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





6.1 The concept of sentence

I have already pointed out that grammar is a sine qua non for classical as well as modern linguistics. I will now discuss some salient features of modern grammar theory. A convenient point would then be the Chomskyan definition of a language as a definite (infinite but enumerable) set of grammatical sentences (cf. V.3). There are at least three debatable features of this definition. The first is the view that a language is considered to be a set of products (written or spoken sentences) rather than underlying activities, conditions on such activities, resources utilized in such activities, or the like. This has been discussed in V.3. The second point, the issue of grammaticality, will be treated in VI.3. Presently I will turn to the third aspect, the status of sentences. [1]

Historically, there are two categories of linguistic units that have been considered much more important than others, i.e. words and sentences. Why is it that Chomsky and many others, indeed the majority of modern linguists, take the sentence to be the self-evident basic unit? This question can hardly be answered unless we attempt an analysis of the concept itself: How should 'sentences be defined?

It is quite obvious that rather different conceptions of what a sentence is have dominated different époques in the history of linguistics. Nowadays, the sentence is usually taken to be a form-based concept, but traditional characterizations are often more vague and content-based. I will first consider some of these conceptions.

A time-honored definition amounts to regarding the sentence as 'the expression of a complete thought' or the like. Is this a workable definition? For example, does it make sense to say that a discourse comprises a sequence of (expressions of) complete thoughts? Is a language really an infinite but enumerable set of such expressions? The answers must be no, and the reasons are quite well-known from the linguistic literature. I will therefore be content with noting two difficulties. One has to do with the fact that it is hard or even impossible to tell what is meant by saying that a thought is "complete". Instead a more reasonable thing would be to say that any given 'thought', whether overtly expressed or not, can always be further specified, made deeper or more precise. No linguistic expression is ever semantically autonomous or complete in itself; in order to be interpreted and understood it always needs a certain amount of pragmatic specification (more about this in VII.l). Accordingly, we can hardly rely on notions like '(expressions of) complete thoughts'.

Another objection to the classical definition of the sentence is that it seems counter-intuitive to argue that linguistic expressions, or utterances, always express 'thoughts' (unless one is willing to subsume under this term everything indicative of some mental act on the part of the speaker):


'As applied to many samples of speech, the description is... grotesque. Consider a mother anxiously asking for news of her son, or a tradesman driving a hard bargain. Or again, imagine an angry traveler hurling words of abuse at an uncomprehending porter, or a judge pronouncing sentence of death upon a murderer? Shall we say that these persons are expressing thoughts?' (Gardiner 1951:17)


It seems inaccurate to argue that utterances such as 'Fire!', 'How do you do, Mr. Swanson' and 'I christen Thee Andrea Doria' are expressions of thoughts, whether complete or incomplete. Moreover, the theory of sentences as expressions of thoughts is embedded in a general theory of utterance, which says, in essence, that the speaker first forms a thought, and then verbalizes it as an overt utterance. This theory is unrealistic for several reasons (IX.l).

Accordingly, Gardiner and many others opt for a more general and still very vague definition of sentence. In fact Gardiner uses the term more or less synonymously with 'utterance'. He proposes (op.cit.:98, 182) that a sentence is an utterance which is "completed" "to the satisfaction of the listener", i.e., the listener feels that he has been told something when he has heard and understood a 'sentence'. Again, we are faced with an unduly vague notion of sentence. Therefore, one is tempted to opt for a more formed-based concept of sentence. This is all the more motivated, because we will need several concepts, including a distinction between 'utterance' and (something like) 'sentence' [2] (more about this need below).

The concept of 'utterance' is not easy to define, and I cannot give but a very loose characterization here. Thus, I shall mean by 'utterance' a stretch of speech (by one speaker) which is formally relatively independent; it is grammatically and prosodically marked off from the context and does not contain any major internal prosodic or grammatical breaks. From a grammatical point-of-view an utterance may have various structures (phrase, clause, sentence etc). Some scholars prefer other terms instead of 'utterance', e.g. 'locution' (Kendon l980), 'macro-syntagm' (Loman & Jorgensen 1971), 'macro-segment' (Hockett 1958) etc. [3]

However, we are still not finished with the content-based definitions of 'sentence'. There are still more variants, e.g. those which are based on notions like 'proposition' or 'subject-predicate structure'. It seems reasonable to argue that every message, however simple, must logically contain at least a logical (or psychological) subject and a logical (psychological) predicate, i.e. a 'theme' and a 'rheme', a function with at least one argument, a proposition, [4] or else it would simply not be a message.

The primacy of propositions is endorsed by many linguists, logicians, and psychologists. It is part of Wundt's theory of "the psychology of the sentence", and one of the great pioneers of aphasiology, Hughlings Jackson, declared that "the unit of speech is the proposition - to speak is to propositionize" (1878). Perhaps, this may be interpreted as evidence for the sentence as a unit of discourse? Again, the answer has to be in the negative, and this briefly for the following two reasons. First, if the proposition is the semantic unit, then the corresponding expression would be the clause (or simplex (main) sentence), i.e. a simple association of a logical subject and a logical predicate, rather than the sentence (as the term is normally understood, cf. below). Furthermore, it must be emphasized that the proposition is a semantic or communicative unit rather than a unit of expression. Its parts need not be overtly expressed. In the following fragment of a dialogue, B's utterance may be said to embody or presuppose a proposition (roughly 'I live in London'):


	1.	A: Where do you live? 
		B: In London 


In other words, the expression In London must be further specified (using the overall context) in order for the listener to "reconstruct" a proposition. However, such pragmatic specification is always necessary; it applies to (1B) and (2B) alike, although we would say that only the latter is a sentence (and then we presuppose, of course, a form-based notion of sentence):


	2.	A: Do you live in London? 
		B: (Yes,) I do.


It is implicit in the preceding discussion that 'sentence' should be defined by reference to formal-structural properties, and that such a concept, distinct from that of 'utterance', is in fact needed. Briefly, sentences are a category of linguistic expressions with the following characteristics:


A) a sentence must be limited by an onset and an end that are structurally clearly manifested, [5]
B) a sentence must, at least in some languages, e.g. English, contain both a subject and a predicate [6]
C) a sentence must exhibit certain other properties which enable it to occur as an independent unit. Compare the difference between a matrix sentence (main sentence, Hauptsatz) which is a true sentence, and a subordinate clause (or subordinate sentence, Nebensatz) which is not a true sentence (but a clause).
D) a single sentence may contain several clauses, e.g. subordinate clauses.


Since utterances are basically stretches of speech, whereas sentences (in the normal modern sense adopted here) are abstract grammatical units (configurations of grammatical conditions), the two units are actually incommensurable:

 'As a grammatical unit, the sentence is an abstract entity in terms of which the linguist accounts for the distributional relations holding within utterances. In this sense of the term, utterances never consist of sentences, but of one or more segments of speech (or written text) which can be put into correspondence with the sentence generated by the grammar.' (Lyons 1968:176)

 Therefore, with this type of definition, sentences are strictly speaking not performance units. However, we could try to analyze language use at the grammatical level and then find out whether a given text or discourse can be characterized in terms of sequences of spoken or written words (text sentences, "sentence tokens") fulfilling the conditions of sentencehood at the abstract grammatical level. Actually, one should keep 'sentences' qua abstract grammatical entities and 'text sentences' terminologically apart, but for convenience we often refer to text sentences (sentence tokens) simply as 'sentences'. Thus, a text or discourse can be analyzed in (text) sentence if there are clearly identifiable boundaries between the units in question, and if these units display a clear internal structure, including clear differences between main clauses and subordinate clauses. It turns out that these conditions are often simply not fulfilled in normal spoken discourse. In spoken communication we often find a sequence of clauses and phrases which are rather loosely strung together without there being any clear boundaries or differences between clauses of different kinds. In other words, the clause is undoubtedly one of the units of grammatical analysis applicable to casual speech, but the sentence is only characteristic of certain discourses, especially those which are heavily influenced by the speaker's knowledge of elaborated written language.

This means that the sentence is a unit of analysis which is best suited for written discourse. An essential part of the laborious task of learning to write properly consists precisely in the requirement to become capable of organizing one's linguistic output in complete and well organized sentences. In casual speech such structures are not natural. Yet, the linguists analyses of spoken discourse are typically constrained by the search for such structures even where they cannot be found. This approach has several unfortunate consequences. For one thing, many structures are condemned as being ungrammatical, faulty, illogical etc, simply (it seems) because they do not meet the requirements for sentencehood. Secondly, many non-sentences are treated as sentence fragments, and the grammar is typically made to derive them from underlying sentences. After all, the only initial symbol of an orthodox generative grammar is the S. Thus, for example, 'In London' of (1B) is obligatorily derived from a deep structure corresponding to 'I live in London'. This type of analysis is in many ways unfruitful:

'Conversation is a text type replete with surface structures which are not in fact sentences (McCalla:14). Consequently, the study of conversation in the framework of sentence grammars both (a) misses the real issues (actions and goals) and (b) creates difficult if not unresolvable non-issues (how to convert everything into well-formed sentences)'. (Beaugrande 1979:1)

Therefore, if we attempt something like a generative grammar of spoken discourse, [7] it should (base-)generate many different kinds of phrases and clauses that are not sentences. Sentences in the technical sense are only seldom encountered in many natural forms of talk. It seems that the ideal of using only or mainly sentences is something which has gradually evolved with the development of written standard languages. It may be interesting to observe that sentences (in our sense) were not as obligatory in the written language of the past. For example, Plato s dialogues contain a lot of anacolutha, which would now be rejected as ungrammatical, and the same applies to Shakespeare s plays. Early Middle High German texts were replete with structures, which were later relegated from written Standard German. However, the same types of constructions have existed all the time in spoken German (Sandig 1973).

There have been strong elements of regularization in the regimentation of language which has been performed by language cultivators (i.e. prescriptive linguists) when they have tried to develop written standards on the basis of spoken vernaculars (and various classical languages). In general, this has involved a predilection for complete, symmetrical, non-elliptic sentences (or clauses) with a clear grammatical structure (one subject, one predicate verb, a full verb complementation, a certain basic word order etc) and clear differences between matrix sentences and subordinate sentences (clauses). For example, in the grammar tradition for German and Swedish one can discern an endeavor to define various clear, grammatical differences between matrix and subordinate clauses, although these differences have never been very clear-cut in natural spoken discourse. [8] Today, similar trends towards regularization of data appear in the linguists' theoretical analyses, e.g. when non-sentences are derived from underlying sentences (c.f. above).

The preoccupation with sentences in contemporary linguistics has made it practitioners unable to distinguish properly between sentences, which are, or should be, defined in terms of certain formal traits, and utterances, which may take on various kinds of overt forms. It is commonplace in contemporary psycholinguistics, particularly in the U.S., that the use of the language, speech performance, is talked about in terms of the production and comprehension of sentences, not of utterances (e.g. Fodor et al. 1974, Rosenberg 1977, Foss & Hakes 1978). One may argue that this is a rather innocent example of sloppy terminology, but it seems to me that it invites the idea that sentences (in the strict, form-based sense) are indeed produced and understood in spoken discourse. At least it is implied that the underlying structures are sentences, since the occurring non-sentences are treated as sentence fragments and grammatically analyzed as derived from underlying sentences.

If some linguists still believe that spoken discourse consists of sentences, this is of course even more true of laymen. After all, one learns at school that only complete sentences count as correct language.

Even if it is a wide-spread view that spoken vernaculars are wrong (cf. XI.2), it is still generally believed that it would be relatively easy to identify the spoken counterparts of such things as the sentences and punctuation marks of the written code. Hence, when this is not borne out, the tendency is to assume that the language being analyzed is deviant from normal language. For example, such a view seems to be inherent in Maher's (1972) analysis of the language of schizophrenia.


6.2 The neglect of prosody

There is hardly any doubt that prosodic phenomena like stress, tone, intonation and durational phenomena are very important cues in speech, for example as regards the foregrounding vs. backgrounding of various elements of meaning. In addition, emotions and attitudes are expressed by the use of related, so called paralinguistic phenomena, i.e., variations in melody, tempo, pausing, loudness, register, articulatory accuracy etc. However, most of this has no counterpart at all in written texts. Note, incidentally, that the communicative functions of prosody (in the wide sense) are emotive, evocative and social rather than cognitive and referential, and only the latter, more "intellectual" functions have traditionally been regarded as typical of human language. Again, this is in full agreement with the written language bias (III:l).

Accordingly, one is tempted to say, linguists have been very little interested in prosody, and our current theories of prosody are clearly insufficient. It is true that prosody seems to be extraordinarily hard to come by, and the analysis could hardly be performed without modern instrumental equipment like tape recorders (sound and video taping) and computers, but there is hardly any doubt that the written language perspective is also partly responsible for the neglect of prosody.

The effects of the neglect of prosody can be seen in several subfields of linguistics. It is evident in grammar theory, where our data are almost always only written sentences (considered out of context), although this restriction is seldom pointed out. As an example, consider the numerous cases of ambiguous sentences, the analysis of which has played such a crucial role in the argumentation for deep structures and transformations in syntax. In many of these cases, the different readings (note the word 'reading'') have, or at least can have different prosodic patterns, and therefore we presumably have distinct surface structures in speech. Here are some well-known examples:

	(3)	A: Flying planes can be dangerous 
		B: He has Plans to leave 
		C: What disturbed John was being disregarded by everyone. 
		D: The police were ordered to stop drinking after midnight.

 The orthodox generative literature maintains that these sentences have several readings, and these readings with their distinct meanings are disambiguated in speech. Note the written language perspective underlying this reasoning; the linguistic objects, i.e. the sentences are treated as being ambiguous as such, but distinct readings may be signaled in speech. But if speech is primary (c.f. V:2), it would seem natural to talk about ambiguation in writing, rather than disambiguation in speech.

What is perhaps even more remarkable is the fact that the neglect of prosody characterizes psycholinguistics just as much. In research on the perception and comprehension of linguistic messages, one normally uses written sentences (in isolation), as if the perception and comprehension processes involved in reading were only trivially different from those of listening to natural speech. The same is true of research on memory effects, in artificial intelligence etc; prosody and its possible effects are almost never taken into consideration.

In his discussion of the merits and drawbacks of some current models of perceptual strategies, Welin (1979:45) notes: [9]

'A fourth problem is the implicit orientation of these models towards the perception of written text, although they are claimed to be valid as models of the listener as well. The use of intonational patterns and pauses in natural speech is not discussed, and it is generally assumed that the input is already segmented into words'.

 Yet it is of course far from true that the isolation of words can be performed automatically and in the same way as graphic words are separated in writing (cf VI:4). Again we see that the only problems which are seriously considered by linguists tend to be those which are evident from the study of written communication.


6.3 The concept of grammaticality

The functions of linguistics in past times provide a natural explanation for the fact that grammar has always been the focus of linguistic theory (V.1) The practical goals of traditional linguistics also made a concentration on the norms of grammaticality seem motivated and indeed necessary; it was important to propose more or less explicit norms for how a standard language should be written.

In present-day theoretical linguistics, especially in the Chomskyan tradition, the rules for grammaticality have a somewhat different but in no way less important function. If we assume a language to be a definite set of linguistic products, i.e. a set that can be exactly characterized (generated) by a finite set of formal rules (V.3), then these rules (or norms) of grammaticality will be of the utmost theoretical importance. In fact, generative linguistics is in a way more normative than most of traditional linguistics. Whereas most traditionalists probably regarded the living language as a rather diversified array of linguistic products, in which the boundaries between grammaticality and ungrammaticality were at least partly quite fuzzy and not amenable to any exact characterization - after all, this was one of the reasons for establishing explicit norms for the written standard language -, generative linguistics assumes that it is indeed always and everywhere possible to determine the exact limits of grammaticality on empirical grounds, at least if one is prepared to allow for a certain amount of idealization.

There is today a growing awareness also among theoretical linguists that the traditional assumption of fuzzy boundaries was after all well motivated, and that the idealization involved in establishing the Chomskyan construct of competence with its exact rules is intolerable because it distorts the empirical facts beyond recognizability. Of course, no one would deny that there are a considerable number of rules of grammaticality underlying spoken language. These rules clearly distinguish between grammatical and ungrammatical strings in countless simple cases:

	(4) 	A: He saw her 	B: He did not see her 
	(5)	A: Her saw he	B: Her see he not did

 The sentences of (4) are undoubtedly correct English, whereas the strings of (5) can just as easily be dismissed as incorrect. This of course holds for spoken language too, and most native speakers of English, even young children, would no doubt react negatively and correct the use of them (by e.g. foreigners). Thus, we should be able to characterize a number of conditions on correct English which would distinguish clearly grammatical strings (such as (4))from clearly ungrammatical strings (like (5)). However, the Chomskyan idealization is much more far-reaching; it demands that it be possible to define rules of grammaticality for any types of strings, no matter how long and structurally complex they are. This follows from the Chomskyan definition of a language (V.3). Furthermore, since the clear cases are never decisive in the testings of various proposed systems of sophisticated generative rules, it is precisely the more complex strings which have come to play the crucial role in the grammarians' efforts to develop a complete generative grammar for a given language. There are innumerable examples in the generative literature of very complex strings, the grammaticality of which is very difficult to decide. Here are some items: [10]


(6) I watched the Indians who the man who had been my advisor
in my freshman year had advised me to study when I got to Utah talk,
because I was fascinated by the way their view of the world seemed
to be constrained by the structure of their language. (Ross 1967:39) 
(7) The contract which I want to peruse after copying by treating in
milk while pressing between two Pieces of marble in flattening out
is a beautiful piece of art. (ibid:106)
(8) Reports of which the government prescribes the height of the
lettering on the covers are invariably boring. (ibid.: 111)
(9) Which packages is it possible that Sam didn't pick up until it
had stopped raining which are to be mailed tomorrow? (ibid:159)
(10) Which packages is it possible that Sam didn't pick up until it
had stopped raining which he arranged with his agents in Calcutta to
send to him here in Poplar Bluff because of his fear that someone in
Saint Louis might recognize him? (ibid:159)
(11) Var det Soders~ukhuset, som Halsovardsmyndigheterna lat undersoka
vilka farskvaror- det var oklart vilken grossist som hade levererat_:till
'Was it the South Hospital that the Health Department investigated which fresh
produce it was unclear which firm that had delivered (Engdahl 1979)

 It may seem very difficult to comprehend and accept complex sentences like these. Their grammatical status is hard, if not impossible, to determine on anything like "objective" grounds. Even the judgements made by professional linguists seem to depend on rather accidental events. Sometimes, the linguist succeeds in finding a context in which the sentence under consideration makes sense (and, hence, could possibly be used), and then it seems tempting to dub it grammatical. In other cases, the linguist has not (yet) succeeded in finding a suitable context, in which case the sentence will normally be starred as being ungrammatical. [11]

For such reasons (cf. also below), controversies over the grammaticality or ungrammaticality of very complex sentences have often turned out to be fruitless. Indeed, they seem to be unsolvable without recourse to rather arbitrary principles. One such way out is the assumption of different idiolectal grammars; when linguists cannot agree on the grammaticality of some given sentences, they may simply assume that they have "internalized" different grammars (different norms), which accounts for their differing judgements (cf. what Botha, 1973:219-9, calls the ' "my dialect"-"your dialect" gambit'). Another rather dubious principle is part of Chomsky's early proposal of "letting the grammar decide" ('Clear Cases Principle', cf. Chomsky 1957:13-13, Fillmore 1972). These attempts have met with a lot of criticism. A reasonable conclusion is that the whole issue of grammaticality, when applied to complex cases and in general to spoken language, is a pseudo-problem. It is futile to search for the precise set of grammatical rules underlying spoken discourse, simply because there is in many cases no such set of rules.

Grammaticality is a rather peripheral property of structurally complex utterances in natural spoken discourse. There are of course clear rules for how simple words, phrases and sentences may be constructed, but when speakers are forced - due to complex contents or other situational factors - to use long and complicated linguistic expressions, these constructions are combined in various ways, and the results become more or less clumsy, more or less comprehensible, they may be more or less good vehicles for getting the messages across. Many such expressions are hard to assess for grammaticality; instead, one must concentrate one's efforts on finding plausible interpretations. When linguists try to use their intuitions for determining grammaticality, they very often fail. As a result, many utterances are discarded as impossible or ungrammatical, unless - by some coincidence - someone comes up with a situational context where the utterances may in fact occur, in which case the judgements are often changed. In other cases, the linguists are forced to rely on more or less arbitrary rules of thumb (c.f. above).

One of the reasons why the assessment of grammaticality becomes such an insecure endeavor is the fact that many spoken utterances look very clumsy, unnatural and difficult to interpret when they are given in written form. The context, which made their production possible, is gone, and the same holds for the prosody. Both may be quite difficult to hit upon in the absence of direct clues. Furthermore, there are fewer and less clear restrictions on word order in speech than in standard written language. Once a natural prosody is found many seemingly odd word orders turn into something quite natural.

It might be added here that the data which are discussed by linguists are most often made-up sentences and not utterances taken from authentic spoken discourse. This means that the examples are quite often edited in some respects; for example, the complex structures considered by e.g. Ross (1967) are typically coherent and complete sentences, not such often incoherent structures (like anacolutha) as occur in actual spoken language. Here we see how the linguist s norms may even distort the very data, i.e. that which is to be used as empirical evidence. We witness the practice of regularizing data (V.1)

The boundaries of grammaticality are fuzzy at many points. As I noted earlier, the main explanation for this is probably the fact that grammaticality is often rather unimportant at least in the life of spoken language. After all, normal dialogues do not consist in an exchange of arbitrarily chosen "grammatical sentences". On the contrary, utterances must, first and foremost, be situationallv adequate and relevant at the particular stage of discourse where they occur. This raises the question of whether sentences can be considered well-formed in themselves, i.e., as independent formal objects (as is usually assumed if language is defined as a set of sentences), or if they should be considered "well-formed", "acceptable", "correct", adequate" or "plausible" (there are many such predicates that could be used) only relative to (classes of) meanings, interpretations and contexts.

Let us therefore briefly discuss the relative importance of the three conditions of situational relevance (which is basically a matter of semantics in the wide sense), coherence (and/or cohesion) [12] (which has to do with both semantics and syntax) and grammaticality (grammatical accuracy)(which is primarily a question of outer form). The first ment1oned condition specifies that utterances must be adequate and relevant in the particular discourse contexts in which they are embedded. The second condition means that utterances and discourses must be sufficiently coherent as regards both content and form. Thirdly, utterances must in general exhibit certain features typical of what is conventionally regarded as correct in the language in question. In addition, speech must often also exhibit a considerable amount of fluency on the part of the speaker [13]. Without this he would run the risk of the listeners taking over the floor [14].

Since communicative purposes and goals vary a great deal, the inherent importance of these conditions also varies across situations. Thus, for example, narratives, which tend to be relatively self-contextualizing, have to rely very much on coherence, whereas informal interviews and small talk are often seemingly quite incoherent. Conventional constraints on the outer form are socially important in some cases, e.g., in solemn speeches and other rituals. In spite of this, however, it seems that situational relevance, coherence and grammaticality should be rated in this order with regard to their overall communicative import. Thus, in most situations situational relevance is much more important than grammatical correctness. Without that the speaker will not be understood, whereas an utterance that is poorly formulated from a grammatical point of view, may very well be comprehended as long as it fits the communicative context reasonably well. Indeed, in most situations language users do not pay any attention to the question whether their utterances are grammatical or not. Conscious attempts to judge the grammatical acceptability of utterances may sometimes occur, but they are in no way obligatory aspects of utterance production and/or comprehension. Evidently such metalinguistic judgements require an extra effort (cf. Hakes 1980:ch. 2).

Despite all this, the linguist's normal attitude has been to reverse the order of priority for the three properties under discussion. For him the issue of grammaticality is the fundamental question, whereas situational relevance is not considered at all, at least not as long as he is following the traditions of contemplating linguistic constructions, most often sentences, without any situational context at all. This in turn brings us back to the facts that linguists still use written sentences and texts as their subject matter, and that written texts are more autonomous and less context-dependent than most spoken language. For reasons which are largely conventional, and hence not entirely functional in nature, most cultures attach a great deal of importance to the correctness, including the grammaticality, and more generally, to the formal features of written texts. Thus, we see that the linguist's preoccupation with grammaticality is closely related to the sequels of the written language bias.


6.4 Words

 In traditional linguistics as well as in the layman's understanding of language the word is beyond comparison the basic unit of language. However, words are conceived of in two different ways, on the one hand as the building-blocks used by the speaker in constructing linguistic products, i.e. as lexical items, and on the other hand as the most important components of actual utterances and sentences, i.e. as syntactic constituents. It is quite obvious that we are dealing here with two different concepts. I will first comment on the latter.

There are several ways of making the notion of 'word' in the sense of 'syntactic constituent' more precise and workable. The main alternatives are those of 'morphological word' and 'phonological word' (e.g. Ullman 1951:43 ff, Lyons 1968:194 ff). It is rather typical that the morphological word form is the type of unit which is preferred in most linguistic analyses. A morphological word form consists of one of several roots plus their various bound morphemes (affixes etc).(This covers both simplex words and compounds). Now, this notion can of course be at least provisionally defined in structural terms (cf. the notion of 'minimal free form' etc), but it is hardly a mere coincidence that the notion of the morphological word corresponds rather well with the unit of conventional writing, the graphic word, i.e., a sequence of letters surrounded by empty spaces but containing no internal spaces. Thus, at yet another level we see the intimate coupling between the linguist's theory and the layman's views, and between these two and the conventions for correct writing: 


		unit of linguistic	unit of standard
		analysis: 		written language:
phonology	segment (vowel or 	alphabetic letter
morphology	morphological word	graphic word (unit 
		form			delimited by empty
syntax		sentence		graphic sentence
					(unit delimited by major
					punctuations, i.e. full 


I am of course aware of the fact that the graphic word is not an arbitrary unit from the point of view of the structure of language (including spoken language). Obviously, it represents a rectification of a unit which by and large is the morphological word. But it may still be true that once the orthographic convention was firmly established, this circumstance served to reinforce the status of the morphological word quite strongly. After all, for spoken language there is another quite plausible candidate available, i.e. the phonological word (phonemic phrase, tone group) [15] consisting roughly of a stretch of continuous speech unified by one main stress and an intonational contour. Such a unit normally involves several morphological words; typically function words such as prepositions, articles, weak pronouns and adverbs are enclitically adjoined to a nucleus consisting of one content word (in some cases several such words). If we seriously consider the dynamics of speech behavior, i.e. the processes involved in the production and perception of utterances, it is quite probable that such units are of primary importance. The fact that they have been - until recently - largely ignored by linguists is no doubt related to the neglect of prosody (VI.2).

Let me now turn for a while to linguist s conception of lexical items (lexemes). If we consider a broad spectrum of empirical evidence, it will turn out to be necessary to postulate several different types of lexical items, e.g. stems, word forms and phrases (see e.g. summary in Linell 1979a:ch4). Yet, in modern theoretical linguistics morphemes have been assigned the status of lexical primes; some scholars seem to be prepared to exclude the possibilities of all other types of lexical items. Such a view implies an overly abstract view of language, especially in certain applications of orthodox generative phonology [16]. By straining our arguments just a little, we might connect this preference for morphemes to the written language bias.

The metalinguistic contemplation of abstract or ephemeral meaning relations underlying many abstract morpheme identities postulated in generative phonology seems typical of someone who is working with words as isolated from the ever changing and perhaps bewildering array of different situational contexts of spoken discourse especially; from a communicative point of view such lexical and morphophonological regularities are hardly primary. Furthermore, although it is sometimes not frankly admitted, such a linguist normally draws quite a lot upon his knowledge of the etymology of the words involved and of the entire history and genetic relationships of the languages under consideration, and this knowledge is ultimately derived from written records. Orthodox generative-phonological analyses (cf. fn. 15) tend to recapitulate history in synchronic description, as has been pointed out by e.g. Maher (1969).

There is another aspect of the structuralist view of morphology which is worth mentioning in this context, viz. the tendency to regard root morphemes and grammatical morphemes as segment-based expression units of the same kind. This seems natural enough given the orthographic conventions according to which they are represented in basically the same way. However, evidence from studies of utterance production and perception strongly suggests that content words and grammatical morphemes are functionally quite different. For example, data on speech errors and aphasia indicate that most grammatical morphemes should be treated together with other aspects of syntactic construction, e.g. word order (cf.Garrett 1980, Saffran et al 1980, Linell 1982 b).


6.5 The structure of texts and discourse

As part of the immensely increased activities within text linguistics, discourse analysis and conversational analysis in the 1970-s,attempts were of course made to apply linguistic methods of analysis also to larger written texts as well as to spoken dialogues. Linguists started to search for the structure of texts and conversations. Some of these efforts ended in disappointment; one did not find any good examples of well organized structure, particularly not in certain forms of spoken discourse. However, the main reason for this may well have been that most scholars started with a preconceived notion of what kind of structure they expected to find. They tried to use a normative concept of structure, one which was based on rules for how written texts should be composed and organized. Both outsiders and insiders in text linguistics now seem inclined to agree that the attempts to set up generative text grammars (van Dijk 1972, Thorndyke 1975 and others) were failures. It was useless to try to define possible vs impossible texts in analogy with grammatical vs ungrammatical sentences. tIn fact, even the latter task is dubious when applied to natural spoken language, as we have seen (VI.3)). Halliday & Hasan (1976) address the issue in this way:

'A text is a unit of language in use. It is not a grammatical unit, like a clause or a sentence; and it is not defined by its size. A text is sometimes envisaged to be some kind of super-sentence, a grammatical unit that is larger than a sentence but is related to a sentence in the same way that a sentence is related to a clause, a clause to a group and so on: by constituency the composition of larger units out of smaller ones. But this is misleading. A text is not something that is like a sentence, only bigger; it is something that differs from a sentence in kind.... A text does not consist of sentences; it is realized by, or encoded in, sentences. If we understand it in this way, we shall not expect to find the same kind of structural integration among the parts of a text as we find among the parts of a sentence or clause. The unity of a text is a unity of a different kind. (Halliday and Hasan 1976:1-2)

The upshot of the argument is that texts and discourses do not have grammatical structures (like sentences and smaller units), and texts cannot be assessed with regard to grammaticality. Instead, texts convey meaning in contexts, and what might be called discourse structure should be explained with reference to the dynamics of the whole communication situation. to the processes of production and comprehension which can hardly be treated adequately without recourse to the intentions, expectations and partially shared worlds Of the communicating parties (Levy 1979). But the linguistic approach, which was criticized earlier, simply followed the tradition of studying language as products, as static objects, which are searched for a structure of well-defined internal relations.






1. Note also the characterization of a natural language as if it were a formal language (7.8).


2. In fact, Gardiner himself confuses these two concepts, as he concedes himself (1951:336-7)


3. The term 'utterance' is often used in another sense, viz. 'the entire stretch of speech produced by a speaker in one single turn of speaking'. This sense of the term will not be discussed here.


4. I am, of course, aware that these concepts are not equivalent, but the differences seem immaterial to my present argument.


5. One might say that there has to be some type of counterparts of the initial majuscle and full stop (point) of the written language.


6. I abstain from discussing whether or not imperatives are complete sentences.


7. The idea of such a generative text grammar is actually, if taken seriously, a dead end (cf. e.g. VI:3)


8. On matrix vs. subordinate sentences in German and Swedish, see e.g. Sandig 1973 and Andersson 1975, respectively.


9. Welin here alludes to the work of Frazier & Fodor 1978, but his points are generally valid for most of the standard psycholinguistic research on the perception and comprehension of linguistic texts.

10. Stars and question marks are Ross's own. They seem to be used about both ungrammaticality and unacceptability. Although Ross subscribes in theory (1967:59, fn.8) to Chomsky's (1965) definition of grammaticality vs. acceptability, he works - in practice - primarily with unacceptability. (As we will point out below, the distinction is indeed hard to maintain). Ross (op.cit.) characterizes (6) as relatively acceptable (of "reduced unacceptability"), (8) as "ungrammatical", and (9) as "grammatical but of low acceptability". (7) is said to be a "tortured" and "less than felicitous".


11. Ross (1967) declared e.g. (7,8) ungrammatical, while Encdahl (1979) found (11) to be grammatical in Swedish. Engdahl did so after providing a context where (11) could possibly occur, while Ross aid not discuss contexts explicitly. However, in all fairness it must be pointed out that, in general, Swedish does seem to be more liberal in allowing so-called extractions out of relative clauses, i.e. in the type of examples considered here (Engdahl & Ejerhed 1982). Such differences between languages are obviously not uncommon. But this does not change the main points of our arguments in this paragraph.


12. Many scholars make a distinction between, on the one hand, coherence (coherency) meaning conceptual (semantic) connectedness and, on the other hand, cohesion having to do with formal markers of connectedness (anaphora, ellipsis etc).


13. For an analysis of the concept of fluency, see Sajavaara & Lehtonen (l978)


14. Obviously these four conditions are not mutually independent, but we need not indulge in a comprehensive conceptual analysis here.


15. I ignore here any differences between these concepts.


16. Some extreme examples of orthodox generative-phonological analysis are Chomsky & Halle 1968, Schane 1968, Foley 1965, Vennemann 1968. For a general critique of the treatment of morphemes as abstract phonological forms, see Linell 1979a: ch 13.