component for their semantics according to Putnam. But, how do we know when something is a natural-kind term? Putnam is aware of the problem: “Some words which were intended to be natural-kind terms turn out not to refer to natural kinds. “Phlogiston” was intended to be the name of a natural kind, but it turned out that there was no such natural kind. And similarly for “ether” and “caloric”. In these cases, it does seem that something like conceptual role is the dominant factor in meaning, for obvious reasons; we don’t want to say that the words “ether” and “caloric” and “phlogiston” are synonymous just because they have the same (empty) extension. […] Indeed the conceptual role theory comes closest to being true in the case of words with an empty extension.” (1988: 50)

However, here he seems to rely on some form of realist essentialism. If “phlogiston” could turn out not to be a natural-kind term, so can “water” and “gold,” unless one assumes that natural kinds exist independently of language and cognition. And this is the kind of essentialist assumption Putnam needs to make in order to argue against the cognitive approach. But this is putting the

30

PETER GÄRDENFORS

cart before the horse: He assumes that a term is defined in terms of realist notions in order to show that it cannot be given a purely conceptual meaning. In contrast to Putnam, I claim that the meaning of natural-kind terms like “gold” and “water” do change because of changes in the linguistic power structure (see Langacker 1987: 154-166). I believe that this kind of meaning change is common in science in connection with scientific revolutions. For example, before the Copernican revolution “the earth” meant something that did not move, and before Einstein “mass” was something that was a constant of an object. Perhaps even clearer examples are found in the social sciences. In particular, Foucault’s “archeological” investigations of the terminologies in various areas (“madness” for instance) show how modifications in the power structure in a society can result in radical changes in the language. II. Linguistic modality as expressions of social power Let me finally turn to another aspect of semantics that, in my opinion, requires taking social interactions into account. From the early works of Kripke and on, one of the major successes of intensional semantics was that it provided a formal semantics for modal expressions. However, the modals being analyzed were of an abstract philosophical nature, and hardly any attention was paid to linguistic data on modal expressions. Within cognitive semantics, there are now several works on modals, where perhaps Talmy’s (1988) work on “force dynamics” and Sweetser’s (1990) extension to the “epistemic” use of modals have been the most influential. In Talmy’s analysis, physical forces are seen as more fundamental than the social ones for describing the semantics of modals. By metaphorical extension, the expressions used to express physical forces are used in the “psychological, social, inferential, discourse, and mental-model domains of reference and conception” (Talmy 1988:49, the abstract). In a recent study (Winter and Gärdenfors 1995), a cognitive semantics (or rather pragmatics) is proposed that covers the whole field of (Swedish) modal verbs. In contrast to Talmy’s physically oriented forces, the basic notions in the analysis are, firstly, the power relations between the speaker and the hearer (and sometimes a third party), and secondly, the expectations of the actors with respect to the action governed by the modal verb. In the analysis of Winter and Gärdenfors (1995), the central elements of the speaker’s and the listener’s mental representations are the social power

SOME TENETS OF COGNITIVE SEMANTICS

31

relations that hold between various agents. The objects of power are actions, for example, the action of blasting a safe. I can blast it myself, but if I have power over you, I can also command you to do it. The second important factor of a speech situation is the agents’ attitudes to the relevant actions.13 For example, I may want to blast the safe, while you may not want this action to be performed. To give an example of how these notions are applied to modals, let us consider the expression “want to p,” where p is an action. The expression occurs typically when the speaker wants p but is uncertain whether the hearer has the same attitude. Thus, if the speaker is the subordinate of the two, and the hearer the one in power, the speaker can say “I want to p” to mark his attitude and to signal that he is uncertain of the hearer’s attitude. Or, in the reverse power relation, the speaker can, instead of directly exerting his power, say “Do you want to p?” when he expects the hearer not to be aware of his attitude. The speaker will then, by conversational implicature, expect the hearer to understand his attitude (and, consequently, perform the action p). Another example is the use of “shall.” The typical use of this modal occurs in situations where the speaker is in power, when the speaker wants p, he expects the hearer not to want p, and he does not expect the hearer to have a correct expectation of his, i.e., the speaker’s, attitude. In this situation, “You shall p!” is used to inform the hearer about the speaker’s attitude, and to remind him of the power relation. Thus, this analysis of modal expressions is also an example of a cognitive model containing a social element. In brief, it is proposed that the primary meaning of modals is to express power relations. Within the philosophical tradition, earlier analyses of modal expressions have, almost exclusively, been based on possible worlds and relations between worlds as semantic primitives. Indeed, the first modal notions to be analyzed were those of necessity and possibility. However, there is nothing in the structure of possible worlds semantics that is suitable for describing social power relations, but such features must be added by more or less ad hoc means.

5.

Conclusion

Cognitive semantics is still rather undeveloped. Its most detailed applications have been areas where language is tightly bound to perception as, for example,

32

PETER GÄRDENFORS

in spatial prepositions. Cognitive semantics has also offered new insights into the mechanisms of metaphors. Its strength lies mainly in the analysis of lexical items, even though there are interesting attempts to explain syntactic features by cognitive means (e.g., Langacker 1987, Holmqvist 1993, 1994). In this paper, I have tried to summarize the foundations of cognitive semantics in the form of six general tenets, and I have presented the skeleton of a formal cognitive semantics based on conceptual spaces. This kind of semantics has been contrasted with the more traditional extensional and intensional types of semantics. Putnam has argued that a pure cognitive semantics, which puts meanings in the heads of individual speakers, is impossible. I have attempted to refute his arguments by pointing out that, if the social aspects of language, in particular different kinds of power relations, are taken into account, Putnam’s arguments are no longer valid. There are areas where traditional semantics is strongly developed and where cognitive semantics is weak, for example, quantifiers and modal expression. However, I have presented a recent undertaking to supply a cognitive analysis of modals, again in a social setting including power relations. Furthermore, a cognitively oriented analysis of quantifiers has recently been proposed by Moxey and Sanford (1993).

Acknowledgments Research for this paper has been supported by the Swedish Council for Research in the Humanities and Social Sciences. The second and third sections draw partly on Gärdenfors (to appear b), and section 4 borrows an argument from Gärdenfors (1993a).

Notes 1.

A rich source for this purpose is Lakoff’s (1987) book, which is a lengthy criticism of what he calls “objectivist semantics.”

2.

A slightly more detailed presentation can be found in Gärdenfors (to appear b).

3.

Cognitive semantics should be separated from Fodor’s (1981) “Language of Thought” hypothesis. Fodor also uses mental entities to represent linguistic information. This is his

SOME TENETS OF COGNITIVE SEMANTICS

33

“language of thought”, which is sometimes also called “Mentalese.” However, the mental entities constituting Mentalese form a language with syntactic structures governed by some recursive set of rules. And when it comes to the semantics of Mentalese, Fodor is still a realist and relies on references in the external world as well as truth conditions. 4.

For a discussion of the implications of this translatability for semantics, see Jackendoff (1987) and (1990).

5.

See e.g. Broström (1994), Brugman (1981), Gärdenfors (to appear a), (to appear b), Indurkya (1986), Lakoff (1987), (to appear), Lakoff and Johnson (1980), Sweetser (1990), and Thorangeau and Sternberg (1982).

6.

See e.g. Rosch (1975), (1978), Mervis and Rosch (1981), Smith and Medin (1981), and Lakoff (1987) for extended discussions of the theory.

7.

A more precise and powerful idea is the following criterion (see Gärdenfors 1990, 1991) where the topological characteristics of the quality dimensions are utilized to introduce a spatial structure into properties: A natural property is a convex region of a conceptual space. A convex region is characterized by the criterion that for every pair of points v1 and v2 in the region and all points in between v1 and v2 are also in the region. The motivation for the criterion is that if some objects which are located at v1 and v2 in relation to some quality dimension (or several dimensions) and both are examples of the property P, then any object that is located between v1 and v2 on the quality dimension(s) will also be an example of P.

8.

For a criticism of the notion of a property within intensional semantics, and for an alternative account based on conceptual spaces, see Gärdenfors (1991).

9.

Relations can be treated in a similar way (see Holmqvist 1993).

10.

This partitioning is supported by a couple of theorems in Gärdenfors (1993a).

11.

See Lakoff (1987: 122-125).

12.

The only remark in this direction isthe following: “It would be of interest, in particular, to discover if extremely primitive peoples were sometimes exceptions to this hypothesis (which would indicate that the division of linguistic labor is a product of social evolution), or if even they exhibit it” (1975: 229).

13.

Attitudes to actions concern the agents’ preferences, and should not be confounded with so-called propositional attitudes, e.g., believing or hoping.

References Broström, S. 1994

The Role of Metaphor in Cognitive Semantics, Lund: Lund University Cognitive Studies 31.

34 Brugman, C. 1981 Burge, T. 1979

Declés, J.-P. 1985

PETER GÄRDENFORS

Story of Over. Bloomington: Indiana Linguistics Club. “Individualism and the mental”. Midwest Studies in Philosophy, vol. 4: Studies in Metaphysics, 73-121. Minneapolis: University of Minnesota Press. Représentation des connaissances (Actes Semiotiques - Documents, VII, 69-70) Paris: Institut National de la Langue Française.

Fauconnier, G. 1985 Mental Spaces. Cambridge, MA: MIT Press. Fodor, J. A. 1981 Representations. Cambridge, MA: MIT Press. Gärdenfors, P. 1988 “Semantics, conceptual spaces and the dimensions of music”. Essays on the Philosophy of Music (Acta Philosophica Fennica, vol. 43), ed. by V. Rantala, L. Rowell, & E. Tarasti, 9-27. Helsinki. Gärdenfors, P. 1990 “Induction, conceptual spaces and AI”. Philosophy of Science 57. 78-95. Gärdenfors, P. 1991 “Frameworks for properties: Possible worlds vs. conceptual spaces”. Language, Knowledge and Intentionality (Acta Philosophica Fennica, vol. 49) ed. by L. Haaparanta, M. Kusch, & I. Niiniluoto, 383-407. Helsinki. Gärdenfors, P. 1993a “The emergence of meaning”. Linguistics and Philosophy 16, 285-309. Gärdenfors, P. 1993b “Induction and the evolution of conceptual spaces”. Charles S. Peirce and the Philosophy of Science ed. by E. C. Moore, 72-88. Tuscaloosa: The University of Alabama Press. Gärdenfors, P. 1996a “Mental representation, conceptual spaces and metaphors”. Synthese 106, 21-47. Gärdenfors, P. 1996b “Conceptual spaces as a basis for cognitive semantics”. Philosophy and Cognitive Science, ed. by A. Clark et al., 159-180. Dordrecht: Kluwer. Gärdenfors, P. 1997 “Meaning as conceptual structures”. Mindscapes: Philosophy, Science, and the Mind, ed. by M. Carrier and P. Machamer, 61-86. Pittsburgh: Pittsburgh University Press. Herskovits, A. 1986 Language and Spatial Cognition: A Interdisciplinary Study of the Prepositions in English. Cambridge: Cambridge University Press. Holmqvist, K. 1993 “Implementing cognitive semantics”. Lund: Lund University Cognitive Studies 17.

SOME TENETS OF COGNITIVE SEMANTICS

35

Holmqvist, K. 1994 “Conceptual engineering I: From morphemes to valence relations”. Lund: Lund University Cognitive Studies 28. Holmqvist, K. “Implementing cognitive semantics - Overview of the semantic composition processes and insights into the grammatical composition processes”. Indurkhya, B. 1986 “Constrained semantic transference: A formal theory of metaphors”. Synthese 68. 515-551. Jackendoff, R. 1983 Semantics and Cognition. Cambridge, MA: MIT Press. Jackendoff, R. 1987 “On Beyond Zebra: The relation of linguistic and visual information”. Cognition 26. 89-114. Jackendoff, R. 1990 Semantic Structures. Cambridge, MA: MIT Press. Johnson, M. 1987 The Body in the Mind: The Bodily Basis of Reason and Imagination. Chicago, IL: University of Chicago Press. Kripke, S. 1959 “A completeness theorem in modal logic”. Journal of Symbolic Logic 24. 1-24. Lakoff, G. 1987 Women, Fire, and Dangerous Things. Chicago, IL: University of Chicago Press. Lakoff, G. “The contemporary theory of metaphor”. Metaphor and Thought (2nd edition) ed. by A. Ortony. Cambridge: Cambridge University Press. Lakoff, G. and Johnson, M. 1980 Metaphors We Live By. Chicago, IL: University of Chicago Press. Lang, E., Carstensen, K-U, & Simmons, G. 1991 Modeling Spatial Knowledge on a Linguistic Basis. Berlin: SpringerVerlag. Langacker, R. W. 1987 Foundations of Cognitive Grammar, Vol. 1. Stanford, CA: Stanford University Press. Mervis, C. & Rosch, E. 1981 “Categorization of natural objects”. Annual Review of Psychology 32. 89115. Montague, R. 1974 Formal Philosophy, ed. by R. H. Thomason. New Haven: Yale University Press. Moxey, L. M. & Sanford, A. J. 1993 Communicating Quantities: A Psychological Perspective. Hove: Lawrence Erlbaum Associates.

36

PETER GÄRDENFORS

Petitot-Cocorda, J. 1985 Morphogenèse du Sens I. Paris: Presses Universitaires de France. Putnam, H. 1975 “The meaning of ‘meaning’”. In Mind, Language and Reality. Cambridge: Cambridge University Press. Putnam, H. 1988 Representation and Reality. Cambridge, MA: MIT Press. Rosch, E. 1975 “Cognitive representations of semantic categories”. Journal of Experimental Psychology: General 104. 192-233. Rosch, E. 1978 “Prototype classification and logical classification: The two systems”. In New Trends in Cognitive Representation: Challenges to Piaget’s Theory ed. by E. Scholnik. 73-86. Hillsdale, NJ: Lawrence Erlbaum Associates,. Saussure, F. de 1966 Course in General Linguistics, New York: McGraw-Hill. Smith, E. & Medin, D. L. 1981 Categories and Concepts. Cambridge, MA: Harvard University Press. Stalnaker, R. 1981 “Antiessentialism”. Midwest Studies of Philosophy 4. 343-355. Sweetser, E. 1990 From Etymology to Pragmatics. Cambridge: Cambridge University Press. Talmy, L. 1988 “Force dynamics in language and cognition”. Cognitive Science 12. 49-100. Tourangeau, R. & Sternberg, R. J. 1982 “Understanding and appreciating metaphors”. Cognition 11, 203-244. Winter, S. & Gärdenfors, P. 1995 “Linguistic modality as expressions of social power”. Nordic Journal of Linguistics 18, 137-166.

FUNCTION, COGNITION,

AND

LAYERED CLAUSE STRUCTURE

37

Function, Cognition, and Layered Clause Structure Peter Harder English Department, University of Copenhagen

1.

Introduction

This article deals with two different, but congenial, perspectives on language description: the “functional” and the “cognitive”. The aim is to suggest a general formula for how the two perspectives in combination can throw light on the organization of meaning in the clause. The proposal builds on suggestions made within two frameworks which have adopted these terms as brand names: Functional Grammar in the tradition of Simon Dik and Cognitive Grammar in the tradition of Ronald Langacker. People who use the words “functional” and “cognitive” about the work they are doing generally see each other as allies, and have the same opponents. But there is a difference of emphasis in that the word “functional” is oriented towards processes going on outside the head, in the realm of communicative interaction, whereas “cognitive” involves an orientation towards mental structures and processes. In particular, there is an orientation towards those aspects of language that are covered by the word “conceptual” (cf. below on Cognitive Grammar). This article is based on the assumption that it might be useful to attempt to clarify what the relationship is between the two approaches. I shall begin by looking at the two central notions in a biological context.

2.

Function and cognition in a biological perspective

Both function and cognition can be regarded as biologically based phenomena. Cognition is a bodily skill attributable to neural processes in the brain;

38

PETER HARDER

functional explanation has been taken for granted in biology since Aristotle (cf. Millikan 1984, Givón 1989: 383f). However, the standard uses of the two terms do not immediately qualify as foundations for scientific approaches to the study of language. I shall take the problems in turn. The problem of function as a basic concept has to do with basic assumptions in the philosophy of science: it is not clear that “function” has any status in the generally accepted modern world picture. Searle (1992) explicitly argues that, scientifically speaking, we only have effects; the promotion of one (type of) effect among others to the status of “function” is due solely to the observer. To pump blood (thus keeping the animal alive) is an effect of the heart’s activity, but we only call it “the function”, because we think that survival is important. In other words, function is only in the eyes of the beholder; in biology it is just a way of saying that certain effects have survival value. The reason for this skepticism is the danger of allowing an implicit teleological presupposition into one’s descriptive framework, sliding into preDarwinian modes of thinking, according to which the course of nature is shaped by inherent goals. However, there is a way to see function as an intrinsic property of the ecosystem even if we stay within the world of cause and effect. This notion of function has to do with an extra level of complication in causal chains (cf. Wright 1973) that arises with reproduction and may thus be understood as coming into being at the point where we ascend from the chemical to the biological domain. In the pre-biological world of physical and chemical processes, causes bring about effects; they may then function as causes and bring about new effects, etc. — but reproduction, with the attendant mechanism of evolution, adds a type of feedback mechanism to this one-way sequence. Because of reproduction, survival is not merely important in the eyes of the beholder, it is also a prerequisite for the persistence of the species as such. The fact that an animal is still around requires a different, more complex explanation than the fact that a rock is still around — regardless of what the observer may personally think of survival. Hence, among the causal powers of an organ or behavior there are intrinsic reasons why contributions to the survival of the organism have special status. By securing the persistence of the animal, they simultaneously secure the persistence of the organ or behavior itself. In other words, functions are effects which bring about the persistence of the causal factor itself: effects bring about (the persistence of) causes, not just the other way round.1

FUNCTION, COGNITION,

AND

LAYERED CLAUSE STRUCTURE

39

This definition applies to functions as familiar as those from the intentional world of human beings. The function of a lamp is to shed light to see by — since this is the reason why we put lamps in our rooms, it is that effect which causes the persistence of lamps. But the definition covers more than the intentional sense: in looking at the equipment of an animal, such as a lengthened toe or a hairless patch, the functionalist is entitled to ask why, and seek an answer in terms of the job they do. Since function comes into being with reproduction, it is present earlier and lower in the ontological hierarchy than cognition. The selectional pressure which is responsible for organ function also applies to plants and does not depend on cognitive skills. Thus, cognition must be understood within the wider context of biological function: the evolution of cognitive skills must be assumed to occur within the same functional constraints that apply to the rest of the evolutionary process. Not only does this apply to the phylogenetic process whereby gradually more sophisticated cognitive skills evolved in the origin of species — it also applies to the ontogenetic process whereby the cognitive system of an individual develops, as argued by Gerald Edelman (1992). Functions determine cognitive organization, rather than vice versa — which reflects the fact that cognitive science is part of biology, rather than vice versa.

3.

Cognition: A broad and a narrow definition

In making precise the relation between function, thus conceived, and cognition, we now need to be precise about what cognition is. I would like to suggest that there are basically two senses in which the word “cognitive” is used at present, reflecting a broad and a narrow definition. Under the narrow definition, cognitive processes are only those which are associated with the ability to solve problems independently of stimuli from the immediate environment. Cognition is therefore associated with mental content and with an intentional relation between mental content and external world states; it requires an inner, situation-independent environment (cf. Gärdenfors 1991, 1992). A cognitive animal is thus one which has the ability to envisage a state of the world and let it affect its actions independently of environmental stimuli. The narrow definition remains close to the process that everybody understands as the prototype, namely the reasoning process.

40

PETER HARDER

The broad definition can best be understood against the background of the cognitive revolt against behaviorism in the late 1950s and early 60s. The great step forward was the realization that there were things going on between input and output in human beings: instead of resulting from direct, mechanical cause-effect chains, human responses were dictated by inner states which must be described before human reactions could be accounted for. There were things going on in the black box which could be scientifically investigated. Artificial intelligence, which later developed into cognitive science, understood this in analogy with Turing machine states. The nature of computational simulation as a descriptive tool made it possible to sidestep a debate on what kinds of things were involved: input-output relations can be simulated regardless of the actual nature of the processes themselves. On this understanding of cognition, however, any mechanism mediating between input and output counts as cognitive. This has become especially evident in relation to the recent breakthroughs in connectionist modeling launched as constituting “the microstructure of cognition”. Rumelhart et al. (1986) used the ability to reach out and turn a knob under difficult conditions as their presentation example of the cognitive skills that connectionist modeling could aim at modeling. This sort of ability, however, occurs at a fairly lowly evolutionary level: in order to eat, an animal must be able to exercise sufficient motor skill to get at its food. It may be stimulus-controlled, and it may not even be accessible to consciousness; insects can do some of those things much better than human beings. If we base our definition of cognition on that type of modeling and describe human cognition on that basis, it will therefore include all complex neurally based skills. Salivation at the sound of the dinner gong, sexual intercourse, and digestion would only be arbitrarily excluded from the domain of cognitive events. All are dependent on inner mechanisms that transform input to output in a complicated way dependent on patterns of neuron firings. We will also, of course, capture those reasoning processes that stand as the prototype of cognition; but this description will fail to capture the way in which they are different from the motor skill that a dragonfly exercises when catching its prey in mid-air. Worse than that, however, is the risk of confusion between “real” cognition and pre-cognitive skills. This confusion occurs if we permit ourselves to conclude from a certain skill (i.e. the ability to distinguish a particular type of food) that the animal possesses the concept that a human observer formulates

FUNCTION, COGNITION,

AND

LAYERED CLAUSE STRUCTURE

41

in order to describe the skill. If we do that, all animals possess the concept “food”. Again, it may be regarded as merely a terminological issue — but there is a risk that discussion on human conceptualization will be muddled up by such a broad use. As an example, Langacker’s cognitive grammar (1987b, 1991) understands meaning in terms of conceptualization, providing a range of subtle and convincing examples of how “objectivist semantics” is insufficient and human mental structures are necessary to understand meaning. There is also a mental dimension with respect to speech sounds: it is not the physical sounds, but the way the human speaker organizes sounds that matter in linguistic structure. From this, Langacker concludes that the expression side of language is part of the semantic subdomain (1987b: 78–79): it deals with “sound concepts”, and these are part of the general domain of conceptualization that constitutes the area of semantics. I think this is a case of the confusion I described above: if we have the word expression mother and the semantic content ‘mother’, we need a sense of the word “concept” according to which it is only the content side which involves a real concept. In the absence of such a sense, we cannot tell the difference between the expression and the content side of language: there is no reason why one concept is more contentful than the other. What is involved on the expression side, more specifically on the level of phonology, is the ability to make a certain range of perceptual distinctions as a precondition for reacting appropriately to them (as a moth can perceptually distinguish the clicks of a bat and react accordingly). Categorial perception is, of course, a distant relative of conceptualization proper, but the central theoretical basis of a semantic theory should not encompass both on an equal footing. The narrow sense I shall call “concept-cognitive”, and the broad sense “neuro-cognitive”.

4.

Functional and cognitive perspectives on meaning

Linguistic meaning has always been understood primarily in conceptual terms, apart from the time of the invasion of formal logic in linguistics from the 1970s onward. In seeing linguistic meaning in terms of conceptualization, cognitive linguistics therefore has tradition on its side (cf. also Geeraerts 1992). And with respect to the types of meaning that everybody considers basic, I think this tradition is wholly sound: the feature of human language

42

PETER HARDER

which, as far as we know, makes it unique in the animal kingdom is precisely the existence of conceptual meaning, dependent on an inner, stimulus-independent environment (cf. also Ulbæk 1989). However, not all meaning involves concepts. As an example of a kind of linguistic meaning that feeds directly into the situation, one can take greetings. The meaning of hello does not consist in invoking the ‘hello’ concept in the same sense that the meaning of horse consists in invoking the ‘horse’ concept. Rather, the meaning of hello can be described in functional terms as a signal that conveys recognition plus lack of immediately hostile intentions. “Why is that not a concept?” one may ask. After all, the ability to respond with an appropriate greeting presupposes inner mechanisms of recognition and intention-formation, which we can capture in a description that would in fact amount to a ‘hello’ concept. The reason is that these inner mechanisms can be directly triggered by the relevant stimulus — and the concept would then only exist as a meta-level description of factors that are not in themselves conceptual, any more than a thermostat has a temperature concept. Greeting-like signals go down quite far in the animal kingdom and have obvious evolutionary advantages with respect to avoiding unnecessary panic and fighting as well as maintaining a sense of fellowship and well-being in conspecific groups. No inner, situation-independent environment is required to support a system of greetings. As against that, the meaning of the word horse draws on a concept inside the speaker: a person who was only able to produce the word horse when a certain situational stimulus was present could not be said to know the meaning of the word. Cognitive types of meaning are also functional, of course: there is an evolutionary advantage in being able to conduct mental experiments without actually testing them out situationally. As pointed out by Karl Popper (1972), we human beings can test out our hypotheses mentally, and let the hypotheses die in our place: we can look at the thickness of the ice and draw conclusions about its ability to support our weight instead of trying it out in practice every time. Thus, all meanings are functional, but only some are conceptual in the narrow sense. Since the processing of greetings requires discriminatory skills, the meaning of greetings is obviously cognitive in the neuro-cognitive sense. But this is not so interesting as the distinction between cognitive and noncognitive meanings in the narrow sense.

FUNCTION, COGNITION, 5.

AND

LAYERED CLAUSE STRUCTURE

43

Meaning and clause structure in Cognitive Grammar

The basic view of the nature of linguistic semantics in cognitive grammar (=CG) “equates meaning with conceptualization” (cf. Langacker 1987b: 5). This position is formulated in opposition to the (then) dominant view based on an objectivist, truth-conditional orientation in semantics, and emphasizes the importance of aspects of meaning that cannot be captured in a semantics based on formal logic, pre-eminently figurative language. One of the distinctive features of Langacker’s semantics is closely associated with his reliance on visual images as ways of illustrating conceptual structures and processes. This method of representation (which should not be understood as implying that meaning is necessarily pictorial in nature) has the great advantage of making it easy to convey a number of semantic properties that are not easily captured in truth-conditional paraphrases of the “male, adult, human” kind. One basic example is the relationship of profiling, where a lexical item carves its meaning out of the domain in which it belongs; another is the structural configurations that are often central both in metaphors and in processes of bleaching, where one can demonstrate how the semantic skeleton remains while truth-conditions change. The view of grammatical structure in CG is consciously opposed to the structure embodied in generative grammar, and places great emphasis on the individual meaning-bearing elements. A grammar is seen as a “structured inventory of conventional linguistic units” (Langacker 1987b: 57). This “list” character of the grammar is also emphasized by the explicit rejection of the “process” or “constructive” character of grammar that marks generative theory, in which the grammar in itself is capable of specifying well-formed combinations of items. This CG leaves to the speaker rather than the grammar, emphasizing the creative, problem-solving nature of the combinatory process, as opposed to the mathematical character of the generative model. Grammar involves (Langacker 1987b: 82) “the syntagmatic combination of morphemes and larger expressions to form progressively more elaborate symbolic structures”. This might be thought to go against the “list” character of the grammar. However, the list approach can be preserved by the addition of rather abstract items to the list: for each possible type of syntagmatic combination CG posits an abstract item consisting in the construction schema itself. Thus, “plural formation” is itself an item, represented as [[[THING]/[..] - [[PL]/[z]]]. This schema, where “thing” stands for the common semantic

44

PETER HARDER

property of nouns, and [..] for the expression slot (since there is no common denominator on the expression side), states that you can form the plural of a noun by adding [z]. Any plural form thus represents (at least) three linguistic items: a noun, a plural morpheme, and the combination formula above. This view of syntagmatic relations is essentially bottom-up. The fundamental assumption is that one unit can do the same as a complex unit; if, more or less accidentally, we do not have a unit that will do the job on its own, we have to build one out of existing units, cf. Langacker (1987b: 279): ..finding no single morpheme or other fixed expression to convey the desired notion, I construct the novel sentence your football is under the table. I can achieve appropriate linguistic symbolization only by isolating and separately symbolizing various facets of my unified conception..

There is a top-down element in the reference to a unified conception as the goal which the speaker is working towards. The theory also provides a description of how clauses are built up that account for essentially the same elements that enter into the layered clause structure described below. However, there is no clear theory of how syntagmatic combinations create elements that are essentially different from elements that we find on the level of individual lexical items. Clauses inherit the profile of the verb, describing processes; noun phrases (NPs) inherit the profile of nouns and designate things — and this fundamental distinction is not followed up at higher levels of organization by anything of comparable status in the theory. This I see as a consequence of the conceptualist semantics. Essentially, what we do when we combine words is to combine conceptual elements into more complex conceptual wholes until we have something matching the unified conception that we want to convey. This assumption is also reflected in the doctrine that “there is no meaningful distinction between grammar and lexicon” (Langacker 1987b: 2). The overarching concept is “symbolic structure”; such structures differ in various respects, including compositional complexity, but cannot be factored out in distinct subcomponents. The central concept in defining the potential of one conceptual structure for combination with another is “valence”; valence relations are established when elements of conceptual structures are brought into “correspondence” with one another. This explores the familiar chemical metaphor in that just as chemical compounds share electrons, so do linguistic complexes share sites in a conceptual whole. Thus, when we use the units tall and man to form the phrase tall man, the meaning of tall has a “trajector” element (reserving a

FUNCTION, COGNITION,

AND

LAYERED CLAUSE STRUCTURE

45

place for the carrier of the property designated by tall); and this trajector element functions as the “site” where we put the meaning of man. In other words, an unsaturated element in the meaning of one item is filled out with the meaning of another (“elaborated” by it), so that the two isolated meanings now form a composite meaning. This illustrates one of the important differences between types of items in cognitive grammar. Syntagmatic combinations are not typically between equal partners, as expressed in the distinction between autonomous and dependent constituents: One structure, D, is dependent on the other, A, to the extent that A constitutes an elaboration of a salient substructure within D (Langacker 1987b: 300) The point is that if one item has a substructure that requires elaboration, then the element that elaborates it is “autonomous”, whereas the other is “dependent” (i.e. on the elaboration). The central example of this type of dependence is the distinction between things and relations, as reflected in the syntactic distinction between nouns (as autonomous items), on the one hand, and verbs or adjectives (as dependent), on the other. This is illustrated by the “billiard-ball” (1991: 13) model, where the nouns are like the balls in being conceptually autonomous, whereas the movements and relations between the balls are conceptually dependent: you cannot conceive of interaction in isolation from interacting objects — but you can think of the objects themselves in abstraction from any interaction. Langacker states that the notion of dependence he suggests is nearly the opposite to the notion employed in dependence grammar. Instead of letting this stand as just one more source of confusion between schools, I think the oppositeness can be revealingly analyzed as a straightforward consequence of opposing points of view. If one is interested in structure and consequently looks for the paths of determination that create clausal structure, clearly the structural position of argument terms is dependent on the main verb of the sentence. This is so for exactly the same reason as Langacker would say that the verb is the dependent member in the relationship between verb and argument noun: the verb “needs” argument nouns around it. It is therefore the semantic dependence of the verb (on elaborating arguments) that gives rise to the structural dependence of the arguments on the main verb (because the argument positions are created by the semantic “needs” of the verb). As already indicated, the point on which I think this picture of linguistic

46

PETER HARDER

semantics is incomplete has to do with the situational and interactive aspects of meaning. The type of meaning, that is, the type of dependence, described by Langacker is not the only one that plays a role in language: it accounts for distinctions and relations between aspects of a conceived world, not between aspects of a communicative utterance. Before elaborating on this criticism, we shall look at the syntactic model I would like to base my own account on.

6.

Meaning and layered clause structure in Functional Grammar

Inspired by Foley and Van Valin (1984), Functional Grammar (FG) as practiced by Simon Dik and associates (Dik 1989; Hengeveld 1989, 1990) has developed a theory of the clause in which the clause is seen as consisting of a series of “layers” superimposed upon one another. The notion of layering involves a central idea which can be illustrated by a diagram of the earth cut in half. With reference to such a picture, the earth can be described in a movement from the core outwards, such that each successive layer contains the previous layer and adds something to it. The idea is that complications can be described by successive additions to a nuclear element that remains inside the superimposed layers. A central concept is the “scope” relationship: “outer” layers take “inner” layers in their scope. Although the term “operator” in the theory is reserved for grammatical as opposed to lexical items, I shall use the distinction between “operator” and “operand” for the relation that obtains as you move outward from an “inner” layer (the “operand”) to the next higher element, which functions as an “operator” that transforms the operand into a more complex, higher-level entity. Hengeveld (1990) provides a general introduction to the layered format of description under the title “The Hierarchical Structure of Utterances”; but the notion of layering is not identical to that of a hierarchy. Standard examples of hierarchies include administrative organization in terms of departments with subdepartments, giving rise to a chain of command with one head at the top, executives who function as heads of their departments, and so on down to the lowest tier of employees. The most obvious example of this in language is traditional constituent structure; generative grammar, as also revealed in the pervasive metaphor of regimentation (cf. the notions “command”, “government” and “binding”), started off with a clearly hierarchical as opposed to a

FUNCTION, COGNITION,

AND

LAYERED CLAUSE STRUCTURE

47

layered clause structure. However, there is no absolute conflict between layered and hierarchical structure. When a new layer is added, this operation can be seen as creating a new constituent on a higher hierarchical level; and since there is an element of sub-layering associated with noun phrases (cf. Rijkhoff 1990, 1992), there is more than one core item involved in clause structure, giving rise to the characteristic hierarchical configuration. The change from phrase structure rules to X-bar syntax in generative syntax reoriented the structure in the direction of layering: each “bar” level can be seen as a superimposed “layer”; and, as described in Siewierska (1992), there is considerable similarity between the actual structural levels postulated by current versions of GB and FG. As implied in the movement from the center of the earth outwards, there is a natural “bottom-up” movement built into the notion of layering. Foley and Van Valin’s layering (cf. 1984: 78) was clearly focused on the innermost parts, involving a distinction between the nucleus (containing the predicate), the core (containing the argument NPs) and the periphery (containing the “circumstantial” or, in FG terminology, the “satellite” NPs). In motivating the layered structure within FG, however, Hengeveld emphasizes the division into two super-layers: an “interpersonal” layer (in Halliday’s terms, cf. 1970, 1985), which is superimposed upon a “representational” layer (following the terminology of Bühler 1934, rather than Halliday’s “ideational”). The interpersonal layer contains elements inspired by the speech acts philosophy, consisting of an illocution and a proposition, conforming to the formula F(p) (cf. Searle 1969). In the linguistic context, the notion of illocution is anchored in the distinction between sentence types, centrally on the coded distinction between declarative, interrogative and imperative clause constructions. This linguistic narrowing of the concept shows some parallelism with the more recent views of Searle (1991), where interest is focused on the “bare bones” of an illocutionary act, centering on the notion of “direction of fit” and “illocutionary point”, notions which are easier to relate to linguistic categories than the full panoply of illocutions familiar from Austin, and also less vulnerable to criticisms made of the notion of illocutionary force itself (cf. Allwood 1977, Harder 1978). Any adequate account of the distinction between interrogatives and declaratives (the distinction that will be used as the example below) must include these two aspects. They share a world-to-word direction of fit (it is the state of the world that determines our affirmative/ negative response to interrogative and declarative sentences), but they differ

48

PETER HARDER

in illocutionary point: one conveys (the speaker’s commitment to) the existence of the fit, the other raises a question as to the existence of the fit (prototypically, but not necessarily, to be answered by the addressee). The distinction between illocution and proposition splits the interpersonal “super-layer” into two. Similarly, the representational content of the proposition can be subdivided; there is no settled agreement in FG theory as to the precise manner in which this should be done (cf. for comparison Dik 1989: 46, Hengeveld 1990: 4 and Hengeveld 1992: 35); opinion may be seen as converging on a version resembling Foley and Van Valin’s original model, where the naked predicate itself is the basic layer, which at the next stage is combined with terms to form a predication. Each of the syntactic layers corresponds to a level of semantic complexity, reflecting a gradual build-up from a property or relation up to the speech act conveyed by the utterance as a whole. The skeleton of the layered model can thus be outlined: Table 1. Illustration of layers Syntactic layer

designation

linguistic rendering

clause proposition predication predicate term

speech act possible fact state-of-affairs property/relation (typically) individual

“Did Jim go?” (that) Jim went for Jim to go go Jim

Terms have been added at the bottom, in separation from the other categories. The problem of where to put terms is due to a set of related circumstances. To begin with, they probably do not belong in a diagram of “layers”, because argument terms (pace Foley & Van Valin) do not obviously constitute a syntactic layer in themselves. They enter into the layering by being combined with the basic predicate to form a predication; but they do not take the predicate inside their scope, as a “higher” layer should. From that point of view they should rather be the “lowest” layer, with the predicate in the higher position, since the predicate takes arguments in its scope. However, this is difficult to reconcile with constituent order: arguments as a group are not syntactically central, with the predicate in a more marginal position. But terms occupy a natural position at the bottom of the metaphysical (as opposed to the syntactic) hierarchy, by being the sole designators of individuals. Where the syntactic and the metaphysical hierarchy mirror each other at the higher levels,

FUNCTION, COGNITION,

AND

LAYERED CLAUSE STRUCTURE

49

they diverge when it comes to terms — hence the uncertainty about where to put them in the canonical diagram. In addition to describing the different layers of semantic organization in the clause, the layered model is also motivated by a series of distributional facts; for an overview cf. Hengeveld 1989. What I see as missing in the theory of Functional Grammar is an explicit semantic component. As exemplified above, there is a wealth of semantic reflection in the theory, but it is so to speak translated into the structure. Semantic intuitions are used to motivate structures, essentially in the following manner: “because we have semantic intuition x, let us set up structural item y — which may then also be correlated with formal expression z”. However, once the structural distinction is set up, there is no distinct place for the semantic information that motivated it. In a diagram of the theory one looks in vain for a compartment that can be labeled “semantics” (cf. Harder 1990, 1992); the only place where we find separate semantic information is in relation to the lexicon. Semantics therefore stops short of being fully part of the theoretical apparatus; among the points on which I agree with cognitive grammar (as well as Saussurean linguistics) is in the adoption of an explicitly bipolar model of description, with the relation between significant and signifié in the center of description at all linguistic levels. The insights developed in Cognitive Grammar and FG I think are essentially compatible. Thus, the compositional build-up of symbolic structures in CG, which also makes reference to scope relations, to some extent mirrors the layered structure, while providing in a number of cases a richer semantic apparatus to underpin it. For instance, in FG, the predicate is described as having a “predicate frame” specifying the number and kind of core arguments that it combines with, but the mechanism is not further described in semantic terms. On this point, the CG “trajector” and “landmark” elements, constituting potential “elaboration sites”, could be seen as providing semantic microstructure for the notion of “predicate frame”. The point I am pursuing, however, has not to my knowledge been explicitly made in relation to either cognitive or functional approaches to clause structure. It concerns the way in which conceptual and functional elements co-operate in the clause.

50 7.

PETER HARDER Human language: Cognition embedded in interaction

I shall now try to suggest what is involved in the evolutionary transition to the stage of human language; it is an illustrative parable rather than a hypothesis — a “Just so” story in Kipling’s style as suggested by Gärdenfors (1991), or, in Nixon’s terminology, a “scenario”. Let us imagine, in the hypothesized pre-language situation, a communicative system with a limited set of holophrastic signals (“Wholese”), whose function is to influence group behavior in the actual situation. Let us take as an example the vervet monkey (cercopithecus aethiops), which has three alarm calls: one for eagles, one for snakes and one for leopards (cf. Ulbæk 1989: 408). There is a conceptual element in the language, involving a distinction between three categories of predators; but the difference in relation to human concepts as involved in communication is clear. First, there is no distinction between the invocation of the “leopard” category and the “warning” element in the message: the signal expresses the descriptive and the interactive element at the same time. Secondly, this combined meaning relates directly to the situation: “flee + leopard” always applies here-and-now; old, seasoned monkeys cannot tell stories of hair’s breadth escapes to their spellbound young. The decisive step towards a conceptual component of language is the step from a situational-manipulative Wholese to a language with sub-utterance constituents that are not directly tied to any specific contextual function. This involves changes in all three essential properties of Wholese at the same time. The step to sub-utterance coding by definition eliminates the holophrastic character of the language. Situational boundness begins to disappear because the existence of sub-utterance lexical items means that we code something that in itself has no ready-made function in the actual context (otherwise it would not be “sub-utterance”). The step to sub-utterance meanings is therefore inevitably a step towards meanings that are partially context-independent — and therefore must survive in the inner environment (the conceptual world) between instantiations. In order to have a lexical item “leopard” as distinct from a Wholese signal meaning ‘leopard!!’, the speaker must “entertain” the concept leopard as distinct from its situational presence, including the need to react in a particular way. The distinction is parallel to the distinction between an air-raid warning and the concept “air-raid”. The limited repertoire also gradually begins to disappear; sub-utterance items create the possibility of combinations, and although these do not at once

FUNCTION, COGNITION,

AND

LAYERED CLAUSE STRUCTURE

51

become effectively unlimited in number, the automatic character that goes with response by holophrases is lost as soon as there is a combining operation involved in making a message. The central element of creativity, i.e. choosing one’s own syntagmatic combinations (cf. Langacker as quoted above), is already present. By the same monumental step, we also get the necessity of linguistic structure: once we have individual meanings that do not have their own situational function, we need to be able to organize those meanings in such a way that we can put together whole utterances that do have a situational function. Structural complexity, implying a distinction between “langue” and “parole”, is the other side of the transition from situational to cognitive meaning. Since the meaning of a Wholese utterance can be described exhaustively in terms of direct situational function, there is little point in distinguishing potential from actual meaning: there is no difference except the type-token distinction. As opposed to that, once we have sub-utterance meanings, we are forced to have a distinct level of “langue” that is not reducible to simple stimulus generalization from utterance tokens. I shall now attempt to develop the scenario describing the transition from Wholese to sub-utterance coding and clause structure in a way that matches the layered structure. Let me make it perfectly clear that I do not think it developed that way, and this is purely for the purpose of demonstrating the way the layered model can illustrate the result of this monumental transition. Since we start with the Wholese signal, the natural expository path is topdown. The first distinction, as we have seen, is one between an indication of illocutionary type and a propositional content. In the evolutionary perspective, this step would be accomplished if some band of ancestors once developed a system where there were differentiated reactions to potential threats: apart from the alarm (eeek!), for example, an indication of disgust (yuck!). More humanoid, we might also have interrogative illocution, usable in case of doubtful identification of the predator. Instead of the choice between ‘snake!!’ ‘eagle!!’ and ‘leopard!!’, there would thus be a choice of ‘?’, ‘eeek!’ and ‘yuck!’ in the function-slot and ‘snake’, ‘eagle’ and ‘leopard’ in the descriptive slot. Snake, etc., would now be functionally incomplete utterances, because the addressee would not know in what manner the animal was relevant to the situation; for that the speaker would need to add, for instance, eeek!. The next step downwards is the distinction between the descriptive content of a proposition and its application to a situation in the world of which

52

PETER HARDER

we speak. In the example given above, a snake would still mean a snake in the situation; in human language this is not necessarily so. The linguistic locus of the “application” element is deictic tense, i.e. past or present (cf. Harder 1990); clauses in the past tense are understood as applying to the past world of which we are speaking, whereas clauses in the present tense apply to the world as it is at the time of speech. In our proto-language one could, for instance, choose between ‘yuck (past (snake))’, remembering yesterday’s green mamba, or ‘yuck (present (snake))’, meaning the python right over there that is not dangerous because it is eating a goat at the moment, but is disgusting anyway. The element designating ‘snake’ would now be purely conceptual, the situational relations being coded separately. The distinction between a purely conceptual “state-of-affairs” (abbreviated SoA, following Functional Grammar practice) and a proposition is often ignored; the first place I have seen it clearly set forth is in Leech (1981). The central point is that the descriptive content of a clause in itself cannot be true or false of anything. In this, it is like a picture hanging on the wall, showing, for instance, a sturdy fisherman smoking a pipe. It makes no sense to ask whether this picture is true or false, unless we see it as an attempt to portray a particular person. The deictic tense codes this element of application, in essential similarity to definiteness as expressed in a noun phrase: that ferry involves an instruction to invoke the ferry-concept and match it with an object in the situation (cf. below), just as a past tense form, as in John went, involves an instruction to match the description “for John to go” with an event in the world. The descriptive content in the examples of the proto-language above corresponds to something like an existential sentence (or more generally what Strawson (1959) called “feature-placing” sentences). As suggested by Strawson, this may be seen as a more primitive form than the one in which there is a distinction between objects, on the one hand, and property/relations, on the other: our conceptualization of objects can be seen as a development dependent on an earlier notion of “feature instantiated in a place”. The concept of ‘a stone’ may develop as “a bounded instantiation of stonehood in a place”; the stage of having a proper concept of an individual, compare the title of Strawson (1959), may depend on an earlier conception of pure feature instantiation. At the stage of warning cries, a warning against leopards is probably not a warning against an individual-concept any more than a warning against fire would be; it would just indicate that the feature [+leopard] was instantiated.

FUNCTION, COGNITION,

AND

LAYERED CLAUSE STRUCTURE

53

True predications, which involve the “billiard-ball model” of individuals with relations between them, can therefore be seen as a more subtle and complex way of organizing descriptive content. With the existence of individuals or, more generally, “things” as denoted by noun phrases, there are three different ways of using conceptual meaning. Apart from its use in feature-placing sentences and as part of a thing-concept, there is the role associated with the central predicate of a predication. In this role, the conceptual content is conceived in (relative) independence of things (cf. the discussion on dependence below), but used to ascribe a property/relation to its argument “things”. Obviously, the complications of differentiated coding are vastly greater than the basic lay-out of the layered model as outlined here can even begin to hint at; it unfolds ultimately into the whole of linguistics. The point I hope to make here is just that coding differentiation, with holophrases as the point of departure, is a revealing approach. In order to be more explicit about the distinction between functional and conceptual aspects of meaning, I shall pick out a couple of examples below — but clearly they must stand as, hopefully, reasonably central illustrations of some general principles rather than anything more ambitious.

8.

A closer look at functional meaning

The notion of communicative function that is important in relation to the point of this article can be described in continuation of the discussion of the situational nature of holophrastic languages, where meaning always relates directly to the situation. We now need to look at the role of the situational relation in human language, once conceptual meaning begins to arise. Among the representatives of this type of meaning in human language I shall take deixis, which is perhaps the most obvious example, as a demonstration example. Deictic elements have always been understood as exceptions; Jespersen’s term “shifters” points to the problematic property of referring to different things in different situations, which is an anomaly if you think of meaning in terms of what words stand for. The central point in this context is that, from the point of view of situational, interactive function, they do NOT change. The first person pronoun I can always be used to refer to oneself, and thus

54

PETER HARDER

exemplifies the direct situational relation that was also characteristic of holophrases. As a result of the coding differentiation, it is not a complete message (except in special circumstances); but it preserves that situational link which purely conceptual meanings lack. In CG, there are two related notions which account for the peculiarities of such elements: “grounding” and “subjectification”. Grounding occurs when a symbolic structure is located not in relation to the canonical “objective” perspective, but in relation to the “subjective” scene with the speaker in the center. Subjectification is the process whereby meaning elements become reoriented from the objective scene to the subjective scene, as often occurs in processes of grammaticalization, cf. Langacker (1990). More technically, grounding and subjectification involve a special type of profiling: instead of the usual profile on the concept itself, it is the relation between concept and situation-of-speech that is profiled. Within this picture, the pronoun I can be described as invoking the subjective rather than an objective domain; it refers to the speaker as a participant in the speech event itself, rather than a speaker viewed as part of an objective scene. Thus, if we replace the “objective” conceptual content with a representation of the “subjective” situation, we also get a constant meaning for the “shifty” deictic elements. This account I see as true and valid for that part of the function of situational elements which involves the mental process of the individual himself; in order to be able to understand and use deictics, a necessary condition is that the speaker is able to conceive of himself as part of the “grounding scene”, as described by Langacker. However, there is no functional, interactive dimension in this picture. One type of conceptual structure is invoked instead of the other, but it is all a matter of getting one’s conceptualization right. What is missing is the actual process of establishing a link between the ongoing situation and the conceptualization process in the mind the element that was automatic at the holophrastic stage. The distinction is analogical to the difference between having a fully functional electrical device and plugging it in; the interactive element is the element of actually plugging the utterance into the situation. The notion of “near” or “distant” as involved in prototypical deictics can further illustrate what is lacking in a purely conceptual account. The issue can be seen in the light of the etymology of the word “deixis”, which comes from a Greek word for “pointing”. The meaning of deictic items involves something similar to a gesture of pointing - which brings a feature of the situation to

FUNCTION, COGNITION,

AND

LAYERED CLAUSE STRUCTURE

55

the attention of the “addressee” without essential reliance on conceptual resources, and thus establishes a relation between mind and situation. This also affects the way we should understand the contrast between “near” and “distal” deictics. The word here invokes the ground and points to where the speaker is; the word there invokes the ground and points away from the speaker. But the “nearness” and “distance” do not presuppose a decontextualized concept of “nearness” vs. “distance”: you can point to something without possessing a conceptual, decontextualized notion of “distance”, of which deictic distance comes out as a special case. Once you have both the concept and the ability to point, you can generalize, setting up a superordinate concept of “distance” and a subdivision into “deictic” and “objective” distance. But before one has achieved, by evolution, the cognitive level where one can make this generalization, only pointing is available — so a purely conceptual account of pointing is an account based on the hindsight of evolutionary superiority. Essentially the same element is involved in the account of definiteness. In conceptual terms, cf. Langacker (1991: 98), the meaning of the definite article can be described as involving the elements of uniqueness in current discourse space, mental contact by the speaker, and mental contact by the hearer (either previous or as a consequence of the use of the definite NP itself). Thus, an NP with a definite article, as in the ferry, designates a ferry satisfying the three conditions described above. The element that is missing according to the functional perspective is the establishment of a link between the conceptual ferry and the situational ferry. This link is not conceptual — a concept can never get us beyond the conceptual world - but an act of opening the door for a concrete, situational element to be referred to by means of the conceptual construct. The definite article does not predicate a concept — it triggers an action. Another example of interactive meaning types is the words yes and no. Both constitute complete speech acts, and the speaker by using either of the two indicates his own position with respect to something in the situation: does he support it or oppose it? In comparison with deixis, these are clearer cases of purely interactive, situational meaning, because they do not designate or denote anything — they only function as signals of assent or negation. Viewed in isolation from a concrete instance, no conceptual content can plausibly be assigned to them, even in relation to a subjectively construed grounding situation.

56

PETER HARDER

The relation between no! (spelled with an exclamation mark to emphasize its situational function) and negation in general can be seen as exemplifying some of the ways in which functional and cognitive types of meaning interact. One might hypothesize a developmental path for negation that mirrors the evolutionary scenario above. A likely source situation for negation is the desire to reject something one does not like: to some extent, no! stands to “doing away with” as a warning display stands to “physical onslaught” in the animal kingdom. However, no! is not the only form in which negation appears; it also appears in forms (not, and no as in no music) in which it interacts structurally with conceptual types of meaning (cf. Langacker 1991: 132f). The analysis suggested by Langacker is again perfectly convincing as an analysis of the conceptual aspects involved in understanding negation. His analysis sets up an understanding where the point of departure is the un-negated item, which is then contrasted with a configuration where the item is absent. To illustrate this account, a parallel is suggested with the analysis of the preposition towards, which evokes a completed path but only designates the unfinished trajectory. Just as with negation, we need a situation to compare with in order to understand the conceptual import. However, I think the complexity of negation is different from the complexity of towards. This word designates part of a trajectory, essentially as a hand designates part of an arm. Negation, by contrast, does not designate either the item itself, or the missing item, or the pair consisting of both. What happens is better described by a word that Langacker uses repeatedly in the context, namely cancellation: not is used to cancel whatever is negated. However, the word “reject”, which preserves the link with situational rejection, may be preferable. The word cancel might suggest that the description is simply withdrawn, but what happens, as argued at length in Millikan (1984), is rather that a description is replaced by an alternative description — which can be captured by saying that not is used to reject a description, essentially as no! is used to reject a potential event in the situation. In other words, negation is not a concept in the same sense as ‘hand’, ‘arm’ or ‘towards’ is a concept: there is an interactive root in it. And data from language acquisition would appear to be compatible with an assumption that the interactive element is still basic in negation: the child first learns the holophrastic no!, using it whenever there is a danger that events in the situation take an undesirable turn; much later comes the application to conceptual items.

FUNCTION, COGNITION,

AND

LAYERED CLAUSE STRUCTURE

57

In relation to a purely conceptual approach to linguistic meaning, this discussion is an attempt to show that human language has retained the situational anchoring characteristic of previous systems of communication. In relation to FG, what I say can be seen as an attempt to make the functionality of the layered model more transparent. The layered structure describes one way of factoring out subcomponents of meaning, in which the uniqueness of human language consists in its tapping conceptual resources, but without eliminating the situational embedding of communication. Instead of throwing away situational embedding, what has emerged is a structure where the top or “outer” layers establish the role of the utterance in communicative interaction — by anchoring it (deictic tense) and assigning it an illocutionary type (declarative), while the bottom or “inner” layers supply the conceptual meanings which provide the utterance with its content. Situational and conceptual aspects are thus both part of human language. The way this fits with the “mental models” view of mental representation is by assuming that linguistic meanings function as instructions that trigger mental operations in the addressee. Both production and reception of utterances are likely to work by parallel processing, so no simple compositional process is psychologically realistic; but as pointed out by Dik (1989: 52), this does not exclude the possibility that the different steps in a grammar may actually be involved in the processing, even if the descriptive procedure and the psychological process work differently. If that is the case, we can add a sub-scenario whereby the meanings involved in the layered structure specify cognitive routines that addressees must (somehow) perform: invoke conceptual structures, ascribe them to entities, apply them to past or present situations, and construe them as statements or questions about the way the world is. Thus, human language works interactively, by enabling addressees to reconstruct cognitive representations inside their own heads, rather than by simply transmitting pictures directly from brain to brain.

9.

Functional and conceptual dependence

Function is almost by definition something that must be described top-down. As with the function of an organ, the function of an utterance must be described by seeing the object of investigation in relation to the context in which it belongs. Thus, the basic functional fact about an utterance is the

58

PETER HARDER

function served by the utterance as a whole. The function of sub-utterance items must similarly be described by a top-down procedure of the kind followed above: the function of a constituent element is its contribution to the job done by the whole utterance. This approach provides a perspective on dependence that is different from the one described by Langacker. The basic motivation for it is that one linguistic element needs another because it cannot do the whole job on its own; when you code a sub-utterance item, there is always something missing before you have a fully functional utterance. In this formulation, it is true of all utterance fragments; but the central asymmetry of the elements in the layered structure suggests a differentiation between two types of “incompleteness”. As we saw, the coding differentiation embodied in the layered model distinguishes between “operands” and “operators”; and the defining mark of operators is that they “use” the operands in order to create a new and more complex item. At the top (or “output”) end we found the functional types of meaning (that relate content to the situation); at the bottom (or “input”) end we find conceptual content, which is “used” in various ways by higher-level operators. On the basis of this dichotomy, we can set up two complementary types of incompleteness, giving rise to two types of dependence relations. The incompleteness of operators consists in the lack of a content to operate upon. Starting from the top, we began by differentiating between the illocution (for example, declarative or interrogative) and the propositional content. The illocution operator specifies function, for example, that the utterance is a question, but in isolation it would lack a content. With a slight overgeneralization, I shall call the dependence of an operator upon its operand conceptual dependence. The motivation for this name is that the dependence points downward in the structure, towards the conceptual end, and that what is missing therefore includes the conceptual content. The overgeneralization is due to the fact that there may be something else apart from conceptual content missing. The operand is incomplete in the opposite way. What is missing is a specification of what to do with it — in the example we have a proposition, but we do not know whether it is to be used to make a statement or ask a question. Therefore the dependence of operand upon operator will be called functional dependence. Seen from the purely conceptual point of view, there is nothing incomplete about a proposition — we are perfectly capable of entertaining a proposition mentally, as part of an inventory of “issues”, without incorporat-

FUNCTION, COGNITION,

AND

LAYERED CLAUSE STRUCTURE

59

ing it into an utterance and without making a decision as to its truth value. The incompleteness, and the dependence, only arises when it is invoked in connection with an utterance, i.e. called upon to serve a communicative function. It follows from the definition that operators below the top level are dependent in both ways: thus, deictic tense is functionally dependent on the illocution and conceptually dependent on the SoA. It also follows that all operands are conceptually independent in relation to the operator. If we look at this distinction in relation to Langacker’s notion of dependence, which is sometimes also-called “conceptual dependence”, it appears that it tallies with what I have called conceptual dependence. The billiard-balls example covers the relationship between the predicate and the arguments in creating a SoA: the predicate takes the arguments inside its scope, and is therefore conceptually dependent on them, whereas the arguments (at the bottom of the scope hierarchy) are conceptually independent of everything else. In the case of negation, too, it fits Langacker’s analysis: negation is conceptually dependent on what it negates (cf. Langacker 1991: 132). The argumentation behind Langacker’s notion of dependence is based on the conceptual content of the items discussed, and is therefore not completely identical to the reasoning that is behind the definition above. But I think the same basic facts are involved. From the point of view embodied in the approach from above, the conceptual content of individual items must be seen in relation to a division of labor: each item gets its conceptual content because of the kind of job it does. Seen from “below”, the reasoning goes the other way: each item has the job it has because its conceptual content makes it suitable for that job. Neither of these two perspectives is the “right” one; sometimes one is more revealing than the other, but both are necessary. To take two examples: With respect to the relation between verb and arguments, the verb is dependent because it involves a trajector (and possibly landmark) site that requires elaboration. Seen from above, the development of concepts that evoke trajectors without designating them must be understood in connection with the function of “predicating” properties and relations of objects that one wanted to talk about: unless we could factor out properties from the objects that carried them, this job could simply not be carried out. Seen from below, we get the picture argued by Langacker: concepts that fit each other because one fills out something that is missing in the other. In the reasoning of CG with respect to negation, Langacker says that “it makes salient (though schematic) internal reference to the situation whose existence

60

PETER HARDER

it denies”. As far as I know, “make reference to” is not a technical term in cognitive grammar, but I take it to mean that the understanding of not is like the understanding of a verb in requiring the existence of something else: a trajector in the case of a verb, and a negated situation in the case of negation. Seen from above, we can say that the fact that not is designed to reject something (which functions as its operand) means that in using not one is simultaneously presupposing an operand to operate on. In this case, I think the picture from above is more intuitively striking, because the conceptual content of negation is less substantial than in the case of verbs; but to the extent that one can factor out a conceptual content in “not” on its own, it is clearly dependent on the content of the negated item. In other words, conceptual and functional differentiation are two sides of the same coin in creating word meanings. The notion of functional dependence is, however, absent in CG. It is not explained why the billiard-balls are incomplete as the content of an utterance — why full utterances in the form of noun phrases are deviant; or, more generally, why all layers up to the illocution are incomplete from a functional point of view (we find no free-floating predications or propositions either). Both types of dependence have traditionally been handled in terms of dependence between linguistic items alone. Sometimes this is the case; but typically the picture is less clear-cut. The dependence is basically between meanings, and meanings may be situationally present in such a way that items that are not functionally complete can nevertheless occur on their own. The phenomenon of “ellipsis” should be understood as involving a process of drawing upon previous utterances rather than knocking out elements of later utterances. As long as what is missing in an elliptic utterance can be specified in precise linguistic terms, it makes no great difference to think of the dependence as linguistic. An interesting borderline case is zero anaphora (cf. Givón 1990, Tomlin 1991). In the terms described above, one would say that the basic dependence is from a verb meaning to a meaning elaborating its trajector. In languages like English, the dependence in most situations manifests itself linguistically in the need for an NP; in Spanish or Mandarin Chinese, the dependence manifests itself in a drawing upon previously introduced referents, in a manner that is very like pronominal reference (cf Tomlin 1991). The mechanism by which a more or less linguistic enrichment of meaning takes place in virtue of the slot into which an utterance is inserted is the same as we find in the case of selectional restrictions giving rise to metaphors. Time

FUNCTION, COGNITION,

AND

LAYERED CLAUSE STRUCTURE

61

flies works by inserting “time” in an agent slot, just as gladly turns into a commitment to perform an action because it is inserted into a slot where the alternative is acceptance or rejection.

10. Two possible ways of keeping language purely cognitive There are two ways in which the theory that meaning is entirely distinct from communication might be upheld even in agreement with the basic picture I have outlined. First of all, it would be compatible with the evolutionary scenario to have a language with purely conceptual, de-contextualized meaning — provided that all the work of plugging meaning into the context was left to the inferential abilities of the speaker. This would imply that function was not coded in human language at all, but left solely to inferential, pragmatic interpretation — or, alternatively, left to the paralinguistic system that we have essentially inherited from the apes. Language would then feed descriptive, conceptual meaning into the situation without any indication of how it was to be related to the situation. This view is extremely resilient: the distinction between “illocutionary” and “locutionary” act in Austin preserves more or less the same dichotomy between coded description and pragmatic function, with the performative verbs as the privileged exception; and it is still customary to speak of “pragmatic” as opposed to “semantic” types of meaning, even in cases where the so-called pragmatic type of meaning is clearly coded, cf., e.g., the notion of “pragmatic particles”. There is a (deceptive) plausibility about this picture which is due to the fact that the functional, situational types of meaning are by their nature the most easily inferable. It is not typical to be able to guess the conceptual concept of the next utterance — but the nature of the situation may constrain its situational function considerably. This means that if a clever addressee gets the right content words, he can do the job of organizing them and assigning a situational function to them on his own. This is why Schank on behalf of the AI community can say that he can do virtually without syntax, managing everything by inference schemas. That this is communicatively feasible even with a limited vocabulary is exemplified by the tourist situation. And some languages do not code deictic tense, or definite articles. In spite of all this, the human language prototype does not leave all functional organization either to paralinguistic signals or to the unaided infer-

62

PETER HARDER

ences of the addressee. Exactly how the functional dimension manifests itself is a matter of language-specific organization of meaning; but indications specifying the situational application of the conceptual meaning are always present to some degree. Since this does not in any way detract from the centrality of the conceptual richness of language, I see no reason to exile the situational embedding from the basic picture of human language. Another way of demonstrating the points of relatedness between human language and pre-cognitive types of communication is to look at holophrastic types of linguistic communication, such as greetings. As noted above, ways of showing “recognition + lack of hostile intentions” go down quite far in the animal kingdom; and human beings can convey this type of meaning either by gestures (arm-waving), or by signals which, even if they stand out by being holophrases, are phonologically integrated into the linguistic system as a whole. This overlap can be seen as one way in which human language does not stick to its privileged territory of conceptual meaning, but is used also for communicative purposes that can be understood in terms of pre-cognitive stages of development. The second way of cutting off the functional or situational aspect of meaning involves a way of using the word “cognitive” where it covers everything that goes on in language. The idea is that language must be entirely explicable in terms of cognitive process simply in virtue of the fact that nothing that plays a role in language can do so outside the cognitive apparatus of a human language user. Where understanding stops, language must stop. I regard this as trivially true. In this sense, language is a cognitive phenomenon just as seeing the Taj Mahal, fighting in the Second World War and playing football are cognitive phenomena: our experience is the result of cognitive processes going on inside our heads. If the word is to be sensibly contrasted with anything, however, I think it is better to narrow it down to cover mental phenomena as distinct from external events. This would be the complementary error of behaviorism: rather than reducing everything to the simplicity of pre-mental processes, one would assign mental status to everything. It is probably true that once cognition is there, it interferes with everything; but this does not imply that cognition is all there is. If we tentatively distinguish between events in the mental world and events in the external world, we have to add that what we understand as the external world is something also created by means of cognitive processes — which might seem to bring us back to square one. The reason why it does not

FUNCTION, COGNITION,

AND

LAYERED CLAUSE STRUCTURE

63

do so is that everything in our lives is organized round the fundamental distinction between what goes on inside as opposed to outside the individual. Inability to distinguish between the two types of events is the crucial criterion of insanity. Relating to the situation around you is therefore very different from playing around with possibilities on the mental level only; and linguistic meaning covers both types of activity. The price of calling both types of event “cognitive” is that we fail to identify an interesting sub-area of human activity: there is too much cognition going on and too little else.

11. Conclusion It is now time to gather the threads in the argument. The general picture should now have become clear. Communicative interaction is evolutionarily older than cognition. With the development of cognition, for a time (from the first mammals up to and including the pre-human apes) communicative resources did not keep up with the evolutionary pace of cognition. Human language, by contrast, stands on the shoulders of the cognitive advances. In relation to a purely conceptual picture of linguistic meaning, this perspective emphasizes that the situational embedding that characterizes pre-human systems of communication was not utterly lost with the advent of human language: rather, it became factored out into separate coded meanings (clustering at the top of the structure) by the same division of labor that factored out purely conceptual meanings (clustering at the bottom end of the structure). The evolution of human language can be seen as a two-way process: communication expanded inward, tapping the cognitive resources of the mind; and cognition expanded outward into the sphere of social co-operation. The general formula for the language structure that superseded the holophrastic stage was “cognition embedded in interaction” — as reflected in the basic format of the layered model of the clause.

Notes 1.

The story is somewhat more complicated, and crucially involves a part-whole relation between the functional item and that whose survival it promotes; for a fuller discussion, cf. Harder (1996).

64

PETER HARDER

References Allwood, J. 1976 Allwood, J. 1977

Linguistic Communication as Action and Cooperation. (Gothenburg Monographs in Linguistics 2.) University of Göteborg: Dept. of Linguistics. “A Critical Look at Speech Act Theory”. Logic, Pragmatics and Grammar ed. by Ö. Dahl, 53–69. Lund: Studentlitteratur.

Bühler, Karl 1934 Sprachtheorie. Jena: Fischer. Dik, Simon C. 1989 The Theory of Functional Grammar. Vol 1: The Structure of the Clause. Dordrecht: Foris. Dawkins, Richard 1981 “Communication”. The Oxford Companion to Animal Behaviour. ed. by D.McFarland, 78-91. Oxford: Oxford University Press. Foley, W. A. & Van Valin, R. D. 1984 Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. Fortescue, M., Harder, P. and Kristoffersen, L. eds. 1992 Layered Structure and Reference in a Functional Perspective. Amsterdam: John Benjamins. Geeraerts, Dirk 1992 “The return of hermeneutics to lexical semantics”. In (ed), Thirty Years of Linguistic Evolution ed. by M. Pütz, 257–282. Amsterdam: John Benjamins. Gibson, J. J. 1966 The Senses Considered as Perceptual Systems. Boston: Houghton Mifflin Company. Givón, T. 1989 Mind, Code and Context. Essays in Pragmatics. Hillsdale, New Jersey: Lawrence Erlbaum. Givón, T. 1990 Syntax. A Functional-Typological Introduction. Vol. 2. Amsterdam: John Benjamins. Gulz, A. 1991 “The Planning of Action as a Cognitive and Biological Phenomenon”, Lund: Lund University Cognitive Studies 2. Gärdenfors, Peter. 1991 “The Emergence of Meaning”. Lund: Lund University Cognitive Studies 5. Gärdenfors, P. 1992 Blotta Tanken. Nora, Sweden: Nya Doxa. Halliday, M.A.K. 1970 “Functional diversity in language, as seen from a consideration of modality and mood in English”. Foundations of Language 6, 322-361.

FUNCTION, COGNITION,

AND

LAYERED CLAUSE STRUCTURE

65

Halliday, M. A.K. 1985 An Introduction to functional grammar. London: Edward Arnold. Harder, P. 1978 “Language in action. Some arguments against the concept ‘illocutionary’”. Papers from the Fourth Scandinavian Conference of Linguistics ed. by Kirsten Gregersen. Odense: Odense University Press. Harder, P. 1990 “Tense, semantics and layered syntax”. In Nuyts, Bolkestein & Vet 1990. 139-63. Harder, P. 1992 “Semantic Content and Linguistic Structure in Functional Grammar. On the Semantics of ‘Nounhood’”. In Fortescue, Harder & Kristoffersen 1992. 303-27. Harder, P. 1996 Functional Semantics: A Theory of Meaning, Structure, and Tense in English. Berlin: Mouton de Gruyter. Hengeveld, Kees 1989 “Layers and Operators in Functional Grammar”, Journal of Linguistics 25: 127-57. Hengeveld, Kees 1990 “The Hierarchical Structure of Utterances”. In Fortescue, Bolkestein & Vet 1990. 1-23. Hengeveld, Kees 1992 “Parts of Speech”. In Fortescue Fortescue, Harder & Kristoffersen 1992. 29-55. Johnson-Laird, P. N. 1983 Mental Models. Cambridge: Cambridge University Press. Langacker, R. W. 1987a “Nouns and verbs”, Language 63,1. 53-94. Langacker, R. W. 1987b Foundations of Cognitive Grammar, vol.1, Theoretical Prerequisites. Stanford: Stanford University Press. Langacker, R. W. 1990 “Subjectification”. Cognitive Linguistics 1: 5-38. Langacker, R. W. 1991 Foundations of Cognitive Grammar, vol.2, Descriptive Applications. Stanford: Stanford University Press. Leech, G. N. 1981 Semantics. Second edition. Harmondsworth: Penguin. Lepore, E. & Van Gulick, R. eds. 1991 John Searle and his Critics. Oxford: Basil Blackwell. Millikan, R. 1984 Language, Thought and Other Biological Categories. Cambridge, Mass: MIT Press.

66 Nuyts, J. 1990

PETER HARDER

“Linguistic representation and conceptual knowledge representation”. In Nuyts, Bolkestein & Vet 1990. 263-93. Nuyts, J., Bolkestein, A. M. and Vet, C., eds. 1990 Layers and Levels of Representation in Language Theory. Amsterdam: John Benjamins. Popper, K. 1972 Objective Knowledge: An evolutionary approach. Oxford: The Clarendon Press. Putnam, H. 1988 Representation and Reality. Cambridge, MA: MIT Press. Rijkhoff, J. 1990 “Explaining Word Order in the Noun Phrase”. Linguistics 28. 5-42. Rijkhoff, J. 1992 The Noun Phrase. A Typological Study of its Form and Structure. Amsterdam: University of Amsterdam. Rumelhart, D. E., James L. McClelland & The PDP Research Group. 1986 Parallel Distributed Processing. Explorations in the Microstructure of Cognition. Cambridge, MA: MIT Press. Searle, J. R. 1969 Speech Acts. New York and London: Cambridge University Press. Searle, J. R. 1991 “Response: Meaning, Intentionality, and Speech Acts”. Lepore & Van Gulick 1991. 81-102. Searle, J. R. 1992 The Rediscovery of the Mind. Cambridge, MA.: MIT Press. Siewierska, Anna 1992 “Layering in FG and GB”. In Fortescue, Harder & Kristoffersen 1992. 40932. Strawson, Peter F. 1959 Individuals. London: Methuen. Tomlin, R. S. 1991 “The management of reference in Mandarin discourse”, Cognitive Linguistics 2 - 1: 65-93. Ulbæk, I. 1989 “Evolution, sprog og kognition”, unpublished PhD Dissertation. University of Copenhagen. Wright, L. 1973 “Functions”, The Philosophical Review, Vol. LXXXII, 2. 136–68.

FROM VISION TO COGNITION

67

From Vision to Cognition A Study of Metaphor and Polysemy in Swedish Sören Sjöström Department of Linguistics, Göteborg University

1.

Introduction and background

One part of human communication is concerned with internal experience. Naturally, a great many expressions have as their primary function to describe such experience. However, a number of studies have shown that words from other domains are also used in this function. One of the earliest studies to show this was made by Kurath (1921), who demonstrated that words expressing emotion in the Indo-European languages are often derived from expressions referring to the physical actions or experiences which accompany these emotions in question, or to the body parts affected by the physical reactions. One example of this is that the expression heart, besides being used to refer to a bodily organ, is also used to refer to love, courage, etc., depending on the fact that the function of the heart is affected noticeably in amorous and dangerous situations (see Sweetser 1990). Writers such as Lakoff and Johnson (1980), Johnson (1987), Lakoff (1987), Sweetser (1990), and others have also shown that expressions for perception are used metaphorically in the description of internal experience. According to Sweetser (1990), it seems to be the case that, when expressions for perception are used metaphorically, there is a correlation between different modes of perception and different kinds of internal experience. If we limit this observation to vision, hearing and tactile perception (tasting and smelling seem to be more problematic), there seems to exist correlation like the one shown in Table 1.

68

SÖREN SJÖSTRÖM

Table 1. Correlation between perception and cognition. mode of perception

internal experience

linguistic examples

sight (visual) hearing (auditive) feeling (tactile)

objective and intellectual subjective and communicative emotional

inse ‘realize’ hörsamma ‘obey’ känna ‘feel’

These correlations can be explained rationally. Sight, Sweetser points out, is our primary source of objective data. Studies of child language acquisition also suggest (Eve Clark 1976) that visual perception is perhaps the most important aid for the developing child to characterize its environment. Hearing, although it plays a central role in the understanding of what is said, has a different function. Human language is typically auditive and requires susceptibility on the part of the interlocutors. This susceptibility may be understood metaphorically as “mental susceptibility” and, ultimately, as compliance. The connection between tactile perception and emotional experience is perhaps best explained by Kurath (1921) as based on the close connection between physical experiences and emotional ones.

2.

Some observations

With reference primarily to Sweetser’s (1990) study of the metaphorical use of expressions for visual perception, I will make a few comments. First, Sweetser’s study treats the metaphorical use of perception verbs in the IndoEuropean languages. It is clear, however, that nouns, adjectives, etc. are also used metaphorically: att ha insikt ‘to have insight’ att vara synsk ‘to be clairvoyant’ In my survey of Swedish expressions I will give examples to show that the metaphorical use of expressions is not limited to verbs, but also includes nouns, adjectives and, by derivation, adverbs. Secondly, in Sweetser’s analysis of which cognitive aspects are expressed metaphorically, knowledge appears to be the basic concept. This seems to be — mainly — true, but we should observe that, at least in Swedish and English, we also find expressions for intention, attitude to

FROM VISION TO COGNITION

69

knowledge, and non-understanding, etc. sikta på ‘aim at’ framsynt ‘far-sighted’ vara blind ‘being blind’

intention attitude to knowledge inability to understand

Visual expressions may thus be employed to express very many aspects of cognition where “knowledge” is perhaps only one of the most abstract concepts. My last comment concerns the conceptual analysis of vision. Sweetser seems to consider ‘sight’ as a primitive concept. This is somewhat inadequate as a semantic analysis. Sight presupposes the ability to perceive light. Thus, we use the expression he could no longer see to refer to a situation where an individual has lost the ability to perceive light. In my semantic field analysis (section 4), I will accordingly understand the concept ‘light’ as more basic than ‘sight’ or ‘seeing’.

3.

Aim of the study

Taking the critical remarks above as a starting-point, my aim was to explore what different aspects of cognition are expressed metaphorically, or polysemically, in Swedish by the aid of expressions connected with visual perception. The study was not restricted to verbs but included all kinds of lexemes connected with light. The investigation was entirely based on information in Swedish dictionaries concerned with synonymy. The difference between the present study and studies by writers such as Viberg (1980) is that, while Viberg’s interest is focused on the Swedish expressions from a typological point of view, my study is clearly more language-specific with the ultimate aim of formulating some hypotheses which can be investigated in other languages.

4.

The semantic field of visual perception

What follows in this section is an attempt to understand how language structures the domain of visual perception. I would like to characterize my analysis as “folk semantic”, in that it makes an abstraction away from much of the

70

SÖREN SJÖSTRÖM

physical, physiological and psychological knowledge of the phenomena in question. Visual perception is best understood as a two-place relation between a perceiver (usually understood as an agent) and a perceived object. Thus, a great number of lexemes in natural language, typically transitive verbs, denote this relation itself: relation of visual perception: see, look at, stare at, etc These verbs may be considered as the best examples of expressions for visual perception. There are, however, other types of expression related to visual perception, as shown below: far-seeing, blind, visible, light, dark, illuminate, etc. In short, these examples suggest that a semantic field analysis of visual perception can be extended to include at least the following notions: property of the perceiver: far-seeing, blind; property of the perceived: visible, dark, light; external cause of perception: illuminate, darken, dazzle. Expressions related to visual perception fall into four classes: (1) transitive verbs (or intransitive verbs combined with prepositions) which express visual perception as a relation; (2) adjectives and (3) intransitive verbs which express properties of the perceiver or the perceived object; (4) transitive verbs which presuppose an external cause affecting the relation, the perceiver or the perceived. So far, I have not found one single expression denoting a cause for the relation itself. The various notions are summarized in Table 2. Expressions like those in Table 2 can be utilized in various ways. Productive rules of Swedish morphology allow derivations like betrakta ‘look at’ -> betraktelse ‘reflection, meditation’, where the derived noun has a more abstract meaning than the verb. Such examples are important to record because it is often in these derivations we may observe metaphorical meaning. Some verb inflections also allow the shift from relation to property. Thus, the verb förutse ‘foresee’ (which is related etymologically to se) in its present participle form expresses a (relational) property of the perceiver: förutseende ‘having foresight’.

71

FROM VISION TO COGNITION Table 2. Visual perception

PERCEIVER ha syn ‘have sight’ se ‘see’ vara blind ‘be blind’

öppna ögonen på ‘open the eyes of’ förblinda ‘blind, dazzle’

RELATION see ‘see’ titta på ‘look at’ iakttaga ‘observe’ åse ‘witness’ skåda ‘behold, see’ betrakta ‘look at’

CAUSATIVE

PERCEIVED vara synligt ‘be visible’ synas ‘be visible’ vara osynligt ‘be invisible’ vara mörkt eller ljust ‘be dark or light’

belysa ‘illuminate’ mörklägga ‘darken’

In my account of Swedish data (Section 7), I will include examples of inflection, derivations and also compounding. Using the semantic field analysis as a starting-point, we are now in a position to formulate some hypotheses or predictions about what cognitive meanings visual expressions for visual perception may express. If the verb se ‘see’ is used to express understanding, and if the semantic structure which contains this verb is systematically mapped into the cognitive domain, we may predict that: not seeing will express ‘nonunderstanding’ being blind will express ‘inability to understand’ shut one’s eyes will express ‘avoiding understanding (information)’ illuminate will express ‘explaining’ darken will express ‘making information inaccessible’ light will express ‘knowledge’ darkness will express ‘ignorance’ The examples in Section 6 confirm these hypotheses on the whole. We will also meet cases not predicted by the analysis above, which, nevertheless, seem to be coherent with the predictions.

72 5.

SÖREN SJÖSTRÖM A note on data and method

My data consist of information that can be extracted from dictionaries. Ideally, though, a study like the present one should be supplemented by studies of corpuses of written and spoken language data. One problem for a student of polysemy is that such dictionaries simply do not exist. I was therefore obliged to use a more indirect method: looking up, for example, the word see in a dictionary of synonymy, one will find that one of its synonyms is understand. Obviously, then, see and understand can express a common concept. So, using the semantic field analysis as a starting-point, I looked up words for ‘visual perception’, ‘properties of the perceiver’, ‘properties of the perceived’ and ‘cause of perception’. Then I recorded which synonyms expressed cognition. There are problems connected with this method. One is that it is difficult to formulate criteria for what counts as an expression for cognition. Typically, however, cognition should include “mental processes connected with understanding, formulation of beliefs, and acquisition of knowledge” (Flew 1979). I have seen no strong reason for being dogmatic at this point, but have also included expressions for intention, attitudes to knowledge, etc. I soon observed that most expressions of the visual domain are strongly polysemous. Nevertheless, there are exceptions where verbs which denote visual perception are not polysemous: glo glutta kisa skela bliga plira snegla

‘stare, glare, gape’ ‘take a glance at...’ ‘peer’ ‘squint’ ‘stare, glare’ ‘peer, screw up one’s eyes’ ‘ogle, glance’

The works I have relied on are Strömberg’s Synonymordboken (1979), Bring’s (1930) Svenskt ordförråd (which is best characterized as a Swedish version of Roget’s Thesaurus), Hellqvist’s (1980) Svensk etymologisk ordbok and Svensk ordbok (1986). It should be pointed out that I find no reason to make a distinction between polysemy and metaphor in this study. However, the fact that the secondary meanings I have found have indeed entered the dictionary may suggest that we are dealing with polysemy.

FROM VISION TO COGNITION 6.

73

Results of the lexical survey

This section presents the results of my investigation of Swedish lexical expressions for visual perception. It is clear that some of the categories may overlap. The verb se, for example, can be used to express both a relation and a property. The examples are presented in the four categories suggested in the semantic field analysis above, namely, ‘visual perception’, ‘properties of the perceiver’, ‘properties of the perceived’ and ‘cause of perception’. It would be possible to subclassify these four categories into finer categories. The reader will see that the real analytical work has consisted in classifying the examples in groups and providing a suitable name, representing an abstraction, for the different groups. I believe there is only one group of examples which demands a special comment, namely the expressions in section III:9. Why do noun expressions like sken ‘glare’ express “fallacy” when the corresponding verb — skina ‘shine’ — does not? The difference between äktenskap ‘marriage’ and sken-äktenskap ‘pro forma marriage’ is that the latter (Swedish) example explicitly invokes the perceptive aspect, while the former example does not. I now suggest that the perception expression — ‘glare’ — helps to introduce a reservation: “it looks like a marriage to me”. In the former example, there is no such reservation, and the expression is therefore more likely to invite an interpretation where truth is not questioned. I. Visual perception → cognition Even though I have suggested that verbs are the typical means of expressing the perceptual relation, we must allow for the fact that a number of nouns are understood as reifications of this relation. Thus, we find in I.4 below the expression revision. 1. Seeing → understanding SWEDISH ENGLISH se synonyms: inse förstå begripa fatta

‘see’ ‘realize’ ‘understand’ ‘comprehend’ ‘grasp’

74 besinna betänka anse betrakta iakttaga observera skåda åse

SÖREN SJÖSTRÖM ‘consider, bear in mind’ ‘consider, hesitate’ ‘think, consider’ ‘look at, contemplate, regard’ ‘observe, notice’ ‘observe, note’ ‘behold, see’ ‘watch, witness’

The Swedish verb se is not used in exactly the same way as English see. Swedish has inse “in-see” to denote cognitive processes, and this verb is not used to denote perception. Etymologically inse is directly related to se (cf Hellqvist 1980). We may also include a number of idiomatic phrases used metaphorically like, for example, se för sitt inre öga ‘see with one’s inner eye’, gå upp ett ljus för ‘dawn upon’, få en snilleblixt ‘get a flash of genius’, få upp ögonen för ‘have one’s eyes opened to’. 2. Not seeing → not understanding The negation of a sentence which contains the verb se expresses ‘nonunderstanding’: jag kan inte se hur detta hänger ihop med vad du säger ‘I cannot see how this fits in with what you are saying’ 3. Avoiding seeing → avoiding information (and thus responsibility) SWEDISH

ENGLISH

blunda för inte vilja se göra sig blind (och döv) för se genom fingrarna med (eller mellan)

‘shut one’s eyes to’ ‘not wanting to see’ ‘making oneself blind (and deaf) to’ ‘wink/connive at, be lenient with’

4. Seeing actively (=looking) → seeking information SWEDISH se efter titta på syna titta besiktiga bese avsyna överse utföra brandsyn

ENGLISH ‘see if...’ ‘look at’ ‘inspect, survey, examine, scrutinize’ ‘look’ ‘inspect, examine’ ‘see, look at’ ‘inspect’ ‘overlook’ ‘inspection of fire prevention arrangements’

FROM VISION TO COGNITION genomse granska inspektera introspektion mönstra revidera revision skärskåda

75

‘look through’ ‘examine, scrutinize’ ‘inspect’ ‘introspection’ ‘inspect, scrutinize’ ‘revise’ ‘revision’ ‘examine, view, scrutinize’

5. Intentionally not looking at an object → disregarding certain facts SWEDISH bortse ifrån lämna utan avseende frånse ej ta i betraktande oavsett

ENGLISH ‘disregard’ ‘leave out of consideration’ ‘leave out of account’ ‘not take into consideration’ ‘irrespective of’

6. Way of seeing → way of understanding SWEDISH syn på helhetssyn grundsyn överblick synsätt synvinkel perspektiv synpunkt åsikt

ENGLISH ‘view of’ ‘total view’ ‘basic view’ ‘overview’ ‘outlook, approach’ ‘visual angle, angle, aspect’ ‘perspective’ ‘point of view’ ‘view, opinion’

7. Seeing beyond one’s ordinary field of vision → interpreting as possibilities or having hopes SWEDISH utsikter att ha vyer vy vision (hålla) utkik

ENGLISH ‘views, prospects’ ‘to hold views’ ‘view’ ‘vision’ ‘keep a look-out’

8. Seeing backwards → remembering SWEDISH återblick retrospektion retrospektiv revy

ENGLISH ‘retrospect, flashback’ ‘retrospection’ ‘retrospective’ ‘review’

76

SÖREN SJÖSTRÖM

9. Seeing past an object → overlooking something SWEDISH förbise

ENGLISH ‘overlook’

10. Seeing in a definite direction → having a special intention SWEDISH syfta syfta till syfta på åsyfta sikta ha i sikte avse ha i kikaren kasta blickar på

ENGLISH ‘aim’ ‘aim at’ ‘refer to’ ‘refer to’ ‘aim’ ‘be in sight of’ ‘intend, refer to’ ‘intend’, “have in the binoculars” (literal translation) ‘cast a covetous eye in the direction of’

11. Keeping an object in the visual field → controlling that object SWEDISH ha uppsikt över överse utöva tillsyn över se till ha ögonen på utse observera överblicka överinseende eftersyn

ENGLISH ‘to supervise’ ‘oversee’ ‘supervise, look after’ ‘see to’ ‘have one’s eyes on’ ‘choose, pick out’ ‘observe’ ‘survey’ ‘supervision’ ‘closer inspection’

12. Seeing into or through an object → understanding that object SWEDISH (ha) insikt (i NP) synonyms: kännedom förståelse inseende inblick urskillning omdöme medvetande förstånd rön lärospån

ENGLISH ‘(have) insight’ ‘knowledge, acquaintance’ ‘understanding’ ‘supervision’ ‘glimpse, insight’ ‘discrimination’ ‘judgement’ ‘consciousness’ ‘intellect’ ‘observation, experience,discovery’ ‘first experience’

FROM VISION TO COGNITION kunskaper vetande komma till insikt om (vara) insiktsfull (ha) insyn se tvärs igenom ha inblick i

‘knowledge’ ‘knowledge’ ‘realize’ ‘insightful’ ‘(have) insight’ ‘look through’ ‘have insight into’

13. Looking forward → predicting, expecting SWEDISH förutse motse = frukta se fram emot

ENGLISH ‘foresee, anticipate’ ‘expect, await, fear’ ‘look forward to’

II. Properties of the perceiver 1. Being blind, unable to see → unable of understanding SWEDISH blind synonyms: förblindad förstockad oförstående ensidig okritisk tanklös besinningslös oförnuftig vara blind för att ha skygglappar

ENGLISH ‘blind’ ‘blinded’ dazzled’ ‘hidebound’ ‘unsympathetic (towards) ‘one-sided, biased, prejudiced, narrow-minded’ ‘uncritical’ ‘thoughtless’ ‘rash, unreflecting, reckless’ ‘unreasonable, irrational’ ‘be blind to’ ‘wearing blinkers’

2. Being enlightened → being gifted SWEDISH upplyst med andlig synvidd med andlig horisont ljushuvud klassens ljus lysande begåvning briljant

ENGLISH ‘enlightened, educated’ ‘with spiritual horizon’ ‘with spiritual horizon’ ‘genius’ ‘genius of his class’ ‘brilliant talent’ ‘brilliant’

77

78

SÖREN SJÖSTRÖM

3. Reflecting light → thinking, meditating SWEDISH reflektera reflektion

ENGLISH ‘reflect, think, meditate’ ‘reflexion’

4. Not reflecting light → having poor general understanding SWEDISH oreflekterad

ENGLISH ‘unreflecting, spontaneous’

5. Having good sight → having good understanding SWEDISH vara klarsynt vara skarpsynt vara genomskådande

ENGLISH ‘be clear-sighted’ ‘be sharp-eyed’ ‘have a penetrating eye’

6. Seeing what others cannot see → having supernatural knowledge SWEDISH synsk sia siare sierska fjärrskådare klärvoajant

ENGLISH ‘second-sighted, clairvoyant’ ‘prophesy (of) ‘seer, prophet’ ‘female seer, prophet’ ‘clairvoyant’ ‘clairvoyant’

7. Having a vast field of vision → planning well, being tolerant to novelties/ innovations SWEDISH förutseende framsynt vidsynt kringsynt överse med överseende fjärrsynt ha tillförsikt

ENGLISH ‘far-seeing, far-sighted’ ‘far-seeing, far-sighted’ ‘tolerant’ ‘wide-sighted’ ‘tolerate’ ‘indulgent, tolerant’ ‘clairvoyant’ ‘being self-reliant’

8. Having a limited field of vision → planning badly, being intolerant to novelties/innovations SWEDISH kortsynt lite förutseende inte se längre än näsan räcker

ENGLISH ‘short-sighted’ ‘little ‘not see longer than the tip of one’s nose’

FROM VISION TO COGNITION trångsynt med små vyer närsynt

‘narrow-minded’ ‘with small views’ ‘short-sighted, near-sighted’

9. Having a special way of seeing → having a special talent SWEDISH ha blick för ha öga för

ENGLISH ‘have an eye for’ ‘have an eye for’

10. Looking forward → being cautious SWEDISH försiktig förutseende

ENGLISH ‘careful, cautious, prudent’ ‘far-sighted, far-seeing’

11. Being large-eyed → showing wonder SWEDISH storögd

ENGLISH ‘large-eyed, round-eyed’

12. Being a friend of darkness → being hostile to knowledge SWEDISH obskurant

ENGLISH ‘obscurant’

III. Properties of the perceived object 1. Being visible → giving evidence of SWEDISH synas visa sig synbar synbarligen så vitt man kan se till synes visuell åskådlig åskådning överskådlig synbar-lig-en ansenlig iögonfallande välsedd ohöljd siktbar

ENGLISH ‘be visible’ “appear to be” ‘it will show’ ‘visible’ ‘evidently’ ‘as far as you can see’ ‘visible’ ‘visual’ ‘clear, lucid’ ‘outlook, opinions, views’ ‘clear, lucid’ ‘obviously, evidently’ ‘considerable, good-sized’ ‘conspicious’ ‘well-seen (literally), appreciated’ ‘unconcealed, unveiled, open’ ‘visible’

79

80

SÖREN SJÖSTRÖM

2. Being visually more perceptible than → dominating SWEDISH överglänsa överskugga skymma

ENGLISH ‘outshine’ ‘overshadow’ ‘block, dim , obscure’

3. Light → knowledge SWEDISH ljus ljus (idé) förklaring klarhet glimt upplysning rampljus

ENGLISH ‘light’ ‘bright (idea)’ ‘explanation’ ‘clearness, clarity, transparency’ ‘gleam, flash, glimpse’ ‘lighting, illumination, enlightenment’ ‘limelight’ “public knowledge”

4. Being light → being comprehensible SWEDISH ljusna dagas

ENGLISH ‘get light, dawn’ ‘dawn’

5. Being transparent → being easily comprehensible SWEDISH klar solklar klarna klart (intellekt) genomskinligt skönjbar sikt

ENGLISH ‘clear’ ‘as clear as daylight’ ‘getting clear’ ‘clearly’ ‘transparent’ ‘discernible’ ‘visibility, view’

6. Darkness → ignorance SWEDISH mörker synonyms: okunnighet oupplysthet barbari blindhet töcken dunkel

ENGLISH ‘darkness’ ‘ignorance’ “unenlightenment” ‘barbarism’ ‘blindness’ ‘mist, haze’ ‘dusk, gloom’

FROM VISION TO COGNITION 7. Being opaque → being abstruse, obscure SWEDISH ogenomtränglighet ogenomskinlig diffus dimbildning disig det blir oklarare oförklarlig oklar otydlig

ENGLISH “impenetrability, imperviousness” ‘non-transparent’ ‘diffuse, blurred’ ‘smoke screeing’ ‘hazy’ ‘getting more unclear’/’less transparent’ ‘inexpicable, unaccountable’ ‘unclear’ ‘unclear’

8. Being dark → being incomprehensible SWEDISH mörk synonyms: oviss gåtfull svårförståelig obskyr mulen det mörknar skumt skymning oöverskådlig töcknig

ENGLISH ‘dark’ ‘uncertain, doubtful’ ‘mysterious, puzzling, enigmatic’ ‘obscure’ ‘obscure’ ‘cloudy, gloomy’ ‘getting darker’ ‘obscure’ ‘ignorance’ ‘confused, incalculable’ ‘misty, hazy’

9. Glare, which is not direct light → fallacy SWEDISH skenbar skenbild sken skenhelig skenfager skengravid skenäktenskap skendöd skenmanöver bländverk skuggbild drömsyn hägring

ENGLISH ‘apparent, seeming’ ‘phantom, shadow’ ‘light, glare, gleam’ ‘hypocritical’ ‘fraudulent’ ‘apparently pregnant’ ‘pro forma marriage’ ‘apparent death’ ‘diversion, feint’ ‘delusion, illusion’ ‘shadow picture, silhoutte’ ‘dream-vision’ ‘mirage, illusion’

81

82

SÖREN SJÖSTRÖM

10. The observed part → the understood part SWEDISH hänseende avseende utseende se ut som uppsyn ansikte synfält synpunkt glimt aspekt spegelbild uppseende

ENGLISH ‘aspect’ ‘aspect’ ‘outlook, appearance, looks’ ‘look like’ ‘facial expression, countenance’ ‘face’ ‘field of vision’ ‘point of view’ ‘gleam, flash, glimpse’ ‘aspect’ ‘reflected image’ ‘attention, stir’

IV. Causes of perception → causes of understanding 1. Making something visible → conveying information SWEDISH förklara klargöra klarlägga lysa synonyms: upplysa kungöra tillkännage förklara offentliggöra belysa kasta ljus över sprida ljus över sätta strålkastaren på spegla utlysa

ENGLISH ‘explain’ ‘make clear’ ‘make clear’ ‘shine, glare, gleam’ ‘enlighten’ ‘make known’ ‘make known, notify, announce’ ‘explain’ ‘make public’ ‘enlighten’ ‘throw light on’ ‘spread light over’ ‘put the spotlight on’ ‘mirror’ ‘give notice, advertise, proclaim’

2. Making something invisible → making certain information inaccessible SWEDISH mörklägga mörkman förmörka

ENGLISH ‘darken’ ‘obscurantist’ ‘darken’

FROM VISION TO COGNITION

83

3. Causing blindness → preventing from understanding SWEDISH förvända synen på förblinda kasta sand i ögonen på blända

ENGLISH ‘turn somebody’s sight’ ‘blind’ ‘throw sand into the eyes of’ ‘blind, dazzle’

4. Creating glare → creating a fallacy SWEDISH låta påskina ge sken av

7.

ENGLISH ‘make pretence of’ ‘pretend, make a show of...’

Summary and conclusions

My lexical survey has shown that the polysemous use of lexical expressions connected with vision characterizes a large part of Swedish vocabulary, both with regard to the actual number of expressions and the number of lexical categories involved: not only verbs but also nouns and adjectives are used. In principle, the relation between the visual domain and the cognitive domain is quite straightforward. From the assumption that light metaphorically represents ‘knowledge’, it follows that perception of light expresses ‘understanding’, non-perception of light non-understanding, illumination ‘explanation’, etc. The various findings are summarized in Table 3 a-d. The findings of the study can be used to formulate interesting empirical questions. First, the relative ease of translating the Swedish examples into English indicates a high degree of similarity between Swedish and English in the use of expressions related to light for the expression of cognition. But we do not know exactly how similar the two languages are in this respect. Secondly, it would be interesting to investigate languages which are linguistically and culturally very remote from Swedish and other Germanic languages like, for example, Chinese. Without having carried out any systematic comparison with Chinese, I understand that most of the Swedish examples translate into corresponding metaphors in Chinese (Fanglan Chen in a personal communication). It is my hope that a comparative study comprising languages such as Chinese (culturally and linguistically remote), Finnish (culturally closely related, but linguistically remote) and English (culturally

84

SÖREN SJÖSTRÖM

and linguistically closely related) can further our knowledge of universals in this area of semantics and pragmatics. Table 3a. The perception relation → the cognitive relation seeing not seeing avoiding seeing seeing actively (= looking) intentionally not looking at an object way of seeing seeing beyond one’s ordinary field of vision seeing backwards seeing past an object seeing in a definite direction keeping an object in the visual field looking forward

understanding not understanding avoiding information seeking information disregarding that object way of understanding seeing possibilitities remembering overlooking that object having special intention controlling that object predicting, expecting

Table 3b. The perceiver → the cognitive agent being blind being enlightened reflecting light not reflecting light seeing what others cannot see having a vast field of vision having a limited field of vision having a special way of seeing looking forward being large-eyed being a friend of darkness

incapable of understanding being gifted thinking, meditating unreflecting, spontaneous having supernatural knowledge planning well, tolerant planning badly, intolerant having a special talent being cautious showing wonder being hostile to knowledge

Table 3c The perceived object → the cognitive object being visible being visually more perceptible light being light being transparent darkness being opaque being dark glare observed part

giving evidence of dominating knowledge being comprehensible being easily comprehensible ignorance being abstruse, obscure being incomprehensible fallacy understood part

FROM VISION TO COGNITION

85

Table 3d. A cause of perception or non perception→ a cause of cognition or non cognition making visible making something invisible causing blindness creating glare

promoting understanding making information inaccessible preventing from understanding creating a fallacy

References Abelin, Å. 1988

Bring, S. C. 1930

“Patterns of Synaesthesia in the Swedish Vocabulary”. I: Studies in Computer-Aided Lexicology. Department of Computational linguistics. University of Göteborg. Svenskt ordförråd ordnat i begreppsklasser. Hugo Gebers Förlag. Stockholm.

Clark, Eve V. 1976 "Universal categories: on the semantics of classifiers and children's early word meanings". In Alphonse Juilland (ed.). Linguistic studies offered to Joseph Greenberg (Studia Linguistica et Philogica 4:3). Saratoga; CA: Anma Libri, Vol. 3, pp. 449-462. Gärdenfors, P. 1992 Blotta tanken. Bokförlaget Nya Doxa. Jackendoff, R. 1983 Semantics and cognition. Cambridge. MA: MIT Press. Johnson, M. 1987 The body in the mind. Chicago: University of Chicago Press. Lakoff, G och M. Johnson 1980 Metaphors we live by. Chicago: University of Chicago Press. Miller, G. A. and P. Johnson-Laird 1976 Language and perception. Cambridge University Press. Strömberg, A. 1979 Stora synonymordboken. Strömbergs. Stockholm. Sweetser, E. 1990 From etymology to pragmatics. Cambridge studies in linguistics 54. Viberg, Å. 1983 “A universal lexicalization hierarchy for the verbs of perception”. Institute of Linguistics, University of Stockholm.

POLYSEMY AND DIFFERENTIATION

IN THE

LEXICON

87

Polysemy and Differentiation in the Lexicon Verbs of Physical Contact in Swedish Åke Viberg Department of Linguistics, Lund University

1.

Introduction: The lexical-semantic organization of verbs

This study is part of an ongoing investigation of the semantic structure of the verb lexicon in Swedish seen from a crosslinguistic perspective (Viberg 1981, 1983, 1996). As a point of departure for the analysis, verbs are classified into semantic fields on the basis of their prototypical meaning. Table 1 shows a classification of the 100 most frequent Swedish verbs. (The numbers show the rank when the verbs are ordered according to descending frequency.) The fields have been grouped into three larger classes. The class termed “Concrete (physical action) verbs” comprises fields such as Motion (‘go’, ‘put’) and Production (‘make’, ‘build’), which typically refer to situations immediately accessible to the sense organs. Most of these verbs denote canonical actions in the sense that they take an Agent surfacing as subject in the unmarked case. Mental verbs comprising fields such as Cognition and Perception describe psychological processes of various types which can be subjectively experienced by oneself but not directly observed in others. Typically, one of the arguments is an Experiencer. Very often, there are pairs of verbs which contrast primarily with respect to base-selection (Viberg 1983; cf. flip-flop, psych-movement etc.), e.g. Experiencer-based: I could hear that Peter was happy vs. Source (Phenomenon)-based: Peter sounded happy (to me). The last class, Grammatical verbs, includes various groups of verbs with meanings that tend to be grammaticalized in many languages. They express

88

ÅKE VIBERG

dynamic (aspectual, causative) or modal meanings. The classification is based primarily on meaning without regard to morphosyntactic characteristics. Table 1. The 100 most frequent verbs in Swedish classified into semantic fields GRAMMATICAL VERBS GENERAL DYNAMIC

ASPECTUAL

CAUSAL

MODAL

MODALITY

1 vara ‘be’ 6 bli ‘become’ 41 använda ‘use’

23 börja ‘begin’ 51 fortsätta ‘continue’ 89 bruka HABITUAL

26 låta ‘let’ 59 leda ‘lead (to)’ 71 bero ‘depend’ 95 tvinga ‘force’

3 kunna ‘can’ 4 ska ‘shall’ 15 måste ‘must’ 19 böra ‘ought’ 35 behöva ‘need’

49 försöka ‘try’ 54 lyckas ‘succeed’ 85 töras ‘dare’ 100hinna ‘be in time to’

“CONCRETE ACTIONS” POSTURE

MOTION

POSSESSION

EXISTENCE & PRODUCTION

Subject-centered Object-centered 17 stå ‘stand’ 21 ligga ‘lie’ 61 sitta ‘sit’

7 komma ‘come’ 12 gå ‘go’ 34 följa ‘follow; accompany’ 50 lämna ‘leave’ 70 hoppa ‘jump’ 72 nå ‘reach’

37 sätta 2 ha ‘put=set’ ‘have’ 38 ställa 5 få ‘put=stand’ ‘get’ 42 lägga 10 ta ‘put=lay’ ‘take’

göra ‘make’ 9 finnas ‘there is’ 39 bygga ‘build’

47 dra ‘pull’ 64 föra ‘lead’ 80 bära ‘carry’ 92 samla ‘gather’

58 ske ‘happen’ 62 skapa ‘create’ 77 hända “happen’ 88 bestå ‘last’ 90 förekomma ‘occur’

13 ge ‘give’ 81 köpa ‘buy’ 87 sakna ‘lack’ 94 äga ‘own’

8

POLYSEMY AND DIFFERENTIATION

IN THE

ORGANIC LIFE

QUANTITY

PHYSICAL CONTACT

46 leva ‘live, be alive’ 86 växa ‘grow’

45 öka ‘increase’

53 slå 22 hålla ‘strike/hit/ ‘hold’ beat’ 69 röra ‘touch’ (‘move’)

LEXICON

89

MANIPULATION

MENTAL VERBS METALINGUISTIC

VERBAL COMMUNIC.

20 gälla ‘apply; be valid’ 40 kalla ‘call’ 44 betyda ‘mean=signify’ 57 innebära ‘mean=imply’

11 säga ‘say’

14 se ‘see’

28 tala ‘speak’ 29 skriva ‘write’ 65 berätta ‘tell=narrate’

18 visa ‘show’ 32 höra ‘hear’ 33 finna ‘find’

73 heta ‘be called’

74 nämna ‘mention’ 78 kräva ‘demand’ 79 svara ‘answer’

48 söka ‘look for’ 76 verka ‘seem’ 83 betrakta ‘look at, regard’ 93 förefalla ‘seem’

98 fråga ‘ask’

PERCEPTION

COGNITION DESIRE

24 tycka 16 vilja ‘think=be ‘want’ of the opinion’ 25 anse ‘consider’ OTHER 27 veta MENTAL ‘know’ 30 känna 68 välja ‘know’; ‘choose’ ‘feel’ 31 tro 91 uppleva ‘think=believe’ ‘experience’ 36 tänka 99 intressera ‘think=reflect’ ‘interest’ 52 räkna ‘count’ 55 lära ‘learn; teach’ 63 läsa ‘read’ 66 förklara ‘explain, declare’ 67 mena ‘mean=think’ 84 förstå ‘understand’

OTHER FIELDS: 43 spela ‘play’; 56 vänta ‘wait’; 60 arbeta ‘work’; 75 utgöra ‘constitute’; 82 möta ‘meet’; 96 fylla ‘fill’; 97 omfatta ‘comprise’

90

ÅKE VIBERG

The organization of the lexicon can be approached within either a componential or a relational framework. The present study attempts to combine various strands from both approaches. From a componential point of view, the internal structure of a semantic field may be looked upon as the outcome of the interaction of a set of more or less field-specific components and a number of general field-independent components that cut across all verbal semantic fields. To take just one example, verbs of Perception are organized according to field-dependent components such as the sense modalities (see/hear/touch etc.) and the field-independent dynamic system comprising lexical aspect (e.g. stative: see; dynamic: look) and causative distinctions (be visible vs. show ‘make visible’). The distinctions within the dynamic system are relevant within all verbal semantic fields. Certain components play a central role in the structuring of a field, while others represent more peripheral modulations. Following Miller and JohnsonLaird (1976), verbal semantic fields are organized around a core predicate, e.g., Motion verbs are organized around TRAVEL(x) and Possession verbs around POSSESS(x,y). In this paper, I will suggest that Verbs of Physical contact such as hit, strike, beat, punch, knock, bump into, touch, rub etc. are organized around the core predicate CONTACT(x,y). Lexical relations such as hyponomy, meronomy, antonomy and synonymy form the cornerstone of the relational approach (Miller 1993). Due to their generality, such relations are very useful as organizing principles and provide a firm skeleton in the construction of a structured lexicon due to their relatively clear applicability. However, to continue the metaphor, the flesh and blood of the semantic system are provided by more substantial concepts based on our everyday understanding of biology, physics and psychology, as will be demonstrated below.

2.

Physical contact verbs: An overview

We will turn now to the internal structure of one verbal semantic field, the verbs of Physical contact, and the patterns of polysemy that are characteristic of the individual verbs belonging to the field, in particular the most basic and frequent ones, which show the most varied possibilities. In Table 2, the verbs belonging to this field in Swedish are listed, grouped into broadly defined subfields.

POLYSEMY AND DIFFERENTIATION

IN THE

LEXICON

91

Table 2. Physical contact verbs in Swedish. A simple taxonomy.

MOTION röra

PHYSICAL CONTACT Nuclear verb slå ‘strike/hit/beat’ träffa ‘hit a target’

Basic level:

Subfields: Various Bodygroups part krocka ‘collide’ kollidera ramma törna emot knuffa puffa skuffa dänga knäppa pricka påta sticka

‘move;touch’

stryka ‘stroke’

vidröra ‘touch (lightly)’

Tool

Soundsource

Punishment/ Battery

Moving contact

Light contact

HAND smocka

klubba

‘punch’ klappa knocka

hamra piska trumma

banka ‘bang’ bulta dunka daska

prygla ‘flog’ klå

gnida ‘rub’ gnugga gno skrubba

nudda ‘brush snudda against’

massera frottera skrapa smeka kittla klia skava

ta på fingra på tumma på beröra tangera

/klippa till ‘club’

FOOT sparka ‘kick’ trampa stampa krama slicka skalla peta bita nafsa stånga klösa kyssa

spöa risa

knacka smiska smälla gissla klatscha hudflänga örfila kindpusta

ARMS TONGUE HEAD FINGER TEETH TEETH HORN CLAWS LIPS

toucha komma åt

92

ÅKE VIBERG

The hierarchical structure of the field is indicated in a schematic way. A small number of verbs form a basic level. At the level below this, we find various groups of hyponyms forming subfields, the members of which contrast along one or more dimensions. A subfield such as Body-part contains verbs specifying the body-part with which the contact is effected (cf. slap, punch, kick, lick), Sound-source specifies the sound produced from the contact (cf. bang, thump) and Moving contact specifies motion along something simultaneously with a contact which is hard enough to have some effect on the surface (cf. stroke, rub, scrape, graze, scratch, tickle). As a first example, I consider stryka ‘stroke; rub’, which is the most basic verb within the subfield Moving contact: Maria strök lille Peter över håret.

Maria (z) stroke little Peter’s (w) hair(y). (lit. ‘M stroke P over the hair’).

For certain purposes, it will be illuminating to use a formalism based on Miller and Johnson-Laird (1976) to represent the meanings, even if no attempt will be made to give a completely explicit account of the semantic representations discussed in this paper. i. ACT(z,S) ii. CAUSE (S, (ALONG(TRAVEL))(x,y) & CONTACT (x,y))) iii. INTEND(z, SHOW(z,w,AFFECTION)) The representation above states that z performs an act S, which causes x to travel along y at the same time as there is contact between x and y. What travels along Peter’s hair in this example is not expressed explicitly but is normally understood to be Maria’s hand: HAND(x) is thus the default interpretation. Most verbs of physical contact also have an attitudinal component like (iii), when both the Actor and Patient are human beings. The semantic representation is linked to a specific syntactic frame: Syntactic frame: NPz ___ NP w prep NP y (med ‘with’ NPx) default: HUMAN(z), HUMAN(w); BODY-PART(x), BODY-PART(y) The hyponyms of stryka ‘stroke’ are related to various body-parts such as SKIN(y) for smeka ‘caress’ and kittla ‘tickle’ and skava ‘abrade’. These verbs contrast primarily with respect to purpose or result: ‘show affection’, ‘make laugh’ and ‘hurt’, respectively. The differentiation between the verbs belonging to the subfield Moving contact will be discussed further in Section 8.2.

POLYSEMY AND DIFFERENTIATION

IN THE

LEXICON

93

The verbal semantic fields tend to be organized around one or at most a few nuclear verbs (Viberg 1993), which are dominant within their fields with respect to frequency of occurrence, the number of secondary senses and the range of constructions they can enter into. The nuclear verbs include the core component of the field but in addition contain some more specific components. They represent the most typical verbs within their fields rather than being the direct exponents of the general meaning shared by all members of the field. Such exponents, when they exist, usually appear only in formal or specialized registers. For example, the nuclear perception verb is see, whereas perceive is more or less a technical term. Within the field Physical contact, slå ‘strike/hit/beat’ is the nuclear verb in Swedish. The verb is polysemous to an unusually high degree, but its use in examples as the following can be regarded as prototypical since the majority of the other uses can in a natural and systematic way be accounted for as extensions from the semantic representation underlying this use. In addition, this prototypical meaning reaches a high frequency of occurrence in the corpus data (see below). Per slog Pål i ansiktet.

Per hit Pål in the face.

The lexical representation of slå in this use is shown in Table 3. (The semantic representation is simplified in certain respects. Most notably, the predicate CAUSE has been left implicit a number of times.) The meaning of the verb can be understood as a series of events at different levels. The first is the Table 3. Lexical representation of slå ‘strike/hit’ in its prototypical use. Syntactic frame:

NPz ___ NPw (prep NPy) (med NPx)

Example:

Per slog Pål (i ansiktet) (med knytnäven/en käpp) ‘Per(z) hit Pål(w) in the face(y) with his fist(x)/a stick(x)’

Semantic representation: Default: HUMAN(z); HUMAN(w); HAND(x) or HOLD(z,x); BODYPART(y) (i) (ii) (iii) (iv) (v) (vi)

ACT(z,S) STRETCH(ARM) ((FAST)TRAVEL)(x) HAPPEN(CONTACT(x,y)) FORCE(x, y, d1) INTEND(z, DEFEAT or HURT(z, w))

z performs an act S the arm is stretched x travels fast x and y become contiguous A force with strength d1 is directed tow. y z intends to defeat or hurt w

94

ÅKE VIBERG

Intentional level. In its prototypical use, slå is a goal-directed action represented by the predicate ACT, which generally appears in most verbal semantic fields. In less prototypical uses, the intentional meaning is missing, as in Per slog huvudet i taket ‘Per hit his head against the ceiling’. Similar to many basic motion verbs, slå also involves a characteristic Body movement with a characteristic ‘shape’. In certain uses, this part of the meaning is focused as in Per slog ut med armarna ‘Per spread his arms’. Characteristic motion patterns of the limbs constitute important organizing principles for motion verbs. Subject-centered motion verbs like walk, run, jump, in particular, refer to motions of the legs and feet, while many of the most basic objectcentered motion verbs such as put, throw, pull and push refer to characteristic motion patterns of the arm and hand. (See the analysis of Swedish dra ‘pull’, ‘draw’ in Viberg, 1996b). The characteristic motion pattern of the arm is indicated loosely as STRETCH(ARM) in Table 3, which needs further specification. The planning and control of every-day motion patterns such as lifting a cup to one’s mouth are extremely complex (Hollerbach 1990a,b). The most important aspect of the body motion from the point of view of slå as a physical contact verb, the concomitant motion of the hand, is indicated as TRAVEL(x). Motion is so central to the verbs of physical contact that most of them could be regarded as specialized types of motion verbs. Actually, the most general motion verb in Swedish, röra ‘move’, might be regarded as superordinate to many of the verbs of physical contact. This applies in particular to the subfield called Light contact in Table 2. Verbs of motion can be divided into subject-centered verbs of motion such as walk and run, which describe the displacement of the subject, and object-centered verbs of motion, such as throw and put, which describe the displacement of the object. Basically röra is an object-centered motion verb (often with a directional spatial particle such as ut ‘out’): Han rörde ut pannkakspulvret med kakao i vatten.

He stirred some water into the pancake mix and cocoa.

But in the reflexive form it is a subject-centered motion verb corresponding to intransitive ‘move’: Klockan hade en visare som rörde sig mycket långsamt,

The clock had a hand which moved very slowly,

In its uses as a verb of Physical contact, it usually refers to light contact (often in combination with the preposition vid ‘at, near’):

POLYSEMY AND DIFFERENTIATION Men Åke ville att de skulle röra så lite som möjligt vid tältet.

IN THE

LEXICON

95

But Åke wanted them to touch the tent as little as possible.

The most central part of the meaning of slå is the physical contact, in the prototypical case between the hand and some concrete object. From the perspective of spatial perception, Contact can be described in terms of the distance between x and y. In the ideal case, the perceptible distance between x and y is Zero. As will be demonstrated in Section 8.1, this constraint can be weakened to Proximity (very short distance) in certain uses of some Physical contact verbs. However, the contact between the moving hand and another object involves something more than spatial contiguity, such as an impact and a transmission of Force. Michotte (1946/1963) carried out a number of now classic experiments where subjects were shown moving colored spots and asked to describe what they saw. In one experiment, a black square moved towards a stationary red square. When it came into contact with the red square, it stopped moving and the red square started moving. Subjects reported that the black square bumped into the red square and launched it (Fr. lancer), gave it a push or set it in motion. This was called the launching effect (l’effect lancement) by Michotte. Leslie (1994) argues that Force presupposes the solidity constraint which is acquired by infants only a few months old. Solidity, which is central in naive mechanical reasoning, cannot be reduced to mere spatiotemporal patterning. There are visiospatial illusions such as the Pullfrich double pendulum illusion, where subjects wearing a special filter over one eye see solid rods passing through one another. The centrality of Force in the conceptual system underlying lexical semantics has been stressed by Johnson (1987) and Talmy (1988). Force can be quantified, a fact which is linguistically reflected in adverbial modifications such as strike hard, strike lightly and in the selection of synonyms or hyponyms of the nuclear verb: Om någon slår till mig så slår jag tillbaka, bara mycket hårdare.

If someone slaps your left cheek then you slap his right, only harder.

Hon slog honom lätt på handen.

She slapped his hand. (lit. struck ... lightly)

Och Hanna Frankenstein slog med kraft sin stora magra hand i kärrans räcke.

And Hanna Frankenstein banged her big skinny hand on the cart railing with great force.

The default interpretation of Swedish slå is that the force is relatively strong,

96

ÅKE VIBERG

but there is also a subfield of Physical contact verbs indicating Light contact such as nudda and snudda ‘touch lightly’, ‘brush against’: De nedhängande grenarna nuddade vattenytan.

The overhanging branches lightly touched the surface of the water.

Body movement has its own set of modifiers. Bodily activity in general requires Effort, which is also quantifiable: Med stor ansträngning slog Per hål på isen ‘With great effort, Per struck a hole in the ice’. Effort, which is primarily experienced as a physiological condition of our bodies, is also reflected in resultative adjectives like trött ‘tired’: Per slog sig trött på boxbollen Lit. ‘Per hit himself tired on the punching bag’. Like all motion, the motion of the arm and hand can be quantified with respect to Speed. The default for slå is fast motion. The most basic components and the associated quantifiable parameters that can be expressed by characteristic modifiers can be summed up briefly as follows: Experiential level

Concept

Quantifiable Parameter

Cognitive Sensorimotor Spatial perception

ACT(x,S) Body movement TRAVEL(x) CONTACT(x,y) FORCE(x,y,d1)

(Intention) Effort Speed Distance Strength

Mechanical reasoning

3.

Patterns of polysemy of slå

Slå is the most frequent Physical contact verb in Swedish, and it also belongs to the group of most frequent verbs in general. (Actually it has rank 53. See Table 1.) Like frequent verbs in general, it is polysemous to a very high degree. I.

English translation equivalents

The polysemy of slå is reflected in a striking way in its translation equivalents in English. Table 4 shows the English correspondences of Swedish slå in a corpus of bilingual texts consisting of Swedish and English originals with their respective translations. The corpus is being compiled by Aijmer, Altenberg and Johansson (1996). Most of the translated Swedish examples that will be

POLYSEMY AND DIFFERENTIATION

IN THE

LEXICON

97

presented in this paper are taken from the same corpus. (At the time of the preparation of the table, only part of the corpus was available. Primarily texts taken from novels have been used and, due to the relatively limited number of occurrences of slå, both Swedish and English originals have been included.) The English equivalents have been grouped into semantic fields. In total, there were 166 occurrences of slå. The English physical contact verbs strike, hit and beat, which intuitively appear to be the closest correspondences, are among the most frequent equivalents but with only a small margin in relation to other alternatives. Strike is the equivalent of slå in only 8% of the cases and hit and beat do not even reach this modest percentage. Totally, slå Table 4. Equivalents of slå in translations between Swedish and English Physical Contact

Motion

strike hit beat slam knock pound slap crash kick bang bash bump clap clasp drum hammer slop smack thump whip wipe

13/8% 10 8 5 3 3 3 2 2 1 1 1 1 1 1 1 1 1 1 1 1

turn throw settle fling come go put cover bring carry cast cross fasten flap fly gush jig-a-jig land pour rear walk wrap

TOTAL

61/37%

Other Fields 5 5 3 2 2 2 2 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1

37/22%

Posture sit

9

Open/Close open close

13 3

Disconnection break 3 cut 1 Odd ex. 3 Mental look (up) notice amaze consider please

3 1 1 1 1

Various fields take 2 get 2 switch 2 bet 2 work 2 Odd ex. 19 68/41%

166/100%

98

ÅKE VIBERG

corresponds to an English physical contact verb in 37% of the cases. In as many as 22% of the cases, the correspondence is a motion verb, which supports the claim made earlier that motion is a rather prominent aspect of the meaning of the verb. The cases where the translation equivalent belongs to some other field are quite impressive, 41%. This reflects the great proportion of cases where the meaning is extended outside the basic domain of the verb. (In the table, ‘odd ex.’ refers to the number of verbs with a single occurrence within the groups Disconnection and Various fields. This has been done only to save space.) II. The system of related meanings An important step in the analysis consists in establishing the relationships between the prototypical semantic representation and various extended meanings. This part of the analysis was originally based on an intensive textual study of the most frequent Physical contact verbs in “The Bank of Swedish” (Språkbanken, Dept. of Swedish, Gothenburg University). In particular, the corpus Novels 1980 was used (4 million running words from 60 novels, see Gellerstam 1992). A number of actual examples from this corpus can be found in Viberg (1984, 1992, 1994). In this paper many examples are taken from the bilingual corpus mentioned above (Aijmer et al. 1996). Invented examples are in general modeled on actual examples from the Bank of Swedish corpus. When all occurrences of a verb are studied in a corpus, a large number of specific meanings can be detected (intuitively, or with reference to synonyms or paraphrases, or by translating into another language), particularly if the verb has a high frequency of occurrence. To a great extent, these meanings appear to be related. An attempt will be made here to account for these patterns by providing links between the various specific meanings. Some of them involve transfers, which can best be motivated by invoking a number of metaphorical principles, but to a great extent the specific meanings form a continuous chain, where adjacent meanings shade into one another almost imperceptibly. In several respects, the analysis forms a parallel to text-based in-depth studies of individual words, such as the study of risk by Fillmore and Atkins (1992) and the analysis of ask by Rudzka-Ostyn (1989). In the textual study (based on 2354 actually occurring examples), a wide range of non-prototypical or secondary meanings was observed. The majority of these could be described as due to a modification of the representation of the prototypical meaning. Primarily, the modifications are of three types (often used in combinations): (1) Focusing, (2) Resultative strengthening and

POLYSEMY AND DIFFERENTIATION

IN THE

LEXICON

99

(3) Metaphor. Focusing centers the attention on some part of the prototypical meaning, whereas strengthening adds one or more meaning components. Both of these mechanisms are gradual in nature, and the meanings derived in these ways can often be arranged along a continuous scale with almost imperceptible steps. The relationships between the prototypical meaning of slå and a number of the major extended meanings are shown in Table 5.

Table 5. Major meanings of slå Stationary Motion Blommorna slog ut. The flowers came out.

Focusing:

Bodily motion

Social Interaction

Per slog ut med armarna. Per spread his arms.

Per slog Pål i schack. Per beat Pål at chess.

PROTOTYPE:

PHYSICAL CONTACT

Resultative strenghthening: Objectcentered Motion Per slog bollen över nät. Per hit the ball over the net.

Per slog Pål i magen. Per hit Pål in the stomach.

Disconnection

Sound source

Organic life

Postural

Per slog gräset. Per cut the grass.

Det slog i dörrarna. The doors slammed.

Björnen slog ett får. The bear got a lamb.

Per slog sig ner i soffan. Per sat down in the sofa.

Specializedmeanings: meanings: Specialized Open/ Close Per slog upp boken. Per opened the book.

Per slog upp ett ord. Per looked up a word.

Motion: Liquid Per slog upp en grogg. Per poured a drink.

Metaphor Metaphor Symbolic Klockan slog 12. The clock struck 12.

Per slog ihjäl tiden. Per killed time

.

100

ÅKE VIBERG

III. Focusing of various parts of the prototypical representation In certain uses, attention is drawn to some part of the prototypical meaning representation, while the rest is backgrounded or totally suppressed. This process will be referred to as focusing. In a way, one aspect of the meaning is also focused in the prototypical case, i.e. the establishment of physical contact. The term focusing, however, will be used when only part of the prototypical meaning is relevant. Focusing is preferred to other alternatives such as bleaching, since focusing does not necessarily mean that the rest of the representation disappears completely; on this point, there is a continuum. The term generalization cannot be applied either, since rather specialized and not very frequent uses are involved in many cases. There are a number of verbs that primarily describe the movement of a body-part, such as nod, wave and frown. In general, Bodily motion verbs are associated with more or less conventionalized implicature of some emotional reaction or some type of non-verbal communication. In certain uses of slå, the motion of the arm is focused, most clearly in an example such as Per slog ut med armarna ‘Per spread his arms’ (lit. struck out with the arms). Usually, this is interpreted as an expression of resignation or ignorance. One step further removed from this use is the use of slå as a stationary motion verb in combination with the particle ut ‘out’ in expressions such as Blommorna slog ut ‘The flowers came out’. This is a conventionalized use, but slå can also appear productively in this meaning as in the following example from the corpus: Stockholmarna märker det ofta först när främmande flaggor slår ut på Norrbro.

Stockholmers usually become aware of a state visit only when foreign flags fold out along Norrbro bridge.

The characteristic motion associated with slå can be represented schematically as an oblong object moving approximately 90 degrees in various directions, which are often indicated by spatial particles:

POLYSEMY AND DIFFERENTIATION

IN THE

LEXICON

101

There are in fact a variety of partly lexicalized phrases with slå in combination with various particles and body-part terms, where the motion of the limbs is focused and the physical contact is backgrounded or completely absent. In these cases, a motion verb is very often the closest equivalent in English in these uses: Hon sprattlade lite, slog med benen upp och ner som en fisk på land slår med stjärten.

She struggled a little, kicking her legs up and down the way a landed fish flaps its tail.

Marjorie slår handen för munnen.

Marjorie covers her mouth with her hand.

Hon for upp och sprang runt i köket, slog armarna runt kroppen, och hulkade och snyftade.

She leapt up and ran round the kitchen, flinging her arms round her body, sobbing and sniffing.

Och så fort Mattis fick syn på henne, rusade han fram och slog armarna om henne.

As soon as Matt caught sight of her, he rushed forward and threw his arms around her.

medan jag slår armarna om mig själv och låter hakan sjunka ner mot bröstet,

as I wrap my arms around myself and let my chin drop to my chest,

The bodily motion exemplified above probably serves as the model for the use of slå in expressions like slå in ett paket ‘wrap up a parcel’. The type of motion is the same as that involved in throwing an arm around someone: Så la han skjortan i vattnet och när han fått grepp på ålen igen slog han tyget om den.

Then he put the shirt down in the water and, once he got a hold on the eel again, he wrapped the material round it,

In combination with the particle in, slå can also be used as a subject-centered motion verb indicating a turn in a new direction: De lämnade den smala och sönderregnade vägen och slog in på en stig som ledde in bland granarna.

They turned off the narrow road, ruined by rain, and set off down the path that led through the trees.

The expression slå in has an extended use based on a variety of the Path metaphor: Events/Actions are Paths leading to a Goal. Erik var språkkunnig, musikalisk och begåvad, och med Machiavelli som sin läromästare slog han in på en djärv och aggressiv utrikespolitik.

Eric was a good linguist, musical and talented. With Machiavelli as his guide, he pursued a bold and aggressive foreign policy.

102

ÅKE VIBERG

Another extended sense based on the use of slå as a subject-centered motion verb is found in combination with the particle om. In the following sentences, slå indicates a change in some quality (compare English expressions like turn sour, which are based on the same spatial metaphor): Ljuset slår om till gult och Toyotan skjuter iväg

The lights turn to amber and the Toyota darts forward,

Det hade hastigt slagit om till töväder igen,

It had just started to thaw again

Motion is not the only component that can be focused. Even the implied social intention behind the act of hitting can be singled out and brought to our attention when the object is human. The act of hitting another human being is usually aggressive and carried out in order to hurt or in an attempt to defeat another person. This association is so strong that it was included as one part of the representations of the prototypical meaning of slå in Table 5. This aspect of the meaning of slå is focused in the following example: Genom en rad glänsande aktioner slog Karl XII ryssarna vid Narva år 1700 och polackerna vid Klissow år 1702.

In a series of brilliant actions Charles XII beat the Russians at Narva in 1700 and the Poles at Kliszow in 1702.

The physical contact can be backgrounded to various degrees or completely absent from the interpretation. In examples like Tyson slog ut sin motståndare ‘Tyson knocked out his opponent’, it is still part of the meaning, although ‘defeat’ is the meaning component that is focused. In other examples, no physical contact is involved as in: Per slog ut Pål ur schackturneringen ‘Per beat Pål out of the chess tournament’ or Per slog Pål i schack ‘Per beat Pål at chess’.

4.

Non-prototypical subjects and the subject cline

The choice of subject affects the range of interpretations to a high degree. In the prototypical case, the subject of slå is human. There are, however, a number of cases where slå has a non-prototypical subject. Such subjects represent a continuous departure from the prototype. It is possible to account for many of the semantic modifications by making reference to a hierarchy which can be called the subject cline, as is shown in Table 6.

POLYSEMY AND DIFFERENTIATION

IN THE

103

LEXICON

Table 6. Subject cline (for Contact verbs) Human (instru-ment) (body-part) (tool)

Objects with inherent energy source clock heart (projectiles)

Natural force

Sense impression

wind rain wave

light sound smell

Other physical objects prototypical (literal) intention Responsibility (Self-propelled) Motion Concretely perceptible to the senses

Thought, emotion

metaphorical

With some modifications, this hierarchy is valid for many of the concrete action verbs. At the top, semantic classes of subject appear, and at the bottom characterisitc features are listed. Among these, Intention and Responsibility are relevant only for human subjects. Another important characteristic of human subjects, which is acquired early by infants, is self-propelled motion (Spelke et al. 1995). There is also a decrease in concreteness, defined as accessibility to the sense organs. The reason why sense-impressions are ranked rather low, which at first sight might appear odd, is that such impressions are only available to one sense. Physical objects (and persons), which are highest in concreteness, can be both seen and touched and in principle even perceived by the other senses. Having a definite shape is perhaps the most crucial criterion for concreteness. In some examples, a body-part functions as the syntactic subject of slå: I det djupa mörkret strax efter midnatt slog en hand varligt på Adams sovrumsfönster.

In the deep darkness soon after midnight, a hand gently hit against Adam’s bedroom window.

Syntactically, this example deviates from the prototypical syntactic frame by having the instrument hand in subject position. Semantically, however, an unspecified human agent is present, who is interpreted as having the intention to attract attention. In the following examples, the body movements appear to be increasingly uncontrolled:

104

ÅKE VIBERG Och sen en stark rörelse som av en arm som slog ifrån sig.

Then a strong movement like an arm striking out.

Näsorna och pannorna slår ihop och ben och armar börjar plötsligt fäkta i en absurd pantomim.

Their noses and foreheads bump against each other; there is a sudden flurry of limbs, an absurd pantomime struggle.

There are certain physical objects that are ascribed an inner source of energy and which, as a consequence, are capable of total or partial self-propelled motion. The best example of such an object associated with the verb slå is ‘clock’: Klockan slog ‘The clock struck’. An inner organ such as the heart is also generally conceptualized as having an inner source of energy: Hjärtat slog ‘The heart was beating’. Launched objects such as projectiles of various sorts are probably experienced as self-propelled once they have been set in motion: En iransk robot slog på torsdagen åter ner i Iraks huvudstad Bagdad.

Thursday, an Iranian robot once again struck down in the Iraqi capital Bagdad.

This is not the case for certain other types of physical objects as in the following example, where the leaves are obviously moved by the wind: /---/ och därutanför höll hösten på att installera sig med flygande löv, som emellanåt slog mot fönsterrutorna med små dunsar och smällar, och med ljudet av en allt starkare blåst.

/---/ the autumn was establishing itself in flying leaves that sometimes hit the windowpanes with small thuds and bangs, and in the sound of a rising wind.

Examples like the above are somewhat problematic to place in the hierarchy. The concrete physical objects are in motion but not self-propelled, which should place them to the right of natural forces, whose motion is self-propelled. At the same time, however, natural forces are less tangible and less concretely perceptible than physical objects. In our experience, the motion of natural forces such as rain, wind, waves and lightning is caused by an inherent force (self-propelled motion): Grått regn slår mot glas.

Grey rain batters the glass.

Till synkoperat ackompanjemang av kluckande och suckande vågor som slår mot makligt guppande skrov.

Accompanying syncopation of sucking and splashing waves slopping against our supine hulls.

Eller blixten kan slå ner i det elektriska systemet.

[Someone can drop the aquarium,] or lightning strike the electrical system.

POLYSEMY AND DIFFERENTIATION

IN THE

LEXICON

105

Closely related to natural forces are sense-impressions such as light/darkness, sound, heat/cold and smell. Ligga i puffen och lyssna till tystnaden. Se ljuset från gatlyktan slå dagrar och skuggor mot taket.

Lying on the little couch listening to the silence, seeing the beam from the streetlamp throwing light and shade on the ceiling.

Håkans snarkningar slog emot mig i trappan.

I was confronted with Håkan’s snoring in the staircase. [Literally: H’s snoring hit me..]

Hettan slog emot honom som en fast materia.

The heat slammed into him like something solid.

Lukten som slog upp kunde tagit kål på en gödselstack, vad styrkan beträffade.

A smell fumed back, enough to suffocate a sewage farm.

All these examples describe some type of physical contact and share the component TRAVEL, the core component HAPPEN CONTACT and perhaps even FORCE as in the prototypical case. (The intensity of light and smell etc. represents a variety of force that has a concrete, physical basis.) There are also a number of subjects which are strictly Mental and cannot be concretely perceived by the senses. A clear case is Tanken slog honom att han kunde ha fel ‘The thought struck him that he could be wrong’. Examples such as this one are based on a metaphorical principle such as “an idea is a force” (cf. Lakoff & Johnson 1980). In general, the subject is a that-clause expressing the thought as a proposition. Like most clausal subjects, it is usually extraposed with a dummy subject det ‘it’: Det slog honom att det var inte så många vanliga människor han kände.

It struck him that he didn’t know very many ordinary people

Åter slog det mig att en del av de människor som kom in i vagnen för att önska mig lycka till inte ens var födda när jag redan hade blivit stjärna.

It hit me again that some of the people who walked in the trailer to wish me good luck had not even been born when I was already a star.

Det slog mig att det var mycket länge sedan jag känt mig generad.

It occurred to me it had been quite a while since l ‘d felt embarrassment.

An alternative to using an extraposed subject clause is to passivize slå (signalled with the ending -s in Swedish):

106

ÅKE VIBERG Man slås ju av att Slottet klarar sig med så relativt liten personal och att många finns i flera olika funktioner.

An outsider marvels at the fact that the Palace administration can cope with its relatively small staff and that many on the team have double functions.

Medan han berättade om Tessie slogs han av möjligheten att den där Volvon ute på Nynäsvägen hade kunnat vänta in honom.

As he was telling her about Tessie, Carl considered the notion that the police Volvo out on Nynäsvägen might have been waiting for him.

Even if examples like these can naturally be regarded as metaphorical, the step is actually rather short from some types of sense-impressions or emotional impressions to purely mental phenomena such as thoughts: Deras ovilja slog emot honom som en kall fläkt,

5.

Their animosity struck him like a cold breeze

Resultative strengthening

Concrete action verbs are closely related to various types of result, which often represent the goal of the human agent. Such results represent a special variety of inferences or conversational implicatures, which can be gradually conventionalized and lexicalized. For example, in a sentence such as Per slog bollen över nät ‘Per hit the ball over the net’, the contact results in the motion of the object. Actually, resultative strengthening turns the basic meaning of the verb into a kind of manner component: ‘Per moved the ball over the net by hitting it’. Formally, this strengthening can be represented by the addition of various resultative components at the end of the representation of the prototypical meaning of slå in Table 3. Most of the cases that can be interpreted as resultative strengthening are motivated by a naive physics model (or folk biology and folk psychology, when the object is human). Physical contact (and the associated force) can result in (1) Motion, (2) The production of Sound, (3) Disconnection (‘Breaking’), to mention just some of the most important resulting effects. With a human object, we can add categories such as 1. Non-verbal communication, (2) Expression of Emotion and (3) Hurting or (4) Killing (cf. the Engl. cognate slay of slå). Meaning extensions of this type interact in a complex way with the syntactic frame of the verb and are often formally expressed by the addition of

POLYSEMY AND DIFFERENTIATION

IN THE

LEXICON

107

a wide variety of verbal particles in Swedish. Spatial particles are used in most cases when motion is involved. In addition, there are a number of combinations with more specific particles such as slå sönder ‘break (into pieces)’, one of the closest equivalents of English break and slå ihjäl (etymol. ‘into Hell’) ‘beat to death’, ‘kill’. An example of the function of the syntactic frame is the use of the formal subject det ‘it’, which is characteristic of verbs describing a sound-source as in Det knackar ‘There is a knock(at the door)’ . This syntactic frame can even be used with slå in examples such as: Det slog i dörrarna. (Lit. It struck in the doors.) ‘The doors slammed’. The selection of subject can also result in strengthening in some relatively frequent cases, for example,: Klockan slog. ‘The clock struck’ (Sound-source). The selection of object often affects the interpretation of the verb. If the object means ‘grass’ (or an area that is typically covered by grass such as a lawn or the sides of a ditch) the interpretation that naturally imposes itself is that a scythe is used as an instrument and that the grass is cut (see below). The major cases of resultative strengthening are shown immediately below the prototypical meaning in Table 5. In what follows, a number of illustrative examples from the corpus will be presented and commented on. I.

Object-centered motion

As mentioned above, the verb slå can be used as an object-centered motion verb. This use is particularly characteristic of newspaper reports of ball games such as soccer, where slå is often used instead of sparka ‘kick’: I exakt rätt ögonblick slog Staffan fram bollen till Magnus som slog ett härligt inlägg mot bortre stolpen.

At exactly the right moment Staffan hit the ball to Magnus who hit a magnificent cross center toward the far post.

If one simplifies the first part of the prototypical representation of slå in Table 3 somewhat, the first use of slå in the example above would have a representation like: i. ii. iii. iv.

ACT(z,S) CAUSE (S, TRAVEL(x)) & FOOT(x) CAUSE ((ii), HAPPEN(CONTACT(x,y)) CAUSE((iii), TRAVEL(y))

108

ÅKE VIBERG

The second use of slå in the example illustrates a kind of metonymic extension that is common in sport reporting. The object describes the result of the displacement of the ball in relation to strategies central to the ball game. Another example would be Magnus slog in ett tröstmål ‘Magnus scored (lit. hit in) a consolation goal’. In this use, slå functions as a creation verb with an effected object similar to the Swedish expression göra mål ‘score (lit. make) a goal’. In order to accommodate this meaning in the formal representation, something like the following component must be added: (v) CAUSE((iv), HAPPEN(EXIST(GOAL/CROSS CENTER...))) The use of slå as an object-centered motion verb can also be extended metaphorically. As mentioned above, mental phenomena can be interpretated as metaphorical forces, when they appear as subject. The use of mental phenomena in object position can be based on varieties of the metaphors ‘ideas are physical objects’ and ‘the mind is a container’. Och när han efter något halvår ryckt upp sig så mycket att han förstod på vilka grunder domen gick att överklaga hade advokaten kommit rusande till Säpo och slagit i honom att ett överklagande bara skulle leda till ännu hårdare straff

And when, after six months or so, he had pulled himself together and understood on what grounds he could appeal the conviction, the lawyer had gone rushing to the Security Police and then drummed it into him that appealing would only result in an even harsher sentence,

Parallel uses can be found with some of the hyponyms of slå such as banka ‘pound’: Arma människa, visst är det ett illa ansett och lönat värv att banka ABC i bondungar.

Poor creature, it really is an illreputed and badly paid duty to beat the ABC’s into country-kids.

-- Om ja så ska banka galenskaperna ur dej så ur dej ska dom, skrek han under tiden.

If I have to beat the foolishness out of you, then I will, he shouted meanwhile.

Ideas can also be conceptualized in other ways. In the Swedish version of the following example (from a Swedish novel), the idea is treated like an obtrusive insect. The Swedish form slå bort lit. ‘strike away’ is used to refer to the gesture made to fend off mosquitoes or flies. This gesture can sometimes be seen even when someone tries to relieve him/herself from a painful thought, and thus the expression is partly metonymically motivated:

POLYSEMY AND DIFFERENTIATION Han skulle just slå bort tanken

IN THE

LEXICON

109

[He] wanted to wipe away the thought

When an idea is out of mind, it will not disturb you any more. A reliable idea, on the other hand should be firm and immobile. This is the motivation behind a well-established use of slå. Together with the particle fast, slå can be used concretely in the meaning ‘fasten by hitting’: Per slog fast benet på stolen ‘Per hammered the leg on to the chair’. Actually, the most frequent use of slå fast is a variation on ‘an idea is an object ’. ‘Sanningen är krigets första offer’, slog USA-senatorn Hiram Johnson fast redan 1917

‘Truth is the first victim of war’, US senator Hiram Johnson declared as early as 1917

There are several specialized extended uses of slå as an object-centered motion verb. With a liquid as object, slå means ‘pour’: Han slog upp en rejäl slurk i tandborstglaset och kände den brännande värmen i magen.

He poured a goodly slug into his tooth-brush glass and felt the burning warmth down in his stomach.

The relationship of this use to the prototypical meaning is not completely obvious. This use appears, however, in contexts where the holding of a container in the hand can be inferred from the preceding context. The pouring is caused by the motion of the arm resulting in the tilting of the container. The ‘shape’ of the arm movement is related to that of striking. (The particle upp ‘up’, which often occurs in this use of slå refers to the rising of the liquid in a second container that is the goal of the displacement of the liquid.). The representation of this use of slå is given in Table 7. Table 7. The representation of slå meaning ‘pour’ Syntactic frame:

NPz ___ upp/i (NPx) (prep NPy)

Example:

Perz slog upp/i (kaffex) (i koppeny)

Presupposed:

HOLD(z, CONTAINER(w))

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

(ACT(z, S) CAUSE(S, TRAVEL(ARM)) (CAUSE((ii), TILT(w)) TO(FROM(TRAVEL)))(LIQUID(x),CONTAINER(w), CONTAINER(y )) CAUSE((iii), (iv))

Since the phrases referring to the liquid and to the containers are all optional in the syntactic frame, these elements must often be inferred from the context:

110

ÅKE VIBERG “Kaffe”, säger han och det låter som en order. Alla plockar fram koppar och Lindgren slår upp.

‘Coffee’, he says and it sounds like an order. They all produce cups and Lindgren pours.

Another specialized extended meaning is the expression slå ett telefonnummer ‘dial a telephone number’. Basically, this refers to the object-centered motion of the dial but what appears in the object slot is the resulting number (via a variety of metonomy): Det betydde att han måste ringa det där numret som han aldrig hade slagit och be henne om en tjänst.

It meant dialing that number he’d never used and asking her a favor.

II. Disconnection Verbs of Disconnection such as break and cut share a semantic component, which could roughly be paraphrased as ‘divide into pieces’. A more exact characterization can be based on notions such as connectedness and fitness for use as argued in Viberg (1985), where not CONNECTED(x) is suggested as the core of this field with the default inference notPOSSIBLE(USE(z,x)). Even if there are a relatively large number of disconnection verbs in Swedish, this meaning is often expressed by verbal particles such as sönder ‘broken’, ‘in(to) pieces’ and av ‘off’. These particles can be combined with a great variety of verbs. In a general inchoative sense, gå ‘go’ is used in the expression gå sönder as in Vasen gick sönder ‘The vase broke’ and Repet gick av ‘The rope broke’. The most general causative verb is slå, which has a weakened sense in expressions such as Peter slog sönder vasen ‘Peter broke the vase’. This expression is perfectly natural, even if the vase was dropped or knocked over by accident. However, an element of physical contact must be involved, as shown in the following paraphrase: Peter (unintentionally) caused the vase to hit something and break. If there is no element of physical contact, the more general causative ha ‘have’ must be used: Peter hade av repet ‘Peter broke the rope’ (e.g. by pulling too hard). Nu var han uppe och sjöng medan han lagade saker som han slagit sönder dagen innan.

He was up now and singing as he mended things he had broken the day before.

Torsten hade halva krattskaftet i handen. Det var splittrat och vasst i den brutna änden. Han måste ha slagit av det mot knät.

Torsten had half the rake handle in his hand, splintered and sharp at the broken end. He must have broken it across his knee.

POLYSEMY AND DIFFERENTIATION

IN THE

LEXICON

111

The selection of the object often affects the interpretation of the verb. If the object of slå means ‘grass’ (or an area that is typically covered by grass such as a lawn or the sides of a ditch), the interpretation that naturally imposes itself is that a scythe is used as an instrument and that the grass is cut. No particle is needed to obtain this interpretation. Vem är det som slagit ert hö, sa främlingen.

Who mows your hay?” asked the stranger.

Pentti var det slags typ som man nu, ännu år 1982, kunde komma på med att slå grönt gräs till sina trädgårdskaniner kring traktens vägrenar i vårkvällarna och föra hem det i en säck på pakethållarn.

Pentti was the type you might still come across, even in 1982, cutting fresh grass for his rabbits along the roadside verges in the spring evenings and taking it home in a sack on his carrier.

The lexical representation of this use of slå is shown in Table 8. Table 8. The lexical representation of ‘slå gräs’ (lit. beat grass) ‘cut the grass’ Syntactic frame:

NPz ___ NPy (med NPx)

Example:

Per slog gräset (med lie). ‘Per cut the grass (with a scythe)’.

HUMAN(z), GRASS(y), SCYTHE(x) (i) (ii) (iii) (iv)

ACT(z,S) TRAVEL(x) HAPPEN(CONTACT(x,y)) CAUSE((iii), HAPPEN(notCONNECTED(y)))

Resultative strengthening turns the basic meaning of the verb into a kind of manner component (‘cut by striking with a scythe’). The expression slå gräset contrasts with klippa gräset ‘cut the grass with a lawn-mover’. The extension from ‘strike’ to ‘cut’ is attested historically in a number of languages, for example, in French couper ‘cut (
He pulled into a Texaco station, parked beneath the overhang, and cut off the engine.

Macon slog på höger blinkers.

Macon switched his right blinker on.

112

ÅKE VIBERG

The disconnection (of the electric circuit) in the first example is metonymically related to the stopping of the running of the motor. If this is the correct interpretation, the use of slå in the second example is motivated by the principle of antonymy; antonymous expressions should use the same lexical base. An alternative interpretation would be to relate both expressions to the use of slå as a kind of motion verb as described in 5.1. In combination with the particle igenom ‘through’, slå is used primarily metaphorically to indicate the breaking of an (abstract) barrier: Hon slog igenom som artist ‘She had a break-through as an artist’. With abstract subjects, this use of slå + igenom is rather frequently used in factual writing: Sysselsättningens utveckling är emellertid väl så väsentlig, eftersom den åtminstone på sikt slår igenom i befolkningsförändringar.

The demographic picture is often used as an indicator, but employment is at least as important in that it ultimately affects the population, too.

Dessa föreställningar slår igenom likt fjärran tv-störningar.

These appearances come through like distant, disruptive interference on TV.

Hur starkt de sociala skillnaderna slår igenom kan påverkas av offentliga åtgärder,

The impact of social differences can be influenced by public measures,

III. Additional types of strengthening In certain uses, slå is extended into the field of Organic life comprising concepts related to life and death. The Swedish verb slå is a cognate of English slay. In modern Swedish, there are only a few expressions where slå has this meaning without using a special particle. An example is Björnen slog ett får. ‘The bear got (lit. struck) a lamb.’ The interpretation of such a sentence is that the bear killed the lamb. Usually, however, the meaning of ‘kill’ is expressed with the added particle ihjäl, which originally meant ‘into Hel’ (the kingdom of the dead in old Norse mythology from which the word hell is derived): De slog ihjäl säl med sina klubbor och fångade fisk med krok och nät.

[On the west coast there were fishing tribes] who killed seals with their clubs and caught fish with hooks and nets.

Man faller hela tiden utan att någonsin nå botten och slå ihjäl sig vilket vore en oskattbar lättnad.

One falls and falls without ever hitting bottom and killing oneself which would be an inestimable relief.

A metaphorical extension with a parallel in English is ‘kill time’:

POLYSEMY AND DIFFERENTIATION Den unge desertören /---/ hade utomordentligt trist. Han hade slagit ihjäl ett par timmar med korrespondens;

IN THE

LEXICON

113

The young deserter /---/ was utterly gloomy. He had killed a couple of hours with correspondence.

A very language-specific characteristic of slå is its rather frequent use as a postural verb meaning ‘sit down’. In this meaning, it is constructed with the reflexive pronoun sig in combination with the particle ner ‘down’: Före middagen slog vi oss ner i hennes gröna soffa.

Before dinner, we sat down on her green sofa, /--/

The motivation for this use is probably that one’s behind makes contact with the seat, but this is something that is not transparent in present-day Swedish, where the expression is completely conventionalized. There are some related uses where the contact is more clearly felt as in Fiskmåsarna slog ner på bryggan ‘The sea-gulls landed (‘struck down’) on the pier’. Postural verbs have a general tendency in various languages to extend their meaning from temporal to permanent location (‘live’, ‘settle’). This extension is also possible with Swedish slå sig ner: Svenska och finska nybyggare slog sig ner i kolonin, som kallades Nya Sverige.

6.

Swedes and Finns settled in the colony which received the name of New Sweden.

The role of the linguistic context in comprehension

So far the focus of this paper has been on the semantic links between various uses of slå. However, the large number of interpretations of slå present a problem for comprehension. To what extent is it possible to find clues for the various interpretations in written or spoken utterances containing slå? It turns out that even short sentences without context containing slå can in most cases be interpreted in a straightforward way. The most important clues are provided by the semantic classes of the NP:s (usually their head noun) in the syntactic frame of slå. Another important (and more obvious) factor is the large variety of verbal particles that can be combined with slå. To some extent even the choice of preposition in the PP complements provides information relevant to semantic interpretation. In very short sentences such as Per slog Pål, the interpretation is determined by default ‘Per hit Pål (physically)’. It is possible to change the interpretation by adding a complement such as i poker

114

ÅKE VIBERG

‘at poker’ Per slog Pål i poker ‘Per beat (defeated) Pål at poker’. The same interpretation can also be obtained via the preceding discourse or topic of conversation, but in the concordances that formed the basis for the original study, it was only occasionally necessary to go beyond the sentence containing the key word to determine the interpretation of the verb slå in an unambiguous way. In Table 9, the various kinds of factors are illustrated with short, idealized examples. The various classes of subjects shown in (1)-(6) have already been commented on. A large number of more narrowly defined classes of objects affect the interpretation in a radical way, as illustrated by (7)-(9). Examples (10)-(11) show that the semantic class of the object determines the interpretation even when there is a particle. We will return to these examples shortly. Among the prepositions, i ‘in’ and med ‘with’ allow several

Table 9. Factors affecting the interpretation of slå Subject

Swedish examples

Varying English equivalents

Human Natural force Mental Clock Heart Product of art

1) 2) 3) 4) 5) 6)

Per slog Pål. Regnet slog Pål (i ansiktet). Tanken slog Pål. Klockan slog. Hjärtat slog. Pjäsen slog.

Per hit Pål. The rain hit Pål in the face. The thought struck Pål. The clock struck. The heart was beating. The play was a hit.

Object

7) 8) 9)

Per slog gräsmattan. Per slog telefonnumret. Per slog rekordet.

Per mowed the lawn. Per dialed the number. Per beat the record.

Prepositional complement

Particle

10) Per slog upp dörren. 11) Per slog upp en grogg.

Per opened the door. Per poured a drink.

12) Per slog Pål i ansiktet. 13) Per slog Pål i poker.

Per hit Pål in the face. Per beat Pål at poker.

14) Per slog Pål med knytnäven. 15) Per slog Pål med 3-1. 16) Per slog Pål med häpnad.

Per hit Pål with his fist. Per beat Pål 3-1. Per struck Pål with amazement.

17) Per slog in boken. 18) Per slog upp boken.

Per wrapped up the book. Per opened the book.

POLYSEMY AND DIFFERENTIATION

IN THE

LEXICON

115

semantically delimitable classes of NP:s which affect the interpretation of slå. When a body-part appears after i as in (12), this is in general interpreted as the target of a physical blow, whereas the interpretation ‘defeat’ is obtained when the NP refers to a game as in (13). Examples (17) - (18) show one of the many cases where the interpretation is shifted by the choice of particle when other factors are kept constant. There is often a complex interaction between the verb, the particle and the semantic class of the object NP. This is illustrated very clearly in the translation corpus, where the interpretations of slå in such combinations are usually reflected by contrasting translation equivalents in English. One of the particles that appears in a number of combinations with contrasting meanings is upp ‘up’. In Swedish, this particle frequently has the extended meaning ‘become open’ after a number of verbs: kasta/dra/skjuta/sparka upp dörren ‘throw/ pull/ push/kick open the door’. The primary function of the verb is to describe the manner of opening. When upp is combined with slå, the expectation would be that the opening was more violent than normal. To some extent this prediction is borne out with an object such as ‘door’. The sentence Per slog upp dörren implies that the door was opened in a brisk manner, but the meaning of the verb is usually weakened, since the sentence does not imply that the door was opened by hitting it except in marked contexts, for example, when prosodic prominence appears on the verb instead of the particle. With other objects such as ‘book’, the manner component is often completely missing, and this is reflected — correctly — in the use of ‘open’ as a translation equivalent in examples such as the following: Sedan slog han upp boken och började demonstrativt läsa.

Then he opened the book and ostentatiously started reading.

The phrase slå upp is interpreted as ‘open’ with three semi-productive classes of objects meaning ‘door’(‘window’, ‘lid’...), ‘book’ (‘magazine’, ‘newspaper’, ‘file’...) and ‘eye’. Jag slår upp ögonen på morgonen och tänker: ‘Varför ska jag bry mig om att stiga upp?’”

I open my eyes in the morning and think, ‘Why bother getting up?’ ”

Together with the particle igen, slå can also be used to express the antonymous meaning ‘close’ with the same types of objects (except ‘eye’): DeVasher slog igen dossiern om McDeere och slog upp en annan, betydligt tjockare.

DeVasher closed the McDeere file and opened another, much thicker one.

116

ÅKE VIBERG Han slog ihop matsedeln och vinlistan,

He closed the menu and wine list,

When the object means ‘door’, slå preserves more of its basic meaning and implies that the closing is more violent than normal. In particular, it invites the inference that the closing results in a loud noise, which is reflected in the translation equivalent slam: “När går ni av skiftet ikväll?” frågade han i samma ögonblick som en av dem slog igen bildörren.

”When do you get off your shift?” he asked the one in the back as she slammed the car door.

When slå upp is combined with objects of the type ‘book’, the meaning can be extended further. It can refer to the turning of pages, and the object in this case refers to some section of a printed document: Hon slår genast upp Kvinnosidan,

She turns at once to the Women’s page,

The extension proceeds one step further when the object refers to the information that can be found in the document. A conventionalized expression such as slå upp ett ord ‘look up a word (in a dictionary)’ is based on metonymy. The following actual example illustrates the same meaning but contains the particle efter ‘after’ instead of upp: Han hade på försök slagit efter “Världens största mage”

He had tried looking up The Biggest Stomach in the World

Via these extensions, slå upp covers a large class of nouns in the uses related to the meaning ‘open’. The previous examples also show that a rather detailed representation of nouns including a great deal of ‘encyclopedic’ information is required for the comprehension process As already mentioned in 5.1, slå upp can also be used with the meaning ‘pour’ when it appears in contexts with nouns referring to liquids and containers. However, in this use it is rather common that the crucial nouns do not appear in the same sentence but in the wider context. This is clearly an exception with slå, but the phenomenon is of great interest since it shows that it is necessary to extend the analysis beyond the scope of the individual sentence. Lindgren skrattar ett låtsasskratt och drar fram en flaska Kronvodka ur ryggsäcken.—Om du inte är snäll, så får du inte smaka, säger han och slår upp åt sig själv innan han låter

Lindgren laughs an affected laugh and produces a bottle of Crown vodka from his rucksack. –If you’re not nice, you won’t get any, he says and pours some for himself before he lets the bottle

POLYSEMY AND DIFFERENTIATION flaskan gå runt samma väg som TVkannan. Redan innan dom fått ser Kurt och Göran glada och nöjda ut.—Ta ordentligt nu, säger Lindgren när Mats slår i. Som för att ge sin välsignelse.

IN THE

LEXICON

117

go around the same way as the thermos. Even before they’ve been served, Kurt and Göran look content. –Take a good slug now, Lindgren says when Mats pours. As if he wanted to give his blessing.

In spite of this example showing the importance of the wider context, the major conclusion from the study of slå is that the local linguistic context makes it possible to determine the interpretation in the majority of cases. With some other verbs, the situation seems to be different. To what extent it is possible to use linguistic clues for the interpretation is an empirical question that must be determined from case to case.

7.

Sound source

In Swedish, there are a large number of Sound-source verbs (Viberg 1978) such as susa ‘sigh’, ’sough’, surra ‘buzz’, mullra ‘rumble’. These verbs have a characteristic syntactic frame: det ___ prep NPy, where the NP refers to the source of the sound: Det susar i träden ‘The wind is sighing (lit. it is sighing) in the trees’, Det surrar i kupan There is a buzzing in the hive.). This frame is characteristic not only of sounds but of sense impressions in general. Semantically, the sound-source verbs are distinguished by the shared component: (FROM(TRAVEL))(SOUND, y). There is one group of verbs that could best be characterized as a combination of a physical contact verb and a sound-source verb (see Table 2 for a list). They are all frequentative, describing a series of repeated events acoustically and physically. One of these verbs dunka ‘bang, thump’ will be examined a little more closely. It can appear in the following syntactic frame which is highly characteristic of slå: Per dunkade Pål i ryggen ‘Per slapped Pål on the back (to cheer him up)’. As with slå, a certain emotional attitude is strongly implicated when the verb appears in this frame. The most characteristic parts of the lexical representation are shown in Table 10. In spite of the fact that dunka does not occur with very high frequency, it has a number of extended senses, which are displayed in Table 11. The idealized examples are based on actual examples from novels. The identification of the prototypical meaning is problematic. On the one hand, the most straightforward derivation of extended senses is obtained if the use as a

118

ÅKE VIBERG

Table 10. The lexical representation of dunka ‘bang’ ‘thump’ Syntactic frame:

NPz ___ NPw (prep NPy) (med NPx)

Example:

Per(z) dunkade Pål(w) i ryggen(y) med handen(x) Per (x) slapped Pål(w) on the back(y) with his hand(x)

Semantic representation: (i) ACT(z,S) (ii) TRAVEL(x) (iii) HAPPEN(CONTACT(x,y)) (iv) (FROM(TRAVEL))(SOUND,y) (v) INTEND(z, ENCOURAGE (z,w))

physical contact verb is considered to be prototypical, as has been done in Table 11. On the other hand, the most characteristic feature of all uses is the notion of intense repetitiveness which points to the use of dunka as a Sound source verb as the prototypical one. In the following example, dunka is used as a pure sound-source verb, which means that component (iv) in the semantic representation (Table 10) is focused. Den natten vaknade jag av att det dunkade häftigt i väggen.

That night I awoke from a fierce thumping on the wall. lit. it thumped

In this example (see Table 11), dunka is combined with an impersonal subject det ‘it’ in the syntactic frame that is characteristic of sense-impression verbs. Dunka can also be used as a motion verb, an extension that is very characteristic of Sound-source verbs in general: Tåget dunkade oförtrutet fram.

The train chugged indefatigably along.

The verb dunka can also be used with reference to bodily perception such as the pulse or heart beat. The notions of repetitiveness and high intensity are prominent even in this use. Han kände hur hat och adrenalin bubblade upp inom honom så att pulsen plötsligt började dunka i hela kroppen.

He could feel hate and adrenalin bubbling up inside him, so that his pulse started pounding.

Even in this case, the verb can be used in the impersonal construction characteristic of sense-impressions:

POLYSEMY AND DIFFERENTIATION

IN THE

LEXICON

119

Table 11. The pattern of polysemy of the verb dunka ‘bang, thump’ Subject-centered motion

Mental

Metonymy:

Metaphor:

Sound Source

Bodily Perception

Tåget dunkade fram genom dalen. The train chugged along through the valley.

Orden/Skuldkänslorna dunkade i Pers huvud. lit. The words/feelings of remorse were throbbing in Per’s head.

Det dunkade i väggen. There was a banging on the wall.

Det dunkade i tinningarna. My temples were throbbing.

Focusing: Subject: Engine

Subject: Inner organ

Maskinen dunkade. The machine was pounding.

Hjärtat/Blodet dunkade. The heart/blood was pounding.

Non-human subject: PROTOTYPE:

Strengthening:

Physical contact + Sound Source

Per dunkade näven i bordet. Per banged his fist on the table.

Nonverbal Communication

Per dunkade Pål i ryggen. Per slapped Pål on the back.

Det dunkade vid tinningarna och jag tyckte att jag inte kunde se klart,

Production

Per dunkade fram jazz på pianot. Per produced jazz by pounding away at the piano.

My temples were throbbing and I didn’t seem able to see clearly, lit. it throbbed

Closely related to this use are cases where the subject is a mental phenomenon. The following examples can be regarded both as a kind of animation of ideas or emotions (a type of metaphor) and as an association between the mental activity and the resulting physiological reaction (a type of metonymy): Jag kände hur skuldkänslorna dunkade på mej därinne.

I could feel the guilty feelings pounding at me in there.

120

ÅKE VIBERG Våld våld våld. Ordet dunkade i hennes värkande huvud.

Violence, violence, violence. The word was pounding in her aching head.

As a nuclear verb in the field of Physical contact, slå has extended meanings characteristic of the verbs within most of the subfields. The act of hitting often results in the production of sound, and in certain expressions this sound is realized as the object of slå as in slå takten ‘beat time’. In the following construction, where dörr ‘door’ is combined with the preposition i ‘in’, the production of sound is focused: Per slog i dörrarna.

Per slammed the doors.

Slå can even be used as a pure sound-source verb in the characteristic frame with an impersonal subject: Det slog i dörrarna.

The doors were slamming.

The production of sound is also central when ‘clock’ is used as a subject. In the following example, the clock is striking in a physical sense: Klockan under glaskupan slog tolv snabba slag och herdinnan dansade för sin herde.

The clock under its glass dome struck twelve sharp notes and the shepherdess danced for her shepherd.

In this example, each successive note has a symbolic function to indicate an hour of the day. Often expressions such as klockan hade redan slagit ett ‘the clock had already struck one’ are used simply to indicate the hour of the day even in the absence of any physical clock. The extension goes even further in expressions such as Klockan hade slagit midnatt ‘The clock had struck midnight’, Spöktimmen har redan slagit ‘The ghostly hour has already struck’ Snart slår befrielsens timme ‘Soon the hour of deliverance will strike’. In these examples, only the superimposed symbolic meaning is present. A similar case will be discussed in 8.1 below.

8.

Stroking and rubbing (Moving contact)

I. The polysemy of stryka ‘stroke’ The verbs in the subfield of Moving contact (cf Table 2) can all be regarded as hyponyms of stryka approx. ‘stroke’ (with many language-specific characteristics). The central part of the meaning (ii, below) comprises one motion and

POLYSEMY AND DIFFERENTIATION

IN THE

LEXICON

121

one contact component: i. ACT(z,S) ii. CAUSE (S, (ALONG(TRAVEL))(x,y) & CONTACT (x,y))) iii. INTEND(z, SHOW(z,w,AFFECTION)) Example: Maria strök lille Peter över håret. ‘Maria (z) stroke little Peter (w) over his hair(y)’. What travels along Peter’s hair in this example is not expressed explicitly but is normally understood to be Maria’s hand: HAND(x) is thus the default interpretation. Like so many other verbs of physical contact, stryka also has an attitudinal component (iii), which is actualized in the following syntactic frame involving a relation between two human beings: NPz ___ NPw prep NPy (med ‘with’ NPx) [HUMAN:z,w;BODY-PART:x,y] Hon lutade sig ner och strök honom She bent and stroked his head. över huvudet. Literally: She ... stroked him over head-the

The following examples are closely related to the prototype but lack some of its semantic and/or syntactic characteristics: Han högg tag i hennes ben igen och strök knäet.

He grabbed her leg again and rubbed her knee.

Han strök sig över kinden och lät blicken fara ut genom fönstret och in i den vita vårhimlen.

He ran his hand over his cheek, letting his gaze wander out of the window and into the white spring sky.

As shown schematically in Table 12, the meaning of stryka can be extended in several directions. In some uses, motion along something is focused, and the contact component is backgrounded or weakened into proximity rather than contact: Katten strök längs väggen.

The cat stroked along the wall.

Svalorna stryker över hustaken om sommaren.

The swallows sweep over the roofs in the summer time.

Segelbåten strök längs stranden.

The sailing boat skirted along the shore.

Stryka can also function as a subject-centered motion verb when it has a human subject, but in this case an evaluative component is added: ‘shady’ (cf. the compound landstrykare lit. ‘land stroker’, i.e. ‘tramp’). Vi stängde dörrarna rätt noga om det strök omkring någon välvillig granne.

We closed the door fairly carefully if there were any well-meaning neighbours prowling about.

122

ÅKE VIBERG

Table 12. Major classes of derived meanings of stryka ‘stroke’(‘rub’, ‘wipe’) Focusing +Evaluative strengthening:

Focusing

Subject-centered motion + Evaluation

Subject-centered motion

Peter strök längs gatorna. Peter prowled along the streets.

Båten strök längs stranden. The boat skirted along the coast.

PROTOTYPE:

Physical contact

Peter strök Mary över håret. Peter stroke Mary over the hair.

Strengthening: Application

Peter strök smör på brödet. Peter spread butter on his bread.

Removal

Peter strök svetten ur pannan. Peter wiped the sweat from his forehead.

Strengthening: Application + Symbolic

Peter strök under ordet. Peter underlined the word.

Focusing: Symbolic: ‘Important’

Peter strök under att han menade allvar. Peter emphasized that he meant business.

Application: Paint

Peter strök taket. Peter painted the ceiling.

POLYSEMY AND DIFFERENTIATION IN THE LEXICON Han hade strukit i gränden utanför Martinas hus.

123

He had been prowling around in the alley outside Martina’s house.

There are also many clear examples of of resultative strengthening, for example,: Han strök av handen på byxbaken,

He wiped his hand off on his pant-leg.

When certain types of nouns function as object, a specific instrument and a specific result is understood. A sentence such as Per strök taket means ‘Per painted the ceiling’, while Per strök sina byxor means ‘Per ironed his pants’. The instruments in these examples are BRUSH(x) and IRON(x), respectively. Any object that could naturally be interpreted as a surface that is conventionally painted or a cloth that is usually ironed will strongly suggest these specific meanings. The latter meaning can also be implied when there is no overt object: Det fattades smör, tvålen var slut, hon hade inte tänkt på att stryka.

There wasn’t any butter, we’d run out of soap, she had forgotten to do the ironing.

A further possibility is that the object refers to a word or an utterance, TEXT(y), for short. In this case, the instrument PEN(x) is understood: Peter strök under ordet (‘under the word’) ‘Peter underlined the word’ or Peter strök över (‘over’) ordet ‘Peter crossed out the word’. The semantic representation is shown in Table 13. The drawing (‘stroking’) of a pen over a piece of paper causes a line to exist. In addition, the line has a symbolic function, which shifts depending on its position with respect to the text. A line under the words signals importance, whereas a line through the words means that the text is not valid: Han fick fram en penna ur bröstfickan och satte igång att stryka under diverse meningar.

He slipped a pencil from his pocket and began underlining phrases.

Då skulle han ha strukit över vissa namn och satt in andra i stället och gjort anteckningar i marginalen.

He’d have scratched out some names, inserted others, and scrawled notes across the margins.

124

ÅKE VIBERG

Table 13. The lexical representation of ‘stryka’ in combination with TEXT Syntactic frame:

NPz ___ under/över NPy (med NPx)

HUMAN(z), TEXT(y), PEN(x) (i) (ii) (iii) (iv) a) (iv) b)

(WITHi (ACT)) (z, S, x) CAUSE(S,((ALONG(TRAVEL)) (x,y) & HAPPEN (CONTACT (x,y))) CAUSE ( (ii), HAPPEN(EXIST(LINE))) INTEND (z, SHOW (z,w,IMPORTANT (y))) INTEND (z, SHOW (z,w,notVALID(y)))

In these two examples, the ‘stroking’ produces both a concrete result (a line) and a symbolic meaning. If the noun referring to the text appears as a direct object without any preposition, the symbolic meaning ‘not valid’ is more focused: Han strök passusen om badet och skrev:

He crossed out the sentence about swimming and wrote:

De kanske strök dig från listan.

Maybe they took you off the list.

In the novels, there is an example where only the abstract, symbolic meaning is left: Han strök pojken ur sitt liv.

He wiped the boy out of his life (i.e. forgot about him completely)

The combination stryka under ‘stroke under’ is regularly used in a purely abstract sense, ‘emphasize’, in examples such as: I sitt tal strök ministern under vikten av förhandlingar. ‘In his speech, the minister emphasized the importance of negotiations.’ In this example, only the abstract component (iva) in Table 13 is realized. In this abstract use, the particle is often prefixed as in the following two examples. Such prefixed forms have a general tendency to be more abstract and/or appear in more formal contexts in modern Swedish: Det bör här också understrykas att regeringen avser att slutföra EES-förhandlingarna med all kraft.

It should also be emphasized that the Government intends to pursue with all force the EEA negotiations to their conclusion.

Långväga kontakter understrykes ytterligare av fynden från handelsplatser som Helgö i Mälaren.

Further emphasis on contacts with traders from both near and far comes from finds in such trading centers as Helgö, in Lake Mälaren,

POLYSEMY AND DIFFERENTIATION

IN THE

LEXICON

125

II. Patterns of differentiation between the hyponyms of stryka ‘stroke’ The hyponyms of stryka ‘stroke’ form two groups. The first indicate a motion back and forth in combination with a close and intense contact. The second group indicates motion along the skin. Members of this group are verbs such as smeka ‘caress’, kittla ‘tickle’ and skava ‘abrade’, which contrast primarily with respect to purpose or result: ‘show affection’, ‘make laugh’ and ‘hurt’, respectively. An important parameter is also the intensity of the contact along the scale HARD ↔ LIGHT contact. The differentiation between the verbs belonging to the subfield Moving contact is presented schematically in Table 14. Table 14. Differentiation between the hyponyms of stryka ‘stroke’ (BACK&FORTH(TRAVEL))(x) gnida ‘rub’ gno ‘rub’ (ENERGETICALLY(ACT))(z,S) gnugga ‘rub’ (HARD(CONTACT))(x,y), (FAST(TRAVEL))(x) skrubba ‘scrub’ ROUGH(x) skrapa ‘scrape’ SHARP(x) (ALONG(TRAVEL))(x, SKIN) smeka ‘caress’ (LIGHT(CONTACT))(x,y), INTEND(z,SHOW(z,w,AFFECTION)) kittla ‘tickle’ FEEL(w, WANT(w, LAUGH(w))) klia ‘itch’;’scratch’ INTEND(z, HAPPEN(FEEL(w, notSORE))) skava ‘abrade’ CLOTHES(x), FEEL(w, PAIN) frottera ‘give a rubbing’ INTEND(z, HAPPEN(WARM/DRY(w))) massera ‘massage’ MUSCLES(y), INTEND(z, HAPPEN(notFEEL(w, PAIN)))

The verbs belonging to the first group normally indicate some physical result, in particular application or removal, which is also characteristic of stryka: Nu gnider hon in fuktighetsbevarande kräm i huden som skydd mot den råkalla vinterluften utanför,

Now she rubs moisturiser into her facial skin as protection against the raw wintry air outside,

[de] tyckte att hon såg smutsig ut. De skrubbade och skrapade henne för den händelse hon medförde ohyra.

They thought she looked dirty. They scrubbed and scraped her in case she had brought vermin with her.

However, even change of physical properties realized as resultative adjectives are rather common:

126

ÅKE VIBERG Han rev loss en sida [ur tidningen] /--/och gnuggade det mjukt innan han torkade sig med det.

He ripped out a page [from the newspaper] /—/and rubbed it soft before wiping himself with it.

Focusing some part of the basic meaning is rather common, but which component is focused tends to vary idiosyncratically among the verbs. For example, the most distinctive feature of gno is the intensity of the contact. and this can be focused on in at least two steps, as illustrated in Table 15. Defocusing on the contact, the verb can be used as a subject-centered motion verb stressing the energeticness of the motion. One step further from the prototype, the verb means simply ‘act energetically’, ‘toil’. Table 15. Meaning extensions of gno ‘rub hard’ Prototype Physical contact

Focusing 1 Subject-centered motion

Focusing 2 Act energetically

Han gnodde silverbrickan blank. He rubbed the silver plate bright.

Vi gnodde omkring och letade efter boken. We scurried around looking for the book.

Vi gnodde och slet. We toiled and drudged.

The most distinctive feature of kittla ‘tickle’ on the other hand is the sensation of pleasure, which is focused in extended uses such as the following one: Billy Wilders cyniska humor kittlade oss.

9.

Billy Wilder’s cynical humor tickled us.

Conclusion

In this paper, the interest has been concentrated around the nuclear verb of the field and its patterns of polysemy. The meanings of the non-nuclear verbs in general represent an elaboration or specialization of slå, and in this way the nuclear verb forms the basis also for the differentiation between the rest of the verbs belonging to the field. The types of extensions that appear in the meaning patterns of non-nuclear members of the field tend to represent a subset of the possibilities of slå. However, verbs belonging to some of the subfields shown in Table 1 simultaneously belong to some other field, most notably the verbs that simultaneously pattern like sound-source verbs.

POLYSEMY AND DIFFERENTIATION

IN THE

LEXICON

127

The meaning patterns described in this paper are representative of physical action verbs. The verb slå, which represents the nuclear verb within the field of Physical contact verbs in all respects, represents the prototype of physical action and it is probably not a coincidence that the verb kill, whose meaning also represents a frequent secondary meaning of ‘strike’ across languages has been a favored example in linguistic textbooks. There is a human agent who uses his hands intentionally to effect some perceptually salient change in a patient (cf. Lakoff, 1987, p. 54-55 on prototypical causation). An important category of extended meanings of slå represents the product of resultative strengthening, which means that the result of the physical action is incorporated into the meaning of the verb. It is natural that this type of extension should be prominent with a physical action verb. The focusing on movement of the limbs (slå ut med armarna ‘stretch one’s arms’) is another type of meaning extension that is characteristic of prototypical physical action verbs which are based on various fundamental bodily activities. The extended meanings that are related to the subject cline are very similar to what applies to basic motion verbs, which, however, have a stronger tendency to appear in metaphorically motivated mental meanings and even to develop grammatical meanings. It is only to be expected that the analysis of fields that are more distantly related to Physical contact will turn up further possibilities. The meanings of verbs form a vast semantic network held together with general field-independent concepts. Semantic fields represent particularly densely knit and elaborated parts of the lexical semantic network, a kind of local maxima with many field-specific characteristics which must be mapped each on its own terms. The lexicon is richly structured but so intertwined with all aspects of human experience that a characterization in purely abstract terms will appear bloodless.

Acknowledgments This work has been carried out within the project Crosslinguistic Lexicology (Swed. Tvärspråklig lexikologi) financially supported by the Swedish Research Council for the Humanities and Social Sciences, which is gratefully acknowledged.

128

ÅKE VIBERG

References Aijmer, K., Altenberg, B. & Johansson, M. 1996 “Text-based contrastive studies in English. Presentation of a project”. Aijmer, Altenberg & Johansson 1996. 73-85. Aijmer, K., B. Altenberg & M. Johansson, eds. 1996 Languages in Contrast. Papers from a Symposium on Text-based Crosslinguistic Studies. (= Lund Studies in English, 88.) Lund: Lund University Press. Allén, S. 1971 Frequency Dictionary of Present-day Swedish. Stockholm: Almqvist & Wiksell. Fillmore, C. & B. Atkins 1992 “Toward a frame-based lexicon: The semantics of RISK and its Neighbors”. Frames, Fields, and Contrasts ed. by A. Lehrer & E. Kittay, 75-102. Hillsdale/N.J.: Lawrence Erlbaum. Gellerstam, M. 1992 “Modern Swedish text corpora”. Directions in Corpus Linguistics ed. by J. Svartvik. Berlin and New York: Mouton de Gruyter. Hollerbach, J. 1990a “Fundamentals of motor behavior”. Osherson et al. 1990. Hollerbach, J. 1990b “Planning of arm movements”. Osherson et al. 1990. Johnson, M. 1987 The Body in the Mind. Chicago: The Univ. of Chicago Press. Lakoff, G. 1987 Women, Fire, and Dangerous Things. Chicago: The Univ. of Chicago Press. Lakoff, G. & M. Johnson 1980 Metaphors We Live By. Chicago: The Univ. of Chicago Press. Leslie, A. 1994 “ToMM, ToBY, and agency: Core architecture and domain specificity”. Mapping the Mind. Domain Specificity in Cognition and Culture ed. by L. Hirschfeld & S. Gelman. Cambridge: Cambridge University Press. Michotte, A. 1963 The Perception of Causality. London: Methuen & Co. (Original in French 1946.) Miller, G. A. & P. N. Johnson-Laird. 1976 Language and Perception. Cambridge/Mass.: Harvard University Press. Osherson, D., S. Kosslyn & J. Hollerbach, eds. 1990 An Invitation to Cognitive Science. Vol 2. Visual Cognition and Action. Cambridge/Mass.: The MIT Press. Rudzka-Ostyn, B. 1989 “Prototypes, schemas and cross-category correspondences: the case of ask”. Linguistics 27. 613-661.

POLYSEMY AND DIFFERENTIATION

IN THE

LEXICON

129

Spelke, E., A. Phillips & A. Woodward. 1995 “Infants’ knowledge of object motion and human action”. Causal Cognition. A Multidisciplinary Debate ed. by D. Sperber, D. Premack & A. Premack. Oxford: Oxford University Press. Talmy, L. 1988 “Force dynamics in language and cognition”. Cognitive Science 12. 49-100. Viberg, Å. 1978 “Perceptionsverbens semantik i svenskan och några andra språk”. Republished in the following publication. Viberg, Å. 1981 Studier i kontrastiv lexikologi. (PhD thesis). (= SSM Report 7-8.) Dept. of linguistics, Stockholm University. Viberg, Å. 1983 “The verbs of perception: a typological study”. Linguistics 21. 123-62. Viberg, Å. 1984 “Fysiska kontaktverb i svenskan. En skiss”. Svenskans beskrivning 14. Lund. Viberg, Å. 1985 Hel och trasig. En skiss av några verbala semantiska fält i svenskan. Svenskans beskrivning 15. Gothenburg. Viberg, Å. 1992 “Tvärspråklig lexikologi med svenskan i centrum”. Nordiske studier i lexikografi ed. by R. Fjeld. Oslo. Viberg, Å. 1993 “Crosslinguistic perspectives on lexical organization and lexical progression”. Progression and Regression in Language ed. by K. Hyltenstam & Å. Viberg, 340-385. Cambridge: Cambridge University Press. Viberg, Å. 1994 “Polysemi hos svenska verb”. Nordiske studier i leksikografi. (= Skrifter udgivet af Nordisk Forening for Leksikografi. Skrift nr. 2.) ed. by A. Garde & P. Jaravad, 279-290. (Available from: Nordisk Språksekretariat/Postboks 8107 Dep/N-0032 Oslo.) Viberg, Å. 1996a “Crosslinguistic lexicology. The case of English go and Swedish gå”. Aijmer, Altenberg & Johansson 1996, 151-182. Viberg, Å. 1996b “The meanings of Swedish dra ‘pull’: a case study of lexical polysemy”. EURALEX’96. Proceedings. Part I. 293-308. Göteborg University, Department of Swedish.

SPACE AND TIME

131

Space and Time Elisabeth Engberg-Pedersen Department of General and Applied Linguistics, University of Copenhagen

1.

Metaphors and domains

Metaphor is a cognitive phenomenon, cognitive semanticists claim, in the sense that we understand one set of phenomena in terms of another. The metaphorical relation manifests itself linguistically in that we use the same word for phenomena belonging to different domains; we extend the use of the word for one phenomenon to cover another phenomenon and the extension is motivated by a cognitive metaphorical relation, a so-called cross-domain mapping. Metaphorical mapping is unidirectional in the sense that X is conceptualized in terms of Y, while Y is not conceptualized in terms of X. Moreover, metaphorical mapping is said to involve domain shift: metaphor operates between domains (Sweetser 1990: 19; Lakoff 1993: 203). It is, however, not clear what constitutes a domain. Cognitive semanticists seem to be particularly, but not exclusively, interested in extension from concrete to abstract domains, where abstract means what goes “beyond the literal mirroring, or representation, of external reality” (Lakoff 1987: xiv; see also Sweetser 1990: 18; Lakoff 1993: 205; Johnson 1987). Cognitive semanticists are furthermore mainly interested in metaphorical extension where it is possible to find a system, i.e. not just isolated extension of the meaning of one word to cover something abstract as well, but the extension of the meanings of a whole set of words or the conceptualization of one whole area of experience in terms of another. An example is the use of the vocabulary of perception for intellectual activity and emotions in English and other Indo-European languages (Sweetser 1990: Chapter 2). Such extensions are particularly interesting when they are found crosslinguistically, maybe even universally.

132

ELISABETH ENGBERG-PEDERSEN

In sum, metaphorical extension is unidirectional; it takes place between domains, especially from concrete to abstract domains; and it involves structures so that an entire domain (or a considerable part of a domain) can be said to be understood in terms of another domain. This description leaves many questions unanswered, especially: What is a domain? And what is one domain as opposed to another domain? What is concrete and what is abstract? In what follows, I shall focus on one particular area that is often described in terms of metaphorical extension or metaphorical mapping, namely the use of the same vocabulary for spatial and temporal phenomena. My main point is to question the distinction between space and time in terms of domains and thus the claim that there is metaphorical mapping from one domain to another in this area. The two areas are interwoven in cognition, the more basic perceptual difference being the one between dynamic and static events. I am going to show that the dynamic-static opposition is relevant to descriptions of both spatial and temporal expressions, and that the alleged concept of space behind temporal expressions depends heavily on the notion of event, i.e. a notion that underlies our concept of time as much as our concept of space. To demonstrate this point I shall draw on data from spoken and signed languages and from visual perception. Besides presenting data from a type of language which is relatively little known, my purpose in including data from a sign language is to illuminate the opposition dynamic vs. static. By contrast to spoken languages, sign languages are expressed by gestures in space and use space extensively to express temporal and locative notions.1 There is, however, an important difference between the use of space to express locative and temporal relations. While a configuration of points in the signing space in front of the signer can be understood as denoting a stative locative as well as a temporal relation, the temporal reading presupposes a dynamic conceptualization of time. The static-dynamic opposition is thus relevant at two levels: at the level where we describe the meaning of individual linguistic expressions which may be semantically either static or dynamic, and at a level where we describe the conceptualizations of space and time that constitute the coherent systems behind various linguistic expressions. In section 2 I shall present analyses of spatio-temporal expressions in spoken languages, while section 3 is a description of uses of space to express time in Danish Sign Language. Section 4 discusses cognitive semantic descriptions of temporal expressions in the light of Gibson’s (1975) analysis of

133

SPACE AND TIME visual perception and of data from descriptions of spatial layout.

2.

Time lines in spoken languages

There is a considerable amount of literature on time expressions in spoken languages, especially on temporal expressions that seem to be derived from expressions of spatial relations such as the words and phrases in bold in the following examples. (1)

a. b.

The worst is behind us. I am looking ahead to meeting her.

back = earlier front = later

(2)

a. b.

in the following years in the preceding years

back = later front = earlier

(3)

a.

come = earlier

b.

Il venait de fermer la porte. [lit. He came from closing the door =] ‘He had just closed the door.’ He was going to do it.

a. b.

in the coming years in the years gone by

come = later go = earlier

(4)

go = later

The spatial concept underlying the expressions in (1) is one of an individual placed in the middle of time and facing the future. Events earlier in time are conceptualized as being behind the individual, while events later in time are conceptualized as being in front of the individual. The spatial idea behind the expressions in (2) is that of years in a line approaching an individual such that all of the years are in front of the individual. The first years in the line precede the later years when approaching the individual, i.e. ‘front’ is correlated with earlier in time; the following years are behind the preceding years and thus later in time, i.e. ‘back’ (behind the preceding years) is correlated with later in time. (1b) and (2b) contrast in that ‘front’ represents later in time in (1b) and earlier in time in (2b). Words meaning ‘come’ and ‘go’ demonstrate the same contrast, i.e. ‘come’ can be associated with earlier in time as in (3a), or later in time as in (4a), and ‘go’ can be associated with later in time as in (3b), or with earlier in time as in (4b). This apparent paradox as well as the different values of ‘front’ and ‘back’ in (1) and (2) are usually explained in terms of a difference between two conceptualizations of time (Clark 1973; Benveniste 1974; Fillmore 1975;

134

ELISABETH ENGBERG-PEDERSEN

Fleischman 1982). Either time is seen as stationary with someone moving along a path; this interpretation leads to expressions like the ones in (1) and (3) with ‘earlier’ behind ego and ‘later’ in front of ego. Or time is seen as moving from in front of the individual past him/her; this view of time leads to the expressions in (2) and (4) with ‘later’ related to ‘back’ and ‘earlier’ to ‘front’.2 The two models of time, which are represented by Fleischman (1982: 324) as seen in Figure 1, assign the same basic orientation to time in relation to the individual’s front-back orientation: the individual looks in the direction of later in time (ego represented by S is seen as facing towards the future in both models). But time is represented as having different orientations in the two models. moving-ego S —> .... ——————————————>

moving-time S —> <—————————————....

PAST

FUTURE

PAST

FUTURE

come behind back after(?) following ablative

go ahead front before(?) preceding allative

go ahead front before preceding ablative

come behind back after following allative

Figure 1. Fleischman’s (1982: 324) model of the two conceptualizations of time. The model to the left describes examples (1) and (3), the model to the right examples (2) and (4) in the text.

The moving-time model can be explained in the light of how language users describe situations with other “objects” that do not have an inherent front-back orientation. An entity without an inherent front-back orientation can be described as facing the speaker. We see that in the English utterance The ball is in front of the tree as a description of the situation in Figure 2: the tree is assigned front-back orientation in accordance with what is called the egoopposed strategy (Fillmore 1982: 41). The tree is seen as having a front facing the speaker (in front of the tree) as a person in what Clark (1973) calls the canonical encounter, i.e. the situation where two individuals meet face-toface. The speaker and the tree in Figure 2 are conceptualized as two individuals facing each other. In some languages, e.g. Hausa (Hill 1975), the situation in Figure 2 is described as the equivalent of The ball is behind the tree with the tree assigned front-back orientation such that it faces away from the speaker.

SPACE AND TIME

135

This strategy is called the ego-aligned strategy of imposing front-back orientation on objects without an inherent front-back orientation. In the ego-aligned strategy, the speaker, the ball, and the tree are seen as three “individuals” on a line all facing the same direction with the speaker at the end of the line: The ball is behind the tree.

Figure 2. The positions of ego, a ball, and a tree.

Although the ego-opposed strategy appears to be based on a notion of two items in a static relation (cf. Figure 2), it is basically dynamic as suggested by Clark’s notion of the canonical encounter. Only if we conceptualize the situation as an encounter between two people coming from different directions, do we get the right idea of their orientation: they face each other. An alternative way of interpreting the assignment of front-back orientation to time in the moving-time model is to see it in the light of the assignment of front-back orientation to moving objects irrespective of their position in relation to the speaker: even though they do not have an inherent front-back orientation, moving objects are seen as moving “head first”, i.e. they “face” in the direction they are moving (Lakoff & Johnson 1980: 42-43). Lakoff (1993: 216-218) describes temporal expressions in terms of one metaphor TIME PASSING IS MOTION, with two special cases. Special case 1 is TIME PASSING IS MOTION OF AN OBJECT, and special case 2 is TIME PASSING IS MOTION OVER A LANDSCAPE. Even though the basic idea is one of motion, Lakoff includes examples such as (5). (5)

He stayed there a long time.

The dynamic idea TIME PASSING IS MOTION OVER A LANDSCAPE becomes clear if

136

ELISABETH ENGBERG-PEDERSEN

we see ‘his’ staying ‘there’ as part of a journey over a landscape where the passage corresponding to ‘his’ time ‘there’ includes a vast (i.e. long) part of the landscape. Both Lakoff and Johnson (1980) and Lakoff (1993) avoid including the notion of ego or a specific point of view in their models in contrast with the earlier models, the moving-ego vs. moving-time models. In many cases, however, we have to reckon with the notion of ego — or a specific point of view, since verbs such as English come are primarily deictic and denote motion towards ego or some individual from whose point of view the events are seen (Fillmore 1973). In (4a-b) time can be said to be assigned front-back orientation simply because it is seen as moving, as pointed out by Lakoff and Johnson (1980) and Lakoff (1993). But the direction of movement (come vs. go) is determined by the point of view: time comes towards ego’s front and goes away behind ego’s back. What we have seen up to now is that all models for describing temporal expressions in spoken languages are basically dynamic, no matter whether the expressions are semantically static as in (1) and (5) or dynamic as in (2-4). Time is seen as stationary with an individual moving as if over a landscape. Or time is seen as an object moving, either independently of ego or a specific point of view or in relation to ego. Traugott (1978: 382) criticizes descriptions of temporal expressions in terms of moving-ego and moving-time: according to her not all temporal expressions are dynamic, and static temporal expressions should not be described in terms of motion. Static temporal expressions such as the ones in (1) use ego’s front-back orientation; here neither ego nor time are conceptualized as moving. As an alternative to the moving-time and the moving-ego models, Traugott suggests an analysis using two different types of time lines: the time line of tense and the time line of sequencing. By tense, Traugott means “the semantic category that establishes the relationship which holds between the time of the situation or event talked about and the time of utterance” (1978: 371),3 while sequencing is the “ordering of events or situations talked about” (1978: 372): “the time-reference of tense shifts with the ‘now’ of the speaker, [but] the relative relation of two events does not” (1978: 379). Traugott’s representation of the two time lines can be seen in Figure 3. In Traugott’s description, the difference between tense and sequencing is primarily that expressions of tense have a reference point described as [+Proxi-

SPACE AND TIME

137

mal] with the value ‘now’, i.e. tense is deictic, while sequencing is the ordering of events and situations with respect to each other. With respect to tense, some languages only distinguish [+Proximal] from [-Proximal], i.e. ‘now’ from ‘then’, no matter whether ‘then’ is in the future or in the past. Such languages do not make use of a time line of tense. Correspondingly, Traugott claims, if a language only distinguishes [+Initial] from [-Initial], it does not make use of a time line of sequencing. (For a criticism of Traugott’s attempt to transfer the distinction between expressions with and without a time line from the time line of tense to the time line of sequencing (1978: 379-380), see Engberg-Pedersen 1993: 91.) On the two time lines ‘earlier’ is always to the left, and ‘later’ to the right, but [-Front] is to the left (‘earlier’) on the time line of tense, while [-Front] is to the right (‘later’) on the time line of sequencing. The difference in the value of the feature [Front] on the two types of time lines can most easily be understood as a difference in whose “front” is referred to: on the time line of tense, [Front] should be interpreted as referring to the speaker’s front-back orientation; on the time line of sequencing, [Front] refers to time’s front-back orientation. In order to assign front-back orientation to time on the time line of sequencing, Traugott needs a reference point to the left of the time line of sequencing: when Event1 is before — or precedes — Event2, both must be seen from a

TENSE: [-Proximal] (then) [+Proximal] (now) [-Proximal] (then) T--------------------------------------------------------------------------------------------------T [+Prior] (past) [-Prior] (future) Source Goal come → go → [-Front] (behind, back) [+Front] (forward, ahead) SEQUENCING: [+Initial] (E1, first) [-Initial] (E2, second) T--------------------------------------------------------------------------------------------------T [+Prior] (earlier) [-Prior] (later) [+Front] (before) [-Front] (after) ← preceding ← following Figure 3. The time line of tense and the time line of sequencing (Traugott 1978: 378 and 382).

138

ELISABETH ENGBERG-PEDERSEN

point to the left of the sequence line which permits the functioning of the egoopposed strategy, i.e. Event1 and Event2 are conceptualized as having fronts facing an individual to the left of the line in a canonical encounter, the individual facing right. Traugott claims that this does not mean that sequencing is also deictic, because “the assignment of [+Front] to events in sequence, once established in terms of the encounter, remains constant, wherever the speaker is in terms of events” (1978: 380). It does, however, mean that Traugott implicitly includes a dynamic notion in terms of the canonical encounter in the description of sequencing. The dynamic notion is also implicit in the tense model. Traugott’s tense model does not explain the assignment of the values + and – to the feature [Prior], i.e. it does not explain why the past is behind the individual and the future in front and not vice versa, unless the individual is seen as either moving or having moved over a stretch of time such that what is behind the individual is the past. Traugott’s attempt at explaining temporal expressions without the use of the idea of motion demonstrates that we need to separate the semantic notions of static vs. dynamic linguistic expressions from the use of dynamic and static notions in the models that explain the distribution of past/earlier and future/later on time lines. In the moving-ego model, i.e. Traugott’s tense line, time is assigned past-future orientation by virtue of the idea of the individual moving or having moved over time even in static expressions such as The worst is behind us. In the moving-time model time is assigned past-future orientation by virtue either of its own movement or of the notion of the canonical encounter. The notion of canonical encounter is needed if we want to cover expressions with basically deictic terms such as come and go: (4a-b) and (2a-b) can be explained in terms of the moving-time model, but differ in that (4a-b), and not (2a-b), include a specific point of view.

3.

Time lines in sign languages

I.

Sign languages of the deaf

Sign languages of the deaf have developed where deaf people get together, in families with deaf people, through several generations, and in schools and clubs for the deaf. Sign languages are generally mutually unintelligible.

SPACE AND TIME

139

The articulatory means of sign languages are the hands and arms, the body, the head, and the muscles of the face, in particular the muscles around the eyes, the brow and the mouth, and eye movements. The hands produce the lexemes, often jointly with the mouth. Many manual signs of Danish Sign Language are accompanied by mouth patterns which may look like the lipreading pattern of a Danish word. A mouth pattern accompanying a sign can also be different from any Danish word. The non-manual articulators produce visual rhythm and more specific linguistic information. Clause and major constituent boundaries are usually marked by movements of the head and the body and by changes in facial expression. The non-manual articulators also signal various kinds of dependency between constituents and clauses in ways that are not yet fully understood. Special configurations of non-manual signals indicate, for instance, conditionals and topicalized constituents. All sign languages of the deaf that have been analyzed up to now use space for reference. For deictic reference, signers point to the referent in the situational context. For reference to nonpresent entities, signers may use directions in space to represent the referents. A direction that represents a referent is called a locus (pl. loci). A pointing gesture in the direction of a locus is then understood as a pronominal reference to the referent. The choice of loci for nonpresent referents is not predictable, but influenced by certain conventions that need not concern us here (Engberg-Pedersen 1993: 71-78). II. Time lines in Danish Sign Language For time referents, signers may use loci as for other referents. Loci for time referents differ, however, from loci for other referents in that they can be described as belonging to dimensions, so-called time lines, in the space around the signer. A time line in a sign language can be thought of as a line in space in the sense that the association of a time referent with a certain locus invests other loci with meaning, and these other loci are expressed through points to the left or right of, or forward or backward from the point of the first locus. If, for instance, Tuesday is represented by one locus, and a new referent, Monday, is represented by a locus to the left of Tuesday’s locus seen from the signer’s perspective, then Monday is understood as the Monday preceding Tuesday. If Monday is represented by a locus to the right of Tuesday’s locus, it is understood as the Monday following Tuesday.

140

ELISABETH ENGBERG-PEDERSEN

The time lines in Danish Sign Language, which are represented in Figure 4, all have different meanings. The deictic time line is primarily used for deictic reference, the mixed time line may be used with a deictic or an anaphoric reference point, while the other time lines are used only with an anaphoric reference point. The following example of the use of a time line, in this case the deictic time line, comes from a monologue in which a signer talked about a conference that took place in the month of January preceding the time when she told her story. She made reference to January in the following way:4 (6)

RECENTLY CONFERENCE / JANUARY+deictic-tl-before BEFORE+deictic-tl MONTH+deictic-tl-before DET+deictic-tl-before / ‘Recently there was a conference - I think it was in January.’

The signer modified the signs JANUARY, BEFORE, MONTH and the determiner for a locus behind her right shoulder by making the signs close to her shoulder instead of in neutral space outside her right side. Later she referred to the conference in the following way: DET+deictic-tl-before CONFERENCE, i.e. ‘the January conference’.

Figure 4. Time lines in Danish Sign Language a. the deictic time line, b. the anaphoric time line, c. the sequence line, d. the mixed time line (drawing by Hanna Orlof, reprinted with permission from Signum Press).

What I call the anaphoric time line has a reference point represented by a point a little outside the signer’s center. This reference point always has a value established somewhere in the linguistic context. It seems that the anaphoric time line has no future part. When signers want to talk about the future

SPACE AND TIME

141

in relation to an anaphoric reference point, they may use either the mixed time line or the sequence line. The mixed time line is frequently used with an anaphoric reference point (represented by a locus expressed by pointing close to the signer’s body) to mean ‘from that moment onwards’. In this use it can be seen as the ‘later’ section of the anaphoric time line. But the mixed time line may also be seen as the future part of the deictic time line. Signers claim that there is no difference between signing ‘next Monday’, i.e. a time expression with the present as its reference point, and signing ‘the following Monday’ with an anaphoric reference point. The sequence line runs parallel to the signer’s surface plane with more leftward loci used for ‘earlier’ and more rightward loci for ‘later’. Signers may use the sequence line to talk about, for instance, the Friday following a specific Thursday. Signers of Danish Sign Language use the vertical plane in front of their body for temporal expressions. The two-dimensional vertical plane is used for months in a year and days and weeks of the months as indicated on an imaginary calendar and for hours as if on the surface of an imaginary clock. We find the same general orientation of time lines in other sign languages, e.g. American Sign Language (Friedman 1975; Cogen 1977), British Sign Language (Brennan 1983), and Sign Language of the Netherlands (Schermer & Koolhof 1990). Unfortunately, there is very little evidence of time lines in sign languages from outside the Western world (see, however, Kuschel 1974; Kakumasu 1968; Ferreira Brito 1985). III. The static vs. dynamic distinction in sign languages Loci in sign languages originate in spatial deixis (Liddell 1995), but are used for all types of reference, whether to concrete, non-present referents or abstract referents: locations, entities, time, and abstract concepts. The time lines described above have counterparts in expressions denoting spatial relations: a pointing gesture directed at a point behind the signer’s shoulder may be a pronoun that is used to refer to an earlier moment in time, or to a location or entity behind the signer’s back in the situational context or the context of the event talked about. Correspondingly, a pointing gesture directed to some area in front of the signer followed by a pointing gesture to an area a little to the right of the first area may be used either to refer to two moments in time, one occurring before the other, or to two entities placed next to each other.

142

ELISABETH ENGBERG-PEDERSEN

There is, however, an interesting difference between the use of space for locative and for temporal purposes in signing. Spoken languages make use of two different strategies in assigning front-back orientation to entities without an inherent front-back orientation, as described in section 2 in relation to Figure 2. The ego-opposed strategy is based on the notion of the canonical encounter with two individuals facing each other, and it is relevant to the description of both locative expressions such as The ball is in front of the tree and temporal expressions such as in the following years. The values of ‘front’ and ‘back’ (‘earlier’ or ‘later’) in the moving-time model depend on the canonical encounter or the idea that an entity moves front first. In Danish Sign Language, a situation such as the one in Figure 2 is described without any reference to fronts or backs: signers indicate two loci at two different distances out from their own center; the closer one represents the ball, the one farther away the tree. That is, they establish a relationship between three entities, their own body representing the individual, one locus representing the ball, and a second locus representing the tree. The configuration mirrors the relationship between the three entities in the situation described without reference to the tree’s “front” or “back”, somewhat equivalent to the English expression The ball is between NN and the tree. Apparently, the same type of configuration can be used to express temporal relations. Signers may use a locus at their body and two loci at different distances out from their body to denote a relationship between three moments in time. There is a difference between the locative and the temporal use, however. In order to understand the relationship between the three moments in time correctly, the receiver needs to know what corresponds to earlier and what to later in time. Otherwise the moment represented by the locus in the middle might be understood to be either earlier or later than the moment represented by the locus farther away from the signer. The convention of the mixed time line makes it clear that the moment represented by the locus in the middle is earlier in time than the moment represented by the locus farther away. This assignment of earlier and later can be explained in terms of either the moving-ego model or the moving-time model. If time is stationary and ego moves along the line, s/he reaches the moment represented by the locus in the middle first, i.e. earlier in time, and the moment represented by the locus farther away later in time. The assignment of earlier and later to the mixed time line can also be understood in terms of the moving-time model. If time is seen as moving towards the individual as in the canonical encounter, the

SPACE AND TIME

143

moment represented by the locus closest to the signer is earlier in time than the moment represented by the locus farther away. Even though the same configuration of loci can be used to express locative and temporal relations and both are static, there is thus a difference between the locative and the temporal use of space. In order to get the proper assignment of earlier and later in time, the receiver needs the dynamic notion of either the individual moving in time or time moving towards the individual in the canonical encounter. This difference between the two uses of space in sign language underlines the role of the cognitive difference between static situations and dynamic events in the description of temporal expressions, no matter whether these are semantically static or dynamic.5

4.

Time and space: A metaphorical relation?

The preceding sections give ample evidence that the linguistic devices used to express temporal relations and the linguistic devices used to express spatial relations are related in systematic ways. We cannot explain temporal expressions without reference to the perceiving human being in space, either as an individual moving along or as an individual “facing” time in the canonical encounter. But does that mean that temporal expressions are derived metaphorically from spatial expressions? The answer to this question depends on whether time and space constitute two different domains. In order to maintain that temporal expressions are metaphorically derived from spatial expressions, we need to claim that the two areas constitute different cognitive domains and that the domain of space is more basic than the domain of time. I am going to argue that time and space cannot be analyzed as separate domains, a fact that is reflected in many linguistic expressions. It seems, however, that both time and space are conceptualized at different levels, which makes it possible to talk about space-to-time metaphors as well as time-to-space metaphors at some levels. Langacker defines a domain as “[a] context for the characterization of a semantic unit” (1987: 147). Domains are cognitive entities, and a concept is defined in terms of the contexts or domains needed to characterize the concept. Domains are themselves concepts, and they are hierarchically organized in the sense that a concept is characterized by a number of domains that can themselves be characterized in terms of further domains. Basic domains are

144

ELISABETH ENGBERG-PEDERSEN

domains that cannot be characterized in terms of any other domains. Langacker finds that both time and space are “primitive dimension[s] of cognitive representation” (1987: 148; see also Langacker 1991). He characterizes the use of spatial terms to talk about time as metaphor, but also claims that time is in some sense more fundamental than space: “[T]he conception of spatial relationships involves scanning, which requires processing time, and our notions of spatial extension are intimately bound up with time-extended physical actions (e.g. movement and the manipulation of objects).” (Langacker 1987: 149). The idea that time and space are interwoven in perception is developed by Gibson in an article with the title ‘Events are Perceivable But Time is Not’ (1975). Gibson points out that it has long been a problem to psychologists how human beings and animals can perceive space: how can an observer see the third dimension, depth? Based on results of experiments with perception, he claims that the question is a false one. There is no such thing as depth perception or perception of space. All there is is perception of textured surfaces and their layout. The problem is now how we can perceive that portion of the layout of the world that is temporarily out of sight. The answer is: through the motions of objects, the locomotion of the observer, and the fact that what goes out of sight during one displacement will come into sight during the opposite displacement. All this has to do with time. But we can perceive time as little as we can perceive space. What we perceive are events and locomotions occurring in an environment that is rigid and permanent. Three kinds of events are possible: the repositioning of objects, the reshaping of surfaces, and the annihilation or creation of surfaces. All three involve some persistence and some change. The repositioning of an object such as an apple falling, for instance, leaves the shape and size of the object intact and the background surfaces invariant, but changes the pattern of the environment. Gibson concludes that “[t]ime and space are concepts, abstracted from the percepts of events and surfaces. They are not perceived, and they are not prerequisite to perceiving. Time and space are intellectual achievements, not perceptual categories.” (Gibson 1975: 299). Gibson furthermore points out that the theory of event perception implies a rejection of the division of the stream of awareness into a past, a present, and a future (1975: 300). He does acknowledge the feeling of now and attributes it to the perception of the body of the observer as distinguished from the environment. It comes from the visual perception of the observer’s locomo-

SPACE AND TIME

145

tion through the environment. Travelers perceive the path to be traveled if they look ahead, the path that has been traveled if they look behind, and the position in between is called here. The point here and the moment now coincide. What is perceived is a moving self in a stationary environment. Our perception of a static situation thus presupposes our perception of dynamic events, and it is this relationship between static situations and dynamic events that lies behind spatial as well as temporal linguistic expressions. Its usefulness in relation to the description of temporal expressions has been demonstrated above; the mere fact that the traditional models of describing time, the moving-ego and the moving-time models, are models based on movement demonstrates that we need a temporal dimension in the description of temporal expressions. In relation to spatial expressions, the spatial description of the ball, the tree and the individual in Figure 2 in terms of the canonical encounter presupposes a dynamic event of two individuals approaching each other. We also see the semantic relevance of the static vs. dynamic opposition, for instance, in the difference between verbs such as English come and be in their primarily spatial use or between Danish spatial adverbs such as ind (Han gik ind i huset ‘He went into the house’) and inde (Han er inde i huset ‘He is inside the house’): the differences cannot be described without resort to the opposition static vs. dynamic. The semantic description of certain prepositions in their primarily spatial use also requires inclusion of temporal notions: the use of over rather than on the other side of in He lives over the street implies motion across an obstacle from the reference point (Taylor 1989: 112, 114). Several descriptions of the diachronic development of tense markers from verbs of motion such as come and go are based on the idea of a metaphorical extension from space to time. Sweetser (1988), for example, describes the development of go-future in, for instance, English as metaphorical mapping from the spatial to the temporal domain. There is a partial correlation between experiences of time and path-traversal, she claims, but “[t]he metaphorical mapping of going onto futurity is general, and not partial like the experiential correlation” (1988: 391). Given the fact that time and space cannot be distinguished in our perception of events, we should not talk about experiential “correlation”, however, but about the same experiential elements in both cases and, therefore, not of a mapping relation between two domains. Sweetser further claims that in the mapping, “we lose the sense of physical motion (together with all its likely background inferences). We gain, however, a new meaning of future prediction or intention” (1988: 392). But

146

ELISABETH ENGBERG-PEDERSEN

since there is no mapping, this is not a case of “loss” and “gain” of meaning: in its more spatial meaning, go has the additional meaning of physical motion. Moreover, “future prediction or intention” is as much part of go in its primarily spatial uses as in its primarily temporal meaning. The French Elle est allé fermer la porte (‘She has gone to close the door’), which can only be understood in the primarily spatial sense, implies prediction and intention as well. An alternative description of the diachronic development can be found in Bybee, Perkins and Pagliuca (1994). They describe different mechanisms by which diachronic development of grammatical markers can take place, one being metaphorical extension, another generalization. One reason for saying that some instances of meaning shift are metaphorical is that the lexeme in question cannot occur with both readings in the same context, since “metaphor requires a clear image-schematic structure that crosses cognitive domains” (1994: 285). If there are contexts where both meanings can be present, the mechanism of change is generalization rather than metaphor. An example of the role of generalization in semantic change is the development of the progressive in many languages. The origin of the progressive marker in many languages is an adposition meaning ‘at’, ‘on’ or ‘in’ or an auxiliary from a verb meaning ‘sit’, ‘stand’, ‘lie’, ‘be at’, ‘stay’, ‘live’ or ‘reside’ (Bybee, Perkins and Pagliuca 1994: 129-130). In the development of the progressive, “the temporal meaning is present from the beginning, since to be located spatially in an activity is to be located temporally in that activity. The change that occurs is the gradual loss of the locative meaning” (1994: 137). What happens in the development of the progressive is that the locative meaning is gradually eroded, and the expression with only the temporal meaning can be used in a wider range of contexts. Still, in English, for instance, an utterance with the progressive such as He’s taking a bath makes an appropriate answer to the location inquiry Where’s Lou? (1994: 133). That is, in certain contexts both meanings are present. Bybee, Perkins and Pagliuca take this as a sign that the meaning shift does not imply a clear crossing of a domain boundary. Langacker (1991: 155-56) presents a model where space and time are interwoven at the perceptual level, but where it is nevertheless possible to distinguish primarily spatial from primarily temporal conceptualization. He points out that an expression with go in its primarily spatial sense (e.g. They went from Cambridge to London) and an expression with go in its primarily

SPACE AND TIME

147

temporal sense (e.g. The concert went from midnight to 4 am) share a feature: both are based on perception of (concrete or abstract) motion of a mover making contact with an ordered series of locations/points in time, i.e. the component states of the process are distributed through a continuous span of time, or put differently, both sentences describe processes taking place in time. The difference between the two is that in the primarily temporal expression time serves as the cognitive domain of the relation between the concert and the period from midnight to 4 am; each component state is a relation between the mover and a point in time. In the primarily spatial use of go, each component state is a relation between a mover and a location. Can we then claim that the spatial domain is more basic than the temporal domain and that, as a consequence, the use of go in The concert went from midnight to 4 am is metaphorical? Langacker seems to think so when he claims that physical movement through space is the prototypical case of the schematic concept of motion shared by the two examples (1991: 156). He claims that, in the sentence where go has a primarily temporal meaning, the speaker/conceptualizer ‘traces mentally along the path in order to situate the process in relation to a reference point’ (1991: 332). But in They went from Cambridge to London, the conceptualizer also necessarily traces mentally along the path, since there is no other way to scan the miles of land in the time it takes to understand the sentence. What about expressions used prototypically to express time? Are they ever used to express temporal relations? In a study of how people describe the layout of their flats, Linde and Labov (1975) found that the overwhelming majority of the descriptions were organized as tours of the flats. Very few descriptions were maps seen from a bird’s perspective (I’d say it’s laid out in a huge square pattern, broken down into four units...). The imaginary tours began at the front door and the descriptions then provided a minimal set of paths by which each room could be entered. There were two basic types of tours: the static type (to the right, straight ahead of you, off of the X) and the mobile type (you keep walking straight ahead, now if you turn right) (Linde & Labov 1975: 930). The descriptions of the flats emphasize that the static vs. dynamic opposition is also fundamental to spatial descriptions. Langacker’s model demonstrates that it is possible to distinguish primarily spatial from primarily temporal contextualizations of linguistic expressions. The use of before in He stood before the throne expresses primarily a spatial relation; in He heard the scream before the shot, before expresses a primarily temporal relation. What is less obvious, however, is the “prototypi-

148

ELISABETH ENGBERG-PEDERSEN

cal” sense of a word outside its use in a particular context. In order to talk about metaphorical extension from the spatial to the temporal domain or vice versa, we need to determine the primary domain of the word irrespective of its use in a particular context. In some cases, most language-users would probably agree on the expression’s prototypical sense irrespective of its current use: in They live 40 minutes from the center, a primarily temporal expression, 40 minutes, is used metaphorically to express a locative relation. In I haven’t had a drink since Cambridge, a primarily spatial word, the name of a town, is used to denote a moment in time. At a workshop in Copenhagen, May 1994, George Lakoff pointed to a difference between the following two examples: (7) (8)

I don’t know what is ahead of me. (said when walking down an alley) I don’t know what is ahead of the year 2000.

In (7), Lakoff claimed, the spatial ahead stands metonymically for time, while in (8) there is metaphorical mapping of space onto time. In (7) the two domains, space and time, are brought together to form one domain such that expressions belonging to the spatial “section” can be used metonymically for the whole. The concept behind his analysis of both (7) and (8) is thus that it is possible to distinguish two domains, space and time, and that ahead’s sense is basically spatial. The analysis does not take into consideration the impossibility of distinguishing time and space at the perceptual level. Any theory of a metaphorical relation between space and time has to take into consideration that space and time are not distinguished at the perceptual level, that the difference between static and dynamic is fundamental to linguistic expressions of temporal as well as spatial relations, and that it seems possible to distinguish conceptualizations of time and space at some cognitive levels. In the last resort, however, what is metaphorical extension depends on the language- users’ sense of the basic meaning of individual expressions.

Acknowledgment I would like to thank Michael Fortescue for improving the language of this article.

SPACE AND TIME

149

Notes 1.

In talking about sign languages, I distinguish between location and space such that location (or locative) is a semantic notion, while space denotes an expressive means of sign languages.

2.

An example such as (i) We’re coming up on Christmas. (from Lakoff 1993: 218) demonstrates the use of come to denote someone moving in time, but with ‘come’ used of ‘later’ in time in apparent contrast to both (3a) and (4a). The point is, however, that in (3a) the event is seen from the point of view of the time when the door has been closed, in (i) the event is seen from the point of view of Christmas. In both cases the ‘coming’ is a past event in relation to the moment of the point of view, i.e. both demonstrate the conceptualization of time as stationary with someone moving along a path.

3.

Later in the article, Traugott uses Comrie’s (1976: 1-2) definition of tense, which differs significantly from her first definition: “Tense relates the time of the situation referred to to some other time, usually the moment of speaking.” (Traugott 1978: 374 - emphasis added).

4.

The transcriptions of signed examples in this chapter are very simplified. An English word in capital letters is a gloss for a manual sign. DET, the determiner, stands for a pointing gesture that is part of a nominal, while PRON is a pointing gesture that functions as a pronoun. If a sign is modified, the modification is described by one or more words in small letters initiated by +: +deictic-tl-before means that the sign is modified for a locus indicating a moment in time earlier than the reference point. In the transcriptions, / stands for a boundary marked non-manually or manually by lengthening of the sign preceding the boundary or by lowering of the hands or the like.

5.

The overwhelming majority of temporal expressions using space in Danish Sign Language are static. One exception is an expression with a verb of motion modified for the deictic time line to denote the sudden and unexpected occurrence of an event (somewhat like Event X arrived suddenly and unexpectedly).

References Benveniste, E. 1974 Le langage et l’expérience humaine. E. Benveniste: Problèmes de linguistique générale II. Gallimard, 67-78. Brennan, M. 1983 “Marking time in British Sign Language”. Language in Sign: An International Perspective on Sign Language ed. by J. Kyle & B. Woll, 10-31. London: Croom Helm. Bybee, J., Perkins, R. and Pagliuca, W. 1994 The Evolution of Grammar: Tense, Aspect, and Modality in the Languages of the World. Chicago: The University of Chicago Press.

150 Clark, H. 1973

Cogen, C. 1977

Comrie, B. 1976

ELISABETH ENGBERG-PEDERSEN

“Space, time, semantics, and the child”. Cognitive Development and the Acquisition of Language ed. by T. E. Moore, 27-63. New York: Academic Press. “On three aspects of time expression in American Sign Language”. On the Other Hand: New Perspectives on American Sign Language. ed by L. A. Friedman, 197-214. New York: Academic Press.

Aspect: An Introduction to the Study of Verbal Aspect and Related Problems. Cambridge: Cambridge University Press. Engberg-Pedersen, E. 1993 Space in Danish Sign Language: The Semantics and Morphosyntax of the Use of Space in a Visual Language. Hamburg: Signum-Press. Ferreira Brito, L. 1985 “A comparative study of signs for time and space in São Paulo and UrubuKaapor Sign Language”. Presented at the III. International Symposium on Sign Language Research, Rome, June 22-26, 1983. Fillmore, C. J. 1973 “May we come in?”. Semiotica 9, 97-116. Fillmore, C. J. 1975 Santa Cruz Lectures on Deixis, 1971. Bloomington: University of Indiana Linguistics Club. Fillmore, C. J. 1982 “Towards a descriptive framework for spatial deixis”. Speech, Place, and Action ed. by R. J. Jarvella and W. Klein, 31-59. New York: John Wiley & Sons. Fleischman, S. 1982 “The past and the future: Are they coming or going?”. Proceedings of the Eighth Annual Meeting of the Berkeley Linguistics Society ed. by M. Macaulay & O. D. Gensler, 322-334. Berkeley: Berkeley Linguistics Society. Friedman, L. A. 1975 “Space, time, and person reference in American Sign Language”. Language 51(4), 940-961. Gibson, J. J. 1975 “Events are perceivable but time is not”. The Study of Time II: Proceedings of the Second Conference of the International Society for the Study of Time, Lake Yamanaka, Japan ed. by J. T. Fraser & N. Lawrence, 295-301. Berlin: Springer-Verlag. Hill, C. A. 1975 “Variation in the use of ‘front’ and ‘back’ by bilingual speakers”. Proceedings of the First Annual Meeting of the Berkeley Linguistics Society ed. by C. Cogen, H. Thompson & J. Wright, 196-206. Berkeley: Berkeley Linguistics Society.

SPACE AND TIME Johnson, M. 1987

151

The Body in the Mind: The Bodily Basis of Meaning, Imagination, and Reason. Chicago: The University of Chicago Press.

Kakumasu, J. 1968 “Urubú Sign Language”. IJAL XXXIV. 275-281. Kuschel, R. 1974 A Lexicon of Signs From a Polynesian Outliner Island: A Description of 217 Signs as Developed and Used by Kagobar, the Only Deaf-Mute of Rennell Island. København: Københavns Universitet. Lakoff, G. 1987 Women, Fire, and Dangerous Things: What Categories Reveal about the Mind. Chicago: The University of Chicago Press. Lakoff, G. 1993 “The contemporary theory of metaphor”. Metaphor and Thought. 2nd ed. ed. by A. Ortony, 202-251. Cambridge: Cambridge University Press. Lakoff, G. & Johnson, M. 1980 Metaphors We Live By. Chicago: The University of Chicago Press. Langacker, R. W. 1987 Foundations of Cognitive Grammar. Vol. 1: Theoretical Prerequisites. Stanford: Stanford University Press. Langacker, R. W. 1991 “5. Abstract motion”. Concept, Image, and Symbol: The Cognitive Basis of Grammar, 149-163. Berlin: Mouton de Gruyter. (Reprinted from Proceedings from the Annual Meeting of the Berkeley Linguistics Society 12. 455471, 1986.) Liddell, S. K. 1995 “Real, surrogate, and token space: grammatical consequences in ASL”. Language, Gesture, and Space ed. by K. Emmorey & J. S. Reilly, 19-41. Hillsdale: Lawrence Erlbaum. Linde, C. and Labov, W. 1975 “Spatial networks as a site for the study of language and thought”. Language 51(4). 924-939. Schermer, T. & Koolhof, C. 1990 “The reality of time-lines: aspects of tense in SLN”. Current Trends in European Sign Language Research. Proceedings of the Third European Congress on Sign Language Research ed. by S. Prillwitz & T. Vollhaber, 295-305. Hamburg: Signum-Press. Sweetser, E. 1988 “Grammaticalization and semantic bleaching”. Proceedings of the Fourteenth Meeting of the Berkeley Linguistics Society ed. by S. Axmaker, A. Jaisser & H. Singmaster. Berkeley: Berkeley Linguistics Society. Sweetser, E. E. 1990 From Etymology to Pragmatics: Metaphorical and Cultural Aspects of Semantic Structure. Cambridge: Cambridge University Press.

152 Taylor, J. 1989

ELISABETH ENGBERG-PEDERSEN

Linguistic Categorization: Prototypes in Linguistic Theory. Oxford: Clarendon Press.

Traugott, E. C. 1978. “On the expression of spatio-temporal relations in language”. Universals of Human Language, Vol. 3: Word Structure ed. by J. Greenberg, C. Ferguson & E. Moravcsik, 369-400. Stanford: Stanford University Press.

CONCEPTUAL ENGINEERING

153

Conceptual Engineering Implementing Cognitive Semantics1 Kenneth Holmqvist Department of Cognitive Science, Lund University

1.

Introduction

Although Langacker’s Cognitive Grammar (Langacker 1987) is a very powerful full-scale model of language, there have been few attempts to implement it on computers. We think that the reason for this is twofold. First, cognitive grammar is presented as a theory of language as a static entity. It does not focus on the linguistic and semantic processing that an individual has to perform when understanding and producing speech. (As opposed to, for instance, generative grammar, in which the language description, the language reception model and the language production models are all one and the same.) Therefore, anyone who wants to implement language understanding based on cognitive grammar must first devise a processual counterpart to it. The second reason that implementations of cognitive grammar are and will remain rare is the general complexity and openendedness of the theory. It is simply impossible to make any larger computer implementation of cognitive grammar without first interpreting virtually all elements of the theory into some more rigid form. Domains, constructional schemata, schema types, scanning modes, predications, valence composition etc. must therefore be given computational counterparts. This is not only difficult, but it also involves the danger of completely rebuilding the theory so that it is adapted more to the needs of the computer than to the linguistic reality behind cognitive grammar.

154

KENNETH HOLMQVIST

The reason why a computer scientist would be interested in implementations of cognitive grammar is of course its well-founded integration of grammar and semantics with imagery, especially as there is now a rising interest in spatial imagery as a representation form among AI researchers (Glasgow 1993). A cognitive linguist would probably take most interest in the many implications of language processing that an implementation gives: incrementation, linear time complexity, the continuous and parallel adaptation to semantic and grammatical constraints etc. In our project, the cognitive linguist could find suggestions for problems that need the attention of linguistic research, but s/he could not take our solutions to be truths about the human language. We are simulating human language understanding in the computer, but also making necessary computational assumptions about that which is yet unknown. Our project is based on the preliminary computational model of cognitive grammar developed in Holmqvist (1993). This computational model consists of three main parts: Representations of Activated Lexical Units, Semantic Composition Processes, and Mechanisms for Valence Suggestion and the Incremental Updating of the Schema Population. In the following, we will make a summary description of these mechanisms with examples to clarify them.

2.

Representation of activated lexical units

First, we implemented a computer model of the lexical unit structure in socalled ‘image schemata’. Different authors in the cognitive linguistics tradition attach different meanings to the term ‘image schema’. The advantage of choosing Langacker (1987) rather than, for instance, Lakoff (1987, 1989) is simply that its account is clearer and more systematic (although Langacker never explicitly connects his heuristic diagrams to imagery). Above all, the confusion over Lakoff’s ICMs and other meaning variation structures can then be avoided. Unfortunately, this clarity comes at the cost of a lack of description of lexical meaning variation in Langacker. Our model of the semantic pole of an activated lexical unit has been implemented in the form of: I.

a matrix of domains, ordered by centrality values. Langacker ranks domains according to their centrality in the meaning of a term. We have interpreted centrality to be a numeral ordering.

CONCEPTUAL ENGINEERING

155

II. a list of parts ordered by their saliences. Salience is the ordering of parts and wholes according to their relative importance to the meaning of a term. Langacker does not have this ordering, although it is closely related to the centrality of domains. III. a list of wholes ordered by their saliences Take the lexical unit [KNIFE] as an example, as depicted in Figure 1. [BLADE] and [HANDLE] are clearly parts of [KNIFE]. [KNIFE] has [SILVERWARE] as a whole: [KNIFE] is one of the parts in collections making up [SILVERWARE]. But [KNIFE] can also have [CUT] as a whole, because [KNIFE] can be the agent (TR) part of the cutting process. Parts and wholes like [BLADE] or [CUT] are themselves described with the same three elements. Their saliences, as well as the centralities, should be lexically stored but changed on activation to fit a context where that whole, part or domain is salient (this mechanism is not included in our immediate implementation goals). Of course, just saying that the [BLADE] is a part of [KNIFE] is not sufficient. We must characterize this part-whole relation closer. For instance, the relative sizes of the blade and the knife must not deviate outside certain limits. The relative spatial position of the blade on the knife must also be correct, i.e. the blade must be correctly attached. There are two possibilities of implementing these constraints, both involving the predication (below). Most of the content substance of a lexical unit resides in its domains, which collectively form its matrix. Typical domains are: color, 3D space, material, age, temperature, profession, emotion, etc. The matrix of a lexical unit (like [KNIFE]) allows the unit to involve all these aspects simultaneously and to be identified as the same entity across the domains (that this spatial form belongs to that color etc.). Domains are made up by dimensions. The color domain has three dimensions (hue, saturation and brightness), while temperature has one dimension. 3D space has three dimensions. Profession and emotion domains probably have rather abstract dimensions. The lexical unit predicates in each domain in its matrix. In the spatial form domain, [KNIFE] predicates a 1D-directed form. In the color domain, [KNIFE] may predicate a brown color. A predication such as a 1D-directed form or a semancolor picks a subset of the total number of dimensions in the domain and attributes values to the dimensions in that subset. For example, a 1D-directed form attributes a value to the spatial extension dimension.2

156

KENNETH HOLMQVIST

Dimensions and their values need not refer to measure units in the external world. Their main function is to serve in a mathematical region activation model underlying the representation. In all domains, (the dimensional values of) a predication is (are) equated with an absolute point or a variable prototype. The simplest case is that of points for color sensations and regions for color predications in color space. The model thus allows for variability of predication: there is a range of brown colors just like a range of 1D-directed forms. The variability is constrained by there being other regions than the predicated region in the domain. If the color dimension values (i.e. the color point) of the knife would move outside of the brown region, another region will be more saliently activated and there would be a conflict between that region and the brown predication in the lexical unit. All semantic constraints regarding predications are intended to be withheld by this region activation model.

Figure 1.

CONCEPTUAL ENGINEERING

157

The example here has concerned the representation of the thing unit [KNIFE], but the same representation form also applies to other schema types. A process like [CUT] also has its parts, wholes and domains. The domains of [CUT] are different from that of [KNIFE] in several ways, the most important being that [CUT] has temporal dimensions in its predications. It is these temporal dimensions that make [CUT] processual (which in turn has bearings on its possibility to combine with tense-forming lexical units). The structures described this far are implemented, but we have not filled them with actual lexical content, i.e. the correct domains, predications and values of dimensions. In testing the composition processes to be described, we have used an ‘unembodied’ randomly generated filling in the lexical units. This is ultimately because we have no perceptual system in our computer, and hence nothing that can provide the correct domains and values of dimensions in the lexicon.

3.

Semantic composition processes

The second part of the computational model consists of a semantic composition process (aiming at what Langacker [1987] calls ‘accommodation’). The guiding principle behind the construction of the computational model is that understanders of human language perform semantic composition by means of image superimposition: individual lexical units are superimposed to form a composite structure.

[ROPE]

[STRETCHED].TR

[ACROSS] TR

[STREET]

LM

SUPERIMPOSITION

Figure 2.

SUPERIMPOSITION

158

KENNETH HOLMQVIST

Figure 2 shows how the single units [ROPE], [STRETCHED], [ACROSS] and [ROAD] are superimposed to form [(A) ROPE STRETCHED ACROSS (THE) ROAD]. It is those parts that have been set in a valence relation that can be superimposed, simply because they are then identical. For instance, the LM of [ACROSS] is identical to [STREET]. We can use image superimposition as the guiding principle, because lexical units having predications in different domains can be viewed as images. We normally think of images as something spatial, as in Figure 2, where the domain is 3D space and the predications are spatial configurations. But it is just as easy to think of 21° C as a point predication in the temperature domain or five hours as an extended predication in the time domain. In those domains, we can therefore use the same superimposition mechanism that is employed in the spatial domain. Superimposition constructs the composite by making pieces from different lexical units identical and thus welding together the composite. Superimposition also evaluates the composite in that the pieces have to match. A bad match like “A ball stretched across the street” is detected by superimposition. This means that the content of a lexical unit is its own semantic constraints. During superimposition, constraint satisfaction is only a matter of seeing whether a superimposed unit violates the constraints of one of the other units that it is going to be identical to. (The most common case of valence formation is asymmetric: A relatively vacuous e-site is elaborated by a semantically richer elaborator. However, the existence of symmetric cases, as in Langacker [1991b:177], has led us to develop a model which allows for both symmetric and asymmetric valence relations.) Of course, the computational model of the superimposition process operates on lexical unit structures of the sort presented above. Its goal is to mirror natural superimposition as closely as possible. One possible output from the implementation could be a screen image which changes incrementally to show how superimposition proceeds when a morpheme stream is parsed (but it is still unclear whether we can reach that far during this project). The computational simulation of the superimposition process subdivides into a number of smaller processes on activated lexical units: Domain identification, Predication identification, Value adjustment, Part and Whole accommodation. Additionally, there are a number of alternative generation processes that alter the flow of the superimposition mechanism. It is a major goal in our project to investigate these processes and specify an invocation order between them.

CONCEPTUAL ENGINEERING

159

The first of these operations is called domain identification. It takes the matrices of the input units and finds those domains that are present in all input units, such as 3D space in Figure 2 above. For instance, if we superimpose [SALLY] onto [UNDER].TR, the matrices of the two units and the composite would be: [SALLY]: Emotional, Spatial, Age, Color, Profession, … [UNDER].TR: Spatial → [SALLY UNDER]: Spatial That is, in the composite only the spatial domain remains. [SALLY]: Emotional, Spatial, Age, Color, Profession, … [FURIOUS].TR: Emotional, Color → [SALLY FURIOUS]: Emotional, Color If [SALLY] is superimposed onto [FURIOUS].TR instead, the composite probably contains the emotional and color domains. [STONE]: Spatial, Material, Age, Color, … [FURIOUS].TR: Emotional, Color → [STONE FURIOUS]: Color In Holmqvist (1993), a mathematical function for the calculation of the composite matrix is discussed. Here it is sufficient to point out that the domain centralities play the main role. By reducing the number of domains in activated lexical units, domain identification helps in disambiguation, as the lower number of domains restricts the number of possible meanings.3 But domain identification also serves for anomaly detection. If only a non-central domain remains in the composite, as in the third example, then the two input units are too different to be superimposed, and an anomaly will then be signaled. Once the domains of the composite have been sorted out, the actual superimposition can start. In a common domain, each of the input units has a predication. These predications must now be joined into one, as in Figure 2. A number of operations combine to perform this welding task. Sometimes it is only a matter of seeing that two predications are both 1D-extended, as in the temporal domain examples of Figure 3.

160

KENNETH HOLMQVIST

tvelwe o´clock

a…ago

TIME

day

TIME

a…ago

TIME

TIME

SUPERIMPOSITION

NOT SUPERIMPOSABLE

TIME

Figure 3.

However, it often happens that two predications cannot be superimposed until one of them has been turned properly, as in Figure 4. the pole . .. ...

HORIS

HORIS

VERT

VERT

VERT

VERT

along

HORIS

SUPERIMPOSITION

HORIS

VERT

NOT SUPERIMPOSABLE

HORIS

Figure 4.

Also, the superimposition must map the dimensions in one predication onto the proper dimensions of the other predication. When there is only one dimension of the same type, there is no problem. See Figure 5. long

rope

Schema

Dimensions

Figure 5.

Extension

Extension Diameter Position(s) Direction

CONCEPTUAL ENGINEERING

161

But when there are several dimensions of that type in the other predication, as in Figure 6, another mapping principle must be adopted.

Schema

A C

Dimensions

Extension

B

Extension A: 2:nd largest value Extension B: largest value Extension C: least value Direction A: vertical ••

Figure 6.

The solution suggested in Holmqvist (1993) is based on the thorough investigation of the dimensional mapping principles constructed by Lang et al. (1991). Lang presents a complete model of spatial configurations based on how dimensional adjectives apply to them, and a part of our project is to translate Lang’s model into a model that can function as a domain-general mechanism for superimposition of configurations (Holmqvist 1995). The turning and tilting mechanisms of Figure 4 are derived from this work. For figure 6, the application of Lang’s model means that the extension dimension in [LONG] should map onto the extension dimension in [WALL] having the largest value. John walk.TR input level

part level

Figure 7.

snake

walk.TR

162

KENNETH HOLMQVIST

Let us now turn to Part accommodation. Even if we are successful in mapping dimensions, turning and adjusting the predications at the input (matrix) level, there are also predications to be superimposed at the part level. Figure 7 illustrates how an unproblematic superimposition of 1D-directed spatial predications at the input level is followed by different superimpositions of parts. Of these two cases, only the left one is successful: Every part of [WALK].TR is assigned a part in [JOHN]. There is no mismatch like there is between [SNAKE] and [WALK].TR. If one looks at a larger number of cases where parts are superimposed (cf. Holmqvist 1993), it seems as if superimposition of parts consists of two operations: first, each part in one lexical unit must be mapped onto a part in the other lexical unit; second, each such pair of parts must be superimposable. Thus, the legs and feet of [JOHN] must be attached at the same place and have the same directionality and extension as the legs and feet of the generic walker [WALK].TR, as well as having basically the same domains in their matrices. The wholes are in a way the opposites of the parts. Superimposing wholes in a similar way to that used above for superimposing parts is a mechanism for dealing with important but slightly subtle semantic effects. Take as an example the difference between “John used to row above the shoals” and “John used to row above the hills”.

row input level

parts at the whole level

Figure 8.

the shoals

CONCEPTUAL ENGINEERING

163

Figure 8 depicts the situation. [ROW] predicates a processual motion path, placed spatially above the [SHOAL] or [HILL] predications by [ABOVE]. At the input level, this works fine in both cases, because there [HILL] and [SHOAL] have the same predications. However, when we enter the whole level of each, they differ. A whole could be said to be the surrounding of each of its parts. The whole of [SHOAL] thus also has [(SEA-)WATER] as a part (i.e. [SHOAL] and [(SEA)WATER] are siblings), while [HILL] has [AIR] as a sibling. [ROW] has a number of sibling units, among them [(SEA-)WATER]. When [ROW], [ABOVE].TR and [SHOAL] superimpose, the [(SEA-)WATER] unit of [ROW] will coincide with the [(SEA-)WATER] unit of [SHOAL]. This is of course fine, since [(SEA-)WATER] easily superimposes onto itself. However, when [ROW], [ABOVE].LM and [HILL] superimpose, [(SEA-)WATER] will coincide with [AIR], and these two units do not superimpose easily. The basic principle here is the same as when parts were superimposed (Figure 7): When lexical units are mapped onto each other, it is done because they coincide in the composite structure. When two lexical units coincide, it must be possible to superimpose them into being one unit. To simulate this is a central goal in our implementation project. Along with the superimposition operations we have seen so far, Domain identification, Predication identification, Part and Whole accommodation, there are also a number of operations that alter the processing in various ways. They often appear after anomalies occur, but some seem to be invoked without having to be triggered by anomaly. In our project we mean to investigate the order of invocation between these semantic operations.

Figure 9.

164

KENNETH HOLMQVIST

One such operation is the ascent, which can be illustrated by the question what “it” in (1) refers to. (1)

I saw headlights coming straight at me, but I was able to get out of its way.

One answer would claim that it cannot be the headlights, because they superimpose badly onto [COME].TR. However, the whole [CAR] of [HEADLIGHTS] has the domains that allow it to superimpose onto [COME].TR. The ascent operation substitutes a lexical unit by its whole, as illustrated in Figure 9. Another altering operation is the split. The splitting operation appears in different situations, where there is need to split or divide a lexical unit. The hedge [FAKE] performs a split of type 2 by cutting off the processual wholes of units, as in Figure 10.

Figure 10.

The remaining meaning of a [GUN] to which [FAKE] has been applied is a gun of the same spatial form, material etc. and with the same parts (trigger, barrel etc.) as any gun. However, the fake gun will have no ability to shoot, because it lacks the processual wholes. There is no room here for further exemplification of the semantic operations for altering lexical units, but Holmqvist (1993) discusses several more: descent, the reversed ascent, forced installation of structure, predication scaling, correspondence reassignment and lexical revocation (which occur in reinterpretations such as Garden-Path sentences) and predication mapping metaphors.

CONCEPTUAL ENGINEERING

165

Let us finally point out that, in this process, there is no classical typechecking of the kind that the direct object of a verb must be a noun. An error like “Sam hit goes” is detected during domain identification because [GOES] has domains very dissimilar from the domains of [HIT].LM; the two units can not be superimposed easily.

4.

Valence suggestion and the schema population

In order to put the semantic operations into a parser, we need mechanisms for incoming morphemes, suggestion of valence relations and management of the composite constructional schemata. These mechanisms, drawn from Holmqvist (1993a), have already largely been implemented and further described in Holmqvist 1994. The input to the parser is a stream of morphemes. The morphemes may originate either from text or from speech, although at this stage we have not yet included phonetic information in the stream. It is important that the stream should be viewed as such, because the rate of incoming morphemes must be decided from the stream. It may not be the case that the parser decides the rate of incoming morphemes by what speed it is able to consume them at. The parser must consume morphemes at the external rate.4 Placing this simple demand on the time complexity of the parser has as a consequence that most present parser algorithms are excluded. One natural algorithm which satisfies the complexity demand is incremental parsing combined with interrupts caused by incoming morphemes. The algorithm that we are implementating can be described in the following way: (t0) A new morpheme just arrived. Interrupt all processing. (t1) Perform a generation shift. The result is that the processed lexical units up until the previous morpheme are kept in the previous generation. We set up a new generation for the current morpheme, in which we place only the lexical unit (with its parts, wholes, grammatical expectations etc.) evoked by the current morpheme. The important question about the nature of lexical evocation is being left open for the moment. Actually, at this point (June 1995) we do not even have a lexicon. We have tested this first stage of the implementation with random-generated semantic schemata (the only non-random structures were the grammatical expectations, which were entered along with the morpheme). (t2) The parser splits into two (pseudo-parallel) processes, of which the

166

KENNETH HOLMQVIST

first has been implemented: (A) Suggestion of new valence relations by grammatical expectations; (B) Accommodation of valence relations by means of simulated superimposition and, additionally, suggestion of new valence relations by semantic expectations. A suggested valence relation between two lexical units means that a new composite structure is entered into the current generation of the schema population, where it competes for space with a survivability value derived from how well it abides by grammatical and semantic constraints. Section 2 of this chapter discussed how activated lexical units are represented. They are the structures undergoing processing. Section 3 described parts of the B-process in step t2. We are now going to discuss the expectations that operate both in the A- and the B-processes. Their job is suggest valence relations, that is what (parts of) what units are identical to what (parts of) what units. Valence relations keep activated lexical units stuck to one another, that is they form composite structures. Now let us sketch the valence suggestion mechanisms. There are two of them: Semantic and grammatical expectations, the latter being a special case of the former.5 Semantic expectations suggest a valence relation between two lexical units if their predications coincide in one or more central domains. We have already seen examples of this during part and whole superimposition (Figures 7 and 8): Parts and sibling units of the two inputs were mapped onto each other if they coincided. The same kind of mapping seems to occur in many examples: In “John waded across the river”, the liquid part of [WADE]6 will coincide with [RIVER] and the two will therefore superimpose, the result being that in “John waded across the river” there is only one liquid, not two. As the examples show, an implementation of semantic expectations requires a previous implementation of the superimposition mechanism, in order for coinciding units to be possible to detect. Grammatical expectations are a special case of semantic expectations in that the semantic expectation of, for instance a TR elaborator (agent) for [GIVE] is coupled with additional information that the TR elaborator can be found leftward of the morpheme. Figure 11 shows a number of grammatical expectations stretched out in a morpheme stream. The hexagons mark the position in the stream where the expectations have been evoked. The rectangles mark positions where it is likely that unit

167

CONCEPTUAL ENGINEERING

can be found that suit the parts of the evocating lexical unit. For example, [MARY] seems proper to superimpose onto the LM1 (‘recipient’) part of [GIVE].

Morpheme stream order give <John>

<+s> <Mary> <who> <see> <+ed> +s who +ed

a



<some> some

<money>

see

Figure 11.

Grammatical expectations are “stretchable”. Finding the correct stretch is a crucial matter for suggesting the correct valence relations. The so-called Behaghel’s principle claims a correlation between closeness of morphemes and closeness in valence relations. Based on this principle, Holmqvist (1993) develops an incrementally calculable measure describing the behaghelian distance between morphemes and lexical units that we have now implemented, Holmqvist (1994). In parsing the morpheme stream of Figure 11, the distance from the evocating morpheme evolves as in Figure 12. There we can see that, at the shortest distances, we find [GIVES] (distance 0), <+s> (1), <Mary> (1), [MARY WHO] (1) and [MARY WHO SAW A HOUSE] (1). Distance from

4

[A HOUSE] 3 <who>

2

<see>

[SOME MONEY] [SAW]

<+s>

<some>

1 [GIVES] <John>



Figure 12.

<+s>

[MARY WHO]

[MARY WHO SAW A HOUSE]

<who> <see> <+ed> <money> Current morpheme <some> <Mary>

168

KENNETH HOLMQVIST

The process A of t1 below suggests all of those, and all other units as well, for valence relations to the LM1 of [GIVE]. The resulting composites are entered into the schema population, and a value for its grammatical obedience is calculated. For instance, the composite [(JOHN) GIVE(S) MARY WHO] gets a high value, because it establishes a valence relation at short behaghelian distance. The composite [(JOHN] GIVE(S) HOUSE], however, gets a low value for its establishment of a long-distance valence relation between [GIVE].LM1 and [HOUSE]. The precise calculation of behaghelian distance and the value for grammatical obedience can be found in Holmqvist (1994). However, once the value is calculated, it is used in the calculation of the survivability value of the composite. As the survivability value also draws on the semantic superimposability of the units, it weighs grammatical and semantic obedience against one another. The result is that the parser can not only make the correct grammatical assignments of valence relations for most morpheme streams, it will also be able to find many semantically correct but non-grammatical valence relations. All lexically evoked schemata and composites are kept in a schema population. Together with grammatical expectations and the parameters (distance, density and superimposition result), each entry in the schema population models corresponds to what Langacker (1991a) calls a grammatical construction. The schema population functions as a sort of semantic short-term memory, where composites are stored during composition (see Figure 13). Because of the large number of composites being suggested, the parser continuously evaluates them, using superimposition and behaghelian distance calculation. This evaluation is then used to sort the composites so as to always prune off those with low survivability, i.e. those that are unlikely to serve as a part in the final interpretation of the morpheme stream.

CONCEPTUAL ENGINEERING

169

Suggesting

Suggesting

Evaluating

Evaluating

Params

Pruned off because of low survivability

Params

Params

Params

Params

Params

Params

Params

Params

Params

Params

Params

Params

Params

Params

Figure 13.

5.

Project goals and current project work

The project ‘Conceptual Engineering’ has as its goal an experimental implementation of the model described above. In particular, we are interested in investigating the processual order between the superimposition mechanisms in the implementation, such as domain identification, predication mapping, ascent etc. It is also highly relevant that the valence suggesting mechanisms and the schema population have the proper processuality. During the first six months of our project, we have been implementing the morpheme stream and schema population mechanisms. Although we still have no lexicon and much of the semantics is randomly generated, the grammatical expectations combined with the calculation of behaghelian distance, with correspondence assignment but completely without type-checking of any sort, gave outputs for several morpheme streams where the composite with the highest survivability was composed of the correct valence relations. This mechanism then served as the frame within which implementations

170

KENNETH HOLMQVIST

of the superimposition mechanisms could be investigated. We implemented domain representation and the representation of configurations based on Lang’s work. Although we were running only on randomly generated lexicons, the program appeared to behave correctly in the domain and predication identification processes. In parallel with our implementation work, we also try to refine the basic model. At this stage we are looking at the specific input patterns (intonation, pauses, stress etc.) in spoken language. The idea is to reduce the number of extra calculations that have to be made because the behaghelian distance measure does not take phonetics into account. Also, superimposition should be less costly when we no longer work with contextually detached single sentences. We also hope that this work will allow us to specify an updating mechanism for the schema population that will also take the discourse structure into account (Holmqvist and Holsánová 1995).

Notes 1.

‘Conceptual Engineering’, sponsored by the Swedish Council for Research in the Humanities and Social Sciences for 1993 – 1996. Currently the author and Jana Holsánová are working in the project.

2.

Holmqvist (1993a) gives a survey of different predications and their dimensionality. For spatial form predications, Lang et. al. (1991) gives a similar and very systematic formalization, which can be translated into the more general framework of predications and region activation, see Holmqvist (1995).

3.

This connects to a number of questions concerning lexicon and lexically activated structures involving schema polysemy that have been left out here for the sake of simplicity.

4.

This means that the parser should have a roughly linear time complexity in terms of the number of incoming morphemes.

5.

Both of these are based on Langacker’s (1987) conceptual dependency. Harder (1993) suggests the existence of an additional functional dependency which could be developed to a similar valence suggestion mechanism.

6.

For the existence of this liquid part, cf. “John waded across the street”.

CONCEPTUAL ENGINEERING

171

References Glasgow, J.I. 1993 “The Imagery Debate Revisited: A Computational Perspective”, Computational Intelligence (Taking issue position paper), Vol. 9, No. 4, pp. 309333. Harder, P. 1993 “Cognition, Interaction and Structure”, presented at the workshop Cognitive Semantics. NAL 1993: Göteborg University Holmqvist, K. 1993 Implementing Cognitive Semantics. Lund: Department of Cognitive Science. Holmqvist, K. 1994 “Conceptual Engineering I: From morphemes to valence relations”. LUCS 28. Lund: Department of Cognitive Science. Holmqvist, K. 1995 “Two Dimensional Representations Compared”, manuscript, Lund: Department of Cognitive Science. Holmqvist, K., and Holsánová, J. 1997. “Focus Movements and the Internal Images of Spoken Discourse” in W.-A. Liebert, G. Redeker, L. Waugh. “Discourse and Perspective in Cognitive Linguistics”, Amsterdam, Philadelphia; John Benjamins. Lakoff, G. 1987 Women, Fire, and Dangerous Things, Chicago: University of Chicago Press. Lakoff, G. 1989 “Some Empirical Results about the Nature of Concepts”, Mind & Language. Vol. 4. Nos. 1 and 2 Spring/Summer 1989. 103–129. Lang, E. with Carstensen, K.-U. and Simmons, G. 1991 Modelling Spatial Knowledge on a Linguistic Basis. Heidelberg: SpringerVerlag. Langacker, R. 1987 Foundations of Cognitive Grammar, Vol. I. Stanford: Stanford University Press. Langacker, R. 1991a Foundations of Cognitive Grammar, Vol. II. Stanford: Stanford University Press. Langacker, R. 1991b Concept, Image, and Symbol, Berlin. New York: Mouton de Gruyter.

SITUATED EMBODIED SEMANTICS

AND

CONNECTIONIST MODELING

173

Situated Embodied Semantics and Connectionist Modeling Jordan Zlatev Department of Linguistics, Stockholm University

1.

Introduction

Possibly the most interesting claim of cognitive semantics is that concerning the embodiment of meaning. A succinct expression of this claim is the following. Thought is embodied, that is, the structures used to put together our conceptual systems grow out of bodily experience and make sense in terms of it; moreover, the core of our conceptual systems is directly grounded in perception, body movement, and experience of a physical and social character. (Lakoff 1987: xiv)

Lakoff (1987) as well as Johnson (1987) present evidence from categorization and metaphor that support the claim that meaning is embodied. However, relatively little has been said about how the required “grounding” is realized. So, unsurprisingly, there have been objections that the notion of “embodiment” is either hopelessly vague or boils down to empiricism, which has generally been considered discredited in linguistics since Chomsky’s (1959) famous attack on behaviorism. Recently, however, successes in the field of neural networks / connectionism (cf. Rumelhart & McClelland 1986, Hertz et al. 1991) have given rise to some enthusiasm that connectionist models would fill this vacuum and help to provide some account of the “grounding of concepts in experience”. According to Lakoff (1993), for example, the connectionist system of Regier (1992) yields a kind of embodied category structure. For Harnad (1993), on the other side, neural nets can at most perform the grounding of individual (basic) symbols

174

JORDAN ZLATEV

which need to be combined recursively before there can be anything resembling concepts. The result has been something of a debate between proponents of “pure connectionist models” and “hybrid connectionist / symbolic models” of the grounding of meaning. However, linking embodiment and connectionism directly is likely to lead to confusion. For one thing, connectionism covers a whole spectrum of models that vary on properties such as pre-structuring and biological realism. Secondly, and more importantly, connectionism is a new kind of technology and thus provides a new kind of metaphor for the mind; but it is not a theory about the mind and can be used even by conflicting theoretical standpoints. Therefore, I will take a step back and characterize independently from neural nets one particular approach to linguistic meaning and embodiment. For reasons that should become clear in Section 2, I will be referring to this approach as situated embodied semantics. In Section 3 I briefly describe Regier’s system, mentioned above, and use it as an illustration of (some aspects of) the situated embodied approach. Section 4 discusses how “the creativity of language”, i.e. the ability to utter and comprehend novel expressions, could be explained from this perspective, and in Section 5 I present what I think is the main reason for the shortcomings of the given connectionist model and sketch a direction for future research.

2.

Situated embodied semantics, grounding and neural nets

For the sake of reference, rather than with the ambition of introducing one more “semantic theory” into the field, I call the approach to explaining linguistic meaning that I am pursuing situated embodied semantics (cf. Zlatev 1997). In brief, it emphasizes the following aspects of human language. Language is situated, i.e. interwoven into the practices in which it is used; it takes place in the world, it is not just about the world, i.e. not a “picture of reality”. Situatedness requires interaction with an environment through sense and motor organs, i.e. embodiment. Learning a language is achieved through active participation in practices, “language games”, and much of this participation can be seen as “training”. Knowing a language is practical mastery (know-how), not the internalization of an underlying system (“la langue”, “competence”). To know the meaning of an expression is thus to be able to use it in appropriate situations.

SITUATED EMBODIED SEMANTICS

AND

CONNECTIONIST MODELING

175

An example may be in place. Imagine the “language game” of a parent playing with a child by requesting the child to pass a toy. The utterances of the parent, “Pass me the ball”, “Thank you” etc., are embedded in the activity of playing the game (“situatedness”). The child is learning to respond appropriately, i.e. understand the utterances, not by definitions and explanation, but through “participation”. If it were to pick up a tractor when the parent said “car”, the parent’s disapproving behavior would function as negative feedback. The words that the parent utters are not ostensively defined but rather used in the context of the game as a whole. Now, by playing the game for some time the child will pick up certain regularities: regularities between the utterances of the parent and the objects pointed to, between its own actions and the responses of the parent, between the utterances themselves, between the overall game and the kinds of steps taken within it, etc. It is the implicit awareness of these regularities that constitutes the “knowledge of language” of the child in this domain (and it is this awareness which I call “meaning”). It is practical, constitutes a “readiness-to-hand”, and having it in connection with an expression is tantamount to knowing how to use the latter. But “meaning” in this sense is nothing static: it changes constantly due to new uses in concrete situations. The influence over this view from the late Wittgenstein (1953) should be obvious. However, situated embodied semantics differs from more philosophically oriented Wittgenstein-inspired work (e.g., Segerdahl 1993) in actually attempting to answer some questions rather than “dissolving” them by exposing conceptual inconsistencies. Such are the questions concerning “grounding” that were brought up in Section 1, which can be given the following answers within the proposed approach. • What is “grounding”? The linking of linguistic expressions with the situations in which they are appropriate. • How is grounding realized? By using language in concrete situations, made meaningful through active perception and everyday activity, under conditions of social feedback and in a prolonged history of learning. To elaborate: The language learner is exposed to and required to respond with utterances in concrete situations. The social character of these situations implies positive or negative feedback depending on the appropriateness of the

176

JORDAN ZLATEV

learner’s behavior. In early language acquisition (at least), linguistic expressions are not given any other meaning than the situations (contexts-of-use) in which they are performed: there are no explicit “semantic representations” involved. During the learning process, however, there will inevitably be a process of categorization, so that the child will learn to associate utterance and situation types, rather than instances. The situations themselves are at least partially meaningful prior to language because of “embodiment”. Perception is not the passive recording of “sense data”, but active in the sense that relevant aspects of the environment are “brought forth”. As for activity, the emphasis on everydayness is due to the phenomenological observation that we immerse ourselves into habitual patterns of interaction which provide the necessary background for more intentional, purposive action (cf. Dreyfus 1991). A likely objection is that this is still, however, very vague and/or “behaviorist”. Much more has to be said about how the relevant features from the environment can be extracted and adequate categorizations achieved before situated embodied semantics can stand up against such accusations. It is for this purpose that I think neural nets can be used with success. Connectionism provides the following features that seem to fit very well with the situated embodied approach: neural nets deal with analog activation patterns, which is in unison with the continuous, open-ended character of the environment. They can generalize from particulars and display a form of categorical perception. Their adaptability makes them suitable for the study of learning processes. The supervision that a large class of models requires in training can be interpreted as deriving from social and environmental feedback. The “knowledge” that a neural net model has acquired, reflected in the pattern of connectivity and weight matrix, is more akin to practical mastery than to a “theory” or “grammar” in the generative sense.1 (Bates and Elman [1992] discuss a similar set of features.) The similarities between the situated embodied approach and neural nets should not, of course, blind us to the fact that the latter can only be regarded as very rough simplifications, since their “situatedness” is of a very contrived nature: they are spoon-fed through “training sets” rather than being actually involved in a physical, much less social, world. Nevertheless, I maintain that through connectionist modeling, we can address questions pertaining to the feasibility of situated embodied semantics such as the following:

SITUATED EMBODIED SEMANTICS

AND

CONNECTIONIST MODELING

177

• Is it possible to achieve adequate linguistic categorization of situations, without any explicit (symbolic) “semantic representations”? • What kind of pre-structuring (i.e., “innate bias”) is required for this to be achieved? These two questions will be the primary focus of the remaining part of this paper. I will deal with them in the context of a concrete connectionist model described in the following section.

3.

Experimenting with Regier’s system

Regier’s (1992) system, graphically presented in Figure 1, is a feed-forward structured connectionist net, using the backpropagation learning rule (Rumelhart et al. 1986). It takes perceptual input from its environment, processes it within a number of internal modules, and “learns” to associate the perceptual scenes with one or more output nodes by comparing the actual output with an externally given “correct” output and gradually minimizing the discrepancy. The environment the system interacts with is one of 2-dimensional geometric objects. The first major simplification with respect to embodiment is that this interaction takes the form of perception only, with no motor activity whatsoever. A second simplification is that the system “sees” only 2 objects at a time and furthermore, in order to avoid the empiricist problem of deciding which object is being related to which, one of the objects is explicitly labeled trajector (TR) and the other landmark (LM), in correspondence with the terminology from cognitive linguistics.2 A third simplification is the way situations in which there is motion are dealt with: as a sequence of “snapshots” or “frames” during which the landmark remains static, while the trajector varies its position with respect to it. A static scene is a special case where the frames in the sequence are identical. Regier refers to each of these sequences as a “movie”. I will be using the more general term situation since, despite limitations such as the above, I will be using the model as an (impoverished) illustration of the situated embodied approach. The internal modules contain considerable pre-structuring, which I do not intend to describe here, but it should be noted that their purpose is to extract relatively high-level perceptual features such as contact, inclusion,

178

JORDAN ZLATEV

Figure 1. Regier’s system. Input is presented as sequences of 2-dimensional pictures (“frames), processed and associated with labels for the sequences of frames. (Reprinted from Regier 1992 :36)

SITUATED EMBODIED SEMANTICS

AND

CONNECTIONIST MODELING

179

alignment with upright vertical (or some other orientation), direction of motion etc. As to the output nodes of the system, Regier viewed his model as one of lexical acquisition, so he made what might seem to be the most natural assumption: that every node corresponds to a certain word/morpheme in the language modeled. This is the interpretive scheme (a). (a) each situation associates with one output node; each output node corresponds to one word/morpheme. However, as described in (Zlatev 1992), a series of experiments with Regier’s system demonstrated that the model could learn to associate sets of situations (“training sets”) with single output nodes only when the situations within a set were not too different from each other, i.e. when they did not present an equivocal category, a category which can not (in any natural way) be defined through a set of necessary and jointly sufficient conditions for membership. For instance, consider the examples of over-situations presented in Figure 2. In examples 0-5 (see the upper left corner of each picture) the trajector is static and “hanging over” the landmark. In the next six examples (6-11) it is instead “hovering over”. In examples 12 and 13 as well as in 22-24 the trajector is in contact with the landmark and is “going over” or ”climbing over”. Finally in examples 14-21, the trajector seems to be “flying over” or “jumping over” the landmark. The training set that consisted of these examples could not be learned by Regier’s system under interpretive scheme (a). The system was having difficulty in forming equivocal categorizations. Put another way, it could not learn to use polysemous words such as “over” in an appropriate way. An easy way out of this problem is to change the interpretation of the output nodes: instead of corresponding to actual words, e.g. “over”, they could correspond to separate “senses” of words, e.g., OVER1, OVER2, OVER3 etc. Then one could separate the equivocal categories into univocal sub-categories and have the model learn to associate the sub-categories with the different “senses”. This yields the scheme (b). (b) each situation associates with one output node; each output node corresponds to one sense.

180

JORDAN ZLATEV

Figure 2. A training set of situations presented to Regier’s system as examples of over. The set could not be learned with the original architecture and interpretive scheme (a), cf. text.

Figure 3 illustrates this modified system, which indeed had no difficulty in performing the association task. However, this is achieved at the expense of what I have called “the subscript problem” (cf. Zlatev 1992). The point is that subscripts and other technical devices used to separate “senses” have no physical realization. For instance no one says “Go to the bank-sub-1 along the bank-sub-2”. Thus, they give no clue as to how a given “sense” is associated with the right kind of entity or situation during learning (cf. the “association line” between the sub-type 1 situation and OVER1 in Figure 3). Furthermore this misleading solution leads to another problem, which has tormented “natural

SITUATED EMBODIED SEMANTICS

AND

CONNECTIONIST MODELING

181

language processing” for the past few decades, namely “disambiguation”: Once the existence of separate senses is granted, comprehension implies finding the “right one”, which can be quite a formidable task (cf. Small et al. 1988).

"over"

"disambiguation" ?

OVERÿ 1

learning ?

OVER 2

OVER 3

Regier's system

sub-type 1

sub-type 2

sub-type 3

Figure 3. An illustration of “the subscript problem” as posed to Regier’s system if the output nodes are conceived of as separate “senses”: what provides the correct association between univocal sub-categories (“sub-types”) and “senses” during learning?

182

JORDAN ZLATEV

Notice that, according to this interpretation, Regier’s model is not a situated embodied one. The nodes representing “senses” constitute symbols, i.e. explicit formal entities, while, as discussed in Section 2, there are no explicit semantic representations according to the premises of situated embodied semantics. What is it then? I would call it a grounded symbolic model, to separate it from situated embodied models, but also from the various “pure” symbolic models where there is no sensorimotor grounding whatsoever. It is also interesting to observe that the “hybrid” models advocated by Harnad (1993), and mentioned in Section 1, seem to presuppose exactly such a model of “symbol grounding”.3 In this case they are faced with “the subscript problem”. If, instead, they are to ground the individual words of a language, as in interpretive scheme (a), then they must face the problem of equivocal categories. Thus they are confronted with a true dilemma. Situated embodied semantics suggests a straightforward way out of this dilemma. In describing the training set of over-examples in Figure 2, I could not help but use a verb in order to distinguish between the different kinds of situations. Given that the trajector and landmark objects are pre-given in Regier’s model, it appears that the obvious thing to do is to associate the situations not with single labels but with verb-preposition pairs (in a language like English). In this case the interpretation of the output nodes is the following. (c) each situation associates with two output nodes; each output node corresponds to one word. I have experimented with a number of small extensions to Regier’s system that follow this scheme. One of the most successful is presented in Figure 4. The pre-output hidden layer of Regier’s system (cf. Figure 1) feeds into a number of layer-segments, one for each word. Each layer segment for its part feeds forward to a single output node. High activation (> 0.8) of an output node is interpreted as the utterance of the corresponding word. Thus, in Figure 4 the particular situation is associated with the description “be over”, i.e. TR is over LM. According to this interpretation, the model performed very well on equivocal categories such as over and under. Furthermore, the “co-grounding” of the words in an utterance seems to (dis)solve many of the problems of “choosing the right sense” in the appropriate context. The context-dependence of word-meaning (i.e. polysemy) becomes a natural consequence of the fact that words are learned in different contexts. Once these contexts are activated,

SITUATED EMBODIED SEMANTICS

AND

CONNECTIONIST MODELING

183

the appropriate sense of an “ambiguous” word will emerge of itself. For example, when the semantic contribution of “go” is activated, “over” will be automatically constrained to its “contact, side-to-side sense”. In my opinion, this shows an empirical advantage of situated embodied semantics to grounded symbolic models which must ground their univocal, internal symbols on their own (with the problems this implies) before they can be combined according to formal rules. Therefore, I disagree with Plunkett et al. (1992, cf. below) on this point when, in their otherwise excellent article, they conclude that “there seems to be no empirical reason why such a [“hybrid”] theory of cognitive architecture could not adequately account for semantic development” (1992:309).

" be"

" go"

" f ly "

" ov e r "

" under "

la y e r s egments

C o m m o n h id d e n la y er

R e g ie r's s y st e m

Figure 4. Extending Regier’s system (cf. Figure 1) in order to associate verb-preposition pairs (instead of single labels) with situations. In this case, the “be” and “over” nodes become activated in response to the input situation.

184

JORDAN ZLATEV

However, the version of Regier’s system with which these experiments were performed has an obvious deficiency: there is nothing corresponding to the sequential nature of language in the activation patterns of the output nodes. For instance, the output pattern in Figure 4 can correspond to “be over” just as well as to “over be”. The easiest solution is to “map time onto space”: e.g., the highest activated of the verb-nodes is regarded as “first”, the highest activated preposition-node as “second”. But, as shown by, e.g., Elman (1990), this is not

"hover" "go" "fly"

"over

" "above" "under"

Regier's system

Figure 5. A situation which activates two verb nodes and two preposition nodes but which can be appropriately described with only three (not all four) of the verb-preposition pairs. This indicates the necessity of expressing the (temporal) interdependence among the output units (words).

SITUATED EMBODIED SEMANTICS

AND

CONNECTIONIST MODELING

185

a good solution. It fixes in advance the length of the sequence and misses interdependencies between the elements. Even at the two-word level and in the much simplified spatial domain under discussion, this can present itself as a problem. Figure 5, for example, shows how a given perceptual situation can lead to the activation of a number of output nodes, while not all corresponding verb-preposition combinations are appropriate: in this case “fly over” is not a good description. This deficiency could be seen as an instance of the “binding problem”, which arises when continuous activation patterns constituting “distributed representations” have to be associated with (connectionist implementations of) symbolic structures representing syntactic trees, thematic grids, etc. Many researchers (e.g., Smolensky 1990) consider the discovery of methods for performing this “binding” in a reliable and systematic way to be of vital importance for connectionism. Without intending to belittle such research, I think we should first consider if such symbolic structures are really necessary. Even though this seems to be a minority position held by relatively few people (e.g., Dorffner and Prem 1993), I regard it as a sound methodological principle in neural net modeling of language to assume as little symbolic structure as possible, and to leave as much room as possible for self-organization. This principle is especially valid in modeling situated embodied semantics. Admittedly, there is a need for structure, but I believe this can be derived from temporality itself. For example, there are means of “finding structure in time” by using a simple recursive network (Elman 1990, 1993). Instead of having a “verb slot” and a “preposition slot”, to which patterns for different verbs and prepositions will “bind”, we can have the output presented in different time steps, with the (pre-)output activation pattern of the previous step serving as “context”. I believe that this method can be extended well beyond the “two-word level”. (It will be a matter for future research to demonstrate this empirically.) In the meantime, I regard the “binding problem” as irrelevant to situated embodied semantics.

4.

Is “the creativity of language” a problem?

A more pressing concern is to see what situated embodied semantics can say about the ability to produce and comprehend novel linguistic expressions on the basis of expressions already learned. In linguistic parlance this ability is

186

JORDAN ZLATEV

often referred to as “the creativity of language” owing to the writings of Chomsky, who in the following characteristic passage brings up this feature and uses it as a weapon against theories that make recourse to some notion of “similarity” in explaining novel usage. The most striking aspect of linguistic competence is what we may call the ‘creativity of language´, that is, the speaker’s ability to produce new sentences that are immediately understood by other speakers although they bear no physical resemblance to sentences which are ‘familiar’. (Chomsky 1971 :8)

In the onslaught on behaviorism mentioned earlier, Chomsky is somewhat more explicit in his criticism, arguing for the necessity of a generative grammar as (part of) an explanation of “creativity”: It is easy to show that the new events that we accept and understand as sentences are not related to those with which we are familiar by any simple notion of formal (or semantic or statistical) similarity or identity of grammatical frame. Talk of generalization is entirely pointless and empty. It appears that we recognize a new item as a sentence not because it matches a familiar item in any simple way, but because it matches the grammar that each individual has somehow and in some form internalized. (Chomsky 1959 :56).

However, a generative grammar cannot in principle account for the understanding and producing of novel expressions since it is simply a syntactic calculus with recursive functions: a technical device that can at most separate the “grammatical” from the “ungrammatical sentences” and associate the first with syntactic analysis (and in more recent versions “logical form”). This makes Chomsky even less convincing when he in an off-hand way dismisses any appeal to similarity and generalization in the quotations above. A generative grammar can of course be complemented with a compositional semantic theory where the basic idea is that the meaning of the composite expression is a function of the meaning of the parts and their means of combination. Once we know the “meanings” of the words in a sentence, disregarding whether it is new or not, all we need to do is to combine them according to semantic rules that are parallel to the syntactic rules in order to derive the meaning of the sentence, i.e. “understand” it. Working out the details of such “a theory of meaning” has been a major task in more (formal) semantically oriented linguistics and analytical philosophy, often under the motto of “meaning-as-truth-conditions”. However, the compositional approach to explaining creativity has serious problems and below I list some of the foremost while I omit discussion

SITUATED EMBODIED SEMANTICS

AND

CONNECTIONIST MODELING

187

because of space limitation. • While it is necessary to identify the meanings of the “parts” before they can be combined in a “whole”, this cannot be done, since the meaning of words is open-ended and context-dependent (cf. section 3, Lakoff 1987, Anderson 1990). • Speaking and understanding take place against a background of shared practices, which cannot be fully spelled-out (cf. Winograd & Flores 1986, Dreyfus 1991). • Creativity involves the use of new expressions, which requires a social situation and feedback (cf. Wittgenstein 1953, Segerdahl 1993). These problems do not arise in the situated embodied approach, since “meaning” is an ability, not some kind of object, “mental” or other, and “extralinguistic knowledge” (the background of practices) combines naturally with this ability. As to “creativity”, it consists in the fact that we not only learn to associate an utterance type with a situation type, but (on the basis of what we have learned) can match novel utterances with novel situations. This ability can be seen as the result of processes such as generalization, analogy, and feedback. A novel situation would match with some familiar situation in some respects, and with another in others. These familiar situations are linked with corresponding expressions. The result would be a novel utterance. Chomsky’s criticisms are not (a priori) valid, since “sentences” are not “strings of words” but an integral part of a language game. Hence generalization and analogy include an infinitely richer domain than symbolic strings. During early learning, and occasionally thereafter, this process will not always be successful — use is not fully determined by meaning — but then social feedback would provide help. As an illustration of the above, let us look again at the “language game” from Section 2. Say there are two cars (one red, one blue) and two balls (one red, one blue). The child is requested to pass the “blue car” and after some help — it is still learning the color terms — it passes it over. The parent then asks for the “red ball” and again, after some help, the child passes the red ball. Next the parent asks for the “red car”, but does not offer any help. Will the child respond appropriately, given that it has never heard the expression before? It depends on how successful the “training” has been, i.e. on how well the child has generalized its ability to play the game. The new “move” of the game is bound to be compared to the old ones and the child is quite likely to

188

JORDAN ZLATEV

respond by giving the red car, in analogy to the red ball and the blue car. But, then again, the child may make a mistake, and the parent will have to correct it: “No, not the blue car, the red car” — and the process of training will continue. Once the child has mastered the distinctions in this game, it can generalize and use this ability in other games: it will be able to draw red suns and blue moons, but not because it has formed some “context-independent” concept of REDNESS, but because new uses constantly draw on old ones, in the accumulative ability to use language appropriately in ever new contexts. A number of experiments with the extended version of Regier’s system (cf. Figure 4) were performed in order to lend some empirical support for this view and against Chomsky’s “anti-generalization” position. The system was trained on some, but not all verb-preposition pairs, e.g., on the pairs “be over”, “be under”, “go over”, “go under” and “fly over”, but not on the sixth remaining pair “fly under”.4 The hypothesis was that from using “under” in two other contexts and “fly” in one other, the system would generalize and when presented with a novel kind of situation, a trajector flying under a landmark the system would (by analogy) “light up” the two appropriate nodes: “fly” and “under”. The first results were very encouraging. When presenting a set of 30 novel fly under-situations, we found that only 2 were classified as “go under”, all others were correctly classified as “fly under”. However, when testing the ability of the net to generalize to other novel word-pairs, e.g., “go over” (after training on all the rest), we found that the net performed less well — there were many overgeneralizations (“false alarms”) and even worse, many undergeneralizations (“misses”). A natural conclusion was that the environmental “evidence” was insufficient, i.e. the system needed to “see” a larger number of different over and under situations before it could extract a more adequate notion of the semantic contribution of the individual words. Therefore I gradually extended the training to include situations that can be described with other verbs and prepositions. The most extensive training session was performed for the verbs: “be”, “move”, “pass” and “fall” and the prepositions “over”, “under”, “in”, “through” and “on”. The choice of the words was motivated by following the naming of example “movies” by a native speaker of English. However, the results were negative: not only did extending the context of use not make the generalization better, it seemed to make it worse. In most cases a novel situation would be classified not with a novel expression but with the non-novel expression used to label some similar

SITUATED EMBODIED SEMANTICS

AND

CONNECTIONIST MODELING

189

situation. For example, a picture of a trajector being on a landmark would most often be classified as “being in” (both situations involving contact), moving in would be classified as “moving through” (both involving motion and inclusion), etc. So the experiments suggested that simply giving more environmental evidence would not make Regier’s system generalize to novel expressions. Does this lend support to Chomsky’s claim (quoted earlier) that “talk of generalization in this case is entirely pointless”? As I see it, there are two ways in which one can avoid this conclusion from a situated embodied semantics perspective. The first is to regard the improper linguistic categorizations of the novel situations performed by the network in analogy with the case (described earlier) when the child passes the blue car, when requested to pass the “red car” — as an indication that learning has not really been successful, but could be relatively easily improved with more feedback. Had the network performed miscategorizations only occasionally, and if the performance had improved with more evidence, I would have adopted this response. Unfortunately, as described above, this was not the case. Furthermore, the negative results were remarkably stable under a number of different simulations, with various parameters of the learning rule and configurations of hidden layers and number of units. The only conclusion can be that the simple extension of Regier’s system is inadequate as a model of “the creativity of language”. Accounting for the ability to use novel expressions through generalization from concrete examples and analogy would require more elaborate internal structuring than what has been provided.

5.

Conclusions

The questions concerning the feasibility of situated embodied semantics formulated in Section 2, which connectionist modeling was expected to address were, to repeat: • Is it possible to achieve adequate linguistic categorization of situations, without any explicit (symbolic) “semantic representations”? • What kind of pre-structuring (i.e., “innate bias”) is required for this to be achieved?

190

JORDAN ZLATEV

The experiments with Regier’s system, described in sections 3 and 4 have yielded a rather contradictory answer to the first question. On the one hand, the lack of symbolic representations provided an advantage in the treatment of polysemy. On the other hand, the net could not appropriately name novel situations by generalizing from familiar ones and thus failed as a model of “the creativity of language”. Is there a way to preserve the advantages of the situated embodied approach, while dealing with creativity? I believe so, and, in a sense, the fact that Regier’s system was not really capable of dealing with the task can even be seen as an advantage: this could lead to some insight regarding the question about “pre-structuring”. It seems that the main problem with Regier’s model is connected with the fact that it is a feed-forward net in which “situations” are given as input and “words” as output. In having the net adapt only on the basis of how well it succeeds in accomplishing this task, one excludes the possibility of modeling non-linguistic categorization.5 But, as pointed out in Section 2, one perceives and acts in the world even prior to and partially independent of language. This lends situations “meaning” and structure which would allow them to be more easily mapped to linguistic categories. It is interesting that it is exactly this ability and the existence of reentrant (in contrast to feed-forward) connections between the linguistic and conceptual centers of the brain that, e.g., Edelman (1992) claims to be the necessary prerequisites for the emergence of grammar and “true language”. Thus, to build syntax or the bases for grammar, the brain must have reentrant structures that allow semantics to emerge first (prior to syntax) by relating phonological symbols to concepts. [...] When a sufficiently large lexicon is collected, the conceptual areas of the brain categorize the order of such elements, an order that is stabilized in memory as syntax. (1992:130)

It is still an open question whether Edelman’s sketch of an “epigenetic theory” of the acquisition of language will withstand more detailed modeling. However, the ideas expressed in his book (and the connections he makes with cognitive semantics) seem intriguing. They are consistent with what I have said about situated embodied semantics, and connectionist modeling could provide a fruitful synthesis (despite Edelman’s summary dismissal of neural nets in the Postscript). Some initial steps in this direction have been taken by Plunkett et al. (1992) who have an “epigenetic approach to connectionism” (which is similar to the “situated embodied approach” presented here) and show some nice

SITUATED EMBODIED SEMANTICS

AND

CONNECTIONIST MODELING

191

results where the behavior of their net follows some of the steps in the learning of language by children. However, as in all kinds of modeling, they are forced to various kinds of simplifications. One is that they, as in another impressive connectionist model of “grounding” — that of Dorffner (1989), stop at the level of “single symbols”. The major conclusion from the research described in this paper is that situated embodied semantics and other similar approaches to meaning and language acquisition will stand or fall depending on whether they can provide a satisfactory account of how “symbols”, or rather words, are meaningfully combined.

Acknowledgments Though I am not sure that any of the people that I would like to acknowledge would readily associate themselves with “situated embodied semantics”, I feel I need to do so anyway. If it hadn’t been for Terry Regier, who generously left his system at my disposal and performed some of the experiments together with me in Berkeley and Stockholm, there would not have been any connectionist modeling to report on. Sarah Williams gave useful comments from a developmental perspective, which was the first impetus for me to look more seriously at child language acquisition - more so since this paper had taken form. Sven Öhman provided the impulse for most of the reasoning about “creativity” and was generally supportive despite disagreeing with my “solutions”. I also thank to the two anonymous referees for their insightful comments on form and content, which I have tried to address as best as I can, and to the many people who read and commented on drafts of this, perhaps a bit overedited, article.

Notes 1.

It is controversial to what extent neural nets also display “brain-likeness” and can, thus, help answer the question of how the mind is “embodied” also in the sense of “how it is realized in the human brain”. I will not deal with this question, but I basically agree with Sampson (1987) that “when models are as successful as connectionist models appear to be in predicting observable human behavior, they establish a claim to our attention, even if the mechanisms from which they are built seem a-priori entirely implausible” (ibid :873).

192

JORDAN ZLATEV

2.

Thus the picture at the bottom of Figure 1, which the system receives as input, should be described as, e.g., “The circle is above the rectangle.” not as “The rectangle is below the circle.”

3.

In Harnad’s “dualism” the connectionist/symbolic dichotomy roughly corresponds to perception/conception, while in the perhaps better known “hybrid” proposals of Pinker (e.g., Pinker and Prince 1988) the symbolic part is to model the “rules of language”, while the connectionist - the analogies. Harnad can be said to want to cut the cake horizontally, Pinker vertically.

4.

The difference in the training sets involving “go” and “fly” (somewhat arbitrarily decided on) was that the first but not the second involved contact between the objects.

5.

Note that it is not the technical aspect of “feed-forwardness” (nor the backpropagation learning rule) that is to blame, but the input/output interpretation of situations and expressions. Instead situations and expressions should be treated in parallel, e.g., as they are in the “autoassociative” feed-forward system of Plunkett et al. (1992).

References Anderson, R. 1990 “Inferences about word meanings”. Inferences and Text Comprehension ed. by In A. Graesser & G. Bower. New York: Academic Press, 1-16. Bates, E. & Elman, J. 1992 “Connectionism and the study of change”. Brain Development and Cognition: A Reader ed. by In M. Johnson, 623-642. Oxford: Blackwell Publishers. Chomsky, N. 1959 “A review of Skinner’s Verbal Behaviour”. Language 35. 26-58. Chomsky, N. 1971 Selected Readings, ed. by J.P.B. Allen & P. van Buren, Oxford: Oxford University Press. Dorffner, G. 1989 A sub-symbolic connectionist model of basic language functions. Dissertation, Indiana University, Computer Science Department. Dorffner, G. & Prem, E. 1993 “Connectionism, symbol grounding, and autonomous agents”. 15th Annual Meeting of the Cognitive Science Society Symposium on “Grounding, Situatedness, and Meaning”. Boulder, June 1993. Dreyfus, H. 1991 Being-in-the-World. A Commentary on Heidegger’s “Being and Time, Division I”. Cambridge, Mass: MIT Press. Edelman, G. 1992 Bright Air, Brilliant Fire, On the Matter of the Mind. New York: Basic Books, Harper Collins Publications.

SITUATED EMBODIED SEMANTICS

AND

CONNECTIONIST MODELING

193

Elman, J. 1990 “Finding structure in time”. Journal of Cognitive Science 14. 179-211. Elman, J. 1993. “Learning and development in neural networks: the importance of starting small”. Cognition 48. 71-99. Harnad, S. 1993 “Symbol grounding is an empirical problem: neural nets are just a candidate component”. 15th Annual Meeting of the Cognitive Science Society Symposium on “Grounding, Situatedness, and Meaning”. Boulder, June 1993. Hertz, J., Krogh, A. & Palmer, R. 1991 Introduction to the Theory of Neural Computation. Redwood City: Addison-Wesley. Johnson, M. 1987 The Body in the Mind. Chicago: University of Chicago Press. Lakoff, G. 1987 Women, Fire, and Dangerous Things. Chicago: University of Chicago Press. Lakoff, G. 1993 “Grounded concepts without symbols”. 15th Annual Meeting of the Cognitive Science Society Symposium on “Grounding, Situatedness, and Meaning”. Boulder, June 1993. Pinker, S. & Prince, A. 1988 “On language and connectionism: Analysis of a parallel distributed processing model of language acquisition”. Cognition 28. 73-193. Plunkett, K., Sinha, C., Møller, M. & Strandsby, O. 1992 “Symbol grounding or the emergence of symbols? Vocabulary growth in children and in a connectionist net”. Connection Science, Vol 4, 3 & 4. 293312. Regier, T. 1992 The Acquisition of Lexical Semantics for Spatial Terms: A Connectionist Model Of Perceptual Categorisation. Ph.D. Thesis. TR-92-062. International Computer Science Institute. University of California, Berkeley. Rumelhart, D., McClelland, J. & the PDP Research Group. 1986 Parallel Distributed Processing, Explorations in the Microstructure of Cognition, Volume 1: Foundations. Cambridge: MIT Press. Rumelhart, D., Hinton, G. & Williams, R. 1986 “Learning internal representations by error propagation”. Parallel Distributed Processing, Explorations in the Microstructure of Cognition, Volume II ed. by J. McClelland & D. Rumelhart. 318-362. Cambridge: MIT Press. Sampson, G. 1987 “Review article of Parallel Distributed Processing, Explorations in the Microstructure of Cognition” Language, Vol 63 4. 871-886. Segerdahl, P. 1993 Language Use, A Philosophical Investigation into the Basic Notions of Pragmatics. Ph.D. Thesis. Uppsala University, Dept. of Philosophy.

194

JORDAN ZLATEV

Small, S., Cottrell, G. & Tanenhaus, M. eds 1988 Lexical Ambiguity Resolution in the Comprehension of Human Language. Los Altos: Morgan Kaufmann. Smolensky, P. 1990 “Tensor product variable binding and the representation of symbolic structures in connectionist systems”. Artificial Intelligence Vol. 46, 1 & 2. 159216. Winograd, T. & Flores, F. 1986 Understanding Computers and Cognition. Norwood, N.J.: Ablex Publishers. Wittgenstein, L. 1953 Philosophical Investigations. Oxford: Basil Blackwell. Zlatev, J. 1992 A Study of Perceptually Grounded Polysemy in a Spatial Microdomain. TR-92-048. International Computer Science Institute. Berkeley, California. Zlatev, J. 1997 Situated Embodiment: Studies in the emergence of spatial meaning. PhD. Thesis, Stockholm University, Dept. of Linguistics.

INDEX OF NAMES

195

Index of Names

A Aijmer, K. 96, 98 Allwood, J. viii, x, 1, 3, 6, 7, 47 Altenberg, B. 96 Anderson, R. 187 Aristotle 4, 25, 38 Atkins, B. 98 Austin, J.R. 47 B Barwise, J. and Perry, J. 4 Bates, E. and Elman, J. 176 Behaghel, O. 167 Benveniste, E. 133 Berlin, B. and Kay, P. 6 Brennan, M. 141 Bring, S.C. 71 Burge, T. 27 Bühler, K. 47 Bybee, J. 146 C Chomsky, N. 24, 173, 186, 187, 188 Clark, E. 68 Clark, H. 133, 134, 135 Cogen, C. 141 Comrie, B. 149 D Desclés, J.-P. 21 Dik, S.C. 37, 46, 48, 57 Dorffner, G. 191 Dreyfus, H. 176, 187

E Edelman, G. 39, 190 Elman, J. 184, 185 Engberg-Pedersen, E. ix, 137, 139 F Fauconnier, G. 21 Ferreira Brito, L. 141 Fillmore, C. 16, 98, 133, 134, 136 Fleischman, S. 134 Foley, W.A. and Van Valin, R.D. 46, 47, 48 Fortescue, M. 148 Friedman, L.A. 141 G Gärdenfors, P. viii, x, 22, 23, 26, 27, 32, 39, 50 Geeraerts, D. 41 Gellerstam, M. 98 Gibson, J.J. 132, 144 Givón, T. 38, 60 Glasgow, J.I. 154 H Haiman, J. 2 Halliday, M.A.K. 47 Harder, P. viii, 47, 49, 52 Harnad, S. 173, 182, 192 (note) Hengeveld, K. 46, 48, 49 Hertz, J., Krogh, A. and Palmer, R. 173 Hill, C.A. 134 Hollerbach, J. 94 Holmqvist, K. ix, 24, 32, 154, 159, 161, 164, 165, 167, 170

196

INDEX OF NAMES

Husserl, E. 4 J Jackendoff, R. 21 Jespersen, O. 53 Johansson, M. 96 Johnson, M. 23, 67, 95, 105, 131, 136, 173 Johnson-Laird, P.N. 90 K Kakumasu, J. 141 Kant, I. 4 Koolhof, C. 141 Kripke, S. 19 Kuschel, R. 141 L Labov, W. 147 Lakoff, G. 6, 21, 67, 105, 127, 131, 135, 136, 148, 149, 154, 173, 187 Lakoff, G. and Johnson, M. 23, 67, 135, 136 Lang, E. 24, 161 Langacker, R.W. 2, 21, 23, 30, 41, 43, 44, 45, 51, 54, 55, 56, 59, 143, 144, 146, 147, 153, 154, 157, 168 Leech, G. 47 Leslie, A. 95 Liddell, S.K. 141 Linde, C. 147 M Michotte, A. 95 Miller, G.A. 90 Millikan, R. 38, 56 Montague, R. 20 Moxey, L.M. and Sanford, A.J. 32 P Pagliuca, W. 146 Perkins, R. 146 Petitot-Cocorda, J. 21 Pinker, S. and Prince, A. 192 (note) Plunkett, K. et al. 183, 190, 192(note)

Popper, K. 42 Putnam, H. 27-29, 32 Q Quine, W.V.O. 2 R Regier, T. 173, 177ff, 188ff Rijkhoff, J. 47 Rommetveit, R. 1 Rosch, E. 6 Rudzka-Ostyn, B. 98 Rumelhart, D.E. and McClelland, J.L. 73 S Sampson, G. 191(note) Saussure, F. de vii, 20 Schank 61 Schermer, T. 141 Searle, J.R. 38, 47 Segerdahl, P. 175, 187 Siewierska, A. 47 Sjöström, S. viii Small, S., Cottrell, G. and Tanenhaus, M. 181 Smolensky, P. 185 Spelke, E. 103 Stalnaker, R. 25 Strawson, P.F. 52 Strömberg, A. 72 Sweetser, E. 21, 30, 67, 68, 69, 131, 145 T Talmy, L. 21, 30, 95 Taylor, J. 145 Tomlin, R.S. 60 Traugott, E.C. 136, 138, 149 U Ulbæk, I. 42, 50 V Viberg, Å. viii, 69, 87, 110, 117

INDEX OF NAMES W Winograd, T. and Flores, F. 187 Winter, S. and Gärdenfors, P. 30 Wittgenstein, L. 175, 187 Wright, L. 38

Z Zlatev, J. ix, x, 174, 179

197

INDEX OF SUBJECTS

199

Index of Subjects

A abstract 132 Abstract agent 11 abstraction 6, 14, 15 accommodation 157 activation potential 2 active participation 174 actual meaning 2 adaptability 176 adversative construction 15 anchoring in time and space 3 anomaly detection 159 B backpropagation learning rule 177 bank of Swedish 98 basic categories 5 basic level 92 basic semantic-epistemic categories 16 biology 38 C categorematic 2 categorical perception 176 causality 6 choice of subject 102 co-grounding 182 Cognitive Grammar 37, 154 cognitive linguistics 177 cognitive operations 2, 6 Complex State 5 compositional semantic theory 186 comprehension 113 computer implementation 24, 153 conceptual (semantic-epistemic) operations 16

conceptual dependence 58 conceptual or cognitive operations 1 conceptual role semantics 29 conceptual roles 8 conceptual space 22, 23, 25, 26, 32, 33 conceptual structure 3 concreteness 103 conducting vehicle 15 conflict 9, 12 connectionism 173, 174, 176, 185, 190 Container constructions 15 contrast class 26 core predicate 90 corpus of bilingual texts 96 course of events 5, 13 creativity of language 185, 186, 187, 189 D deixis 53, 136-141 Derived categories 5 diachronic 145 disambiguation 159 disconnection 110 division of linguistic labor 28, 29, 33 domains 23, 131, 143, 155, 159 E effected object 108 effort 96 embodiment of meaning 173, 174, 177 encyclopedic’ information 2, 116 entity 4, 5 environment 177 epigenetic approach to connectionism 190

200

INDEX OF SUBJECTS

epistemic modals 30 equivocal categories 179 events 144 evolution 39 expectations 30 experience 6 extensional defined semantics 19 external cause of perception 70 F feed-forward structured connectionist net 177 field-independent components 90 field-specific components 90 focusing 100 force dynamics 30, 95 functional dependence 58 Functional Grammar 37 G geometric 22 grammatical expectations 166 greetings 42 grounded symbolic model 182 grounding 173, 175, 191 H hedge 28 hierarchical structure 92 holophrastic 50 hybrid models 182 hyponyms 92 I ideal types 6 image schemata 23, 24, 25, 154 image superimposition 157 implicit relation 14 innate predispositions 16 intensional semantics 19, 20, 26, 30, 32, 33, 94 interpretive scheme 179 J Joint venture construction 15

L landmark 177 language game 175, 187 layered clause structure 37 “lexical” information 2 lexical meaning 20, 32 lexicon 25 linguistic categorization of situations 189 linguistic constructions 16 linguistic power 28, 30 linguistic support 16 localization 11 M matrix 155 meaning potentials 1, 2, 3, 13 Mentalese 22 Metaphorical mapping 131 metaphors 23, 32 metonymy 23 modal expressions 30, 31, 32 modality 30 N naive physics 106 natural-kind terms 29 negation 55 neural networks 173 non-linguistic categorization 190 nuclear verbs 93 O object-centered motion 94, 107 operations on information 1 operators 58 over 179 P parts 155 peace 9, 12 perception 6, 21, 32, 144 point of view 136, 138 polysemic structure 13 polysemous words 179 possible individual 26

INDEX OF SUBJECTS possible worlds 19, 21, 26, 31 postural verb 113 practical mastery 174 predicates 155 predications 159 process/product ambiguity 12 product 12 properties 3, 4, 5 property of the perceived 70 prototype 6, 25 prototypical causation 127 Q quality dimensions 22, 33 Quantity, modality and evaluation 3 R reference 139, 141 referents 141 Regier’s system 184, 190 reification 2, 6, 13-15 relation of visual perception 70 relation/process 7 relational approach 90 Resultative strengthening 106 role relations 16 role types 11 roots 6 S scope 59 self-organization 185 self-propelled motion 103 semantic class of the object 114 semantic classes of subject 103 Semantic expectations 166 semantic field 69, 87, 90

201

semantic interpretation 113 semantic-epistemic categories 3, 4, 9, 12 semantic-epistemic operations 2, 3 sign languages 132, 138 simple recursive network 185 situated embodied semantics 174, 189 social meaning 21, 27, 28, 29 social power 31 social structure of language 27 Sound-source verbs 117 static-dynamic 138, 141, 143, 145, 147, 148 static events 132 stems 6 stereotypes 6 Subject-centered motion verbs 94 syncategorematic 2 T The Bank of Swedish 98 time complexity 165 time lines 133, 136, 137, 139 top-down 57 topological structures 22 trajector 177 translation 96 truth conditions 19, 21, 33 types 176 typification 2, 3, 6 U univocal sub-categories 179 unpacking 13, 15 V verb-preposition pairs 188 verbs of physical contact 87

In the PRAGMATICS AND BEYOND NEW SERIES the following titles have been published thus far or are scheduled for publication: 1. WALTER, Bettyruth: The Jury Summation as Speech Genre: An Ethnographic Study of What it Means to Those who Use it. Amsterdam/Philadelphia, 1988. 2. BARTON, Ellen: Nonsentential Constituents: A Theory of Grammatical Structure and Pragmatic Interpretation. Amsterdam/Philadelphia, 1990. 3. OLEKSY, Wieslaw (ed.): Contrastive Pragmatics. Amsterdam/Philadelphia, 1989. 4. RAFFLER-ENGEL, Walburga von (ed.): Doctor-Patient Interaction. Amsterdam/Philadelphia, 1989. 5. THELIN, Nils B. (ed.): Verbal Aspect in Discourse. Amsterdam/Philadelphia, 1990. 6. VERSCHUEREN, Jef (ed.): Selected Papers from the 1987 International Pragmatics Conference. Vol. I: Pragmatics at Issue. Vol. II: Levels of Linguistic Adaptation. Vol. III: The Pragmatics of Intercultural and International Communication (ed. with Jan Blommaert). Amsterdam/Philadelphia, 1991. 7. LINDENFELD, Jacqueline: Speech and Sociability at French Urban Market Places. Amsterdam/Philadelphia, 1990. 8. YOUNG, Lynne: Language as Behaviour, Language as Code: A Study of Academic English. Amsterdam/Philadelphia, 1990. 9. LUKE, Kang-Kwong: Utterance Particles in Cantonese Conversation. Amsterdam/Philadelphia, 1990. 10. MURRAY, Denise E.: Conversation for Action. The computer terminal as medium of communication. Amsterdam/Philadelphia, 1991. 11. LUONG, Hy V.: Discursive Practices and Linguistic Meanings. The Vietnamese system of person reference. Amsterdam/Philadelphia, 1990. 12. ABRAHAM, Werner (ed.): Discourse Particles. Descriptive and theoretical investigations on the logical, syntactic and pragmatic properties of discourse particles in German. Amsterdam/Philadelphia, 1991. 13. NUYTS, Jan, A. Machtelt BOLKESTEIN and Co VET (eds): Layers and Levels of Representation in Language Theory: a functional view. Amsterdam/Philadelphia, 1990. 14. SCHWARTZ, Ursula: Young Children’s Dyadic Pretend Play. Amsterdam/Philadelphia, 1991. 15. KOMTER, Martha: Conflict and Cooperation in Job Interviews. Amsterdam/Philadelphia, 1991. 16. MANN, William C. and Sandra A. THOMPSON (eds): Discourse Description: Diverse Linguistic Analyses of a Fund-Raising Text. Amsterdam/Philadelphia, 1992. 17. PIÉRAUT-LE BONNIEC, Gilberte and Marlene DOLITSKY (eds): Language Bases ... Discourse Bases. Amsterdam/Philadelphia, 1991. 18. JOHNSTONE, Barbara: Repetition in Arabic Discourse. Paradigms, syntagms and the ecology of language. Amsterdam/Philadelphia, 1991. 19. BAKER, Carolyn D. and Allan LUKE (eds): Towards a Critical Sociology of Reading Pedagogy. Papers of the XII World Congress on Reading. Amsterdam/Philadelphia, 1991. 20. NUYTS, Jan: Aspects of a Cognitive-Pragmatic Theory of Language. On cognition, functionalism, and grammar. Amsterdam/Philadelphia, 1992. 21. SEARLE, John R. et al.: (On) Searle on Conversation. Compiled and introduced by Herman Parret and Jef Verschueren. Amsterdam/Philadelphia, 1992.

22. AUER, Peter and Aldo Di LUZIO (eds): The Contextualization of Language. Amsterdam/Philadelphia, 1992. 23. FORTESCUE, Michael, Peter HARDER and Lars KRISTOFFERSEN (eds): Layered Structure and Reference in a Functional Perspective. Papers from the Functional Grammar Conference, Copenhagen, 1990. Amsterdam/Philadelphia, 1992. 24. MAYNARD, Senko K.: Discourse Modality: Subjectivity, Emotion and Voice in the Japanese Language. Amsterdam/Philadelphia, 1993. 25. COUPER-KUHLEN, Elizabeth: English Speech Rhythm. Form and function in everyday verbal interaction. Amsterdam/Philadelphia, 1993. 26. STYGALL, Gail: Trial Language. A study in differential discourse processing. Amsterdam/ Philadelphia, 1994. 27. SUTER, Hans Jürg: The Wedding Report: A Prototypical Approach to the Study of Traditional Text Types. Amsterdam/Philadelphia, 1993. 28. VAN DE WALLE, Lieve: Pragmatics and Classical Sanskrit. Amsterdam/Philadelphia, 1993. 29. BARSKY, Robert F.: Constructing a Productive Other: Discourse theory and the convention refugee hearing. Amsterdam/Philadelphia, 1994. 30. WORTHAM, Stanton E.F.: Acting Out Participant Examples in the Classroom. Amsterdam/Philadelphia, 1994. 31. WILDGEN, Wolfgang: Process, Image and Meaning. A realistic model of the meanings of sentences and narrative texts. Amsterdam/Philadelphia, 1994. 32. SHIBATANI, Masayoshi and Sandra A. THOMPSON (eds): Essays in Semantics and Pragmatics. Amsterdam/Philadelphia, 1995. 33. GOOSSENS, Louis, Paul PAUWELS, Brygida RUDZKA-OSTYN, Anne-Marie SIMONVANDENBERGEN and Johan VANPARYS: By Word of Mouth. Metaphor, metonymy and linguistic action in a cognitive perspective. Amsterdam/Philadelphia, 1995. 34. BARBE, Katharina: Irony in Context. Amsterdam/Philadelphia, 1995. 35. JUCKER, Andreas H. (ed.): Historical Pragmatics. Pragmatic developments in the history of English. Amsterdam/Philadelphia, 1995. 36. CHILTON, Paul, Mikhail V. ILYIN and Jacob MEY: Political Discourse in Transition in Eastern and Western Europe (1989-1991). Amsterdam/Philadelphia, 1998. 37. CARSTON, Robyn and Seiji UCHIDA (eds): Relevance Theory. Applications and implications. Amsterdam/Philadelphia, 1998. 38. FRETHEIM, Thorstein and Jeanette K. GUNDEL (eds): Reference and Referent Accessibility. Amsterdam/Philadelphia, 1996. 39. HERRING, Susan (ed.): Computer-Mediated Communication. Linguistic, social, and cross-cultural perspectives. Amsterdam/Philadelphia, 1996. 40. DIAMOND, Julie: Status and Power in Verbal Interaction. A study of discourse in a closeknit social network. Amsterdam/Philadelphia, 1996. 41. VENTOLA, Eija and Anna MAURANEN, (eds): Academic Writing. Intercultural and textual issues. Amsterdam/Philadelphia, 1996. 42. WODAK, Ruth and Helga KOTTHOFF (eds): Communicating Gender in Context. Amsterdam/Philadelphia, 1997. 43. JANSSEN, Theo A.J.M. and Wim van der WURFF (eds): Reported Speech. Forms and functions of the verb. Amsterdam/Philadelphia, 1996. 44. BARGIELA-CHIAPPINI, Francesca and Sandra J. HARRIS: Managing Language. The

discourse of corporate meetings. Amsterdam/Philadelphia, 1997. 45. PALTRIDGE, Brian: Genre, Frames and Writing in Research Settings. Amsterdam/Philadelphia, 1997. 46. GEORGAKOPOULOU, Alexandra: Narrative Performances. A study of Modern Greek storytelling. Amsterdam/Philadelphia, 1997. 47. CHESTERMAN, Andrew: Contrastive Functional Analysis. Amsterdam/Philadelphia, 1998. 48. KAMIO, Akio: Territory of Information. Amsterdam/Philadelphia, 1997. 49. KURZON, Dennis: Discourse of Silence. Amsterdam/Philadelphia, 1998. 50. GRENOBLE, Lenore: Deixis and Information Packaging in Russian Discourse. Amsterdam/Philadelphia, 1998. 51. BOULIMA, Jamila: Negotiated Interaction in Target Language Classroom Discourse. Amsterdam/Philadelphia, 1999. 52. GILLIS, Steven and Annick DE HOUWER (eds): The Acquisition of Dutch. Amsterdam/ Philadelphia, 1998. 53. MOSEGAARD HANSEN, Maj-Britt: The Function of Discourse Particles. A study with special reference to spoken standard French. Amsterdam/Philadelphia, 1998. 54. HYLAND, Ken: Hedging in Scientific Research Articles. Amsterdam/Philadelphia, 1998. 55. ALLWOOD, Jens and Peter Gärdenfors (eds): Cognitive Semantics. Meaning and cognition. Amsterdam/Philadelphia, 1999. 56. TANAKA, Hiroko: Language, Culture and Social Interaction. Turn-taking in Japanese and Anglo-American English. Amsterdam/Philadelphia, 1999. 57 JUCKER, Andreas H. and Yael ZIV (eds): Discourse Markers. Descriptions and theory. Amsterdam/Philadelphia, 1998. 58. ROUCHOTA, Villy and Andreas H. JUCKER (eds): Current Issues in Relevance Theory. Amsterdam/Philadelphia, 1998. 59. KAMIO, Akio and Ken-ichi TAKAMI (eds): Function and Structure. In honor of Susumu Kuno. 1999. 60. JACOBS, Geert: Preformulating the News. An analysis of the metapragmatics of press releases. 1999. 61. MILLS, Margaret H. (ed.): Slavic Gender Linguistics. 1999. 62. TZANNE, Angeliki: Talking at Cross-Purposes. The dynamics of miscommunication. 2000. 63. BUBLITZ, Wolfram, Uta LENK and Eija VENTOLA (eds.): Coherence in Spoken and Written Discourse. How to create it and how to describe it.Selected papers from the International Workshop on Coherence, Augsburg, 24-27 April 1997. 1999. 64. SVENNEVIG, Jan: Getting Acquainted in Conversation. A study of initial interactions. 1999. 65. COOREN, François: The Organizing Dimension of Communication. 2000. 66. JUCKER, Andreas H., Gerd FRITZ and Franz LEBSANFT (eds.): Historical Dialogue Analysis. 1999. 67. TAAVITSAINEN, Irma, Gunnel MELCHERS and Päivi PAHTA (eds.): Dimensions of Writing in Nonstandard English. 1999. 68. ARNOVICK, Leslie: Diachronic Pragmatics. Seven case studies in English illocutionary development. 1999.

69. NOH, Eun-Ju: The Semantics and Pragmatics of Metarepresentation in English. A relevance-theoretic account. 2000. 70. SORJONEN, Marja-Leena: Recipient Activities Particles nii(n) and joo as Responses in Finnish Conversation. n.y.p. 71. GÓMEZ-GONZÁLEZ, María Ángeles: The Theme-Topic Interface. Evidence from English. 2001. 72. MARMARIDOU, Sophia S.A.: Pragmatic Meaning and Cognition. 2000. 73. HESTER, Stephen and David FRANCIS (eds.): Local Educational Order. Ethnomethodological studies of knowledge in action. 2000. 74. TROSBORG, Anna (ed.): Analysing Professional Genres. 2000. 75. PILKINGTON, Adrian: Poetic Effects. A relevance theory perspective. 2000. 76. MATSUI, Tomoko: Bridging and Relevance. 2000. 77. VANDERVEKEN, Daniel and Susumu KUBO (eds.): Essays in Speech Act Theory. n.y.p. 78. SELL, Roger D. : Literature as Communication. The foundations of mediating criticism. 2000. 79. ANDERSEN, Gisle and Thorstein FRETHEIM (eds.): Pragmatic Markers and Propositional Attitude. 2000. 80. UNGERER, Friedrich (ed.): English Media Texts – Past and Present. Language and textual structure. 2000. 81. DI LUZIO, Aldo, Susanne GÜNTHNER and Franca ORLETTI (eds.): Culture in Communication. Analyses of intercultural situations. n.y.p. 82. KHALIL, Esam N.: Grounding in English and Arabic News Discourse. 2000. 83. MÁRQUEZ REITER, Rosina: Linguistic Politeness in Britain and Uruguay. A contrastive study of requests and apologies. 2000. 84. ANDERSEN, Gisle: Pragmatic Markers and Sociolinguistic Variation. A relevance-theoretic approach to the language of adolescents. 2001. 85. COLLINS, Daniel E.: Reanimated Voices. Speech reporting in a historical-pragmatic perspective. 2001. 86. IFANTIDOU, Elly: Evidentials and Relevance. n.y.p. 87. MUSHIN, Ilana: Evidentiality and Epistemological Stance. Narrative Retelling. n.y.p. 88. BAYRAKTAROGLU, Arin and Maria SIFIANOU (eds.): Linguistic Politeness Across Boundaries. Linguistic Politeness Across Boundaries. n.y.p. 89. ITAKURA, Hiroko: Conversational Dominance and Gender. A study of Japanese speakers in first and second language contexts. n.y.p. 90. KENESEI, István and Robert M. HARNISH (eds.): Perspectives on Semantics, Pragmatics, and Discourse. A Festschrift for Ferenc Kiefer. 2001. 91. GROSS, Joan: Speaking in Other Voices. An ethnography of Walloon puppet theaters. n.y.p. 92. GARDNER, Rod: When Listeners Talk. n.y.p. 93. BARON, Bettina and Helga KOTTHOFF (eds.): Gender in Interaction. Perspectives on feminity and masculinity in ethnography and discourse. n.y.p.