Table Of Contents

1. Introduction

2. Speech & Writing

3. Sources of the Written Language Bias in Linguistics

4. On the Scope of our Claims

5. Linguistics & the Overall Theory of Language

6. Grammar

7. Semantics

8. Phonology

9. Linguistic Communication

10.Language Acquisition

11 Linguistic Variation

12 Epilogue





7.1 Introduction: The autonomy of linguistic meaning

 The typical attitude prevailing in most of modern theoretical linguistics is that language can be studied as an autonomous structure without any direct consideration of conditions of use. This applies most straightforwardly to the outer form, the expression side, of language (5.4), but many scholars treat meaning, the content side, in the same way. In short, meanings are held to be

  • objective, i.e. they are not dependent on the ways any given persons happen to understand them, - autonomous and disembodied, i.e. they can and should be treated as independent of what human beings do in speaking, understanding, and acting,
  • compositional, i.e. they are abstract objects with well defined inherent properties, and can be analyzed in terms of components, i.e. "smaller" objects more "primitive" concepts and the like). (Lakoff & Johnson 1980).

Furthermore, it is assumed that words, sentences, texts, and discourses have meanings "in themselves, as linguistic objects as it is often put. They are said to "carry meanings" with them (IX.1). The meaning of a given linguistic object can be unearthed by means of a sophisticated linguistic analysis which aims at finding the correct interpretation or the semantic representation inherent in it (note the singular forms which are used here)(c.f. V.5).

The view just sketched is utterly misconceived, at least if we are interested in the meanings or interpretations relevant in actual communication, or, for that matter, in thinking. The interpretation of an utterance, a discourse or a text is never (I will argue) completely inferable from the linguistic objects alone. It is not inherent in the linguistic signs (the "sound-meaning correspondences") taken as situation-independent entities. In practice, there is always a need for different kinds of background knowledge, i.e., knowledge which is extrinsic to language but usually more or less available to senders and receivers in communication. Consider (1) as a simple example:

(1) He is dangerous.

Such a sentence can never be understood without several types of background knowledge. The listener (or reader) must be able to tackle a number of questions whose answers are in no way inherent in the semantic representation of the sentence as such (however this is conceived): Who is he? (referential specification; note that this is still necessary, even if we use a common noun ('that man') or a proper name ('Ronald Reagan') instead of a pronoun). In which way, to what extent, is he dangerous? To whom and what, when and where is he a threat?(determination of standards of comparison and precision, referential specification, intensional precisation, elimination of vagueness and ambiguity). Who uttered (1), when did he (or she) do it, and why? (assignment of intentions and reasons to the sender). These latter questions are concerned with the so-called why of communication (Ducrot 1972) which is always relevant for the addressee. A reasonable situational interpretation [1] can only be the result of a complex interplay between various situation-specific factors and the inherent meaning potentials (Rommetveit & Blakar,1978) of the linguistic units and constructions. The contribution of the linguistic expressions with their meanings is often very important but it is never enough by itself. Instead, words and utterances should be seen as clues to interpretation, as instructions to the listener to search for an interpretation along certain lines (cf Gardiner 1951, Linell forthcoming.). Utterances allow for, or interpretations, but they do not express or transfer them (IX.1). Accordingly, Merleau-Ponty (1962) speaks of "the essential allusiveness and incompleteness of speech" as a ubiquitous and fundamental feature of all linguistic communication; it is not restricted to "special forms of insinuations, half-sayings, allusions to little tales of an intentionally nonserious character" (Volosinov 1973:97) and similar uses of language.

In discussing formal and semantic autonomy (III.1) I referred to Olson's (1977) pertinent analysis of the written language bias in linguistics and related disciplines. However, even Olson argues that written texts are, or may be, semantically autonomous. We recall that for him this was the most fundamental difference between writing and speech. But such a view can hardly be true in a strict sense. It is a fact that written texts are not semantically autonomous either (cf VII.6). However, they are relatively more autonomous than spoken dialogues, and this is a circumstance which has not failed to exert a great influence on linguists' views of semantics.


7.2 Word meanings

The view that there are immanent meanings or interpretations inherent in the linguistic signs as such, i.e. associated with the linguistic expressions (sign vehicles), has been much cherished by linguists. Sometimes these meanings are talked about as 'literal meanings', and one should carefully note the word 'literal' here' The assumption of literal meanings pertains to several linguistic levels. Here we will consider word meaning first and return to sentence meaning later on (7.5).

Within linguistic semantics there have been several theories of word meaning, and not all of them are equally amenable to characterizations of the kind that I am going to suggest. However, one element that seems to be implicit in many theories is the idea that many ('literal') uses or interpretations of a given word can be directly inferred from an underlying invariant word meaning. (Some would put it even stronger saying that the literal interpretation is identical to the word meaning). Such a conception is compatible with theories based on Grundbedeutunq as well as Gesamtbedeutung.

The standard view is that word meanings are definite, fixed and stable, e.g., a set of invariant semantic features. These meanings are ready-made, they exist "out there" in the language (la langue) as a system, and as such they are available for use by speaker-listeners (writer-readers). An extreme view is held by structuralists who maintain that word meanings should be defined entirely within the system itself without any recourse to extralinguistic experience. According to this view, the meaning of a given word consists in being different from other words; dans la langue il n'y a que des differences (Saussure 1964:166).

Instead of construing word meaning in terms of sets of invariant semantic feature complexes (as, for example, in the semantics of a generative grammar) one could look upon the meaning of a given word as something which is "dynamic, only partially determined" and "open and susceptible to contextual modifications"..."in communication when what is initially intersubjectively shared becomes expanded and/or modified" (Rommetveit & Blakar 1978:354). In Rommetveit's view the word has the potential of eliciting different semantic operations which are partly different in different situations and therefore result in distinct situationspecific interpretations. Another way to put it would be to say that word meanings place certain conditions on possible situational interpretations, or that they point to, allude to, certain interpretations.

Volosinov (1973) argues that the conventional view on word meaning derives from the traditional work by linguists and philologists on written texts in foreign languages ("the grandiose organizing role of the alien word"), which necessitated the development of dictionaries with their standard definitions of word meanings. Obviously, it makes a great difference if someone is acquiring his mother tongue in naturalistic situations, which may be assumed to lead to an implicit recognition of the negotiability and context-sensitivity of meanings, or if he is learning a foreign language by consulting dictionaries as a means of decoding written texts. In general, the system-internal definition of meanings in terms of semantic features or the like seems to fit the latter predicament, i.e. the learning of foreign and learned words and concepts, rather than the acquisition of everyday concepts. Learned vocabulary is typically acquired from written texts and/or by verbal definitions, i.e., the whole process is essentially language-internal.

The conception of word meaning as something fixed and stable recurs in several contexts. It often seems to be presupposed that understanding what a word means simply consists in knowing its linguistic meaning or being able to provide an explicit verbal definition. We are therefore often faced with an all-or-nothing conception; either someone understands what a word means, or else he does not do so. Something like this seems to be implicit in the theory of a linguist who performs investigations of people's knowledge of words by simply testing the understanding of isolated sentences in which the words in question occur. Another case would the parent who observes his little child reacting adequately (or not adequately) to the use of a certain word in a given situation, and then declares that the child understands (or does not understand) the word tout court. This is obviously absurd since the same word can often be understood in many ways and at different levels; we could talk about listeners' varying depth of understanding in analogy with the speakers' depth of intention" (Naess 1953). Therefore, one may perfectly well understand a given word in one situation, where, for example it is perhaps supported by other factors in a fortunate fashion, and at the same time completely fail to understand it in another situation.

In this discussion I have alluded to the layman s theory that "each word has its own proper meaning". It might be pointed out that this "theory" exists in assumed that "each thing has its right name". The view that there is one unique way of expressing something precisely to the point (le mot juste) is the other side of the coin (c.f. Gardiner 1951:174).


7.3 The boundary between semantics and pragmatics

One of the most hotly debated issues in the current theoretical discussion within linguistics concerns the relationship between semantics and pragmatics. Is it possible and/or desirable to split up the study of meaning into two disciplines? This is a complicated problem which is currently being highlighted in a great number of books and articles. For obvious reasons I am unable to discuss here all the pros and cons. Rather what is at stake here is simply the fact that a distinction has been made by many scholars. The dividing line can be drawn in different ways and at different places, and I will mention three different alternatives in the sequel (for further discussion, see Allwood 1981).

I will not argue that the distinction between semantics and pragmatics in any of the three versions or the theory to be discussed here corresponds to, say, what is inherent in the semantic representation of a written text (cf. Olson's hypothesis that texts are semantically autonomous, VII.1) vs. what additional odd factors are involved in the understanding of spoken utterances. There is no such simple relationship between writing and speech on the one hand and semantics and pragmatics on the other. Furthermore, the distinction has never (to the best of my knowledge) been rationalized by linguists on any grounds of that kind. On the other hand, this also applies to almost all the other theoretical issues discussed in this book; linguists generally argue that the theoretical options they take have a clear basis in the linguistic phenomena themselves rather than simply in the tradition of analyzing only written language.

As far as the semantics-pragmatics distinction is concerned, some scholars tend to rationalize it on the grounds that it helps the linguist to delimit language properly. Many linguists have been quite anxious to establish and maintain linguistics as an independent science distinct from, say, psychology, sociology and philosophy. Since one suspects that the actual understanding of utterances and texts on the part of listeners and readers is more or less dependent on their encyclopedic knowledge of the world, and since most linguists would prefer to have a neat dividing line between their (hopefully) well structured science and the whole mess of encyclopedia, the attempt to delineate a more restricted study of meaning, that of semantics, may be a sensible strategy. After all, in comparison with pragmatics, semantics is (in most people's opinions) more closely integrated with the language system itself (la langue, competence).

However, no matter what the explicitly given or consciously admitted reasons for setting up the semantics - pragmatics distinction are, the written language bias is undoubtedly there as a general background, and it may very well be treated as a contributing cause. This can be argued in all the three cases which I will now advert to. [2]

The first, and perhaps most popular, theory (e.g. Lyons 1977: 591) would define semantics as the study of those context-independent meaning elements which are tied to the linguistic expressions, i.e., part of the linguistic signs as such (signs taken roughly in Saussure's sense), and are (considered to be) invariant across all situations of use. Pragmatics, on the other hand, would then deal with the specifying effects of various situational factors, and try to account for the meaning of specific utterance tokens. In other words, semantics would focus on those contributions to the situational interpretations [3] of utterances and texts which are provided by the language itself (the rules for using words and expressions), while pragmatics would encompass the whole process of determining situational interpretations, the entire interplay of all the various linguistic and extralinguistic factors (cf. II). Accordingly, semantics and pragmatics could be seen as concerned with the meanings of linguistic expressions considered as types and tokens, respectively. [4]

It might be noted in passing that my own theoretical preferences (cf. the distinction between linguistic meaning and situational interpretation, VII:5 would, by and large, coincide with these views, although I am aware that it may be difficult to defend a strong version of such a theory. Clearly, there are some meaning potentials associated in the user's mind with the linguistic expressions as types. Thus, even a sentence like 'he is dangerous', which - as an abstract sentence type - is "unspecified" in obvious ways (VII:1), can be presented in abstracto, i.e. without any observable supporting context, and most of us would still be able to (re)construct at least some ingredients of a possible interpretation. We would understand what could or must be the case in a situation in order for the sentence to be meaningfully used there. [5]

Whatever the merits of this first-mentioned attempt at distinguishing semantics from pragmatics are, it is not too farfetched to suggest a possible link between this and the written language bias in linguistics and philosophy. The idea of context-independent, invariant meaning features belonging to the language itself is something which gets easily promoted through the traditional linguistic study of written sentences and texts; I have already pointed to the relative autonomy and explicitness of texts as compared with most spoken dialogues (II, VII.1). Here is Volosinov again:

 "The isolated, finished, monologic utterance, divorced from its verbal and actual context and standing open not to any possible sort of active response but to passive understanding on the part of a philologist - that is the ultimate "donnée n and the starting point of linguistic thought". (Volosinov 1973:73)

The second theory has also been hinted at, e.g. by Lyons (1977: 591). It is based on the different communicative functions of utterances and texts, and on the linguist's assessment of these functions in terms of importance and priority (cf. IX.2, and also VII.4). Thus, it may be suggested that semantics be exclusively concerned with those meaning elements that have referential or cognitive functions, whereas the study of all the various social, emotive, practical and associative aspects be relegated to pragmatics ("the pragmatic waste-basket"). Such a conception is also in keeping with the written language bias, since writing clearly enhances the referential and cognitive aspects, the so-called intellectual functions, at the cost of the other functions. This definition of semantics would also fit the view that there is a close connection between semantics and logic (cf. VII.4, VII.9). Steiner comments on the-attempts to create a universal logic with a fixed semantics in the following way:

"The slippery, ambiguous, altering, subconscious or traditional contextual reflexes of spoken language, the centres of meaning which Ogden and Richards termed "emotive" and which Empson treats under the rubric of 'value' and 'feel', fall outside the tight but exiguous mesh of logic. They belong to the pragmatic".(Steiner 1975:203)

 The third theory ties up with the distinction between normative and descriptive studies, which was applied to semantics by Carnap. According to this view, semantics is normative; it studies meanings and interpretations as they should be, or, alternatively, as they would be if certain well-defined rules were consistently followed. This is Carnap's "pure semantics", a discipline which may be seen as a subpart of logic. Pragmatics, on the other hand, would be the descriptive study of how language is actually used; how concepts are in fact applied to different situations, and what interpretations are arrived at by real language users. This is close to what Naess (1953) termed "empirical semantics". Even this third theory may be associated with the written language bias.

The written language is intimately connected with norms of language and training in an allegedly correct and logical use of language. The pursuit of normative semantics is therefore a rather natural endeavor for someone dealing with the theory of correct (written) language.


7.4 Semantics and truth

 "If we knew what it would be for a given sentence to be found true then we would know what its meaning is". (Carnap 1953)


Linguists and philosophers like to theorize about the functions and goals of semantics in a way that links it up with questions of the theory of truth and truth conditions.

 "Despite the uniquely fragmented state of contemporary linguistic theory at the present time, there is a remarkable consensus with respect to one fundamental issue: the nature of semantic representation. Scholars as disparate as Montague, Chomsky, McCawley, Katz, Sadock and Partee would, I think, now agree that the semantic representation of a sentence is a representation which expresses the logical form of the sentence, or, put differently, a representation which expresses the conditions under which that sentence would be true" (Gazdar 1980:5)

It is significant, especially in the light of the discussion in VII.3, that Gazdar adds to these statements that "other nontruth conditional aspects of the meaning of the sentence belong to pragmatics".

In logic, which may be seen as a special development of -sentence semantics, truth and truth conditions have always been of primary importance. How should one characterize truth vs. falsity? What are the conditions to be fulfilled if a given statement is to be true or false? In modern linguistics, truth conditions have also played a major role, as Gazdar notes. As an illustrative example indicating the primacy of truth conditions in the linguists' conception of meaning we may recall Partee's (1971) discussion of the respects in which transformations (in a standard generative grammar) are, or could be said to be, "meaning preserving"; the proposal given there was that transformations must not change truth values.

A truth-conditional semantics is thus concerned with (mainly or only) the referential ('intellectual') function of language, i.e. aspects of meaning which are emphasized in written texts. The referential-cognitive aspects of language are clearly more independent of the particular contexts of use, more independent of specific senders and receivers, than are the other aspects (emotive, evocative, expressive, social etc, cf. IX.2). The referential association can be depicted as an abstract connection between language and the world:

linguistic expression (e.g. a statement) with its semantic representation(a set of truth conditions)






things referred to in the "real world".


The relation here designated by an arrow can be characterized in terms of truth vs. falsity; either the statement corresponds with reality, or it does not. There is, in other words, a close connection between a correspondence theory of truth, and the conception of linguistic meaning as truth-conditional in nature.

The goal of describing and analyzing objective reality in a true manner is of course rather strongly stressed in certain varieties of mainly written discourse, in particular scientific texts. But there are many other uses of language, especially in spoken dialogues, where the purely referential elements are not of primary importance, and where the truth of the statement is not the main thing. If we take a simple utterance like 'It is raining', there are surely many situations of use, where the determination of the truth of this statement is not central to the speaker's intention or the listener's actual interpretation. Most often the listener would take it for granted that the speaker provides him with a veridical description. Rather than checking whether the speaker's assertion is actually true, the listener asks himself more important questions: What is the point of this assertion? For what purpose does the speaker utter it? In what way should I utilize this piece of information? That is, the why of communication is of utmost importance, and this can only clumsily (if at all) be accounted for in terms of truth conditions.

Another case in point which shows that the question of truth is often downgraded in actual discourse concerns the normal use of referring expressions (noun phrases). If I say 'That man With a camera is dangerous', the point of using the referring noun phrase ('that man with the camera') is not that of issuing a statement, the truth of which the listener should determine, but simply that of using a linguistic expression such that the listener succeeds in identifying the referent that I as speaker have in mind. In fact, the act of reference may be successful even if, on a closer look, either the speaker or the listener, or both, discover that the man was after all not carrying a camera but a pair of binoculars (cf. Donnellan 1971). Thus, the order of priority is this: the superordinate goal is the evocation of a certain act of understanding in the listener (he must understand what the speaker means by his utterance), a means to attain this goal is through reference to a certain person, and this in turn involves the subordinated tactic of describing this person in a certain, veridical way.

Current linguistic semantics thus overrates the importance of truth. The concentration on truth conditions turns into absurdity if it is argued that this is the only significant aspect of meaning. Steiner (1975:211 ff) has attacked this concentration on truth in linguistic semantics and stressed the importance of our ability to imagine and describe things as they are not. The fundamental properties of language, or rather linguistic communication, are incompleteness, vagueness and allusiveness (cf.Merleau-Ponty);the whole truth is never conveyed, it i8 only indicated or alluded to. Thanks to this, we can use language creatively in an ever changing social and physical reality, and we can, if we must, lie or tell half-truths, and this may even be a precondition for survival in some situations (Steiner, op cit).


7.5 Speech acts, sentence meaning and utterance meaning

 Some semanticists (e.g. Grice 1957) have tried to distinguish between sentence meaning and utterance meaning. The former would be some type of invariant meaning which pertains to the linguistic expression as such i.e. without any consideration of particular situations where it may be used (cf. VII.3). After all, most linguists would surely agree with Lyons (1977:35) that "a sentence like "It's raining" has a certain constancy of meaning irrespective of the purpose of the communicative act in which it is used". I would here prefer the term linguistic or structural meaning, since the linguistic expression involved need not formally be a sentence (cf. VI.1). This must be carefully distinguished from the situational interpretations of particular occurrences of the expression (e.g. 'It's raining'), i.e. interpretations made when the expression is used in different specific utterances in particular situations. The term 'situational interpretation', which I use instead of utterance meaning, is applicable to non-verbal acts as well, and is, moreover, meant to imply that the interpretation pertains to a comprehensive communicative act, in which the verbal utterance is just one (albeit important) part. As will be pointed out presently (VII.6), there are always several possible situational interpretations of the same utterance token. For example, we must distinguish between how a speaker wants his utterance to be understood, and what the listener actually understands.

What then is the proper relation between linguistic meaning and situational interpretation? The compilations of this issue can hardly be unraveled here, but it is still possible to contrast two basic theories.

The first type of theory, which has gained a considerable popularity among linguists at least since Searle (1969), implies that there are two kinds of communicative acts, one in which the situational interpretation, what is or should be understood, is simply and precisely equal to the linguistic meaning (this would then be the "literal" meaning or interpretation), and one more complex type in which the understanding involves something more; the speaker does not mean "exactly what he says", and hence we must proceed to a "non-literal" or "figurative" interpretation (indirect requests, metaphors, jokes, irony, sarcasm, and many other implicatures). Thus, this dichotomy corresponds to Searle's distinction between direct and indirect speech acts.

Searle's theory of direct speech acts is based on the thesis of semantic autonomy; it is assumed that there are cases where the task of the listener is "just" that of understanding, or "reconstructing", the structural meaning, usually called the "semantic representation", of the linguistic expression used. Searle also formulates this as the "principle of expressibility": "whatever can be meant can also be said and understood" (1969: 19-20). Thus, it is argued that fixed and finite meanings can (in fortunate cases) be converted into exact expressions, and vice versa. In other words, direct speech acts would be such communicative acts, where the speaker says exactly what he means. I would deny that such acts exist. In fact, considering what has been said in VII.1, it is remarkable that such a theory has been proposed at all. On the other hand, it is not entirely surprising given the written language bias in linguistics and logic (c.f. VII.7). After all, contemporary linguists are accustomed to analyzing sentences in abstracto, thereby figuring out what "their semantic representations" are.

The other type of theory, which seems to me much more plausible, says that the understanding of an utterance always involves something more and something beyond constructing the linguistic meaning. Thus, for example, the why of communication is always more or less important, and this is not something which can be directly or exclusively associated with the linguistic expression or the overt behavior as such. Other situational factors also intervene (see VII.1). Even if we understand a sentence seemingly in vacuo, e.g. if we reflect on the meaning of sentence like He is dangerous or it's raining without any given extralinguistic context - which linguists are used to doing (cf. above) -, we take part in a language game with special presuppositions and expectations and cannot avoid constructing _ some rudimentary aspects of an imaginary context. The same applies to written texts (c.f. VII.6). Thus, Rumelhart is justified in saying:

In summary then, the supposition that conveyed meanings are ever identical to literal meanings (where literal meanings are assumed to be those given by a compositional semantic theory) is surely suspect. The problems of determining conveyed meanings of literal sentences are no less difficult (I believe) than finding those of figurative ones. (Rumelhart 1979:86)

 In concluding this section I should point out that even those who believe in the existence of direct speech acts usually concede that the literal meaning of a sentence comprises only "a set of truth conditions", and that these have to be used "against a background of assumptions that are not explicitly realized in the semantic structure of the sentence" (Searle 1979:95). What Searle apparently alludes to here is assumptions concerning the identity of referents, universes of reference and comparison (e.g. the standard of comparison and precision applied when sentences like 'The box is big' and 'France is hexagonal' are used) etc. It is indeed remarkable that somebody can admit this and at the same time claim that there are cases (of so-called direct speech acts) when n literal sentence meaning and speaker's utterance meaning are the same" (Searle, op. cit. :96). Moreover, apart from this whole complex of referential specification, we must - as I pointed out earlier - add the immensely important aspect of the "why of communication" to the list of items that are usually ignored in linguistic semantics.


7.6 The interpretation of utterances: The roles of the speaker, the listener, and the outside observer

 It is commonplace that linguists talk about the (semantic) interpretation of a given utterance, discourse or text. However, this unique, "correct" interpretation is a fiction. There are always many - in principle infinitely many - interpretations of a given utterance available. (It is an entirely different matter that most of these are never computed or attended to in the concrete communicative situations). For one thing, we may distinguish between the speakers, the listeners and an outside observer's interpretations. Furthermore, in neither of these cases is there any single "correct" situational interpretation.

First of all, it is usually not the case that the speaker has a single intention in mind; there are often many motives, reasons, desires and intentions behind a speaker's utterances. Some of these may be partly contradictory, they vary a lot as regards importance, consciousness etc, and some of them are recognized by the speaker only after the utterance has been produced. The proper interpretation of an utterance is often negotiable, and the listener may often convince the speaker afterwards that a certain interpretation of his utterance is reasonable although it was never (consciously) intended by him.

Similarly, the listener's interpretations are relative to many factors, such as his background knowledge, his intelligence and imaginative power, his degree of attention and interest, his social relations to the speaker and his actual social interaction with the latter, his willingness to think over what has been said and what follows from that. We all know that one single utterance may occasionally give rise to a long - in principle infinite - series of conclusions, associations and other reactions on the part of the listener. There is no non-arbitrary point where this process of understanding of a given utterance stops and some other activities take over. Also if we disregard the off-line reflection occurring after and outside the inter-action itself (afterthoughts etc), and only consider the direct on-line understanding of what is part of the social interaction in the communication situation, we must admit that there is a wide range of possible situational interpretations on the part of the listener.

The more or less "objective" observer, who analyzes a dialogue from outside, naturally does rather different point of view. He is not directly engaged in the social interaction, and is therefore detached from a number of ties, often emotive in character, which hamper certain interpretations on the part of the parties involved in communication. Speakers and listeners have seldom "time or mental capacity to contemplate past behavior, monitor present behavior, and plan future behavior" (Storms 1973:166), and hence they are likely to remain ignorant of many meanings and interpretations that could be attributed to their behavior. It also appears that, in many situations, a given speaker A is more prone to explaining his own behavior by reference to situational factors than is his listener-addressee B, who, in trying to explain the same phenomenon, s behavior, may assign more importance to A's dispositional properties as causal factors. This may be due to differences in information about the event, behavior, and context that is available to the speaker as opposed to the listener, and to differences in how this information is processed (Jones & Nisbett 1971). After all, an actor, e.g. a speaker cannot freely observe himself while acting, and there may also be motivational reasons for avoiding too much self-observation (Storms 1973)

In all these respects, an outside observer is of course much less constrained than both communicating parties; he can explore the whole situation (especially if he has access to videotapes) and exploit his background knowledge (perhaps a vast scholarly expertise). On the other hand, he lacks first-hand knowledge of the actor's actual motives, thoughts and feelings, and may therefore at times misrepresent their intentions and expectations. In addition, it is important to realize that observers are also dependent on differences in background knowledge, interest, willingness to consider different alternatives etc, and therefore we will get a broad spectrum of possible interpretations also on their part.

Linguists seldom find it worth while discussing the multiplicity of situational interpretations of texts and discourses. It is often left unclear whose point-of-view, the speaker's, the listener's, or the observer's, is applied, when meanings and interpretations are assigned and analyzed by linguists (psychologists, anthropologists etc). (This is not to say that it is always an easy task to disentangle the various possible and/or actually occurring interpretations of particular discourses, complexly intertwined as they are). In practice, however, the interpretations are of course mainly those of the expert-observer, i.e., the analyst himself. This specialist (linguist, psychologist, or whoever) must of course have a very profound knowledge of the language and the culture he is analyzing. He must be a native speaker, or, at least, work in close cooperation with native informants (cf. also the participant-observer role in social anthropology). As we noted above, the expert-observer's role is very different from that of a speaker or a listener actively engaged in a social interaction. Yet his assessments are hardly divine, they are not objective in an absolute sense, since they are constrained by particular theories and attitudes, some of which are dependent on specific academic traditions and schools, while others are perhaps more generally typical of the socio-cultural class to which academics belong. In a way the expert-observer may be regarded as a specific type of receiver-interpreter acting under very special conditions. Yet we often regard our own assessments of grammaticality, meaning and meaningfulness etc as the only correct, as-it-were objective evaluations possible (as if they uncover the true essence of language). We tend to regard the primary interpretations made by real speakers and listeners as uninteresting, affected as they are by various distractions, memory limitations, accidental situation-specific needs and impulses. This is, in my view, unfortunate, since we need both empirical studies of how meanings are created and assigned in actual discourse and analyses performed by outside expert-observers. What some of us do, instead, is to regard the expert-observers purportedly correct interpretations as ideal, as something which should be attributed to "ideal speaker-listeners", i.e., those imaginary actors who live "in a completely homogeneous speech-community" and who "know" their "language perfectly" and are "unaffected by such grammatically irrelevant conditions as memory limitations, distractions, shifts of attention and interest, and errors (random or characteristic) in applying (their) knowledge of the language in actual performance" (Chomsky 1965:3) [6].

It would be absurd to argue that the three perspectives of the speaker, the listener, and the observer could be equated, when it comes to the analysis of verbal communication in face-to-face interaction. Note, however, that it is a little less absurd to do so, if we concentrate on the interpretation of written texts. There the reader is much less constrained by time limitations, and the social pressure of other people, including the sender, is much less obvious. The reader can rely on longtime reflection and make extensive use of his encyclopedic knowledge and the various types of expertise that he happens to possess. Therefore, the distinction between the receiver-addressee and the outside observer becomes much less clear. The image of the expert-analyst as a generalized ideal reader turns out to be, if not always fruitful and applicable, so at least sometimes a reasonable first approximation. It is, therefore, understandable that linguists dealing with written language and written texts have not felt any strong need to make a clear distinction between what the expert-observer does and what ordinary readers do (or should do). This automatically leads us over to the practice of literary analysis and text interpretation, which I will discuss in the following paragraph.


7.7 The interpretation of written texts

'Es liegt alles am Wort' (Martin Luther)


In the preceding section I argued against the theory that there is a subclass of speech acts, for which it is possible to assign more or less directly or automatically a unique 'literal' interpretation ('direct speech acts'). I will now advert for a while to written texts and make essentially the same point there. Although written texts are relatively more autonomous as far as the semantic content is concerned, it can never be true that the whole interpretation may inhere in the text itself. One must always draw upon extratextual factors, something which is now becoming more widely recognized among text linguists (cf. e.g. de Beaugrande & Dressler 1981).

The thesis of semantic autonomy shows up not only within linguistics proper, but sometimes also in the theory of literature. Some scholars apparently favor a view that there is, at least in some cases, one unique proper interpretation of a given piece of literature, say a poem. The standard view is then that the correct interpretation is one which experts agree upon after having analyzed the text meticulously and with due attention to all the features in it. The text is thus treated as an autonomous linguistic structure, the understanding of which demands no more and no less than a proper linguistic (or perhaps philological) exegesis.

It should be noted here that there is an alternative conception of the unique and correct interpretation. According to this latter theory, the right interpretation is simply the one that the author intended. However, we have already seen that speakers often post hoc admit that their utterances can be ascribed reasonable interpretations that they never intended them to have, and the same thing applies to writers and texts. The theory under discussion is therefore fraught with difficulties. Accordingly, it is often dismissed with the following argument.

If the poet did not succeed in expressing what he had intended, then his intentions (whatever they were) are uninteresting so far as the reader's analysis is concerned; the only thing that counts is what the author has actually written.

However, the unique correct interpretation in the sense of something intersubjectively shared among expert observers is also a fiction. There is no such thing as the only (correct) situational interpretation. There are lots of interpretations which are more or less justifiable, depending on what kinds of background knowledge the interpreters have but also on such things as their reading strategies, intelligence, imagination, willingness to pursue their interpretive activities for a sufficiently long time etc. Note, however, that the theory of the correct interpretation often serves to turn literary analysis into a normative rather than a descriptive discipline; this theory does not encourage the scholar to investigate what kinds of interpretation people actually make of a given piece of literature.

 The quotation used as a vignette for this section is intended as a reminder that the thesis of a unique and correct interpretation in the text was vigorously propounded by Martin Luther (see III.1). Luther argued that the proper interpretation of the Holy Scripture was to be found in the text itself, in 'God's pure words'. In his opinion the bible was far from being obscure or ambiguous; on the contrary it was a clear and unambiguous message which could be decoded by anyone possessing the necessary knowledge of the language. Here was an enormously strong impetus for doing linguistic analysis and philological exegesis Form and content (in the sense of (situational) interpretation) were directly coupled. Therefore, the study of language and the understanding of the gospels were inseparable activities.


7.8 Logic and formal languages

The idea of the semantically autonomous text is most clearly and radically reflected in the traditional conception of logic and of what is required from a logical argumentation. The language of formal logic is a regimentation of certain forms of written language. In the light of what was reviewed in III.1 it seems unbelievable that logic could have been developed at all in the absence of written language.

The symbols and symbol combinations of formal logic are supposed to have exact semantic interpretations, which must be mechanically computable without recourse to any extratextual n information. Accordingly, in order to perform successfully on tasks of logical argumentation one must be able to draw the right conclusions using only the explicitly given linguistic formulation and its supposedly unambiguous, exact semantic interpretation. One must not ask oneself whether the task squares with one's experience of everyday reality, whether the conclusions make sense in real life situations. However, this is exactly what ordinary people tend to do when faced with logical tasks. This of course holds most clearly for children, who are very dependent on their experience from real life contexts. [7] It is well known that Piagetian logical tasks may well be solved by quite young children, if the problems are applied to subject matters which the children are familiar with, whereas they cannot be solved by much older children if they are couched in abstract symbols (Wason & Johnson-Laird 1972, Donaldson 1978, Rommetveit & Blakar 1978). The same applies to normal adults, as Olson (1972) points out in his review of some studies:

"Formal reasoning has led to a reliance, where possible, on the use of symbols related by a logical calculus. To illustrate the difficulties, I will use three studies from our laboratory. Bracewell (Note 2) has shown that the simple propositional statement employed by Wason and Johnson-Lalrd (1970), "If p is on one side, then q is on the other," is ambiguous in at least two ways: "one side" may be interpreted as referring to n the showing side" or to "either the showing side or the hidden side"; "if...then" may be interpreted as a conditional relation or as a biconditional relation. Differences in subjects' performance can be traced to different interpretations of the proposition. In a similar vein, Hidi (Note 3) has shown that if a simple proposition such as "if you go to Ottawa, you must travel by car" is understood as describing a temporal event, subjects draw quite different inferences than if it is treated purely as a logical statement. In a developmental study, Ford (1976) has shown that, given a disjunctive statement, children (and adults in natural language contexts) treat "or" as posing a simple choice between mutually exclusive, disjoint alternatives (for example, "Do you want an apple or an orange?" "An apple. n ) When children of five or six years of age are presented with Nor" commands involving disjoint events as well as overlapping and inclusive events - the latter being involved in Piaget's famous task "Are there more rabbits or animals?" - Ford found that children's logical competence breaks down only when the known structure of events runs counter to the presuppositions of the language. Rather than revise their conception of events - rabbits and animals are not disjoint classes - children misinterpret or reject the sentence. They say, for example, "There are more rabbits because there are only two ducks!"" (Olson 1977:274)

Stenlund (1980) has subjected the ideas behind formal logic and analytic philosophy to a penetrating and very interesting analysis, which leads up to a rather negative assessment of their "world view". His subject matter is thus the development of logic and mathematics into pure "non-empirical" sciences, disconnected from their practical applications and from their origin in real life problems. In particular, Stenlund's objective is to show that analytic philosophy and logic depicts the world as made up of objects (cf. I). Here are a couple of quotations:

"They (i.e. logicians/PL) structure language in the same way as they imagine the logical form of the world: in quantitative part-wholes. They think as they speak: in outer forms, in writing. They think in terms of writing - begriffsschriftlich' (op.cit.: 99 ).
'Logics as a phenomenon, as that which rests upon the separation between content and form, between living and thinking, is perhaps that subject which better than anything else brings out the "true nature" of object thinking' (op.cit.:116).

Stenlund argues that Frege's logic in terms of arguments and functions blurs a number of distinctions that are important in natural logic; for example, many of Aristotle's categories are collapsed (cf Stenlund, op.cit. :146) [8] As a simple case in point we could here draw an analogy with the linguists now popuular conceptions of semantic structure, according to which such a fundamental distinction as that between attribution and predication is mistreated (cf. VII.4).

"Frege was a formalist - of that there is no doubt. He belonged to the "higher" formalist school where rashness and arbitrariness are abhorred, where not just anyone may determine the rules.
 Thus it is here that our mathematical logic comes into the picture, and never before has logic been so pure, so free from "foreign elements". Questions regarding the origin and development of concepts become irrelevant from the viewpoint of logic; modal, grammatical and linguistic aspects become irrelevant from the viewpoint of logic; (one recalls Frege's bitter polemic in Begriffsschrift and Grundlagen against those who include such considerations in their logical analyses); epistemological considerations become irrelevant from the viewpoint of logic and the same applies to ethical and aesthetic considerations and questions concerning use, interest, aim and intention. In brief, everything that gives human concept formations their content and value becomes irrelevant from the viewpoint of logic! It has been possible to free logical forms from all content, taking the new mathematics as the model. It has been possible to formulate the laws of "pure thought" by undermining the concept of content or meaning itself; by inventing purely logical content (Frege's begriffliche Inhalt), a "formal content" which logical semantics (now also beginning to be set over against linguistic semantics) has later to take care of.
Modern logic, primarily through the discovery of formalization (in the arithmetical-algebraic sense of the word), constitutes the culmination of the separation of form and content, of the "purification" of theoretical thinking that began with Aristotle. One could perhaps say that Frege cuts the last tie - ordinary language. Ordinary language became ein Hindernis, as Frege puts it in his Begriffsschrift, in the search for the true language, the logical language (in the same way as ordinary reality has often also constituted an obstacle to philosophers in their search for true reality). The important thing here is not just the fact of the discovery of formalization; what is essentially new is the spirit that makes possible the acceptance of this discovery - the spirit according to which formalization becomes something respectable, legitimate and praiseworthy (consider, for example, the formalistic ideal for evidence for which Frege constantly agitates in his longer writings and contrast this with Rant s clearer conception of mathematical proofs - even if this too was a "pure" way of thinking). (Stenlund 1980:156-7) [9]

The thesis that logicians tend to think of the world as made up of static objects ties in with the observation that logic is poor at conceptually analyzing change. "A logic which is concerned with the conceptual scaffolding of a dynamic world is still largely a desideratum" (von Wright 1965). Ejerhed (1979) notes that this also holds for attempts at constructing tense logics. In logic, change is consistently represented as the difference between states, rather than taken to be a primitive and basic notion in its own right. Yet, as Ejerhed notes, changes may well be more important than states in animal and human perception and cognition (cf. also Miller & Johnson-Laird 1976). It may be added that change is clearly much more salient as a basic dimension in the production and understanding of face-to-face interaction than in (most aspects of) the communication by written texts.

In concluding this section on logic and formal languages, let me call attention to a completely different point having to do with the written language bias in modern linguistics. Logicians have traditionally argued that natural language - being so vague, ambiguous, elliptical, "illogical" etc - impedes logical thinking (cf. the quotation above). This was, and presumably still is, the main reason for developing the formal languages of logic and mathematics as more exact and reliable vehicles for n logically correct" reasoning. These languages were developed on the basis of natural language in written form, and they are in fact used as very special, strictly regimented variants of written language (various forms of symbolic notation). There are entirely explicit rules for the outer form and the interpretation of these symbol languages. Now, it is important to notice that very influential scholars in modern linguistics have borrowed largely from logic and mathematics their conceptions of what a language is (or should be). We have witnessed, in the last two decades, how linguists and philosophers have argued that natural languages are, after all, not that different from formal languages. In fact, the former should, according to some currently popular doctrines, be amenable to just as exact characterizations as the latter. In our recent history, this argumentation has been carried through in two steps. First, Chomsky (e.g. 1957) argued that the syntax of any natural language should be given in terms of an axiomatized finite system of rules capable of enumerating all and only the grammatical sentences of the language (cf. V.4). Later, Montague (1974) proposed that the semantics of a natural language be given in terms of a formalized, intensional logic. Thus, he declared programmatically: n I reject the contention that an important theoretical difference exists between formal and natural languages" (1974:188). As we all know, Chomsky s and Montague's views have had a tremendous impact on modern linguistics. In view of the fact that formal languages are so dependent on writing and written language, these currents in today's mainstream linguistics have increased the written language bias.



7.9 Semantic representations

A written sentence or text is static and atemporal. It is in principle simultaneously accessible in its entirety, although, its production and the single scannings in its perception are of course distributed in time and subject to time limitations

On e.g. short term memory. The left-to-right organization of a written sentence is spatial rather than temporal. It is typical, however, that linguists, and similarly (but perhaps less consistently) experts in philosophy, psycholinguistics, artificial intelligence etc, regard the various "linguistic representations" (semantic, syntactic, phonological) of sentences and discourses as equally static and atemporal, no matter whether they are concerned with writing or speech. It seems to be tacitly assumed that what is true of a written sentence is also true of a spoken utterance, as if the distinction between sentences and non-sentence-like utterances were uncontroversial, and the transposition from one medium to the other unproblematic.

It is commonplace to argue that the surface-syntactic structure of a sentence or utterance exhibits both linear structure (left to right organization) and hierarchical phrase structure. The latter point is, however, rather debatable; it has not been proved that surface structures must have connex tree structures with single roots, [10] and this of course holds a fortiori for spoken utterances. As regards the left-to-right organization, it reflects the temporal structure of the corresponding stre2m of speech behavior, but this fact is not incompatible with the contention that the linguistic structure is atemporal, i.e. (at some point in time) simultaneously existing, and static. In fact, most theories of speech performance assume that in production the linguistic structure (of the whole clause or sentence) is first constructed in its entirety and hence exists before it is executed in overt speech, and similarly this very same structure is assumed to be assembled at some point in time during the perception process.

It seems, however, that at least some linguists and psycholinguists are becoming increasingly inclined to admit that spatial left-to-right dimension of the written sentence has a temporal rather than a spatial counterpart in the spoken utterance, also as regards its linguistic organization. But this is usually applied only to (surface-) syntactic and phonological structure, not to semantic representations. The atemporality and staticness of the latter seems indisputable in the eyes of most people; after all, many linguists have maintained that the semantic representation of a sentence or utterance displays only hierarchical structure and no linear (left-to-right) organization. The question that now arises is: Is this really self-evident? Let me suggest a few reasons why it need not be that way.

First of all, if we are willing to assign to the outer form of an utterance a temporally distributed linguistic structure, then we should do the same as regards semantic structure. After all, semantic structure and outer form are strongly interrelated and are best considered two aspects of the same activity. Semantic structure may be thought of as consisting of various conditions or structure assignments on continuous mental activities, i.e. the stream of consciousness, the flow of interpreting activities accompanying either overt verbal behavior (utterances) or covert behavior, i.e. the kind of thought that is heavily dependent on verbal means ( n thinking in words n ) Thoughts and verbalizations arise, grow, get completed and disappear with time, and semantic structure and-outer form are aspects of this temporal process (cf. VII.10). Apart from these admittedly abstract characterizations, we can, furthermore, note that there are many semantic aspects, traditionally discussed in semantic theory, which seem naturally amenable to a characterization in terms of dynamic and temporally distributed dimensions; some examples are functional sentence perspective (FSP), theme-rheme, background-focus etc.

Now, linguistic structure is of course something very abstract, and must not be equated with the behavioral processes themselves (VII.10). Therefore, it is a moot point whether this structure is temporal as such; perhaps it would be a category mistake to argue in that vein. What remains true, however, is that the processes which are subject to linguistic structure (and without which there would be no language), are temporally distributed, and this fact must be given due attention in our analysis. [11] In the production and comprehension of speech, the entire configurations of phonological, grammatical and semantic conditions corresponding to the "linguistic structure" of sentences and larger units (i.e. units of the kinds that linguists generally contemplate) are presumably not simultaneously attended to and/ or otherwise mentally available. Rather, they are subject to piecemeal processing.

The reader should note that so far I have only discussed the semantic representations of utterances, i.e. the verbal means used in communication processes. I will presently advert to the nature of the representations of knowledge, the memory models of discourse universe, that are built up, exploited, and altered in the communication process. Could knowledge structures be considered atemporal and static, even if we admit that utterances and thought processes are time-distributed and dynamic?


 7.10 Knowledge structures

 In the fields of semantics, artificial intelligence and cognitive science, an important theoretical issue concerns the ways in which speakers store and use the knowledge that they need in order to communicate, and to perceive and understand language and the world. Scholars have assumed the existence of different types of knowledge structures, some termed "models", some depicted as networks of propositional knowledge. Another distinction is that between procedural knowledge and declarative knowledge (Winograd 1975). However, it is far from clear what is actually implied with these distinctions. Johnson-Laird (1980) wrestles with several of the basic issues and concludes, contrary to some other commentators (e.g. Pylyshyn 1973, Anderson 1978), that there are indeed a few differences in the empirical implications of postulating mental models as opposed to (merely) propositional knowledge. While propositional knowledge is supposed to be language-like in character (discrete and digital rather than continuous and analogical).

 "a model represents a state of affairs and accordingly its structure is not arbitrary like that of a propositional representation, but plays a direct representational or analogical role. Its structure mirrors the relevant aspects of the corresponding state of affairs in the world". (Johnson-Laird 1980:98)

 Mental models are not pictures (such a view leads to the well known absurdities of assuming an infinite number of homunculi), but nevertheless they seem to be some kind of (abstract) images. They allow the construction of subjectively (phenomenologically) accessible images, which are, among other things, amenable to mental transformations such as rotations and expansions (Cooper 1975). Kosslyn (1980) suggests that mental imagery draws on at least two codes, an analogical "surface image" code that arises from some kind of "deep structure" code (the nature of which is unknown) and a propositional code which contains information that can also be used in the process of generating mental images.

Cognitive science is vital field, which shows a lively discussion of theoretical and philosophical issues, and it is of course impossible to do justice here even to the most important arguments. However, a few reflections may be called for in the light of our own major topic. One cannot help observing that most of the knowledge structures of the types postulated in artificial intelligence and cognitive science seem to display some fundamental properties of written or pictorial representation, cf. networks or graphs with nodes containing propositions, and analogical "mental models n respectively. This in turn suggests an interpretation according to which a knowledge structure is conceived of as a more or less fixed, static and context-free object, which is exploited in thinking and communication,-where various parts of it would be retrieved, displayed, combined or otherwise manipulated, and finally perhaps transferred in communication (as if communication were a kind of transportation of ready-made pieces of information, cf. X.1).

The assumption of large-scale integrated coherent knowledge structures is thoroughly unrealistic. It seems unlikely that we have ready-made models of all (or most) of the various parts of the world that we are acquainted with. Our knowledge of any universe of discourse has to be fragmentary and flexible, so that it can be used in many ways and for different purposes. It seems impossible that any normal language user could have a stable model of, say, the government of the United States. The reason is that his knowledge is by necessity multi-faceted and can therefore be organized under many different aspects. One could, e.g., focus on the present U.S. government in relation to the previous ones or possible alternative administrations, or one could think of U.S. governments in general and their functions. This in turn could be looked upon from different angles; what status i8 allotted to the government by the constitution, and what are the actual roles played by the U.S. government in practice? The U.S. government may be seen in relation to foreign governments or to the various sectors and layers of the American society. One could compare the functions of the government as they are with what they should be, one could consider the staff as governmental officers or as particular human beings etc.

There is an infinite set of different aspects, under which we can analyze reality and organize our knowledge of it. Human beings encounter an infinity of situations in life, which are all different with respect to which facts are possibly relevant, which are actually relevant, and, among these, which are essential and which inessential (cf. Dreyfus 1972:168ff). In these situations people behave in ways which are usually orderly but not governed by well-defined rules. Nor can such intelligent behavior be explained by recourse to well-defined "models of the world", because any particular well-defined model, i.e. a specific organization of knowledge of a given section of the world, would be too unadaptable and too dependent on the purposes for which it was originally set up.

The assumption of integrated stable knowledge structures does not square with the openness of the human mind. It seems to underrate strongly the agent s reorganizing activities in recall, association, verbalization and argumentative reasoning; the subject matter is exploited in a creative fashion and in constant interaction with the environment and the specific tasks to which the activities are oriented. No static model of the world could satisfy the needs of such an active subject. In fact, we could say with Dreyfus (1972:212) that human beings utilize not a model of the world but the world itself.

It remains true, of course, that humans make use of memory-stored knowledge. It is also evident that there are various specific associations in this memory-stored knowledge, although the whole structure cannot be very stable and integrated. When we use our knowledge, only very limited parts can be attended to at one time. The stream of consciousness moves from fragment to fragment. Perhaps these fragments are what Chafe (1979) called 'foci', which "are, in a sense, the basic units of memory" and "represent the amount of information to which a person can devote his central attention at any one time" (180):

My guess at the moment is that people have in memory a large number of foci involving knowledge of particular objects, events, and so on plus a relatively small number of principles of coherence by these foci can be organized into the larger units that appear in language as sentences. These principles of coherence have to do with unity in terms of images, schemas, goals, and the like. (Chafe 1979:176)

It may be objected that it might be possible, after all, to build up a well-organized model in at least certain cases, where the tasks demand so, e.g. if one has to read a certain text, memorize it and then later relate it. But again we are probably not faced with a very stable structure, like a chart, a picture or a written text. Rather, our knowledge remains based on fragments ("foci"), but in addition we may have memorized certain tricks that enable us to go from one fragment to the other in a certain more or less fixed order. Thus, though our knowledge may consist only in local pointers (at a given place we know how to proceed to the adjacent fragment), we may still be able to follow consistently a specific path through the whole story. This path in memorizing and recall would then reflect the temporal organization of the discourse.

Obviously, we cannot here make any attempt at solving the problems of what kinds of knowledge structures must be posited in cognitive psychology and artificial intelligence. Such an endeavor is entirely beside our main point, which is merely to suggest that some tendencies in current thinking on knowledge are biased by the traditional written language focus. We often think of knowledge in terms of (more or less) atomic facts organized in charts, diagrams, pictures, lists, tables, networks etc. Such structures can only be produced if we are allowed to use paper and pencil, or, on a much more sophisticated level, a computer. Computers represent a natural extension of the applications of written language. They use written language rather than spoken language, and to the extent that linguistic behavior has been simulated in artificial intelligence, it is not the dynamic social interaction in spoken dialogues that has been aimed at. [12]

Computers work with language and knowledge on a digital basis. Now, it cannot be denied, of course, that all (human or nonhuman) kinds of cognitive processing of sensory inputs seem to involve such things as selection, quantization (perhaps digitalization), categorization, assimilation to schemas etc. Nor should we ignore the fact that our inability to observe directly mental processes as such leaves us with the sole possibility of characterizing the processes in terms of their outcome (IV). But I would maintain that writing strongly reinforces these things and specializes them in certain directions. This has to do with important inherent limitations in the media. For example, the lack of a context must be compensated for in written texts and in computers. Humans constantly interact with their environment during spoken communication, but computers have to have an internal model of the environment with a very large, almost infinite amount of data ("knowledge") stored. Computers do not get anything comparable with normal sensory input. It would therefore be preposterous to assume that computers could simulate the human condition in a veridical way (cf. Dreyfus 1972).

With computers we can produce complex structures, pictures and lists, in which great amounts of richly structured elements can be integrated and systematized. The products are preserved over time and can be surveyed, manipulated, adjusted and improved, reorganized in part-wholes etc, in principle without limits and without the interaction of oblivion and the limitations on normal human attention. Are there any interesting analogies between this and the knowledge structures set up and used in oral communication, where the linguistic products are fugitive, the spans of attention and short term memory limited, and the flow of incoming signals and impressions of the social interaction virtually endless? The "knowledge" involved in the social interaction of spoken dialogues is to a large extent practical know-how, skills of various kinds, and this crucially involves emotive and social (phatic) aspects, i.e. not only cognitive aspects such as intellectual knowledge of atomic facts. Is this at all amenable to simulation on digital computers?

In concluding this section, it would be fair to point out that the attitudes to knowledge that we have attributed to artificial intelligence and to the computer sciences in general, have a much wider societal basis than we have so far assumed. It seems to me that there is in our Western culture a ubiquitous tendency to conceive of "knowledge" as something which has only objective referential aspects; we tend to look upon knowledge as facts, as fixed and stable pieces of information such as can be found in text-books, waiting there to be acquired and used, knowledge as something which is already complete and stored in archives and repositories. Such a conception of knowledge is very narrow. It leads to difficulty in explaining our abilities to combine basic elements in new ways and to apply them to new problems, and it does not take into account all the emotive, practical and social [13] aspects of our social interaction and, in general, in our ways of dealing with our environment in a flexible, intelligent manner. Whatever the merits and limitations of this conception of knowledge are, it reflects the profound impact that the written language bias has had on our entire culture.


7.11 Semantic representations and thought processes

"All the major figures responsible for the classical phrasing of linguistic relativity, Herder, Humboldt, Weisgerber, Trier, Cassirer, Sapir, and Whorf, erred in precisely the same way. Each, under the sway of the Western inclination to objectify proces3es - an inclination characteristic of peoples who embrace alphabetic writing - has distinguished grammar (langue) from speaking. Each has hypostatized a socially shared grammar and has assumed that grammar to be an object which each individual reproduces in his mind. Each in different words and in different ways has claimed that mental life is identical to that grammatical object. The error of linguistic relativity does not lie so much in the claim that mental life has no independent existence outside of language, but that mental life is identical with language as an object rather than with language as a process". Washabaugh 1980:210)


For various reasons, one of them being the long-time insistence that linguistics must be an independent discipline with its own object of study (V.1), linguistics and psycholinguists have been prone to postulate "mental objects" and "mental processes" that are claimed to be specifically "linguistic" in nature. Such an assumption appears to distort reality at several levels. For example, what a speaker produces when he constructs and executes an utterance is not a mental "linguistic object" e.g. a plan) but a stream of phonetic behavior ("vocalizations") such that it meets certain conditions defined by its expression plan ("outer form", cf. Linell 1980). When a listener perceives, say, a word, he does not perceive a linguistic (phonological or whatever) form, e.g. a string of phonemes, but he perceives a section of someone's phonetic behavior as a word, i.e. as a stretch of behavior that is conventionally associated with a set of semantic, grammatical and phonological conditions. Similarly, when someone understands an utterance, he does not construct a semantic representation for the linguistic expression involved, but he uses semantics as clues in the search for an interpretation in the environment, in his inner world or in both (cf. VII.3-5).

Furthermore, it seems to me that many versions of information processing psychology, artificial intelligence etc. easily lead to the hypothesis that mental operations are carried out in "the language of semantic representations" (or deep structures, transformations), that thinking consists in the application of semantic procedures etc. For example, this view is articulated in Fodor's "The language of thought" (1976). [14] Such a view may easily force us to accept a strong version of the Sapir/Whorf hypothesis, something which must be regarded as a rather unfortunate consequence. It is of course impossible to discuss the enormous problem of the relations between language and thought here, but it may be remarked that our traditional theory of linguistic rules as conditions on behavior [15] seems to make a more natural and plausible alternative available. Rules are not processes but conditions on communicative behavior and other kinds of behavior such as perceptual exploration, imagery, problem-solving etc (cf. VIII.4). Thinking is undoubtedly a form of very language-dependent (internal) behavior, and as such it may follow, or be subject to, semantic rules of various kinds, but it does not consist in the manipulation of purely "linguistic" semantic representations. The latter view would turn language into a completely self-contained system without any basic relation to the physical and social world. 







1. More about this notion in VII.5.


2. These three attempts to legitimize the distinction between semantics and pragmatics are all important, but I do not claim that they exhaust the set of possibilities of drawing the boundary line. See for some discussion Allwood (1981).


3. For this term, see VII:5.


4. A somewhat different variant of this theory would involve the claim that semantics deals with conventional meaning, and pragmatics with nonconventional (natural) principles for assigning interpretations (cf. e.g. Grice 1975, Sadock 1979).


5. Many semanticists would prefer to talk of truth conditions (VII:4). Accordingly, the linguistic meaning of a sentence type would define the conditions that situations must fulfil in order for the sentence to be true of them.


6. I quote Chomsky (1965) here in full cognizance of the fact that he is not concerned with hypotheses and claims about situational interpretations of utterances.


7. The same seems to apply to non-schooled individuals in illiterate cultures. Scribner (1977) summarizes several cross-cultural studies on verbal reasoning, and points out the consistent differences between schooled and non-schooled subjects (i.e. subjects acquainted vs. non-acquainted with written language and the premises of written communication) (from several mutually unrelated cultures in Africa, America and Central Asia) with respect to their performance on logical tasks (solving syllogisms). Non-schooled subjects base their judgments on their own everyday experience of factual matters (i.e. matters having to do with the content of the syllogisms), whereas schooled subjects are much better at basing their conclusions exclusively on what is explicitly stated in the premises of the syllogisms.


8. Cf. Also op. Cit.: 152-7.


9. Translation by Brian Beattie.


10. cf. Hetzron 1973, Hudson 1980, Dahl 1980.


11. I am not convinced that the adoption of a procedural semantics instead of, say, semantic networks and dependency trees makes any difference in this regard. On procedural semantics, se e.g. Johnson-Laird (1977).


12. It is true that there are a few studies which attempt at dealing with "artificial intelligence and conversation" or with simulations of dialogue exchanges (e.g. Allen & Perrault 1980, Perrault & Allen 1980). However, apart from those studies of natural spoken discourse which have not led to program formulation or actual simulation (and hence have nothing particular in common specifically with computer sciences) (e.g. Reichman 1978), it seems that this work is entirely based on the medium of written language. m e only property which it shares with spoken discourse, is the interactive character, the relatively bidirectional flow of questions and answers between system and user.


13. On these "functions", see IX.2.


14. For critical discussion, see Olsen 1979, Fortescue 1979, as well as the now classical critique of cognitive simulation and artificial intelligence in Dreyfus 1972. See also Winograd's (1980) reorientation.


15. Thus, e.g., a sentence is an abstract entity consisting of a configuration of grammatical conditions (that may or may not be met by specific stretches of speech or written language (VI.1). Phonological structure is seen as phonetic conditions on the stream of vocalizations (VIII.4). Similarly, in VII.9 semantics is described as conditions on mental activities.