Table Of Contents
5. LINGUISTICS AND THE OVERALL THEORY OF LANGUAGE
"La tâche de la linguistique sera:
5.1 Linguistics and other language sciences
It seems appropriate to start our survey of trends in traditional and present-day linguistics by asking how linguists tend to define (explicitly or implicitly) the scope of their discipline. 'Linguistics' is a term which covers certain types of scientific approaches to language; it does not denote all sciences or scientific activities that are concerned with language and the use of language. On the contrary, both insiders and outsiders would identify linguistics as a humanistic discipline which analyzes language only in some specific ways and respects.  Two historical factors are responsible for this state of affairs. First, there has been a time-honored scholarly tradition directed towards the study of certain aspects of language. This tradition used to be called an art, but today the designation science is more fashionable. Secondly, this tradition has been institutionalized in the traditional academic systems of the Western countries, which means that there are special departments of linguistics and the linguistic study of specific languages.
What then is the traditional content of linguistics? Originally, i.e. through most of our long history almost up to the 19th and 20th centuries, it contained phonology, grammar (especially morphology and syntax (which, however, did not consider word order very much) and some of their semantic aspects), logic, and rhetorics, and also philology (text interpretation etc.). Later, philology was (temporarily?) relegated from linguistics (because it did not study language for its own sake), and something similar happened to logic and rhetorics. Hence, according to our 20th century conception, linguistics is basically phonology, grammar (morphology, syntax) and semantics (especially word semantics; sentence semantics is a rather recent rediscovery). In the 1970's logic and rhetorics have also been reincorporated under the headings of (formal) semantics and pragmatics, respectively.
In a more global perspective, the linguists' approach(es) to language must be considered narrow, and this is also true of today's linguistics. For example, one very important aspect of language which is seldom or never discussed in linguistics (and this also applies to psycho-linguistics) is the question why people use language, for what purposes utterances and written messages are compiled and used (cf. ¤ 7.1 on the why of communication). Instead the typical linguist's attitude is this: Given that people use messages formulated in natural languages, i.e. in terms of linguistic expressions, how are these expressions structurally organized, and what meanings can be attached to the expressions qua items in an abstract, supraindividual linguistic code (= la langue, cf ¤ 5.2). Accordingly, linguistics in the usual traditional (though still widely accepted) sense is narrow in scope, (Lyons, 1981:36, has proposed the term 'microlinguistics' for it). Many scientific studies of the structure, function and use of language simply do not belong to linguistics in this traditional sense. Such fields as the psychology, sociology and philosophy of language are concerned with linguistic phenomena, but they are said to do so from the vantage points of other disciplines (outside linguistics). Such scholars as Wundt, Mead and Wittgenstein do not belong to the history of linguistics, although they made contributions to the study of language which compete quite well with those of, say, Humboldt, Sapir, Saussure and Chomsky. Even such borderline disciplines ("Bindestreichlinguistiken") as psycholinguistics, sociolinguistics and neurolinguistics do not really count as linguistics proper, according to many scholars in the field. (They would, however, be included in a more comprehensive concept of linguistics, Lyons' (op. cit.) 'macrolinguistics', which is now gaining acceptance, cf. below).
Linguistics has traditionally been concerned with written language, i.e. the language of the Holy Scriptures, the great authors, and later on simply with what is considered to be "good" written standrads, educated or "received" English, German (German Hochsprache), Swedish, etc. Furthermore, the focus has been on formal rather than semantic aspects. In fact, grammar has for centuries been the kernel of "real linguistics".
The concentration on grammar was quite natural as long as linguistics served the practical goals set up by the church and its allies, the educational system and the national state (cf. III.2). At school, foreign languages were taught - at first Latin and later on other modern languages (in Sweden German, French and English) - and in this context grammar played an important role, especially so because the means for acquiring knowledge of the foreign languages was always the study of written texts. The study of the mother tongue was also completely focussed on the written language, and it does not seem entirely wrong to say that the written standard taught at school was also something of a foreign language for the pupil coming to school equipped only with his spoken vernacular. At any rate, learning to write correctly involves in no small measure learning to follow a number of grammatical rules which do not necessarily hold for the spoken language (cf. 6:3).
But in today's theoretical linguistics, too, grammar theory indisputably constitutes the kernel of linguistics proper. Some linguists would even argue that linguistic theory equals grammar theory. This is, for example, not far from the position actually held by most generativists. It is true that the term 'grammar' is nowadays used in a rather comprehensive sense. Traditionally, 'grammar' meant morphology and syntax (the latter comprised mainly the study of the grammatical functions of words and phrases), whereas today the term is often used to cover the entire "language structure" including phonology and semantics (though the focus is still more on the form that the content).
Thus, there are historical reasons why linguistics developed into a rather narrow, abstract discipline treating certain aspects of language structure as if they were autonomous (and I will have still more to say about this below). We still live by this conception of linguistics, aptly characterized by Derwing (1973) and others as 'autonomous linguistics'. Yet, it should be outdated by now, incapable as it is of providing a real understanding of the nature and functions of language.  Language is embedded in acts of communication and cannot be properly described and understood out of communicative contexts. This holds a fortiori if we want to understand the nature of spoken language. This means that relevant parts of psychology, sociology, anthropology, ethology, neurophysiology, etc., must be taken into consideration; therefore, many scholars in the field (including myself) would now hold that the Bindestreichlinguistiken should be recognized as linguistics tout court. Fortunately, there are right now strong currents leading in this direction; linguistics would then become a cover term for those scientific approaches which deal with human natural language(s).  Such a linguistics would be characterized by methodological pluralism; it should study all significant aspects of language and its use, including its relations to other types of symbol systems and other types of behavior.
5.2 Structure and Use: language vs. parole, competence vs. performance
"Writing cannot be flatly dismissed as an imperfect, conservative quasi-transcription, as has often been done up to the present day" (Vachek 1949:93)
We noted in Section IV that speech and writing are largely based on the same underlying language system and that many categories of analysis should or must be the same for both types of manifestation. But I also pointed out that there are likewise a number of important differences, and it must be an empirical task to find out how far-reaching these are. Obviously, there are considerable culture-specific differences; the spoken and written variants of Chinese are clearly much more different from each other than speech and writing in, e.g. English. However, I cannot here and now go into the problem of what the actual structural relations between speech and writing are. Instead I wish to call attention to the ways in which modern linguistics tends to treat the relationships between the structure and use of language in speech and writing.
If we have to make a very succinct statement, we should say that linguists tend to ignore the differences between the language systems underlying speech vs. writing. It is assumed that there is one single system, what Saussure termed la langue (e.g. the English language = the langue of the English-speaking culture), which underlies both kinds of manifestations. In Chomsky's terms we would assume one monolithic linguistic competence, not an aggregate of partially distinct capabilities, e.g. one for language used in speaking and listening, and one for language used in writing and reading.
However, in addition, the manifestations of language in speech and writing are assumed to exhibit great similarities. Furthermore, 'it is one of the cardinal principles of modern linguistics that spoken language is more basic than written language' (Lyons 1981:11). Writing is usually treated as if it were merely a secondary manifestation, a relatively unproblematic and theoretically uncontroversial reflection of spoken language. Such a view has been held for long; it was emphatically pronounced by e.g. J.J. Rousseau:
Most modern linguists espouse similar views. Thus, Hockett, in his influential text-book (1958), makes only one distinction, that between "language" and "writing" (p.4), and it is quite clear that "language" here means spoken language; "language excludes writing" (op cit:11). Turning to the founders of structuralist linguistics, we find statements like the following:
'Language et ectriture sont deux systemes de signes dis l'unique raison d'e^tre du second est de representer le premier'
'Writing is not language, but merely a way of recording language by means of visible marks'
'For the linguist, writing is, except for certain matters of detail, merely an external device, like the use of the phonograph, which happens to preserve for our observation some features of tthe speech of past times'
Apart from this, the prevailing attitude implies that the actual use of language is linguistically uninteresting, and that linguistics must be concerned with the theory of the underlying system (la langue) or the ideal speaker-listener's competence, not with the performance of actual language users. Note, however, that when performance is referred to, it almost always is the performance of speech communication that is presupposed. Saussure simply used the term parole (literally: 'speech'). Such an attitude is largely unjustified. Although in general speech takes precedence over writing structurally, historically and ontogenetically, there is no doubt that writing is a medium with partly quite different inherent possibilities and functions (see III.1). Therefore, it is simply not true that writing is "merely an external device" used to "preserve for our observation some features of the speech of past times". I will come back to this issue presently.
We now come to what may be seen as a remarkable paradox, viz. the obvious contradiction between what modern linguists say about the relations between speech and writing, and what they actually do in practice. On the one hand, linguists maintain that speech is primary and, therefore, the most (or only) interesting object of study. On the other hand, their actual practice reveals quite different traditions and values. To a very great extent, linguists still study written sentences (often out of contexts) and written texts, and to the extent that spoken language is actually studied it is done with a conceptual apparatus which is more or less totally derived from the experience of written language analysis.
Returning to Rousseau, we note that he protested rather vigorously against the linguists' one-sided interest in written language. He complained that:
Saussure used quite strong words about what he called the "tyranny of letters" (1962:53) which he considered to be a moral perversion ("un fait pathologique") and a sin of idolatry ("un peche d'idolatrie"). Yet, the traditional practice and attitudes of linguists have stayed on. It is true that the primacy of speech is nowadays verbally acknowledged, but the ways in which linguists construe their theories and seek their applications have not been sufficiently revised. Accordingly, the distinctive characters of spoken vs. written communication are poorly understood. Therefore, what is needed today is not only a more adequate analysis of spoken language but also a general recognition of the independent importance of writing. (The latter is something which Derrida (1967a,b) wants to achieve in his so-called grammatology.)
There are obvious differences between speech as opposed to writing, spoken language as opposed to written language, communication via spoken utterances as opposed to written texts, all of which have been insufficiently attended to and inadequately accounted for in contemporary linguistics. It is most noteworthy that there seem to be some differences not only in the performance but also in the language systems underlying speech vs. writing. That is, spoken and written language have partly different standards of their own . For one thing, the substantial content of the linguistic rules is not quite the same for the two media. Secondly, the impact of norms seems to be different. In the dynamic interaction of spoken discourse, where rapid responses are legion, grammatical norms seem to play a somewhat subordinate role, whereas in written discourse the conventional and explicit rules exert a considerable influence. All this has of course to do with the fact that communication in the two media is subject to quite different conditions (II).
We have already pointed out that speech precedes writing fylogenetically and ontogenetically, and that the written code is largely secondary to the structure of the spoken language from a structural-linguistic point of view. But speech is not primary in all respects, and writing is not always merely a secondary reflection. Some features of the written language are clearly structurally prior to features of the spoken language. To use but one well-known example, written forms sometimes give rise to new speech forms, rather than vice versa; e.g. [ au ke'~l~ for O.K., t ~Łn ~ for 10 p.("10 pence"), t p~j eitS dij~ for Ph.D. ("doctor of philosophy"). In general, there seems to be a bidirectional interaction between the competence to speak and understand speech, and the competence to write and read. It is highly likely that the competence for spoken language of a literate person differs from that of the illiterate. Unfortunately, there are too few empirical studies of what happens to people's total linguistic competence when they gradually become more literate. However, there are some hints that our phonological intuition may change as a consequence of learning the alphabet(for many persons it is, for example, very difficult to imagine the pronunciation of a word without bringing its conventional spelling into conscious attention), and the actual pronunciation also changes at some points (cf. so-called spelling pronunciations). At the grammatical level, the acquisition of written standard language sometimes provides a person with a much more "elaborated code" to be used also in speech (cf. XI.2). Yet, such phenomena are seldom discussed in terms of competence for spoken language vs. competence for written language. Instead, we have been accustomed to talking about one monolithic competence, that of the ideal speaker-listener (note: not the ideal writer-reader). 
By way of conclusion, the study of linguistic performance is certainly not linguistically uninteresting and must be pursued if we want to understand the true nature of language. It is necessary to consider speech and writing as two different media, each with its own characteristics which are worth while investigating.  Thus, Goody (1977:76ff) argues that we cannot be satisfied with only one distinction langue vs parole, or competence vs. performance; he argues for at least three basic concepts, language (langue), speech (parole) and writing (ecriture). One may add to this that it might also be necessary to split up the underlying language system into two distinct but largely overlapping systems, "la langue de la parole" and "la langue de l'ecriture".
Saussure and Chomsky are generally regarded as the most influential theoreticians in 20th century linguistics. Saussure founded modern linguistics, and Chomsky revolutionized it.
No one can have avoided noting that Saussure's concepts of langue and parole has played a major role in our discussion so far. Similarly, some of Chomsky's concepts recur in our discussion. There will be several opportunities to return to their concepts later on, and we will see that the changes brought about by their "revolutions" in linguistics were not as sweeping as we often like to think.
5.3. Language as a set of products
"When words and sentences are written down, they can be readily looked upon as objects"
Speech consists of transient, dynamic behavior distributed and limited in time. The transience of the products of the activities of utterance production and comprehension make a process oriented approach seem natural. In other words, we should focus on the behavioral activities themselves. The interpretation of linguistic behavior is heavily dependent on an on-line interaction with background knowledge, non-verbal signals and various other features of the situational context. Written texts, on the other hand, consist of permanent object-like products which tend to be relatively autonomous, explicit, and subject to less variation than speech (cf. III). It is quite obvious that the linguists' conception of language structure and linguistic items squares rather well with the latter phenomena, whereas its adequacy for the analysis of spoken language and dynamic speech behavior is hardly beyond dispute.
The view that a language consists of a set of thing-like products is a recurrent theme in the linguistic literature. One of the most well-known definitions is the following:
One of the most remarkable things in this statement  is that a language is regarded as the set of all the linguistic products that can be constructed according to certain rules. An alternative, and intuitively more satisfactory, view would be that a language consists of all the units and rules which make up the system underlying the products. From the vantage point of the language user one could propose a more psychologically based definition; the language of a speaker/listener is his knowledge of the underlying language system (la langue), or better, his ability to produce and comprehend an infinite set of utterances, discourses and texts, which fit the underlying system of rules. This, by the way, would tie up with Humboldt's proposal that language be regarded as an activity (and an ability to act linguistically) (energeia) rather than as a set of products (ergon).
We will soon come back to some popular conceptions of the underlying language system and the speaker's linguistic competence. In the meantime I would like to return to the product type of definition. What is the source of the view underlying this? In Chomsky's case the inspiration came, at least partly, from mathematics, where a language is precisely a set of strings of symbols which are subject to a number of specific rules. (Note, by the way, that mathematical symbol languages are a highly specialized kind of written languages). But there are deeper and more traditional roots. Derwing (1979:165) wants to derive some of the characteristics of American linguistics, the emphasis on products (rather than processes and underlying capacities) and the view of language as something autonomous (cf. V.6), from Bloomfield's methodological recommendations. Bloomfield argued that linguists must concentrate on the structure of overt behavior, since we cannot speculate on the underlying processes out of lack of the necessary knowledge of physiology and psychology. However, while this may well have had an impact on American linguistics (including Chomskyan generative linguistics) it is surely not the whole truth.  The fact is that linguistics has always been biased towards analyzing products, viz. written sentences and texts. This deep-rooted tradition of course influenced Bloomfield and his contemporaries, although they were able to legitimize the product view on other grounds too (as Derwing points out).
5.4. Linguistic structures as closed hierarchical systems of things
'Formalism and systematicity are the typical distinguishing marks of any kind of thinking focused on a readymade and, so to speak, arrested object' (Volosinov 1973:78)
My next point is part of the very core of my argumentation. It concerns something which was outlined already in the Introduction (¤ 1); linguistic structures are typically viewed as hierarchically organized systems of objects.
Lakoff and Johnson (1980:204) summarize "the premise of objectivist linguistics from its origins in antiquity to the present" as follows:
The "objectivist" view does not apply only to the linguistic products as such, i.e. what may be seen as the "linguistic data" (words, sentences, texts etc.); it is usually also expanded to cover "the underlying structures" underneath and beyond the "surface data", as well as to the linguistic code itself, i.e. the more or less permanent language system(s) consisting of items and rules and normally conceived of as "the grammar of the language". These last-mentioned constructs, i.e. la langue itself, will be discussed later on (see also V.5). Presently I will advert to the analysis of linguistic data (or the linguistic products) and their underlying structures.
The model which is most often applied to linguistic products, at least by (American) structuralists, is that of hierarchical constituent structure. Thus, the typical approach amounts to dividing linguistic products into successively smaller segments; the whole discourse (or text) is broken down into sentences, these in their turn into constituent sentences (main and subordinate sentences), clauses, and phrases, and phrases are thought to consist of words, words of morphs, morphs of syllables and/or phonological segments (vowels and consonants), and the latter are finally dissected into the "ultimate constituents", i.e. phonological features. A coherent text is thus pictured as a timeless web of part-wholes (proper constituencies) and other structural relations (dependencies of various types). The method of segmentation and classification has of course been applied mainly to intrasentential relations, but it has occasionally been applied to texts and discourses as well. In the analysis of face-to-face interaction, these methods have been termed "structural studies" (Duncan 1979) and considered to be typically "linguistic". Duncan describes this approach to face-to-face interaction in the following way:
It is worth noting that an outside reviewer like Duncan stresses the timeless nature of linguistic structures as particularly salient, and I shall revert to this point below (V.6).
If linguists analyze the observable patterns of linguistic data in terms of hierarchical constituent structures, this holds just as much for the underlying structures. Deep structures, semantic representations and morphophonemic forms are portrayed as hierarchical structures of thing-like (static, discrete, segments of various sizes. Furthermore, in Chomskyan mentalism these things are assumed to be "psychologically real", and some of Chomsky's adherents, most clearly perhaps Katz (1964), have accordingly postulated a machinery of mental things that causally impinge on each other in the course of the "speech communication chain" (¤ IX:l). Katz argues that the linguist "invents a theory about the structure of this mechanism, /i.e., the mechanism underlying linguistic communication/PL)  and the causal chain connecting the mechanism to observable events, to explain how these internal causes produce linguistic communication as their effect" (op.cit.:129). "The events to which the mentalist's constructions refer can stand as links in the causal chain that contains vocalizations and sound waves as other links" (ibid: 129-130). Thus, Katz clearly views speakers as more or less mechanical input-output systems; his "mentalism" is a paramechanism with a certain affinity to multi-stage S-R behaviorism (cf. Linell 1979b).
However, it must be admitted that Katz' paramechanism represents a very radical view which has not been espoused by many linguists, not even among generativists. It is therefore a rather special conceptualization of linguistic objects. Returning instead to more common ground in linguistics, we may say that the view that linguistic structure are hierarchically organized object-like entities exists in its purest forms in structuralism. Saussure regarded language (la langue) as a closed integrated system of units (i.e. abstract objects or "things") with stable internal relations ("un systeme ou tout se tient"). The portrayal of language as an entirely rigid system, a structure within which items are arranged in fixed ways, was brought to an extreme in post-Bloomfieldian American structuralism. Later, generative linguists developed this radical structuralism along partly rather different lines. On the one hand, they preferred descriptions in terms of processes instead of merely items and arrangements, thereby doing justice to the recursiveness of the linguistic rule system. On the other hand, they expanded the universe of putatively existing linguistic objects; in addition to "surface units", i.e. actual word forms, sentences etc., they introduced "deep-structural" units into both syntax and phonology. Thus, abstract relationships such as morpheme identities between actual forms were thought to presuppose the existence of abstract morpheme-invariant phonological forms; i.e. in addition to word forms like English sane /sein/ and sanity /saeniti/ they (in this case: Chomsky and Halle 1968) posited a morpheme-invariant abstract form /saen/.  In syntax, surface structures and deep structures were thought of in analogous fashions. Thus, despite several obvious and important differences, it would be justified to conceive of Chomskyan linguistics as firmly anchored within the American structuralist tradition. (It is instructive to consult Hockett's (1968) characterizations of the "state of the art" in generative linguistics, e.g. pp. 31, 37, et passim).
The theory that a language should be seen as nothing but a closed integrated system is a most unfortunate one, since it denies the significance of some of the most fundamental properties of language. Thus, adaptivity to new situations lies at the heart of language. Linguistic variation is typical of linguistic communities, and this in turn is connected with the fact that all natural languages change over time. It is well known that extreme structuralism has great difficulties in accommodating such facts. According to these theories, "each historical change would have to be conceived of as a willful distortion of the inherited pattern, which is absurd" (Andersen 1969:828). Instead we must admit that languages, as well as speakers' competences in them, are open systems into which new elements can be introduced without becoming fully integrated. Therefore, a certain instability and certain conflicts may characterize them.
How is it that linguists look upon linguistic structures as hierarchically ordered, as it were spatially (two-dimensionally) arranged, static objects? It seems to me that the only really revealing explanation is to be found in the time-honored written language bias. Thus, I very much doubt that linguistic structuralism would have evolved in the way it has, if linguists had studied the varying, quasi-continuous speech behavior instead of the spatially arranged objects of written texts, and if they had not had access to the possibilities of organization and systematization inherent in writing (including listings, diagrams etc.) as a meta-language (cf. III:4). At any rate, the analysis of dynamic behavior might necessitate other conceptions of linguistic structure. Instead of thinking exclusively in terms of hierarchies of objects, other models and metaphors might be necessary. Thus, if we look upon speech as a stream of vocalizations, as a stream of continuously progressing dynamic behavior, we might conceive of phonological and grammatical structure as sets of conditions superimposed on this carrier wave of vocalizations. Similarly, the function of semantic structure in thinking, perception, etc., may be understood in terms of conditions on streams of various kinds of cognitive activities.
One may also raise the question of whether the listener's analysis of behavior performed in actual language use is carried out entirely in digital terms. After all, that is the view that follows from most models in linguistics, cognitive psychology and artificial intelligence. The assumption of analogous processing seems fairly probable in some cases. Prosody, in particular the paralinguistic expressions of emotions (Fonagy 1977), is a case in point, but also many linguistically determined, phonetic elements may build upon dynamic wholes rather than sequentially ordered aggregates of static features. One may also generalize the argument and raise the question of whether perception in general involves global gestalts which cannot be reduced to independent elements and structures (cf. Dreyfus 1972:151 ff.).
The common objection to the arguments that I just hinted at is of course the well-established contention that we always categorize and digitalize the objective reality in perception and cognition. Processing information via categorization has been said to be one of the most important organizing principles of human cognitive systems, and as such it would provide a natural basis also for structuralism in linguistics. While the force of this counter-argument must be admitted, it is still possible to argue on good grounds that structuralism is exaggerated in viewing language as a closed, integrated and surveyable structure in which each thing-like item has its carefully defined place. In order to substantiate this point a little more, I shall now return to the phenomena of linguistic variation and linguistic change for a moment.
5.5 The invariance of linguistic structure
Objects must necessarily have a considerable degree of permanence. Along with the view of linguistic items as thing-like phenomena goes therefore the view that their properties are fixed, stable, and invariant. It is assumed that under the bewildering variation in actual use there are some context-free and stable properties which are invariant over time and space. Hence it must be an important task for linguists to find out what these properties actually are (or should be).
This drive towards (searching for) invariance may appear in rather different disguises. If we look upon language as a set of norms for linguistic behavior, we may ask: What are the rules of proper behavior? What does the word X actually mean? What is the correct pronunciation of the word X (note: pronunciation in the singular)? In earlier times, the linguistic clergy (cf. III.2) often thought of their task as that of preserving the original, i.e. God-given, correct form and meaning of words. Today, the correct pronunciation and meaning is something that linguistic specialists determine by means of adequate analysis and properly constrained use of their own intuition. The normative perspective is therefore somewhat less salient; rather, linguistic items are treated as independently existing objects with objective (sic!) properties. However, the basic philosophy is still there; it is usually taken for granted that there are unique and stable phonological, grammatical and syntactic properties inherent in words and sentences, and, hence, that the search for these properties is a meaningful enterprise. It is assumed that under the rich dialectal, phonostylistic and idiolectal phonetic variation in the pronunciation of a given word, there is one unique phonological form. It is-taken for granted that a given sentence has one well-defined syntactic structure, e.g. a specific constituency or dependency tree (or, in the case of syntactic ambiguity, a finite number of such structures). We like to think that the meaning of lexical items can be adequately described in terms of a finite set of semantic features or a hierarchical organization of such features, rather than, say, a more indeterminate and vague meaning potential (VII.2). Similarly, and perhaps even more questionably, sentences are assumed to have specific semantic representations defining "literal interpretations" of those sentences (VII.l, 3, 5). In mainstream contemporary linguistics, each sentence is assumed to be mappable onto a finite number of distinct meanings. That is, a sentence either does or does not have a particular meaning, a sentence either is or is not ambiguous, two sentences are or are not paraphrases of each other, etc. Yet, the study of the actual use of these sentences calls such an analysis into question; it appears that semantic interpretations are often more-or-less plausible. Rather than being unequivocally correct or absolutely excluded, many interpretations are more-or-less justified (cf. Sadock 1979).
The assumption of underlying invariance is clearly made also in modern psycholinguistics and cognitive psychology; language is regarded as a reified cognitive structure with fixed components. The kernels of linguistic items are constituted by the distinctive invariant properties, which are supposed to be clearly different from the redundant features. Thus, underlying phonological forms comprise a small and fixed set of phonological features, and word meanings are assumed to be simply set of essential features (necessary and sufficient conditions for some exemplar to fall under a certain concept, cf. VII.2). It is obvious that this approach obfuscates the facts that a given language is based on social and cultural norms and conventions, and that these norms are applicable in different degrees to the varying social contexts. Instead, linguistic rules are construed as mental processes (cf. VIII.3).
In spite of this, the currently popular conceptions of linguistic structure involve normativity, although in a much more subtle and less obvious way than the old view that linguistic rules were instigated by God. What happens today is that the linguists, rather than God, dub certain stable features the true, distinctive features and rules of semantics, grammar and phonology, thereby normatively disqualifying a great amount of variation as "linguistically insignificant" ("exclusively pertaining to performance", "linguistically irrelevant" etc.).
Normativity, stability and invariance are features which are much more characteristic of the use of the written language than spoken communication. The linguists' attitudes to their objects of study can thus be at least partially explained by reference to the written language bias. In addition, it may be argued that this bias has been endorsed in recent years by the computer revolution in linguistics, especially as regards psycholinguistics and artificial intelligence. When we try to model linguistic structure and the use of language in reasoning on a computer, we have to use specific implementations and representations; computer programs must not be vague, and therefore, we choose a particular representation (or a finite number of such specific representations). It may then be tempting to assume that human beings are similarly determinate on all specific points. Such an assumption must be considered highly implausible (cf. VII.9). Moreover, disregarding this point for the moment, it is a significant fact that computers basically represent an expansion of the written, rather than the spoken mode of communication.
5.6 The atemporality of linguistic structure
As we have seen, linguistic items and structures are usually regarded as being static and invariant. Along with these properties goes another property, viz. that of atemporality. Linguistic structures, such as semantic, syntactic and phonological representations, lack a time dimension; they are atemporal in nature (cf. V.4).
Now, linguistic structures are indeed abstract, and like other abstract phenomena, specifically various kinds of norms and rules, they are atemporal in an important sense. It would therefore be a category mistake to think of them as, e.g. behavioral processes. Instead, they may be construed as abstract conditions that are superimposed on the stream of behavior (or the stream of consciousness and/or other mental processes).However, in contemporary linguistics the abstractness and atemporality of linguistic structure are conceived of in quite another manner, viz. as mental objects which exist simultaneously in their entirety; exactly as real physical objects, including written texts, they are (in principle) invariant over time. Thus, despite the fact that speech consists in a dynamic behavior distributed over time, linguists (and psychologists) look upon phonological, syntactic and semantic "representations" of individual utterances just as simultaneously existing mental objects.
Most psycholinguists are careful to distinguish between the linguistic structures as such (which are atemporal and simultaneously existing) and the processing involved in the linguistic activities of utterance production and perception. The processing is of course distributed over time. What is important, however, is that (psycho-)linguists tend to conceive of this processing as consisting in the construction, analysis, and manipulation of (simultaneously existing) mental objects, i.e. linguistic forms and structures. This is particularly true of the kind of mentalism that inheres in orthodox generative psycholinguistics. The time-distributed processes involved in utterance production and perception are assumed to involve the accumulation and integration of structures, e.g. syntactic surface structures, in a short-term memory, where they are supposed to be simultaneously accessible for inspection and analysis in their entirety. Many of the common testings in psycholinguistics are characterized by these assumptions: 
These remarks of Marslen-Wilson's seem quite to the point. Notice that this does not amount to a denial that assembling of information in short-time memory does occur in speech processing. The point is simply that the prevailing theory (which, I think, is ultimately dependent on the written language bias) may prevent the psycholinguist from seriously trying out alternative models like those involving interactive on-line processing. See also ¤ IX.l.
5.7 The autonomy of language and linguistic structure
I have already commented upon the common view that language is a stable, autonomous system which can and should be seen as immanent, sufficient in itself, and independent of other factors, e.g. the contexts where language is used. We also noted the corresponding methodological stand according to which linguistics should be pursued as an abstract discipline analyzing the relations between different linguistic items and products without regard to contextual modifications and the interplay with other conditions of use (so-called autonomous linguistics). I have also pointed out that these conceptions are more naturally at hand for a linguistics based on written discourse than for a study of communication by means of speech. I would now only add a few remarks concerning the consequences of this type of linguistics.
Autonomous linguistics implies, first of all, that language structure can be studied without regard to the use of language. According to Saussure and Chomsky, the study of la langue and the ideal speaker's competence are logically prior to any study of parole and performance. This has led to the Chomskyan division of labor between linguists and psychologists: first linguists determine what is assumed to be the speaker's linguistic competence (and this is done by autonomous linguistic-structural methods), and only then are psychologists (and psycholinguists) encouraged to investigate how this 'competence' is put to use in performance. It is nowadays widely recognized that this procedure is impracticable, and that structure and use, competence and performance must be studied in parallel and not in isolation from each other.
A related consequence of the structuralistic ideal of immanence is the hope that language use and linguistic competence can be isolated from other mental activities and capacities. But if we approach the complexity of speech communication with open eyes, prepared to observe the intricate interplay between verbal utterances and the surrounding context of background knowledge, non-verbal signals, and reliance on physical and social-psychological features of the situation, we realize that this is a vain hope. On the other hand, it may seem less absurd as long as we as linguists concentrate on written texts, which are more independent of specific senders, receivers and contexts.
For the study of linguistic communication, the linguists' isolationism has had many important consequences, some of which have - in my view - been rather unfortunate. For example it has enhanced a view according to which the structures of language and speech are radically different from all forms of non-verbal, e.g. paralinguistic, communication, and this has precluded the possibilities of searching for common sources and explanations valid for both verbal and non-verbal communication. Compare, e.g. Chomsky's (1968:70ff.) denial that there are any symbolic systems comparable to language within the domain of non-verbal communication. In particular one should deplore the consequences that this view has had for the theories of the child's language acquisition. Jakobson (1968) argued that there were no interesting relations between babbling and phonological development; we know from developmental psychology that there are successive developments from non-verbal gestures to verbal ones (Werner and Kaplan 1963). Chomsky's belief in the absolute uniqueness of natural language led him to postulate a highly specific, innate faculty of language, when in fact many features of language can be at least partially explained by relating them to various forms of non-verbal communication.
5.8. Norms and normativity in linguistics
When linguists were involved more or less directly and exclusively with teaching people a correct language or with developing new written standards of so far insufficiently standardized national languages etc., their discipline was of course rather explicitly normative; it simply prescribed rules to be followed in the use of language. We can also put the point in quite general terms; a normative focus is natural for a linguistics dealing with written language. Learning to write makes a great deal of education necessary, and this consists in no small measure in the learning of conscious norms. On the other hand, a person's spoken vernacular is acquired without any explicit instruction, and it may be argued that it is in fact constrained by fewer grammatical rules than written language. Yet, as we will see, the norms of written language enter the analysis of spoken discourse too.
In today's linguistics analyses are carried out more for purely theoretical reasons. It is generally held that the aims of linguists are nowadays plainly descriptive (and, possibly, explanatory); rather than prescribe how people should act linguists (and other social scientists) are assumed to describe how they do in fact act. Nevertheless, our linguistic practice is still replete with normative aspects, although this is rather seldom admitted.  One should therefore talk about the hidden normativity of descriptive linguistics.
The hidden normativity shows up at different levels. For example, anyone who really pays attention to the full variability that actually characterizes spoken language, may well be amazed by the fact that many of the actually occurring structures have not been described in the usual grammars, not even in modern grammars compiled by "descriptive" linguists. The most important reason for this is probably the fact that the structures in question are not accepted in written language. Hence the most natural reaction is to regard such structures as ungrammatical or deficient. (The history of language cultivation is full of examples.) But once some such structure is attended to and described by some linguist as part of spoken language, it is thereby assigned another status; it becomes recognized as part of the language in question and thereby, as it were, legitimized. This phenomenon - quite frequent in modern linguistics - clearly shows the normative function of the descriptivist practice.
But normativity cuts both ways in current descriptive linguistics. Perhaps the normative aspects are most perspicuous in other ways, e.g. in our assessments of various constructions of spoken language, as, for example, when so-called anacolutha are discarded as ungrammatical, i.e., as not belonging to the language involved (language defined according to Chomsky, cf. V.3) although they are perfectly natural and function well in their own medium. The norms which are explicitly or implicitly applied by linguists in such enterprises are quite often such norms as have been consciously adopted by normative grammarians of the past in their regimentation of written standards (VI.3). There is always a temptation to regard as ungrammatical such stretches of spoken languages which cannot be subsumed under the generalizations which follow from these rules (or norms).
Furthermore, linguistic analysis is of course subject to norms that belong to the metalanguage used. Another way to put it would be to say that analyses are dependent on the very special language games which are performed by the linguist as he applies his methods to the data. The metatheory to which a given linguist adheres allows only of certain kinds of "linguistic representations", e.g. certain types of syntactic constituency trees or dependency structures, or certain kinds of underlying phonological representations (e.g. linear configuration of segments, cf. VIII.l). Thus, such metatheoretical norms will largely determine the linguist's identification of invariant properties behind the variation in linguistic data.
By way of summary, linguistics is still normative in nature. There is nothing surprising or indecent in that; all sciences are to some extent dependent on norms, and in the social sciences the situation is particularly complex, with norms permeating both data and (meta)theory.
1. The term linguistics borrowed from English (and/or from French linguistique) is nowadays used in most other languages too (e.g. German Linguistik, Russian lingvistika, Swedish lingvistik), but many of these languages also use native terms like Sprachwissenschaft, jazykoznanije, sprakvetenskap, Dutch Taalwetenschap, Italian glottologia, etc. to denote the same traditional humanistic study of language. Sometimes a distinction is made between linguistics proper and philology. Both, however, belong to the linguistic tradition with which I will be concerned here, although I will mainly deal with linguistics proper.
2. Here the term linguistics (or its equivalents, cf. fn. 1) would cover the activities both in general linguistics and in the linguistics of specific languages, e.g. English linguistics, Romance linguistics, Uto-Aztec linguistics, regardless of how these studies are organized in academic institutions in various countries.
3. I have ignored here other attempts to justify an autonomous linguistics. For example, many linguistics since Saussure, not in the least Hjelmslev and Chomsky, have wanted to consolidate linguistics as an 'immanent' discipline independent of psychology, physiology etc. It is true that Chomsky (e.g. 1968:84) argues that linguistics should be regarded as a subdiscipline of cognitive psychology, but his actual theories and methodological recommendations go very much against his. In fact, the intense and rather devastating critique that has appeared in the 1970's, demonstrates the failure of Chomskyan autonomous linguistics as regards its claim for psychological validity (e.g. Derwing 1973, Linell 1979a, Botha 1979).
4. Cf. Lyons (1981:36) 'macrolinguistics' (see above).
5. In modern linguistics, this has been seriously discussed only by very few scholars, e.g. Vachek (1939, 1949).
6. Some modern psycholinguists work with several notions of competence. For example, Fodor, Bever & Garrett (1974) make use of at least three constructs, i.e., the grammar (Chomskyan competence), the idealized sentence recognizer and producer, and the real (sic!) sentence recognizer and producer. However, the differences between speech and writing have nothing to do with this trichotomy.
7. In very recent years, a few books have appeared which bear witness to the fact that this is a view which is currently gaining ground among students of language. See e.g. Baron (1981), Tannen (1982a,b).
8. I will discuss other aspects of Chomsky's conception of a language later on, cf. esp. VI.l, VI.3, VII.8.
9. There are more similarities between Chomsky's and Katz' psycho-linguistics and variants of multi-stage behaviorism than is usually thought. See Linell (1979bJ for some points.
10. Katz (1964) hypothesized that this "mechanism" included a generative transformational grammar, apparently in a rather straightforward fashion.
11. Barring the general problems in treating linguistic elements as objects, this seems to amount to a category mistake (cf. Linell 1979a:ch 13).
12. Cf. also Goldman-Eisler (1968:2) who argues that a tendency to spatialize and detemporalize events appeared when psychology was developed in the direction of experimentation and quantification. In accordance with the main lines of argument of this book I would say that the roots of this "detemporalization" goes much further back in the history of psychology.
13. This is particularly true of conventional introductions to modern linguistics. These very often contain a short introductory chapter on the history of linguistics, in which it is simply asserted that linguistics used to be prescriptive whereas modern linguistics is descriptive. Yet, such textbooks are heavily prescriptive in that they tell beginners how they should understand language and how linguistics should be done (according to the particular author's theoretical preferences).