Cognitive Semantics : Meaning and Cognition

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: andreas.jucker@anglistik.uni-giessen.de 

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 vii 

Semantics as Meaning Determination with Semantic-Epistemic 

Operations 1 

Jens Allwood 

Some Tenets of Cognitive Semantics 19 

Peter Gärdenfors 

Function, Cognition, and Layered Clause Structure 37 

Peter Harder 

From Vision to Cognition. A study of metaphor and polysemy in 

Swedish 67 

Sören Sjöström 

Polysemy and Differentiation in the Lexicon 87 

Åke Viberg 

Space and Time 131 

Elisabeth Engberg-Pedersen 

Conceptual Engineering 153 

Kenneth Holmqvist 

Situated Embodied Semantics and Connectionist Modeling 173 

Jordan Zlatev 

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 

1. Introduction 

Jens Allwood 

Department of Linguistics, Göteborg University 

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


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. Basic semantic-epistemic categories 

ii. Basic conceptual structure 

iii. Anchoring in time and space 

iv. Relations 

v. Processes 

vi. Roles derived from relations and processes 

vii. Properties 

viii. Quantity, modality and evaluation 

Below I will now try to illustrate the approach by discussing two types of 

operations: 

i. Basic semantic-epistemic categories and 

ii. 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


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 

(courses of) events 

properties relations processes 

entities 

Figure 1. Relations between semantic-epistemic categories.


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: objects e.g. tree 

substances e.g. water 

collections e.g. police 

holistic e.g. nature 

Property: e.g. blue, strong 

Relation: static e.g. in, and 

dynamic e.g. give, hug (Dynamic relations 

are also classified as processes.) 

Process: e.g. run, give 

B: Complex 

State: e.g. X is blue 

(course of events) e.g. X closed the door, X built a 

house 

2. Derived categories 

Entity: blueness, strength, inclusion, running 

Property: watery, natural, included, running 

Relation: bluer than, being in love with 

Process: 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


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


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. Cause - motive - reason - origin 

B. Result - function - product - effect 

C. Direction - purpose - goal 

D. Need (of agent, instrument, process, patient) 

E. Object - material 

F. 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


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) John (agent) hired a car (object) from Bill (source) 

(2) 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. 

source 

cause/origin 

instrument 

agent 

Relation/ 

Process 

location object 

Figure 2. Roles derived from Relations and Processes 

goal 

result 

patient

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 krig fred 

(conflict) (war) (peace) 

användas i (used in) X 2 

avspänning i (detente in) X 1 

befinna sig i (be in) X/(be at) X 9 5 

delta i (participate in) X 6 

dra sig ur (pull out of) X 1 

dras in i (*pulled into) X 1 

dödas i (killed in) X 1 

fatta beslut i (make decisions in) X 1 

fungera i (function in) X 1 

få vara i (be left in) X 1 

gå med i (*join in) X 1 

gälla i (hold in) X 1 

hamna i (land in) X 2 

hålla utanför (keep out of) X 1 1 

i (in) X 5 20 3 

inbegripen i (involved in) X 1 

inblandad i (involved in) X 1 1 

indragna i (*pulled into) X 1 

invecklad i (*embroiled in) X 6 1 

klara sig i (survive in) X 1 

komma i (*come in) X 2 1 

komma i X med (come in X with) 23 

komma ur (*come out of) X 1


leva i (live in) X 1 1 

ligga i botten på (lie at the bottom of X 1 

liv i (life in) X 1 

lämna i (leave in) X 1 

rädda ur (save from) X 1 

råka i (*fall into) X 3 

skapa i (create in) X 1 

ske i (happen in) X 1 1 

skede i (phase of) X 1 

stå i (*stand in) X 1 

svåranvändbar i (hard to use in) X 1 

söka i (search for in) X 1 

tillverka i (manufacture in) X 1 

tvingas ut i (forced into) X 1 

uppgift i (task in) X 3 

utveckling i (development in) X 1 

är i (is at) X 5 

∑ 71 49 9 

objects of localization 

(i) process 

(ii) identity 

(iii) possession 

(iv) legal restriction 

(v) attitude 

(vi) speech 

goal 

conflict 

war 

peace 

consequence 

result 

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

(i) spatial container 

(ii) temporal 

(iii) circumstantial 

source 

instrument 

abstract agent of 

(i) inchoative course of events 

(ii) stative relation 

(iii) dynamic course of events 

(iv) causal force 

(v) premise, content


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.


II. 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 

konflikt krig fred 

(conflict) (war) (peace) 

State konflikt råder krig råder fred råder 

(conflict exists) (war exists) (peace exists) 

Event konflikt bröt ut krig bröt ut fred inträdde 

(conflict broke out) (war broke out) (peace came into existence) 

Course of konflikt pågår krig pågår ?fred pågår 

Events (conflict is going on) (war is going on) ?(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


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.


State 

abstract objects 

conflict, war, peace 

course of (events) 

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


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 konflikt 

A is in X with B krig 

(ii) A kommer i X med B konflikt 

A come into X with B krig 

(iii) A drar sig ur X med B konflikt 

A pulls out of X with B 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)


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.

References 


Allwood, J. 

1989 Om begrepp - deras bestämning, analys och kommunikation (ms). Göteborgs 

universitet, Institutionen för lingvistik. 

Aristotle. 

1938 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


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.


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


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


V. 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.


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


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.


4. Some social aspects of meaning 

I. Putnam’s counter-arguments 

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. Every word the speaker uses is associated in his mind with a certain 

mental representation. 

2. 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. 

3. The mental representation determines what the word refers to, if anything. 

4. 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


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


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


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


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,


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). 


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


“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.


Brugman, C. 

1981 Story of Over. Bloomington: Indiana Linguistics Club. 

Burge, T. 

1979 “Individualism and the mental”. Midwest Studies in Philosophy, vol. 4: 

Studies in Metaphysics, 73-121. Minneapolis: University of Minnesota 

Press. 

Declés, J.-P. 

1985 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. 


1990 “Induction, conceptual spaces and AI”. Philosophy of Science 57. 78-95. 


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. 


1993a “The emergence of meaning”. Linguistics and Philosophy 16, 285-309. 


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. 


1996a “Mental representation, conceptual spaces and metaphors”. Synthese 106, 

21-47. 


1996b “Conceptual spaces as a basis for cognitive semantics”. Philosophy and 

Cognitive Science, ed. by A. Clark et al., 159-180. Dordrecht: Kluwer. 


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.


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. 


1987 “On Beyond Zebra: The relation of linguistic and visual information”. 

Cognition 26. 89-114. 


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: Springer- 

Verlag. 

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. 89- 

115. 

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.


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 

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 pre- 

Darwinian 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


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.


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


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


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.


5. 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


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


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


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


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


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 speech act “Did Jim go?” 

proposition possible fact (that) Jim went 

predication state-of-affairs for Jim to go 

predicate property/relation go 

term (typically) individual 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,


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.


7. 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


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


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.


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


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


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.


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.


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


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-


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


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


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-


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


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).


References 

Allwood, J. 

1976 Linguistic Communication as Action and Cooperation. (Gothenburg Monographs 

in Linguistics 2.) University of Göteborg: Dept. of Linguistics. 

Allwood, J. 

1977 “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. 


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.


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. 


1990 “The Hierarchical Structure of Utterances”. In Fortescue, Bolkestein & Vet 

1990. 1-23. 


1992 “Parts of Speech”. In Fortescue Fortescue, Harder & Kristoffersen 1992. 

29-55. 

Johnson-Laird, P. N. 

1983 Mental Models. Cambridge: Cambridge University Press. 


1987a “Nouns and verbs”, Language 63,1. 53-94. 


1987b Foundations of Cognitive Grammar, vol.1, Theoretical Prerequisites. 

Stanford: Stanford University Press. 


1990 “Subjectification”. Cognitive Linguistics 1: 5-38. 


1991 Foundations of Cognitive Grammar, vol.2, Descriptive Applications. 


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.


Nuyts, J. 

1990 “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. 409- 

32. 

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) objective and intellectual inse ‘realize’ 

hearing (auditive) subjective and communicative hörsamma ‘obey’ 

feeling (tactile) emotional 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 Indo- 

European 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

knowledge, and non-understanding, etc. 


sikta på ‘aim at’ intention 

framsynt ‘far-sighted’ attitude to knowledge 

vara blind ‘being blind’ 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


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’.

Table 2. Visual perception 


RELATION 

see ‘see’ 

titta på ‘look at’ 

iakttaga ‘observe’ 

PERCEIVER åse ‘witness’ PERCEIVED 

ha syn ‘have sight’ skåda ‘behold, see’ vara synligt ‘be visible’ 

se ‘see’ betrakta ‘look at’ synas ‘be visible’ 

vara blind ‘be blind’ vara osynligt ‘be invisible’ 

vara mörkt eller ljust 

‘be dark or light’ 

öppna ögonen på CAUSATIVE belysa ‘illuminate’ 

‘open the eyes of’ mörklägga ‘darken’ 

förblinda ‘blind, dazzle’ 

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.


5. 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 ‘stare, glare, gape’ 

glutta ‘take a glance at...’ 

kisa ‘peer’ 

skela ‘squint’ 

bliga ‘stare, glare’ 

plira ‘peer, screw up one’s eyes’ 

snegla ‘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.

6. 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 ‘see’ 

synonyms: 

inse ‘realize’ 

förstå ‘understand’ 

begripa ‘comprehend’ 

fatta ‘grasp’


besinna ‘consider, bear in mind’ 

betänka ‘consider, hesitate’ 

anse ‘think, consider’ 

betrakta ‘look at, contemplate, regard’ 

iakttaga ‘observe, notice’ 

observera ‘observe, note’ 

skåda ‘behold, see’ 

åse ‘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) 


blunda för ‘shut one’s eyes to’ 

inte vilja se ‘not wanting to see’ 

göra sig blind (och döv) för ‘making oneself blind (and deaf) to’ 

se genom fingrarna med ‘wink/connive at, be lenient with’ 

(eller mellan) 

4. Seeing actively (=looking) → seeking information 


se efter ‘see if...’ 

titta på ‘look at’ 

syna ‘inspect, survey, examine, scrutinize’ 

titta ‘look’ 

besiktiga ‘inspect, examine’ 

bese ‘see, look at’ 

avsyna ‘inspect’ 

överse ‘overlook’ 

utföra brandsyn ‘inspection of fire prevention arrangements’


genomse ‘look through’ 

granska ‘examine, scrutinize’ 

inspektera ‘inspect’ 

introspektion ‘introspection’ 

mönstra ‘inspect, scrutinize’ 

revidera ‘revise’ 

revision ‘revision’ 

skärskåda ‘examine, view, scrutinize’ 

5. Intentionally not looking at an object → disregarding certain facts 


bortse ifrån ‘disregard’ 

lämna utan avseende ‘leave out of consideration’ 

frånse ‘leave out of account’ 

ej ta i betraktande ‘not take into consideration’ 

oavsett ‘irrespective of’ 

6. Way of seeing → way of understanding 


syn på ‘view of’ 

helhetssyn ‘total view’ 

grundsyn ‘basic view’ 

överblick ‘overview’ 

synsätt ‘outlook, approach’ 

synvinkel ‘visual angle, angle, aspect’ 

perspektiv ‘perspective’ 

synpunkt ‘point of view’ 

åsikt ‘view, opinion’ 

7. Seeing beyond one’s ordinary field of vision → interpreting as possibilities 

or having hopes 


utsikter ‘views, prospects’ 

att ha vyer ‘to hold views’ 

vy ‘view’ 

vision ‘vision’ 

(hålla) utkik ‘keep a look-out’ 

8. Seeing backwards → remembering 


återblick ‘retrospect, flashback’ 

retrospektion ‘retrospection’ 

retrospektiv ‘retrospective’ 

revy ‘review’


9. Seeing past an object → overlooking something 


förbise ‘overlook’ 

10. Seeing in a definite direction → having a special intention 


syfta ‘aim’ 

syfta till ‘aim at’ 

syfta på ‘refer to’ 

åsyfta ‘refer to’ 

sikta ‘aim’ 

ha i sikte ‘be in sight of’ 

avse ‘intend, refer to’ 

ha i kikaren ‘intend’, “have in the binoculars” (literal translation) 

kasta blickar på ‘cast a covetous eye in the direction of’ 

11. Keeping an object in the visual field → controlling that object 


ha uppsikt över ‘to supervise’ 

överse ‘oversee’ 

utöva tillsyn över ‘supervise, look after’ 

se till ‘see to’ 

ha ögonen på ‘have one’s eyes on’ 

utse ‘choose, pick out’ 

observera ‘observe’ 

överblicka ‘survey’ 

överinseende ‘supervision’ 

eftersyn ‘closer inspection’ 

12. Seeing into or through an object → understanding that object 


(ha) insikt (i NP) ‘(have) insight’ 

synonyms: 

kännedom ‘knowledge, acquaintance’ 

förståelse ‘understanding’ 

inseende ‘supervision’ 

inblick ‘glimpse, insight’ 

urskillning ‘discrimination’ 

omdöme ‘judgement’ 

medvetande ‘consciousness’ 

förstånd ‘intellect’ 

rön ‘observation, experience,discovery’ 

lärospån ‘first experience’

kunskaper ‘knowledge’ 

vetande ‘knowledge’ 

komma till insikt om ‘realize’ 

(vara) insiktsfull ‘insightful’ 

(ha) insyn ‘(have) insight’ 

se tvärs igenom ‘look through’ 

ha inblick i ‘have insight into’ 

13. Looking forward → predicting, expecting 


förutse ‘foresee, anticipate’ 

motse = frukta ‘expect, await, fear’ 

se fram emot ‘look forward to’ 

II. Properties of the perceiver 


1. Being blind, unable to see → unable of understanding 


blind 

synonyms: 

‘blind’ 

förblindad ‘blinded’ dazzled’ 

förstockad ‘hidebound’ 

oförstående ‘unsympathetic (towards) 

ensidig ‘one-sided, biased, prejudiced, narrow-minded’ 

okritisk ‘uncritical’ 

tanklös ‘thoughtless’ 

besinningslös ‘rash, unreflecting, reckless’ 

oförnuftig ‘unreasonable, irrational’ 

vara blind för ‘be blind to’ 

att ha skygglappar ‘wearing blinkers’ 

2. Being enlightened → being gifted 


upplyst ‘enlightened, educated’ 

med andlig synvidd ‘with spiritual horizon’ 

med andlig horisont ‘with spiritual horizon’ 

ljushuvud ‘genius’ 

klassens ljus ‘genius of his class’ 

lysande begåvning ‘brilliant talent’ 

briljant ‘brilliant’


3. Reflecting light → thinking, meditating 


reflektera ‘reflect, think, meditate’ 

reflektion ‘reflexion’ 

4. Not reflecting light → having poor general understanding 


oreflekterad ‘unreflecting, spontaneous’ 

5. Having good sight → having good understanding 


vara klarsynt ‘be clear-sighted’ 

vara skarpsynt ‘be sharp-eyed’ 

vara genomskådande ‘have a penetrating eye’ 

6. Seeing what others cannot see → having supernatural knowledge 


synsk ‘second-sighted, clairvoyant’ 

sia ‘prophesy (of) 

siare ‘seer, prophet’ 

sierska ‘female seer, prophet’ 

fjärrskådare ‘clairvoyant’ 

klärvoajant ‘clairvoyant’ 

7. Having a vast field of vision → planning well, being tolerant to novelties/ 

innovations 


förutseende ‘far-seeing, far-sighted’ 

framsynt ‘far-seeing, far-sighted’ 

vidsynt ‘tolerant’ 

kringsynt ‘wide-sighted’ 

överse med ‘tolerate’ 

överseende ‘indulgent, tolerant’ 

fjärrsynt ‘clairvoyant’ 

ha tillförsikt ‘being self-reliant’ 

8. Having a limited field of vision → planning badly, being intolerant to 

novelties/innovations 


kortsynt ‘short-sighted’ 

lite förutseende ‘little 

inte se längre än näsan räcker ‘not see longer than the tip of one’s nose’


trångsynt ‘narrow-minded’ 

med små vyer ‘with small views’ 

närsynt ‘short-sighted, near-sighted’ 

9. Having a special way of seeing → having a special talent 


ha blick för ‘have an eye for’ 

ha öga för ‘have an eye for’ 

10. Looking forward → being cautious 


försiktig ‘careful, cautious, prudent’ 

förutseende ‘far-sighted, far-seeing’ 

11. Being large-eyed → showing wonder 


storögd ‘large-eyed, round-eyed’ 

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


obskurant ‘obscurant’ 

III. Properties of the perceived object 

1. Being visible → giving evidence of 


synas ‘be visible’ “appear to be” 

visa sig ‘it will show’ 

synbar ‘visible’ 

synbarligen ‘evidently’ 

så vitt man kan se ‘as far as you can see’ 

till synes ‘visible’ 

visuell ‘visual’ 

åskådlig ‘clear, lucid’ 

åskådning ‘outlook, opinions, views’ 

överskådlig ‘clear, lucid’ 

synbar-lig-en ‘obviously, evidently’ 

ansenlig ‘considerable, good-sized’ 

iögonfallande ‘conspicious’ 

välsedd ‘well-seen (literally), appreciated’ 

ohöljd ‘unconcealed, unveiled, open’ 

siktbar ‘visible’


2. Being visually more perceptible than → dominating 


överglänsa ‘outshine’ 

överskugga ‘overshadow’ 

skymma ‘block, dim , obscure’ 

3. Light → knowledge 


ljus ‘light’ 

ljus (idé) ‘bright (idea)’ 

förklaring ‘explanation’ 

klarhet ‘clearness, clarity, transparency’ 

glimt ‘gleam, flash, glimpse’ 

upplysning ‘lighting, illumination, enlightenment’ 

rampljus ‘limelight’ “public knowledge” 

4. Being light → being comprehensible 


ljusna ‘get light, dawn’ 

dagas ‘dawn’ 

5. Being transparent → being easily comprehensible 


klar ‘clear’ 

solklar ‘as clear as daylight’ 

klarna ‘getting clear’ 

klart (intellekt) ‘clearly’ 

genomskinligt ‘transparent’ 

skönjbar ‘discernible’ 

sikt ‘visibility, view’ 

6. Darkness → ignorance 


mörker ‘darkness’ 

synonyms: 

okunnighet ‘ignorance’ 

oupplysthet “unenlightenment” 

barbari ‘barbarism’ 

blindhet ‘blindness’ 

töcken ‘mist, haze’ 

dunkel ‘dusk, gloom’

7. Being opaque → being abstruse, obscure 



ogenomtränglighet “impenetrability, imperviousness” 

ogenomskinlig ‘non-transparent’ 

diffus ‘diffuse, blurred’ 

dimbildning ‘smoke screeing’ 

disig ‘hazy’ 

det blir oklarare ‘getting more unclear’/’less transparent’ 

oförklarlig ‘inexpicable, unaccountable’ 

oklar ‘unclear’ 

otydlig ‘unclear’ 

8. Being dark → being incomprehensible 


mörk ‘dark’ 

synonyms: 

oviss ‘uncertain, doubtful’ 

gåtfull ‘mysterious, puzzling, enigmatic’ 

svårförståelig ‘obscure’ 

obskyr ‘obscure’ 

mulen ‘cloudy, gloomy’ 

det mörknar ‘getting darker’ 

skumt ‘obscure’ 

skymning ‘ignorance’ 

oöverskådlig ‘confused, incalculable’ 

töcknig ‘misty, hazy’ 

9. Glare, which is not direct light → fallacy 


skenbar ‘apparent, seeming’ 

skenbild ‘phantom, shadow’ 

sken ‘light, glare, gleam’ 

skenhelig ‘hypocritical’ 

skenfager ‘fraudulent’ 

skengravid ‘apparently pregnant’ 

skenäktenskap ‘pro forma marriage’ 

skendöd ‘apparent death’ 

skenmanöver ‘diversion, feint’ 

bländverk ‘delusion, illusion’ 

skuggbild ‘shadow picture, silhoutte’ 

drömsyn ‘dream-vision’ 

hägring ‘mirage, illusion’


10. The observed part → the understood part 


hänseende ‘aspect’ 

avseende ‘aspect’ 

utseende ‘outlook, appearance, looks’ 

se ut som ‘look like’ 

uppsyn ‘facial expression, countenance’ 

ansikte ‘face’ 

synfält ‘field of vision’ 

synpunkt ‘point of view’ 

glimt ‘gleam, flash, glimpse’ 

aspekt ‘aspect’ 

spegelbild ‘reflected image’ 

uppseende ‘attention, stir’ 

IV. Causes of perception → causes of understanding 

1. Making something visible → conveying information 


förklara ‘explain’ 

klargöra ‘make clear’ 

klarlägga ‘make clear’ 

lysa ‘shine, glare, gleam’ 

synonyms: 

upplysa ‘enlighten’ 

kungöra ‘make known’ 

tillkännage ‘make known, notify, announce’ 

förklara ‘explain’ 

offentliggöra ‘make public’ 

belysa ‘enlighten’ 

kasta ljus över ‘throw light on’ 

sprida ljus över ‘spread light over’ 

sätta strålkastaren på ‘put the spotlight on’ 

spegla ‘mirror’ 

utlysa ‘give notice, advertise, proclaim’ 

2. Making something invisible → making certain information inaccessible 


mörklägga ‘darken’ 

mörkman ‘obscurantist’ 

förmörka ‘darken’


3. Causing blindness → preventing from understanding 


förvända synen på ‘turn somebody’s sight’ 

förblinda ‘blind’ 

kasta sand i ögonen på ‘throw sand into the eyes of’ 

blända ‘blind, dazzle’ 

4. Creating glare → creating a fallacy 


låta påskina ‘make pretence of’ 

ge sken av ‘pretend, make a show of...’ 

7. 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


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 understanding 

not seeing not understanding 

avoiding seeing avoiding information 

seeing actively (= looking) seeking information 

intentionally not looking at an object disregarding that object 

way of seeing way of understanding 

seeing beyond one’s ordinary field of vision seeing possibilitities 

seeing backwards remembering 

seeing past an object overlooking that object 

seeing in a definite direction having special intention 

keeping an object in the visual field controlling that object 

looking forward predicting, expecting 

Table 3b. The perceiver → the cognitive agent 

being blind incapable of understanding 

being enlightened being gifted 

reflecting light thinking, meditating 

not reflecting light unreflecting, spontaneous 

seeing what others cannot see having supernatural knowledge 

having a vast field of vision planning well, tolerant 

having a limited field of vision planning badly, intolerant 

having a special way of seeing having a special talent 

looking forward being cautious 

being large-eyed showing wonder 

being a friend of darkness being hostile to knowledge 

Table 3c The perceived object → the cognitive object 

being visible giving evidence of 

being visually more perceptible dominating 

light knowledge 

being light being comprehensible 

being transparent being easily comprehensible 

darkness ignorance 

being opaque being abstruse, obscure 

being dark being incomprehensible 

glare fallacy 

observed part understood part


Table 3d. A cause of perception or non perception→ a cause of cognition or non cognition 

making visible promoting understanding 

making something invisible making information inaccessible 

causing blindness preventing from understanding 

creating glare creating a fallacy 

References 

Abelin, Å. 

1988 “Patterns of Synaesthesia in the Swedish Vocabulary”. I: Studies in Computer-Aided 

Lexicology. Department of Computational linguistics. University 

of Göteborg. 

Bring, S. C. 

1930 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. 


1992 Blotta tanken. Bokförlaget Nya Doxa. 


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 ASPECTUAL CAUSAL MODAL MODALITY 

DYNAMIC 

1 vara 23 börja 26 låta 3 kunna 49 försöka 

‘be’ ‘begin’ ‘let’ ‘can’ ‘try’ 

6 bli 51 fortsätta 59 leda 4 ska 54 lyckas 

‘become’ ‘continue’ ‘lead (to)’ ‘shall’ ‘succeed’ 

41 använda 89 bruka 71 bero 15 måste 85 töras 

‘use’ HABITUAL ‘depend’ ‘must’ ‘dare’ 

95 tvinga 19 böra 100hinna 

‘force’ ‘ought’ ‘be in time to’ 

35 behöva 

‘need’ 

“CONCRETE ACTIONS” 

POSTURE MOTION POSSESSION EXISTENCE & 

PRODUCTION 

Subject-centered Object-centered 

17 stå 7 komma 37 sätta 2 ha 8 göra 

‘stand’ ‘come’ ‘put=set’ ‘have’ ‘make’ 

21 ligga 12 gå 38 ställa 5 få 9 finnas 

‘lie’ ‘go’ ‘put=stand’ ‘get’ ‘there is’ 

61 sitta 34 följa 42 lägga 10 ta 39 bygga 

‘sit’ ‘follow; ‘put=lay’ ‘take’ ‘build’ 

accompany’ 

50 lämna 47 dra 13 ge 58 ske 

‘leave’ ‘pull’ ‘give’ ‘happen’ 

70 hoppa 64 föra 81 köpa 62 skapa 

‘jump’ ‘lead’ ‘buy’ ‘create’ 

72 nå 80 bära 87 sakna 77 hända 

‘reach’ ‘carry’ ‘lack’ “happen’ 

92 samla 94 äga 88 bestå 

‘gather’ ‘own’ ‘last’ 

90 förekomma 

‘occur’


ORGANIC QUANTITY PHYSICAL MANIPULA- 

LIFE CONTACT TION 

46 leva 45 öka 53 slå 22 hålla 

‘live, be ‘increase’ ‘strike/hit/ ‘hold’ 

alive’ beat’ 

86 växa 69 röra 

‘grow’ ‘touch’ 

(‘move’) 

MENTAL VERBS 

METALING- VERBAL PERCEPTION COGNITION DESIRE 

UISTIC COMMUNIC. 

20 gälla 11 säga 14 se 24 tycka 16 vilja 

‘apply; ‘say’ ‘see’ ‘think=be ‘want’ 

be valid’ of the opinion’ 

40 kalla 28 tala 18 visa 25 anse 

‘call’ ‘speak’ ‘show’ ‘consider’ OTHER 

44 betyda 29 skriva 32 höra 27 veta MENTAL 

‘mean=signify’ ‘write’ ‘hear’ ‘know’ 

57 innebära 65 berätta 33 finna 30 känna 68 välja 

‘mean=imply’ ‘tell=narrate’ ‘find’ ‘know’; ‘choose’ 

‘feel’ 

73 heta 74 nämna 48 söka 31 tro 91 uppleva 

‘be called’ ‘mention’ ‘look for’ ‘think=believe’ ‘experience’ 

78 kräva 76 verka 36 tänka 99 intressera 

‘demand’ ‘seem’ ‘think=reflect’ ‘interest’ 

79 svara 83 betrakta 52 räkna 

‘answer’ ‘look at, ‘count’ 

regard’ 

98 fråga 93 förefalla 55 lära 

‘ask’ ‘seem’ ‘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’


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 Johnson- 

Laird (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.


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

MOTION 

PHYSICAL CONTACT röra 

‘move;touch’ 

Nuclear 

verb 

Basic level: slå ‘strike/hit/beat’ träffa 

‘hit a target’ 

Subfields: 

Various 

groups 

Bodypart 

Tool Soundsource 

Punishment/ 

Battery 

stryka 

‘stroke’ 

Moving 

contact 

vidröra 

‘touch (lightly)’ 

Light 

contact 

krocka 

HAND 

smocka klubba banka prygla ‘flog’ gnida nudda ‘brush 

‘collide’ 

kollidera 

/klippa till 

‘punch’ 

‘club’ 

hamra 

‘bang’ 

bulta 

klå 

‘rub’ 

gnugga 

snudda against’ 

ramma klappa piska dunka spöa gno toucha 

törna 

emot 

knocka trumma daska risa skrubba komma åt 

knacka smiska massera ta på 

knuffa FOOT 

smälla gissla frottera fingra på 

puffa klatscha hudflänga skrapa tumma på 

skuffa sparka ‘kick’ örfila smeka beröra 

trampa kindpusta kittla tangera 

dänga stampa klia 

skava 

knäppa krama ARMS 

pricka slicka TONGUE 

skalla HEAD 

påta peta FINGER 

sticka bita TEETH 

nafsa TEETH 

stånga HORN 

klösa CLAWS 

kyssa LIPS


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: 

NP z ___ NP w prep NP y (med ‘with’ NP x ) 

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.


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: NP z ___ NP w (prep NP y ) (med NP x ) 

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) ACT(z,S) z performs an act S 

(ii) STRETCH(ARM) the arm is stretched 

(iii) ((FAST)TRAVEL)(x) x travels fast 

(iv) HAPPEN(CONTACT(x,y)) x and y become contiguous 

(v) FORCE(x, y, d 1 ) A force with strength d 1 is directed tow. y 

(vi) INTEND(z, DEFEAT or HURT(z, w)) z intends to defeat or hurt w


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 He stirred some water into the 

kakao i vatten. 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 The clock had a hand which moved 

sig mycket långsamt, 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’):


Men Åke ville att de skulle röra så But Åke wanted them to touch the 

lite som möjligt vid tältet. 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 If someone slaps your left cheek 

tillbaka, bara mycket hårdare. 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 And Hanna Frankenstein banged her 

kraft sin stora magra hand i kärrans big skinny hand on the cart railing 

räcke. with great force. 

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


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 The overhanging branches lightly 

vattenytan. 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 ACT(x,S) (Intention) 

Sensorimotor Body movement Effort 

Spatial perception TRAVEL(x) Speed 

CONTACT(x,y) Distance 

Mechanical reasoning FORCE(x,y,d1 ) Strength 

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


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 Other Fields 

strike 13/8% turn 5 Posture 

hit 10 throw 5 sit 9 

beat 8 settle 3 

slam 5 fling 2 Open/Close 

knock 3 come 2 open 13 

pound 3 go 2 close 3 

slap 3 put 2 

crash 2 cover 2 Disconnection 

kick 2 bring 1 break 3 

bang 1 carry 1 cut 1 

bash 1 cast 1 Odd ex. 3 

bump 1 cross 1 

clap 1 fasten 1 Mental 

clasp 1 flap 1 look (up) 3 

drum 1 fly 1 notice 1 

hammer 1 gush 1 amaze 1 

slop 1 jig-a-jig 1 consider 1 

smack 1 land 1 please 1 

thump 1 pour 1 

whip 1 rear 1 Various fields 

wipe 1 walk 1 take 2 

wrap 1 get 2 

switch 2 

bet 2 

work 2 

Odd ex. 19 

TOTAL 61/37% 37/22% 68/41% 166/100%


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


(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 

Per slog ut med armarna. 

Per spread his arms. 

PROTOTYPE: PHYSICAL CONTACT 

Resultative strenghthening: 

Objectcentered 

Motion 

Per slog bollen 

över nät. Per hit 

the ball over the 

net. 

Disconnection 

Per slog 

gräset. 

Per cut the 

grass. 

Social Interaction 

Per slog Pål i magen. 

Per hit Pål in the stomach. 

Sound source 

Det slog i 

dörrarna. 

The doors 

slammed. 

Per slog Pål i schack. 

Per beat Pål at chess. 

Organic life 

Björnenslogett 

får. The bear 

got a lamb. 

Specialized Specializedmeanings: meanings: Metaphor Metaphor 

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 

adrink. 

Symbolic 

Klockan slog 

12. 

The clock 

struck 12. 

Per slog ihjäl 

tiden. 

Per killed 

time 

Postural 

Per slog sig ner 

i soffan. Per 

sat down in the 

sofa. 

.


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 Stockholmers usually become aware of 

främmande flaggor slår ut på Norrbro. 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:


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 She struggled a little, kicking her legs 

upp och ner som en fisk på land slår up and down the way a landed fish 

med stjärten. flaps its tail. 

Marjorie slår handen för munnen. Marjorie covers her mouth with her hand. 

Hon for upp och sprang runt i köket, She leapt up and ran round the 

slog armarna runt kroppen, och kitchen, flinging her arms round her 

hulkade och snyftade. body, sobbing and sniffing. 

Och så fort Mattis fick syn på As soon as Matt caught sight of her, 

henne, rusade han fram och slog he rushed forward and threw his 

armarna om henne. arms around her. 

medan jag slår armarna om mig as I wrap my arms around myself 

själv och låter hakan sjunka ner mot and let my chin drop to my chest, 

bröstet, 

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 Then he put the shirt down in the 

han fått grepp på ålen igen slog han water and, once he got a hold on the 

tyget om den. 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 They turned off the narrow road, 

sönderregnade vägen och slog in på ruined by rain, and set off down the 

en stig som ledde in bland granarna. 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 Eric was a good linguist, musical and 

begåvad, och med Machiavelli som talented. With Machiavelli as his 

sin läromästare slog han in på en guide, he pursued a bold and 

djärv och aggressiv utrikespolitik. aggressive foreign policy.


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 The lights turn to amber and the 

skjuter iväg Toyota darts forward, 

Det hade hastigt slagit om till It had just started to thaw again 

töväder igen, 

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 In a series of brilliant actions Charles 

slog Karl XII ryssarna vid Narva år XII beat the Russians at Narva in 1700 

1700 och polackerna vid Klissow år and the Poles at Kliszow in 1702. 

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.


Table 6. Subject cline (for Contact verbs) 

Human Objects Natural Sense im- Thought, 

(instru-ment) with inherent force pression emotion 

energy 

(body-part) source wind light 

(tool) rain sound 

clock wave smell 

heart 

(projectiles) 

Other physical objects 

prototypical (literal) metaphorical 

intention 

Responsibility 

(Self-propelled) Motion 

Concretely perceptible to the senses 

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 In the deep darkness soon after midnight, 

midnatt slog en hand varligt på a hand gently hit against Adam’s bedroom 

Adams sovrumsfönster. 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:


Och sen en stark rörelse som av en Then a strong movement like an 

arm som slog ifrån sig. arm striking out. 

Näsorna och pannorna slår ihop Their noses and foreheads bump against 

och ben och armar börjar plötsligt each other; there is a sudden flurry of 

fäkta i en absurd pantomim. 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 Thursday, an Iranian robot once again 

åter ner i Iraks huvudstad Bagdad. 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 /---/ the autumn was establishing 

att installera sig med flygande löv, itself in flying leaves that sometimes 

som emellanåt slog mot hit the windowpanes with small thuds 

fönsterrutorna med små dunsar och and bangs, and in the sound of a 

smällar, och med ljudet av en allt rising wind. 

starkare blåst. 

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 Accompanying syncopation of 

kluckande och suckande vågor som sucking and splashing waves 

slår mot makligt guppande skrov. slopping against our supine hulls. 

Eller blixten kan slå ner i det [Someone can drop the aquarium,] or 

elektriska systemet. lightning strike the electrical system.


Closely related to natural forces are sense-impressions such as light/darkness, 

sound, heat/cold and smell. 

Ligga i puffen och lyssna till Lying on the little couch listening to 

tystnaden. Se ljuset från gatlyktan the silence, seeing the beam from the 

slå dagrar och skuggor mot taket. streetlamp throwing light and shade on 

the ceiling. 

Håkans snarkningar slog emot mig i I was confronted with Håkan’s snoring 

trappan. in the staircase. [Literally: H’s snoring 

hit me..] 

Hettan slog emot honom som en The heat slammed into him like 

fast materia. something solid. 

Lukten som slog upp kunde tagit A smell fumed back, enough to suffocate 

kål på en gödselstack, vad styrkan a sewage farm. 

beträffade. 

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 It struck him that he didn’t know 

många vanliga människor han kände. very many ordinary people 

Åter slog det mig att en del av de It hit me again that some of the people 

människor som kom in i vagnen who walked in the trailer to wish me 

för att önska mig lycka till inte good luck had not even been born when 

ens var födda när jag redan hade 

blivit stjärna. 

I was already a star. 

Det slog mig att det var mycket It occurred to me it had been quite a 

länge sedan jag känt mig generad. 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):


Man slås ju av att Slottet klarar sig An outsider marvels at the fact that 

med så relativt liten personal och att the Palace administration can cope 

många finns i flera olika funktioner. with its relatively small staff and that 

many on the team have double functions. 

Medan han berättade om Tessie slogs As he was telling her about Tessie, 

han av möjligheten att den där Carl considered the notion that the 

Volvon ute på Nynäsvägen hade police Volvo out on Nynäsvägen 

kunnat vänta in honom. 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 Their animosity struck him like a 

en kall fläkt, cold breeze 

5. 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


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 At exactly the right moment Staffan 

fram bollen till Magnus som slog ett hit the ball to Magnus who hit a 

härligt inlägg mot bortre stolpen. 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. ACT(z,S) 

ii. CAUSE (S, TRAVEL(x)) & FOOT(x) 

iii. CAUSE ((ii), HAPPEN(CONTACT(x,y)) 

iv. CAUSE((iii), TRAVEL(y))


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 And when, after six months or so, he 

upp sig så mycket att han förstod på had pulled himself together and understood 

vilka grunder domen gick att överklaga on what grounds he could appeal the 

hade advokaten kommit rusande till conviction, the lawyer had gone rushing to 

Säpo och slagit i honom att ett the Security Police and then drummed 

överklagande bara skulle leda it into him that appealing would only 

till ännu hårdare straff 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 Poor creature, it really is an illansett 

och lönat värv att banka reputed and badly paid duty to beat 

ABC i bondungar. the ABC’s into country-kids. 

-- Om ja så ska banka galenskaperna If I have to beat the foolishness out 

ur dej så ur dej ska dom, skrek han of you, then I will, he shouted meanwhile. 

under tiden. 

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:


Han skulle just slå bort tanken [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’, ‘Truth is the first victim of war’, US 

slog USA-senatorn Hiram Johnson senator Hiram Johnson declared as 

fast redan 1917 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 He poured a goodly slug into his 

tandborstglaset och kände den tooth-brush glass and felt the burning 

brännande värmen i magen. 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: NP z ___ upp/i (NP x ) (prep NP y ) 

Example: Perz slog upp/i (kaffe x ) (i koppen y ) 

Presupposed: HOLD(z, CONTAINER(w)) 

(i) (ACT(z, S) 

(ii) CAUSE(S, TRAVEL(ARM)) 

(iii) (CAUSE((ii), TILT(w)) 

(iv) TO(FROM(TRAVEL)))(LIQUID(x),CONTAINER(w), CONTAINER(y )) 

(v) 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:


“Kaffe”, säger han och det låter som ‘Coffee’, he says and it sounds like 

en order. Alla plockar fram koppar an order. They all produce cups and 

och Lindgren slår upp. 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 It meant dialing that number he’d 

där numret som han aldrig hade never used and asking her a favor. 

slagit och be henne om en tjänst. 

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 He was up now and singing as he 

han lagade saker som han slagit mended things he had broken the 

sönder dagen innan. day before. 

Torsten hade halva krattskaftet i Torsten had half the rake handle in 

handen. Det var splittrat och vasst i his hand, splintered and sharp at the 

den brutna änden. Han måste ha broken end. He must have broken it 

slagit av det mot knät. across his knee.


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 Who mows your hay?” asked the 

främlingen. stranger. 

Pentti var det slags typ som man nu, Pentti was the type you might still 

ännu år 1982, kunde komma på med come across, even in 1982, cutting 

att slå grönt gräs till sina fresh grass for his rabbits along the 

trädgårdskaniner kring traktens roadside verges in the spring evenings 

vägrenar i vårkvällarna och föra and taking it home in a sack on his carrier. 

hem det i en säck på pakethållarn. 

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: NP z ___ NP y (med NP x ) 

Example: Per slog gräset (med lie). 

‘Per cut the grass (with a scythe)’. 

HUMAN(z), GRASS(y), SCYTHE(x) 

(i) ACT(z,S) 

(ii) TRAVEL(x) 

(iii) HAPPEN(CONTACT(x,y)) 

(iv) 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 (


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 The demographic picture is often 

emellertid väl så väsentlig, eftersom used as an indicator, but employment 

den åtminstone på sikt slår igenom i is at least as important in that it 

befolkningsförändringar. ultimately affects the population, too. 

Dessa föreställningar slår igenom These appearances come through 

likt fjärran tv-störningar. like distant, disruptive interference on TV. 

Hur starkt de sociala skillnaderna The impact of social differences can 

slår igenom kan påverkas av 

offentliga åtgärder, 

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 [On the west coast there were fishing 

och fångade fisk med krok och nät. tribes] who killed seals with their 

clubs and caught fish with hooks and nets. 

Man faller hela tiden utan att One falls and falls without ever 

någonsin nå botten och slå ihjäl sig - hitting bottom and killing oneself - 

vilket vore en oskattbar lättnad. which would be an inestimable relief. 

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


Den unge desertören /---/ hade The young deserter /---/ was utterly 

utomordentligt trist. Han hade slagit gloomy. He had killed a couple of 

ihjäl ett par timmar med hours with correspondence. 

korrespondens; 

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 Before dinner, we sat down on her 

hennes gröna soffa. 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 Swedes and Finns settled in the 

sig ner i kolonin, som kallades Nya colony which received the name of 

Sverige. New Sweden. 

6. 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


‘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 1) Per slog Pål. Per hit Pål. 

Natural force 2) Regnet slog Pål (i ansiktet). The rain hit Pål in the face. 

Mental 3) Tanken slog Pål. The thought struck Pål. 

Clock 4) Klockan slog. The clock struck. 

Heart 5) Hjärtat slog. The heart was beating. 

Product of art 6) Pjäsen slog. The play was a hit. 

Object 7) Per slog gräsmattan. Per mowed the lawn. 

8) Per slog telefonnumret. Per dialed the number. 

9) Per slog rekordet. Per beat the record. 

10) Per slog upp dörren. Per opened the door. 

11) Per slog upp en grogg. Per poured a drink. 

Prepositional 

complement 12) Per slog Pål i ansiktet. Per hit Pål in the face. 

13) Per slog Pål i poker. Per beat Pål at poker. 

14) Per slog Pål med knytnäven. Per hit Pål with his fist. 

15) Per slog Pål med 3-1. Per beat Pål 3-1. 

16) Per slog Pål med häpnad. Per struck Pål with amazement. 

Particle 17) Per slog in boken. Per wrapped up the book. 

18) Per slog upp boken. Per opened the book.


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 Then he opened the book and 

och började demonstrativt läsa. 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 I open my eyes in the morning and 

och tänker: ‘Varför ska jag bry mig think, ‘Why bother getting up?’ ” 

om att stiga upp?’” 

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 DeVasher closed the McDeere file and 

McDeere och slog upp en annan, opened another, much thicker one. 

betydligt tjockare.


Han slog ihop matsedeln och He closed the menu and wine list, 

vinlistan, 

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?” ”When do you get off your shift?” 

frågade han i samma ögonblick som he asked the one in the back as she 

en av dem slog igen bildörren. 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 He had tried looking up The Biggest 

“Världens största mage” 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 Lindgren laughs an affected laugh 

och drar fram en flaska Kronvodka and produces a bottle of Crown vodka 

ur ryggsäcken.—Om du inte är snäll, from his rucksack. –If you’re not nice, 

så får du inte smaka, säger han och you won’t get any, he says and pours 

slår upp åt sig själv innan han låter some for himself before he lets the bottle


flaskan gå runt samma väg som TV- go around the same way as the thermos. 

kannan. Redan innan dom fått ser Even before they’ve been served, Kurt and 

Kurt och Göran glada och nöjda Göran look content. –Take a good slug 

ut.—Ta ordentligt nu, säger Lindgren now, Lindgren says when Mats pours. As 

när Mats slår i. Som för att ge sin if he wanted to give his blessing. 

välsignelse. 

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 NP y , 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


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

Syntactic frame: NP z ___ NP w (prep NP y ) (med NP x ) 

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 That night I awoke from a fierce 

dunkade häftigt i väggen. 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 He could feel hate and adrenalin 

bubblade upp inom honom så att bubbling up inside him, so that his 

pulsen plötsligt började dunka i hela pulse started pounding. 

kroppen. 

Even in this case, the verb can be used in the impersonal construction characteristic 

of sense-impressions:


Table 11. The pattern of polysemy of the verb dunka ‘bang, thump’ 

Subject-centered motion 

Tåget dunkade fram genom 

dalen. The train chugged along 

through the valley. 

Metonymy: Metaphor: 

Sound Source 

Det dunkade i väggen. 

There was a banging on the wall. 

Focusing: 

Subject: Engine 

Maskinen dunkade. 

The machine was pounding. 

Non-human subject: 

PROTOTYPE: Physical contact + Sound Source 

Per dunkade näven i bordet. 

Per banged his fist on the table. 

Strengthening: 

Nonverbal Communication 

Per dunkade Pål i ryggen. 

Per slapped Pål on the back. 

Mental 

Orden/Skuldkänslorna dunkade i 

Pers huvud. lit. The 

words/feelings of remorse were 

throbbing in Per’s head. 

Bodily Perception 

Det dunkade i tinningarna. 

My temples were throbbing. 

Subject: Inner organ 

Hjärtat/Blodet dunkade. 

The heart/blood was pounding. 

Production 

Per dunkade fram jazz på pianot. 

Per produced jazz by pounding 

away at the piano. 

Det dunkade vid tinningarna och jag My temples were throbbing and I 

tyckte att jag inte kunde se klart, 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 I could feel the guilty feelings 

dunkade på mej därinne. pounding at me in there.


Våld våld våld. Ordet dunkade i Violence, violence, violence. The word 

hennes värkande huvud. 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 The clock under its glass dome struck 

snabba slag och herdinnan dansade twelve sharp notes and the shepherdess 

för sin herde. 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

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: 

NP z ___ NP w prep NP y (med ‘with’ NP x ) [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 He grabbed her leg again and rubbed 

strök knäet. her knee. 

Han strök sig över kinden och lät He ran his hand over his cheek, letting 

blicken fara ut genom fönstret och his gaze wander out of the window and 

in i den vita vårhimlen. 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 The swallows sweep over the roofs 

sommaren. 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 We closed the door fairly carefully if there 

strök omkring någon välvillig granne. were any well-meaning neighbours 

prowling about.


Table 12. Major classes of derived meanings of stryka ‘stroke’(‘rub’, ‘wipe’) 

Focusing 

+Evaluative 

strengthening: 

Focusing Subject-centered motion 

Båten strök längs 

stranden. 

The boat skirted along 

the coast. 

PROTOTYPE: Physical contact 

Peter strök Mary över 

håret. 

PeterstrokeMaryover 

the hair. 



Focusing: 

Application + Symbolic 

Peter strök under ordet. 

Peter underlined the 

word. 

Symbolic: ‘Important’ 

Peter strök under att 

han menade allvar. 

Peter emphasized that 

he meant business. 

Application 

Peter strök smör på 

brödet. 

Peter spread butter on 

his bread. 

Application: Paint 

Peter strök taket. 

Peter painted the 

ceiling. 

Subject-centered motion 

+ Evaluation 

Peter strök längs 

gatorna. 

Peter prowled along the 

streets. 

Removal 

Peter strök svetten ur 

pannan. Peter wiped the 

sweat from his 

forehead.


Han hade strukit i gränden utanför He had been prowling around in the 

Martinas hus. 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, There wasn’t any butter, we’d run out of 

hon hade inte tänkt på att stryka. 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 He slipped a pencil from his pocket and 

bröstfickan och satte igång att stryka began underlining phrases. 

under diverse meningar. 

Då skulle han ha strukit över He’d have scratched out some names, 

vissa namn och satt in andra i stället inserted others, and scrawled notes 

och gjort anteckningar i marginalen. across the margins.


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

Syntactic frame: NP z ___ under/över NP y (med NP x ) 

HUMAN(z), TEXT(y), PEN(x) 

(i) (WITHi (ACT)) (z, S, x) 

(ii) CAUSE(S,((ALONG(TRAVEL)) (x,y) & HAPPEN (CONTACT (x,y))) 

(iii) CAUSE ( (ii), HAPPEN(EXIST(LINE))) 

(iv) a) INTEND (z, SHOW (z,w,IMPORTANT (y))) 

(iv) b) 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 He crossed out the sentence about 

skrev: 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 It should also be emphasized that 

regeringen avser att slutföra the Government intends to pursue with 

EES-förhandlingarna med all kraft. all force the EEA negotiations to their 

conclusion. 

Långväga kontakter understrykes Further emphasis on contacts with traders 

ytterligare av fynden från from both near and far comes from finds 

handelsplatser som Helgö i Mälaren. in such trading centers as Helgö, in 

Lake Mälaren,


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 Now she rubs moisturiser into her 

kräm i huden som skydd mot den facial skin as protection against the 

råkalla vinterluften utanför, raw wintry air outside, 

[de] tyckte att hon såg smutsig ut. They thought she looked dirty. They 

De skrubbade och skrapade henne scrubbed and scraped her in case she had 

för den händelse hon medförde ohyra. brought vermin with her. 

However, even change of physical properties realized as resultative adjectives 

are rather common:


Han rev loss en sida [ur tidningen] /-- He ripped out a page [from the 

-/och gnuggade det mjukt innan han newspaper] /—/and rubbed it soft 

torkade sig med det. 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 Focusing 1 Focusing 2 

Physical contact Subject-centered motion Act energetically 

Han gnodde Vi gnodde omkring Vi gnodde och slet. 

silverbrickan blank. och letade efter boken. 

He rubbed the silver We scurried around We toiled and drudged. 

plate bright. looking for the book. 

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 Billy Wilder’s cynical humor tickled us. 

oss. 

9. 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.


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. 


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.


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.


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


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. The worst is behind us. back = earlier 

b. I am looking ahead to meeting her. front = later 

(2) a. in the following years back = later 

b. in the preceding years front = earlier 

(3) a. Il venait de fermer la porte. come = earlier 

[lit. He came from closing the door =] 

‘He had just closed the door.’ 

b. He was going to do it. go = later 

(4) a. in the coming years come = later 

b. in the years gone by go = earlier 

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;


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 moving-time 

S —> .... S —> 

——————————————>


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


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-


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 Event 1 is before — or precedes — Event 2 , 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).


point to the left of the sequence line which permits the functioning of the egoopposed 

strategy, i.e. Event 1 and Event 2 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.


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.


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


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.


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


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


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-


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


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


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-


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

(8) 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.

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.


Clark, H. 

1973 “Space, time, semantics, and the child”. Cognitive Development and the 

Acquisition of Language ed. by T. E. Moore, 27-63. New York: Academic 

Press. 

Cogen, C. 

1977 “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. 

Comrie, B. 

1976 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 Urubu- 

Kaapor 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. 


1975 Santa Cruz Lectures on Deixis, 1971. Bloomington: University of Indiana 

Linguistics Club. 


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.


Johnson, M. 

1987 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. 


1987 Foundations of Cognitive Grammar. Vol. 1: Theoretical Prerequisites. 



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. 455- 

471, 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.


Taylor, J. 

1989 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 Semantics 1 

Kenneth Holmqvist 

Department of Cognitive Science, Lund University 

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.


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


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.


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] 

SUPERIMPOSITION 

Figure 2. 

[STRETCHED].TR 

[ACROSS] 

TR 

LM 

[STREET] 

SUPERIMPOSITION


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.


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.


tvelwe o´clock a…ago 

Figure 3. 

However, it often happens that two predications cannot be superimposed until 

one of them has been turned properly, as in Figure 4. 

VERT 

Figure 4. 

TIME TIME 

NOT SUPERIMPOSABLE 

a…ago 

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. 

Schema 

along the pole 

HORIS 

VERT 

. . . 


VERT 

long 

HORIS 

. . . 

HORIS 

rope 

Dimensions 

Figure 5. 

Extension 

Extension 

Diameter 

Position(s) 

Direction 

VERT 

day 

TIME 


HORIS 

TIME 

VERT 

TIME 

HORIS 

NOT SUPERIMPOSABLE


But when there are several dimensions of that type in the other predication, as 

in Figure 6, another mapping principle must be adopted. 

Schema 

Dimensions Extension 

Extension A: 2:nd largest value 

Extension B: largest value 

Extension C: least value 

Direction A: vertical 

• 

Figure 6. 

A 

C 

B 

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. 

input level 

part level 

Figure 7. 

John walk.TR 

snake walk.TR


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”. 

input 

level 

parts 

at the 

whole 

level 

Figure 8. 

row 

the shoals


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.


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.


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: 

(t 0 ) A new morpheme just arrived. Interrupt all processing. 

(t 1 ) 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). 

(t 2 ) The parser splits into two (pseudo-parallel) processes, of which the


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 t 2 . 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


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]. 

 

give 

 

+s who 

see 

+ed a some 

Figure 11. 

Morpheme stream order 

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), (1), (1), 

[MARY WHO] (1) and [MARY WHO SAW A HOUSE] (1). 

Distance from 

4 

3 

2 

1 

 

 

 

 

 

Figure 12. 

 

[GIVES] 

 

[MARY WHO] 

[SAW] 

 

[A HOUSE] 

 

[SOME MONEY] 

[MARY WHO SAW A HOUSE] 

Current morpheme


The process A of t 1 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.

Suggesting 

Figure 13. 

Params 

Evaluating 

Pruned off because of 

low survivability 

5. Project goals and current project work 


Suggesting 

Params 

Params 

Params 

Params 

Params 

Params 

Params 

Evaluating 

Params 

Params 

Params 

Params 

Params 

Params 

Params 

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


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”.

References 


Glasgow, J.I. 

1993 “The Imagery Debate Revisited: A Computational Perspective”, Computational 

Intelligence (Taking issue position paper), Vol. 9, No. 4, pp. 309- 

333. 

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: Springer- 

Verlag. 

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 

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.


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


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:


• 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,


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)


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., OVER 1 , OVER 2 , OVER 3 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.


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 OVER 1 in Figure 3). Furthermore this 

misleading solution leads to another problem, which has tormented “natural


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). 

"disambiguation" ? 

OVERÿ 

1 

"over" 

OVER 

2 

learning ? Regier's system 

OVER 

3 

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?


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,


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 er " " under " 

Common hidden layer 

Regier's system 

la y e r 

segments 

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.


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).


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


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


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


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


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?


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


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. 


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).


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.


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. 293- 

312. 

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.


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. 159- 

216. 

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.

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 

INDEX OF NAMES 195 

Index of Names 

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

W 

Winograd, T. and Flores, F. 187 

Winter, S. and Gärdenfors, P. 30 

Wittgenstein, L. 175, 187 

Wright, L. 38 

INDEX OF NAMES 197 

Z 

Zlatev, J. ix, x, 174, 179

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 

INDEX OF SUBJECTS 199 

Index of Subjects 

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

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 

INDEX OF SUBJECTS 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. 


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 SIMON- 

VANDENBERGEN 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. 


38. FRETHEIM, Thorstein and Jeanette K. GUNDEL (eds): Reference and Referent Accessibility. 


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. 


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. 


58. ROUCHOTA, Villy and Andreas H. JUCKER (eds): Current Issues in Relevance Theory. 


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.

Cognitive Semantics : Meaning and Cognition

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?