13. Linguistics versus language: Retrospects and prospects ¹

13.1 We have now surveyed the discourse of ‘theorizing’ in some major works in this century attempting to establish the foundations of linguistics. Focusing on the works as discourse helps to see them not just as documents, but also as ‘performances’ with characteristic ‘discourse moves’ (cf. 1.11). These moves include claiming scientific status for linguistics; estimating the state of the discipline with its strengths and weaknesses, and situating it in respect to other disciplines; selecting certain aspects for investigation and rejecting others; proposing criteria for constructing theories or discovering data; setting degrees of precision or delicacy; determining what counts as the same or different within one language or among several; deciding how many levels of structure should be postulated for language sequences; presenting and justifying terms or notations; and so on. Many theoretical steps involve tradeoffs, where some advantage is gained by accepting a disadvantage elsewhere in the theory (cf. 9.111; 13.1, 20, 52, 55). But local losses and gains can still add up to a global increase in insight: ‘the history of language research shows that progress comes from shifting and enriching the problematics’ (12.4). We can thereby move beyond the ordinary awareness wherein ‘language is such a familiar, important, and central phenomenon’ ‘that it is hard to even notice all its basic features and their significance’ (12.9; cf. 2.8; 3.1; 4.2; 6.6). Yet linguists also are continually in danger of understanding the data too readily and underestimating their involvement in producing them (cf. 1.9; 2.66; 3.11, 50; 4.4, 31, 72; 5.9, 11, 24, 78; 8.14; 13.36, 49).

13.2 My survey hardly resembles the ‘textbook’ sketched by Thomas S. Kuhn (1970:137ff). Because ‘the scientist's contemporary position seems so secure’ and because they are ‘pedagogic vehicles for the perpetuation of normal science’, Kuhn says, ‘textbooks’ get ‘rewritten in the aftermath of every scientific revolution and then disguise not only the role but the very existence of the revolutions that preceded them’ (cf. 1.4). ‘Textbooks thus begin by truncating the scientists’ sense of their discipline's history and then proceed to supply a substitute for what they have eliminated’. They ‘refer only to that part of past work which can easily be viewed as contributions to the statement and solution of the textbook's paradigm problems’. So ‘instead of forgetting’ their ‘founders’, ‘scientists are able to forget’ ‘or revise their works’.

13.3 By restaging the founders’ works in a complex fabric of individual voices, my survey hopes to dissolve the complacent idea that past trends have been inexorably leading up to some culmination in the present. Indeed, I do not see any obvious current ‘normal science paradigm’ in linguistic theory.² Nor is there currently one general theory capable of subsuming and integrating the available alternatives (cf. 5.9; 6.18; 8.35), which is scarcely surprising given the widespread accentuation of divergences (13.6). Basic works in linguistics typically controvert the Kuhnian ‘textbook’ by their express intent to inaugurate ‘revolutions’, and thus find it strategic to criticize the state of the discipline and to propose new projects in declared opposition to past research.³

13.4 This intent favoured the impression that ‘revolutions’ in linguistics have been ‘frequent and radical in throwing away all that came before’ (Winograd 1983: 8). On closer inspection, the progress of the discipline reveals a more interesting and complicated pattern we might call ‘ancestor-hopping': repudiating one's immediate precursors while reaching further back for sources. Time and again, our theorists directed their sharpest criticism against the more recent segments of prior research, which they no doubt viewed as the nearest competitors. Suppose we made a roughly chronological chart:

1. traditional grammar (classical, medieval, school grammar)

2. philology (historical or comparative grammar, phonetics of sound change)

3. mentalist descriptive linguistics (continental European structuralism)

4. physicalist descriptive linguistics (American and British structuralism)

5. logical linguistics (algebra, calculus, generative or transformational grammar)

6. systemic or functional linguistics (British functionalism)

7. computational research (artificial intelligence)

8. cognitive research (cognitive science and psychology)

We can often see an approach dissociating itself from the one(s) just before it, while approving one or more earlier ones. The mentalist descriptive linguistics of continental European structuralism spearheaded by Saussure turned against philology so emphatically that ‘traditional grammar’, though admittedly ‘unscientific’, was judged more ‘correct’ and ‘less open to criticism’ (2.6). The physicalist descriptive linguistics of American structuralism inaugurated by Bloomfield repudiated mentalism, which ‘still prevailed’ ‘among men of science’, and which he grouped together with the outlook of ‘grammarians’ in ‘our school tradition’; in exchange, philology was lauded as ‘one of the most successful’ ‘enterprises’ ‘of European science in the nineteenth century’, one that ‘replaced speculation’ ‘with scientific induction’ (4.4f, 8, 73, 76). In turn, Chomsky rebuked Bloomfieldian descriptive structuralism as ‘fundamentally inadequate’ and gave a high appraisal both to ‘mentalism’ and to ‘traditional grammar’, which he also associated with each other (7.5, 7, 34, 37, 10, 4). His own ‘generativism’ was in its turn reprimanded by van Dijk and Kintsch's cognitive approach, which in exchange saluted linguistic ‘Structuralism’ and ‘Formalism’, along with ‘classical poetics and rhetoric’, and even ‘literary scholarship’ (11.1f, 58).⁴

13.5 On the British scene, Firth, like Bloomfield, excoriated both mentalism and current school grammar; but he also mistrusted ‘philology’ and reached far back to ancestors in phonetics, grammar, and orthography from the Elizabethans down to Henry Sweet (8.24, 37, 41, 6f, 15, 8¹³). Halliday's ‘systemic’ or ‘functional’ approach purported to be an elaboration of Firthian linguistics, and an alternative to both American structuralism (‘chain grammars’) and Chomskyan generativism (‘formal’ and ‘transformational’ grammars) (9.3-6, 9²⁰).⁵ In fact, his approach entailed extensive revisions of Firth's, particularly by incorporating the Prague school's ‘functional’ strand of European structuralism dealing with communicative topics and focus, which Firth would probably have considered mentalistic (9.47, 56, 9¹⁰, 9¹⁹; cf. 8.25). Only Hartmann seems to have genuinely appreciated all his predecessors, and even he, in doing so, was revising a preceding tradition, namely the isolationist ‘mother-tongue’ linguistics that held sway in Germany before and during World War II (12.19) -- and of course the general fractiousness of the disclipine.

13.6 These patterns of ‘ancestor-hopping’ indicate that linguistics did not so much deny its global history as accentuate its local discontinuities, censuring the more recent or dominant past and commending the more remote or marginal past. Hence, although each ‘revolution’ in linguistics overthrew and supplanted the currently ruling paradigm, it could be hailed as a source again after a later revolution, when it would no longer pose a threat of competition. Paradigms in linguistics have thus proven more resilient and resurrectable than those in many other disciplines. Admittedly, the acute emphasis on local discontinuities can foster an image of linguistics as a contentious field with more periods of ‘crisis’ than of ‘normal science’.⁶ If we totalled up all the criticism raised at some time against theories and methods (13.3), few if any would remain unscathed.

13.7 Of course, explicit discontinuities did not preclude implicit continuities. Despite declared antagonism, our theorists retained some contact with traditional grammar and at times raised the prospect of recycling it into linguistics (cf. 2.6, 15; 6.49; 7.4, 75; 8.38; 12.41, 88, 11²¹). A case in point is the embattled ‘parts of speech’ schemes often subjected to ‘muckraking’ (5.72ff; cf. 2.65; 3.23; 4.51; 6.49; 8.37, 43, 58f; 9.13, 34, 9¹³). Theorists concurred that the criteria for defining these ‘parts’ had been inconsistent, diffuse, vague, and unreliable, but not about which new criteria which should replace the old. Recourses included: not presenting one's own parts of speech scheme at all (Saussure, Sapir, Firth, Hartmann); proposing an entirely new scheme (Hjelmslev); or, most often, maintaining and revising the traditional scheme with added or substituted criteria drawn from fieldwork on lesser-studied languages (Sapir, Bloomfield, Pike), formal logic (Chomsky), communicative contexts (Halliday), or cognitive psychology (van Dijk and Kintsch). The popularity of maintenance and revision suggests a general belief that a totally new scheme wouldn't be accepted, and that the old criteria were not such a serious liability because they could be easily used and because they reflected the fuzziness of the categories themselves (cf. 2.33; 13.59). Linguists can usually agree on what to classify as a ‘noun’ or a ‘verb’, however intuitive their criteria might be, whereas brand-new terms like Hjelmslev's ‘plerematic syntagmateme’ and ‘nexus-conjunction’ for those categories (6.59) lack that advantage, unless they are taken to mean exactly the same as the old terms, which would defeat the purpose. Ironically, it was Chomsky's proposal to create a ‘purely formal basis’ for ‘grammatical theory’ (7.56) that relied most heavily on traditional, very non-formal grammars, and set English up as the model of ‘underlying’ formality in place of Latin with its explicit formality (cf. 13.30, 42).⁷

13.8 Continuity also obtained when structuralists borrowed from philological methods and materials, although they reversed the earlier emphasis by foregrounding the formal diversity of languages rather than the comparison and classification into families based on genetic commonalities (cf. 2.5, 10; 3.40, 45ff; 4.72ff; 5.26, 5⁶). The generativists borrowed the materials and methods, now much expanded by the structuralists, and (though generally ignorant of philology) returned to an emphasis on commonalities -- not genetic but arising from ‘universals’ -- while devaluing formal diversity as a ‘surface’ issue (7.19). All these far-reaching but implicit continuities made the history of the discipline far less cataclysmic, whence Halliday's wry remark that ‘twentieth-century linguistics’ ‘has tended to wrap old descriptions’ inside ‘new theories’ instead of seeking genuinely ‘new descriptions’ (9.24).

13.9 An equally intriguing pattern of alliances and antagonisms appears in the shifting relations between linguistics and the other disciplines. The latter often provided strategic frameworks, the more so when linguists were anxious to dissociate themselves from the prior or current paradigms in their own discipline. So ‘linguistics’ conspicuously illustrates how the ‘scientist looking for a new paradigm is strongly affected by the other sciences currently enjoying successful development’ (Winograd 1983: 8). The pressure to borrow is reinforced when the object domain is too complex to suggest any obviously appropriate theory, and when the scientific climate is too austere to favour entirely novel theories. On the other hand, linguistics sometimes showed a drive to go its own ways and remain aloof from its neighbours (cf. 2.7; 6.6; 8.17).

13.10 Of all the disciplines, psychology had the most varied fortunes as a model. In the early decades of the 20th century, when Saussure and Sapir were working out their conceptions (and the young Bloomfield was writing his 1914 Introduction to the Study of Language), mentalistic psychology was well established, particularly in Europe. It therefore seemed plausible that ‘the concrete object of linguistic science is the social product deposited in the brain of each individual’, and that language is a medium for coordinating discoverable forms (words, word-parts, ‘signifiers’, etc.) with their communally assigned content (‘ideas’, ‘thoughts’, ‘concepts’, ‘signifieds’, etc.) (2.16, 83; 3.1, 17). For Sapir, ‘language’ represented ‘a fully formed functional system within man's psychic or “spiritual” constitution’, and ‘linguistic forms’ ‘had the greatest possible diagnostic value’ for ‘understanding’ ‘problems in the psychology of thought’; ‘perhaps psychologists of the future would be able to give us the ultimate reasons’ for the ‘fundamental form intuitions’ of language (3.9f, 55; cf. 2.7, 17, 32, 35; 3.12, 20, 37, 62; 6.6;). Certain affinities for mechanism and physicalism were detectable (2.31, 83, 2³; 3.20, 3¹¹), but were not felt to disturb the mentalist scenario (cf. also 7.16, 93).

13.11 The dramatic swing from mentalism to the ‘mechanism’ or ‘physicalism’ roundly espoused by Bloomfield, Pike, and Firth, however, transformed the scientific climate and the prospects for cross-discipinary interaction. Psychology lost its model status to ostensibly ‘harder’ disciplines. Bloomfield hopefully suggested that ‘the methods of linguistics’ ‘resemble those of a natural science, the domain in which science has been the most successful’ (BL 509) (4.8; cf. 2.13; 4.18; 7.11; 9.112; 12.14, 49, 99). Whereas Sapir had declared that ‘languages’ were in no way ‘explainable’ by ‘the laws of physics and chemistry’, Bloomfield now declared that the constitution of ‘speech’ follows ‘cause-and-effect sequences exactly like those we may observe, say, in the study of physics and chemistry’ (3¹⁰; 4.8; cf. 2.82; 5.66). Henceforth, all mentalist terms, like ‘“mental images”, “feelings”, “thoughts”, “concepts”‘, ‘“ideas”‘, or ‘“volitions”‘, were deemed ‘merely popular names for various bodily movements’ (4.9; cf. 8.22ff).

13.12 Although Bloomfield's ‘stimulus-response’ model came not from physics or chemistry but from animal-conditioning research, he nowhere expressly proposed biology as a model science. Instead, he usually mentioned ‘biological’ factors as a contrast to language, and averred that ‘the effects of language’ ‘distinguish man from the animals’ -- a view aired also by Sapir, Chomsky, Firth, and Hartmann (4.34, 75, 2; 3.15; 7.35; 8.27; 12.10, 11³⁴; 13.18), though semioticians reject it today. Perhaps biology seemed unhelpful because Bloomfield realized that an explanation of language would not readily come from ‘the working of the nervous system’, which ‘is not accessible to observation from without’, nor even by one's own ‘sense-organs’ (4.18; cf. 8.21). Another, more powerful, reason may have been that nineteenth century language study had drawn elaborate parallels to such biological conceptions as ‘organicism’ and ‘evolution’ (cf. 3.2; 8.6; 12.17). An emergent science might find it politically unwise to advertise its reliance on biology.

13.13 Despite having been a pupil of Sapir's (5.69), Pike made similar moves to Bloomfield's. Pike's ‘particle’, ‘wave’, and ‘field’ scheme was more elaborately modelled after physics than was Bloomfield's sketch of speech as ‘sound-wave’ transfer (cf. 5.31f; 4.10); in return, Pike's ‘Unified Theory’ made no explicit appeals to biology. Firth attacked ‘philology’ for its ‘biological analogies’, but drew some of his own, e.g., advocating the study of ‘linguistic behaviour’ as a way of ‘maintaining appropriate patterns of life’ in analogy to ‘the study of the whole man by biologists, anatomists, physiologists’, ‘neurologists, and pathologists’ (8.6, 20).

13.14 The fortunes of psychology in linguistics also alternated with those of social research. For the early mentalists, ‘language’ was ‘exclusively psychological’ (2.31). In return, ‘society’ and ‘social’ aspects were treated only episodically, chiefly as a regulatory factor that disseminates language and imposes uniformity upon it (2.16, 28, 33, 67; 3.1, 3, 55), and sometimes as a devisive or irrelevant factor, a move repeated by Hjelmslev (2.9, 44; 3.64f; 6.14).⁸ Later, the physicalists and behaviourists declared their disdain for ‘mentalistic psychology’ and for ‘psychological doctrine’, theory’, ‘explanations’, or ‘analysis’ (4.8, 19, 80, 4¹ 4²⁶; 8.17, 24, 28, 54f).⁹ In return, much attention was given to the ‘social’ aspects of ‘language’, the latter being ‘the most fundamental of our social’ ‘activities’ (4.16; cf. 4.9f, 25, 82, 84f, 88; 5.65, 85; 8.10, 16, 28, 47; 9.8f, 18, 40). Firth and Halliday suggested that ‘sociology’ exerted priorities conflicting with ‘psychology’, which they considered too dependent on (non-observable) mental states and individual dispositions (8.17, 25, 28; 9.6f, 9⁹). Van Dijk and Kintsch finally signalled a balanced synthesis: their methods are mainly psychological, but social factors are prominent also (11.1), notably in such concepts as ‘schemas’ of shared world knowledge, and in the appeal to large experimental test populations rather than to themselves, single readers, or ‘ideal speaker-hearers’.¹⁰

13.15 The relation between linguistics and mathematics has also had a peculiar history, being upheld more often in name than in deed. Saussure and Sapir betrayed a taste for ‘formulas’ and envisioned some parallels between ‘language’ and ‘algebra’ (2.70, 82; 3.72f), but made no attempt to work out a full theory or representation on that basis. Bloomfield wistfully admired ‘mathematics’ as the ‘ideal use of language’ and a ‘specially accurate form of speech’, and introduced some of his own ‘formulas’ and ‘equations’ (4.21, 4²⁰), but he too went no further. Firth dourly conjectured that ‘a linguistic mathematics’ would ‘become a dead technical language’ (8.31). Hjelmslev, however, declared the ‘main task’ of ‘linguistics’ to be the creation of ‘an immanent algebra of language’, and his Resume executed such a system in relentless detail (6.8, 29, 42, 59). He also advocated a ‘logical theory of signs’ based on ‘the metamathematics of Hilbert’, whereby we could ‘consider the system of mathematical symbols’ without ‘regard for their content, and describe its transformation rules’ ‘without considering possible interpretations’ (6.56; cf. 12.36). In return, Hjelmslev made no appeal to biology or chemistry, and placed physics, sociology, and psychology firmly outside the scope of his proposed discipline (cf. 6.7, 12, 14, 32, 43, 54, 62).

13.16 Logic played an influential role too in the unsettled relation between linguistics and philosophy. Having been a chief ancestor of linguistics, philosophy was a common target of censure from many theorists except the inclusive Hartmann (12.16f, 20, 33, 38, 94). Saussure's brief overview of the prior ‘stages’ of his ‘science’ omitted language philosophy outright, even its offshoots in philology, e.g. via Humboldt (cf. 2.5; 12.16f, 20). Sapir grouped ‘philosophers’ with ‘romancers’ as people concerned with ‘what lies beyond the demonstrable’ (3.67). Bloomfield blamed ‘philosophy’ for the disarray and confusion in ‘traditional grammar’, and derided the ‘metaphysics’ in such conceptions as ‘universal forms of speech or of human “thought”‘ (4.4ff, 51, 72). Firth similarly chided the ‘philosophically pretentious’ nature of ‘traditional grammatical categories’, the reliance on ‘logic and metaphysics’, and the notion of ‘universals’; he cheerfully forecast that ‘during the next fifty years general linguistics may supplant a great deal of philosophy’, Hjelmslev's work being one foretaste (8.5, 19, 16).

13.17 And Hjelmslev's deliberations did demarcate an important shift, abetted by the continuing march of ‘positivism’ and ‘unified science’ (Vienna Circle, Carnap, Neurath, Morris, Hempel, etc.).¹¹ For Bloomfield and Firth, ‘logic’ had been just a part of ‘philosophy’ and as such a baleful influence on traditional grammar (4.4f; 8.5, 17; cf. 3.23). Hjelmslev too occasionally gave ‘logical’ a negative sense by associating it with ‘psychological’, and strongly rejected ‘metaphysics’, a domain in which he surprisingly included ‘realism’, ‘objects’, and ‘substance’ (cf. 6.3, 7, 12ff, 28, 32, 39, 44). Yet he envisioned a new ‘semiotics’ as a ‘logical theory of signs’, inspired by the work of ‘logicians’ (Tarski, Carnap) (6.56). Evidently, formal logic was one branch of ‘philosophy’ free from all suspicion of ‘metaphysics’ and suited to offer a prestigious example for linguistics, as occasionally signalled also by Firth, Halliday, van Dijk and Kintsch, and Hartmann (8⁷; 9.59; 11.40; 12.64). Still, we find a more widespead undercurrent of scepticism in linguistic theorizing about the usefulness of models from logic.¹² Pike for instance was nonplussed by Hjelmslev's ‘general, logical “grid” of relations of a quasi-mathematical type’, and by proposals for ‘a theory’ as ‘a set’ of ‘postulates, definitions, transformation rules, and theorems’ (Olmstead 1954:106), or for ‘a scientific theory’ ‘constructed’ upon a ‘relevant mathematical system’ (Peterson & Fillmore 1962:477), and so on (LB 285, 71f) (cf. 5.86).¹³

13.18 In Chomsky's Aspects, the reverence for logic was coupled with the admission of philosophy at large, including areas once classed under metaphysics, e.g. ‘universals’, which are much less likely to emerge from empirical findings.¹⁴ At that point, philosophy was set up as a full-fledged framework in opposition to the sociological and anthropological ones favoured by both structuralist and systemic-functional approaches (cf. 9.3f), thanks above all to the famous ‘idealization’ about the ‘completely homogeneous speech-community’ (cf. 7.12, 7.96). Regarding other disciplines, Chomsky was syncretistic, even opportunistic, appropriating symbols and notations from mathematics and algebra, and comparing his own theory to ‘a scientific theory’ in ‘physics’, and his ‘grammar’ to a ‘chemical theory’ that ‘generates all physically possible compounds’ (7.16, 33, 36, 40, 7¹⁸). Or, he brought in biology by appealing to such notions as ‘evolution’ and ‘neural organization’ ‘grounded in physical law’ and by referring to ‘animal learning’, even though he elsewhere dismissed ‘comparisons with species other than man’ and asserted that ‘language’ is ‘a human creation’ (7.33, 35; 13.12). Still, Chomsky's invocations of ‘natural science’ (7.11) were not reflected or pursued very far in his actual proceedings. For example, his treatment of sentences was more elementary than a chemist's treatment of compounds: whereas compounds have emergent properties unlike those of the parts extracted by analysis (e.g. water vs oxygen and hydrogen), the sentences were supposed to be treated by the grammar as the sum of invariant parts and features (cf. 7.82; 13.59).

13.19 Chomsky's revitalized mentalism brought renewed attunement with psychology, but not with its then current paradigm. In effect, he envisaged a revolution in psychology to be steered by remote control from his new linguistics, which rejected the behaviourist inductive-experimental methods in favour of elaborate conceptions of ‘intrinsic cognitive capacities’ responsible for language acquisition (7.10f, 30-35). This intervention helped to hasten the decline of the stagnating behaviourist paradigm within psychology and to lend fresh momentum to the field of ‘psycholinguistics’, which provided new ways to test linguistic claims and, ironically, later uncovered the flaws in Chomsky's own paradigm, consumed so to speak by the outrunners of its own revolution (cf. 11.2f, 34, 40, 81).

13.20 By a curious symbiosis, contact with linguistics seems to cause other disciplines to undergo fractionation. Just as philosophy got polarized into ‘bad’ metaphysics and ‘good’ formal logic by Hjelmslev and the positivists, so also was psychology split into ‘bad’ behaviourist empiricism and ‘good’ innatist rationalism by Chomsky and the generativists. The counter-trend, namely an openness to cross-disciplinary currents despite inner-disciplinary rivalries, has been fairly rare and appears most clearly in Hartmann's synthesis, e.g. when he respected the philosophical approach of ‘registering language as a phenomenon’ rather than ‘treating’ it ‘as an object’ because ‘the original experience should not be disturbed by dividing or combining’ it, or by ‘forcing language into predecided representations’ (12.38).

13.21 The reverse side of this indebtedness to other disciplines was the remarkable eagerness of linguistics to present itself as the model or theoretical centre for the others (2.7f; 5.7, 84; 6.9f, 41, 53; 7.8; 8.16, 29; 12.6, 9, 12, 33, 64; 13.59). Even Hjelmslev, whose theorizing remained in the most expressly preliminary stage, felt ‘led to regard all science as centred around linguistics’, and foresaw a perspective in which ‘no object is not illuminated from the key position of linguistic theory’ (6.10, 53). His reasoning may have resembled Firth's, who claimed that ‘linguistics’ as ‘a social science’ was ‘ahead of the others in theoretical formulation and technique of statement’; or Pike's, who said that ‘formal studies in the linguistic area’ offer ‘a base which’ ‘is easier to build on’ (8.16; 5.7). Evidently, the abstractness, generality, and rigour of linguistic theorizing were construed as advantages over sciences with more concrete empirical methods and more tangible objects (cf. 12.99; 13.60).¹⁵

13.22 Or, the key argument in claiming model status could be the centrality of language, which our theorists enjoy emphasizing (cf. 1.9; 2.8; 3.1, 3; 4.2, 10, 82; 6.2, 20; 8.12, 18; 12.9). Though language is not the only mode of human understanding, it is undeniably the most readily shared and documented mode, and thus ought to be a key domain in a general enterprise of exploring communication, epistemology, and social interaction.¹⁶ However, as the shifting interdisciplinary scene indicates, the centrality of language does not necessarily establish the centrality of linguistics. For one thing, a house divided against itself or under continual reconstruction hardly offers an inviting haven for neighbouring enterprises that already have greater unity and continuity. For another thing, all the factors contributing to the centrality of language comprise too broad an expanse for linguistics to incorporate with any methods prevailing so far (cf. 8.39; 9.1, 24; 13.63). Some of the main issues that make language central to human understanding tend to be considered ‘non-linguistic’.¹⁷ Even the most conspicuous counterexample, Pike's foray into the ‘nonverbal’, brought along a markedly linguistic groundwork (cf. 5.8, 84).

13.23 Decisions about what factors to include or exclude significantly shape every linguistic school or approach and endow it with its peculiar glory and misery. The history of the field indicates that major issues can be left in the background only so long before they exert uncomfortable pressure on the conceptions and practices situated in the foreground. In consequence, linguistics has been prone to undergo ‘gestalt’ switches wherein foreground and background change places. For instance, the social and situational contexts of language use marginalized by Saussure were resolutely brought to centre stage by Bloomfield, Pike and Firth, later pushed behind the scenes again by Chomsky and his school, and then restored again to prominence by Halliday and van Dijk and Kintsch (13.14).

13.24 If language seems highly central at some times, it can also seem marginal or derivative at others. Much though by no means all of the apparent organization of language, including some of the traditional ‘parts of speech’ scheme and large areas of semantics, comes second-hand from the organization of a language community's world-model of ‘reality’.¹⁸ This ‘reality'-factor is reflected in the commonsense belief that language is primarily a means of ‘representing’ things and conveying ‘information’ (3.15; 8.47; 9.15). But the factor has received the most diverse treatment by linguists, from refusing to address it on grounds of ‘arbitrariness’ (Saussure) plus ‘metaphysics’ (Hjelmslev), or ‘autonomous syntax’ (early Chomsky), over to attacking it head-on in terms of behavioural ‘hierarchies’ (Pike), semantic ‘universals’ (later Chomsky), ‘experiential’ organization of clauses (Halliday), and finally actual ‘world’ models (van Dijk and Kintsch).¹⁹ Gradually, everyday knowledge has been recognized as a cogent and powerful resource, not an unmanageable hodgepodge vastly inferior to scientific knowledge (cf. 4.22 and 8⁷; 8¹⁴ vs. 11.24). It cannot have a solely objective relation to language because objective knowledge is not appropriate or even possible for many domains that language must deal with; but language is the major means wherewith the subjective is negotiated into the intersubjective (cf. 12.12f; 13.58).

13.25 If we insist, as Saussure and Hjelmslev in particular did, on addressing only the organization ‘in’ language and not ‘in’ the external world (2.9; 6.64), we face the perennial problem of how to uncover the ‘internal reality’ of language (2.15; 6.12). Hjelmslev adopted the most startling recourse, abjuring reality altogether as a ‘metaphysical’ factor, vowing that ‘linguistic theory cannot be verified (confirmed or invalidated) by reference to any existing texts and languages’, and offering no ‘discovery procedure’ (6.12, 19, 61). His notion of ‘theory’ seems to have had an exceptional sense, i.e. ‘formal system’ or ‘notation’, which are indeed not verifiable but merely more or less insightful and appropriate. Yet to assume that a formal system or notation already is a theory or an explanation would set linguistics outside mainstream science, where theories must be testable (cf. 2.82; 7.86; 9¹²; 11.6, 75, 99, 101f; 13.14, 19, 49, 57, 61). Evading this issue leads to such unusual senses of the term ‘empirical’ as those contrived by Hjelmslev and Chomsky (6.13, 7.85).

13.26 A more moderate and popular recourse for locating the reality in language has been to take the spoken sequence -- the ‘chain’, ‘string’, ‘utterance’, ‘sentence’, etc. -- and segment it into ‘constituents’ until we obtain the ‘minimal distinctive units’.²⁰ This solution enjoyed exemplary success with language sounds: the smallest units were those that could make a difference between two speech events, usually uttered words (cf. 2.69f; 4.29ff, 34; 6.43; 7²⁷; 8.64f, 70; 12.80, 89). These ‘distinctive’ sound-units both constitute and are subsumed by an ‘underlying’ (not directly ‘manifested’) system organized on a small number of criteria. However, these criteria were derived from the reality of articulation: not how the units in fact distinguish words or utterances but where and how they are formed by the speech organs (cf. 2.70, 73; 3.14, 18, 21; 4.29, 34; 5.42, 5¹²; 6.43; 7.20; 8.66, 70; 12.80).²¹ This derivation persisted despite the theoretical division between ‘phonology’ (or ‘phonemics’) versus ‘phonetics’, and remained the most reassuring backdrop for Saussure's vision of pure ‘differences’ ostensibly ‘abstracted’ away from ‘substance’ yet moored in the ‘concrete’ (cf. 2.16f, 26, 68; 6.13, 28ff; 12.25, 50). The concrete events (explosion, implosion, etc.) and sites (dental, labial, etc.) of articulation guaranteed the reality of the abstract system that preserves the identity of sound units (‘phonemes’) against the multitude of variations and accidents involved in acts of utterance, such as loudness, pitch, inflection, tone and quality of voice, emotional colouring, and so forth (cf. 2.68, 70; 3.20; 4.3, 45; 5.42f; 6.42f; 7.43; 8.23, 70; 12.80, 89). The correlation between phonemes and written letters was also reassuring, though usually rejected in theory (cf. 2.69; 3.19 4.38, 45; 6⁸; 8.71)

13.27 Buttressed by these implicit supports, phonology could afford a high abstractness -- above all in its focus on pure ‘difference’, ‘distinctiveness’, ‘opposition’, and ‘relation’ -- and its purportedly clean-cut separation between system and event, without seeming unrealistic or vague. Also, the phonemic system clearly showed the cogency of the notion of ‘linguistic level’ (cf. 4.71; 5.35; 7.46; 8.67; 9.30; 11.35; 12.82). Impressed by this feat, theorists aspired to project the methods and conceptions of phonology (often with phonetics in its wake) over to other domains of language (cf. 2.17, 67, 69ff, 3.18, 58f; 4.30; 5.42, 44, 5¹²; 7.20, 71; 8.66f; 12.80, 82). ‘Morphology’ postulated its own minimal units of form, again those capable of differentiating utterances, and, in the early stage at least, suggested that these ‘morphemes’ were composed of phonemes (cf. 4.50; 5.36, 45; 7.46, 61; 12.82). Admittedly, neither the organization nor the inventory of morphemic systems could be as tidy and compact as those of phonemes, and articulatory criteria could no longer offer any concealed guarantee, though writing was still helpful (4.42). Even the familiar ‘paradigms’ like noun declensions and verb conjugations in conventional grammar seemed to some theorists rather diffuse and artificial, in part because they don't form sequences or chains (cf. 4.57ff, 86; 5.74; 6.34; 7.75f; 8.57, 59; 9.31, 9¹¹; 12.29, 71). I suspect the ‘arbitrary’ quality diagnosed in language is enhanced by the ‘arbitrary’ decisions and classifications linguists must increasingly make as they go beyond language sounds, the more so if communicative contexts are discounted.²² An additional strain was exerted by the demand to keep the ‘levels’ separate, again doubtless inspired by the seeming independence of the phonemic level (cf. 5.34f; 7.20, 46; 8²³). And separating levels provided an argument for keeping them free of ‘semantics’ or ‘meaning’, which was, ironically, an original criterion in recognizing phonemes (cf, 4.14ff; 5.61; 6.43, 56, 60; 7.56ff 8.31, 46, 56, 69).

13.28 The state of affairs became still more unsettling when linguistics moved on toward the ‘levels’ of description ‘above’ morphemics. There, neither constituency nor relative size offered fully reliable criteria, and each unit of a sequence differs from those before and after it in diverse, complex ways (cf. 2.58; 12.51). If we discount forms identifiable only through etymological derivation, many ‘words’ appear to consist of just one ‘morpheme’ (4.53; 5.40, 46, 53; 6.45). Also, although a ‘phrase’, ‘sentence’, or ‘utterance’ usually contains several words, some consist of just one word, and some consist of several words which nonetheless function as a self-sufficient unit (a ‘fixed phrase’) (cf. 2.55, 61; 3.34; 4.60; 5.32, 54; 7³⁴; 9.93; 11.80). In addition, the capacity of larger units for being subjected to interruption or interpolation implied ‘discontinuous constituents’, and the absence of a unit where one appeared in parallel forms implied ‘zero’ elements (cf. 4.10, 60; 5.6; 7.39; 8.61; 11.32, 64; 2²⁶; 4³; 5.46, 5¹²; 6¹⁶; 7.75, 90). For such reasons, many diverse conjectures were made about how phrases, clauses, or sentences might form a system, and which types an inventory should list (cf. 3.37f; 4.68f; 5.57f, 5²⁹; 6.49; 7.50-53; 9.46, 54, 57f, 61-64, 74; 11.40, 67). Also, the borderline between morphology and syntax has remained problematic, due mainly to analogies between morphemes in words and words in phrases; and ‘grammar’ is generally construed to include both.²³

13.29 These perplexities indicate that data should be distinguished not merely by the constituency and size of units, but by the aspect of language placed in focus. In American linguistics, the term ‘level’ was somewhat indiscriminately defined both by size and by aspect, whereas the British proposed the term rank for sizes and ‘level’ for aspects (cf. 4.71; 5.34f; 7.45f; 8.51f; 9.30, 33ff, 75, 83; 11.16f, 35, 56; 12.82). The distinction is important because a ‘higher level’ may not always have ‘larger’ units than a ‘lower one’ (5.41), and because in theory, the levels should be natural and common to all human languages (cf. 4.71; 5.34f; 7.45; 8.51f; 9.30). In practice, ‘rank’ and ‘level’ would correspond roughly like this:²⁴

RANKS LEVELS UNITS

sound, vowel, consonant phonology phonemes

word, stem, prefix, suffix, infix morphology morphemes

word, fixed phrase lexicon lexemes

word-part, word, phrase, clause, sentence syntax syntagmemes

proposition, predicate, argument semantics ememes

utterance, speech act, text, discourse pragmatics utteremes

These correspondences are conspicuously untidy, especially regarding the ‘word’. Hence, theorists were frequently uncertain about the status of the word, even though it is the entity ordinary people probably consider most obvious in language (cf. 2.18, 55, 2⁷; 3.31, 73; 4.54, 60; 5.53; 6.23; 7.70, 7¹⁹; 8.53f; 12.69, 71, 77; 13.32).

13.30 It was therefore predictable that the paradigm shift from structuralist toward generative approaches would involve a fundamental reorientation away from the minimal units obtained by exhaustive segmentation of utterances and placed in complete inventories (7.6). In the new paradigm, units could be postulated deductively and situated in more ‘abstract’, ‘underlying’ configurations of words and morphemes appearing alongside symbols (e.g. ‘“the + man + Aux + V + the + book”‘, SS 39) (cf. 13.51). The notations for ‘rules’ freely mixed items that correspond to discoverable, segmentable units together with items that only describe or classify units (e.g. ‘V --> [V + [+Abstract]-Subject, +[Animate]-Object]’, AS 114) (cf. 7.74). Words were treated in terms of ‘lexical entries’ of ‘formatives’ in which most ‘features’ did not fit segmentable units at all (e.g. ‘boy, [+Common, +Human, + Count, ...]’, AS 166), the plus sign now indicating mere presence, not concatination (cf. 7.70ff; 13.33). This treatment circumvented the problem, latent in Hjelmslev's proposals, of how to handle semantics (‘the content plane’) by a progressive, exhaustive partition of text (the ‘expression plane’) even though meaning is not really linear or isolatable by segments (6.24, 42, 47f; 12.54 cf. 5.76; 7.71f). In return, the new freedom to postulate abstract items fostered an explosion of ad hoc constructions when the description became fairly detailed. No system of ‘sememes’ or ‘semantic features’ could be as neat and complete as the system of ‘phonemes’ anchored in articulation (cf. 13.26, 59).

13.31 This rising need to impose more control on non-manifest items and structures and on explosive, ad hoc inventories created an auspicious setting for computational and cognitive processing approaches. In the computational one, items, structures, and inventories were selected and organized in terms of computability, and the explosion inherent in a generative approach was confined by the development of new programming techniques as guides and heuristics for search and construction (cf. Winograd 1972, 1983). In the cognitive approach, items, structures, and inventories were justified by showing plausible effects of their presumed utilization during tasks like recognition or recall of discourse (cf. 11.27, 33, 42, 46, 50, 58f, 62, 77, 89, 94). Both approaches saw immediate advantages in studying realistic text and discourse, where controls are the most elaborate and semantics becomes more tractable.²⁵ However, the full status of the text will be appreciated only when we overcome the tendency, inherited from ‘static’ linguistics, to treat it as a stable configuration, and can see it as a dynamic process (cf. 2.36; 3.54; 5.31, 33; 6.33; 9.43, 95; 12.56). That step would in turn help in finding ‘a more dynamic interpretation’ of the ‘system’ than ‘the static, set-theoretical view of whole’ and ‘parts’ (12.52; cf. 13.36). If language is action, then the system is a potential for action, and the theory a mode of ‘meta-action’ (cf. 4.88; 5.7ff, 12, 26f, 50; 8.20ff, 26f, 46; 9.7ff; 11.5f, 11f; 13.58).

13.32 As we'd expect, this progression of paradigms drifted steadily away from phonology as the basic model. Hjelmslev's glossematics already proposed an ‘analysis’ of both ‘the content plane’ and ‘the expression plane’ into ‘an inventory’ of ‘virtual elements’ leading to ‘essentially different results from the phonemic analyses hitherto attempted’ (PT 99; cf. 6.42, 49); one big difference would come from postulating only ‘relations’ and ‘dependences’, not ‘units’ (6.25, 28, 44f; cf. 5.20; 12.25). Despite their appeal to ‘universal phonetic features’, again grounded in articulation, generativists relegated phonological aspects to an after-the-fact ‘interpretation’ of strings ‘generated’ by the ‘syntactic component’, which was declared the ‘sole creative’ one because it alone could arrange and move things (7.20, 71, 67; 13.41). From a computational or a cognitive standpoint, language sounds are of interest only insofar as they assist processing in a ‘bottom-up’ manner, whereas the ‘top-down’ controls are more likely to be applied to higher levels and larger ranks (cf. 8.52; 11.13, 19, 25, 32, 55, 59, 73, 77; 13.44). In effect, sounds were now regarded not as a simplification but as a complication that could be postponed by assuming the input or output to consist of written character strings made of separate words (cf. 11.34, 10³⁵). Here at least, the obvious heuristic value of the ‘word’ for both machines and people easily compensated for its troublesome role within purely segmental and classificatory sound-based schemes (cf. 3.31f; 13.28f).

13.33 Again predictably, the status of written language fluctuated in inverse proportion to the status of phonology. For strongly sound-oriented theorists like Saussure, Bloomfield, and Pike, writing was merely derivative, if not misleading (2.21; 4.37, 45; 5⁴⁸). But for theorists more interested in the ‘higher levels’, like Firth, Halliday, and van Dijk and Kintsch, writing was a major medium in its own right for both theory and practice (8.72ff; 9.42f; 11.67, 77, 93; cf. 3²⁶). Firth still accepted language sounds as the procedural base and phonology as a model, but also advocated the investigation of corpuses of written texts, which can be strategically selected to exemplify a ‘restricted language’ (8.33, 65, 72ff, 81). Halliday treated sounds only in terms of the ‘intonation’ of longer stretches of utterance and their communicative intent or impact, not of their minimal differential units, and ranked speech over writing in terms of complexity (9.52, 42). Both Firth and Halliday hoped for a really new ‘grammar of spoken language’, but were content to compromise by revising the traditional one, which they admitted was centred on writing (especially in regard to the ‘sentence’) (8.58, 67, 73; 9.24, 43, 82 9⁴¹; 13.7; cf. 2.21; 4.39; 7.61). Later, formalist approaches relied directly (albeit metaphorically) on the spatial quality of written representations as sentences or symbol chains with ‘left’ and ‘right’ sides, etc., and discounted the unfolding of utterances in time (cf. 2.17, 60, 72; 5.27; 6.50; 7.48; 12.47, 83; 13.33). Even Chomsky, the champion of underlying order, once called for ‘a general theory of linguistic structure’ whose ‘notions’ like ‘phrase’ and ‘transformation’ ‘are defined’ ‘in terms of physical and distributional properties of utterances’ (SS 54).

13.34 For an empirical, experimental approach, the modalities of spoken versus written language must be handled as a concrete factor in tasks such as production, perception, recognition, and recall. Most research for longer stretches of discourse has centred on reading, while listening has been chiefly addressed in phoneme or word recognition tasks in very limited contexts (10³⁵; cf. Beaugrande 1980, 1984a, 1986b). Some research indicates that reading involves phonological recoding, but whether this is obligatory or exhaustive has been questioned (cf. 4¹³; 4¹⁵; 10³⁵) (references in Beaugrande 1984a:224). Van Dijk and Kintsch did not try to resolve the issue of recoding, which does not decisively affect the overall comprehension of discourse in terms of knowledge structures, and they saw a continuity between their own research and the ethnographic work on oral narratives and folktales (11.1, 60ff).

13.35 The foregoing sketch suggests overall that when linguistic theory moves away from language sounds, alternative controls are introduced, such as the frameworks of formal logic or operational processing. Also, the greater the concern for extended realistic communication, the more evident it becomes that exhaustive analysis into minimal, fully distinctive units is a specialized concern, useful for some tasks, such as preliminary discovery of otherwise uninvestigated languages, but by no means an account of language as a human phenomenon, a processing medium, or an interactional domain (cf. 3.24; 7.6, 30; 8.31). However, recent trends also indicate a widening awareness that such an account is unlikely to emerge out of any ‘pure’ linguistics isolated from other disciplines, particularly from sociological, anthropological, and psychological issues (5.7f; 8.16; 9.2; 11.1, 4f, 100, 102; 12.3; 13.14, 53). Saussure's famous demand for a ‘linguistics’ whose ‘unique object’ ‘is language studied in and for itself’ (2.9) ultimately means taking language as a given and abandoning the project of explaining it.

13.36 This recent awareness sends us back to the ‘uses’ of language that Saussure so emphatically set aside, and to a re-examination of his division between ‘langue’ and ‘parole’.²⁶ It now seems more productive to view the two not as a static dichotomy, but as a dynamic dialectic that can be suspended or abstracted out only by lowering control over both theory and practice. Either we investigate how the knowledge of language influences the uses of language and vice versa, or we leave the dialectic as a hidden step within the linguists’ own discovery and analysis and have few real guarantees of consistency or reliability from case to case (cf. 1.9; 13.1, 49; Beaugrande 1984b, 1987a, b, 1988b, 1989b). A study of people using language is also the best foundation for a comparative study of linguists analysing language (cf. 5.9, 13f, 16, 18, 20, 36, 46; 6.58; 8.36, 83; 12.5, 39, 84).

13.37 The dialectic can be pictured as a bi-directional complementarity, wherein each side controls the other by limiting its indeterminacy (Fig. 13.1).

-- INSERT FIGURE 13.1 HERE --

On the one hand, the ‘system’ of the language as known to the communicative participants determines what items, relations, and significances they assign to any instance of language in use. On the other hand, the steadily accruing body of experience with language use is both the source of that knowledge and a continual influence upon it.

13.38 Presumably, language acquisition involves some stage at which the system assumes a ‘critical mass’ in the sense that it can be effectively applied to most instances of use without undergoing any further radical revisions (1.9). How the system might be organized prior to that stage is among the most difficult questions for linguistics. The structuralists left the question aside and constructed their own ‘grammars’ of remote languages from the ground up, though with an enormous headstart over the native child, namely their knowledge about other language systems and about language as a cultural factor (cf. 5.26).²⁷ They were unfairly attacked by the generativists, who openly claimed that language acquisition operates the same way as the linguist's grammar-constructing process, for having created an implausible acquisition model (7.24f, 87). But Chomsky's Aspects offered no account either about what an infantile language system might look like prior to critical mass, nor indeed about how linguists should construct a grammar (7.9). He merely invoked ‘the best information now available’, without saying what it was, as proof that ‘children cannot help constructing’ a ‘transformational grammar’ (7.89). He skipped over the details of this process by adopting an ‘idealized “instantaneous” model’ where ‘successful language acquisition’ happens in one ‘moment’, and by declining to ‘consider’ the ‘order and manner’ in which ‘linguistic data’ ‘are presented’, or the ‘series of successively more detailed and structured schemata corresponding to maturational stages’ (7.89). Finally, Halliday's ‘systemic’ approach postulated a ‘proto'-stage in which children are not yet using ‘grammar’ at all, but going directly to meaning (9.12; cf. 13.53).

13.39 The dichotomy of ‘system’ and ‘use’ can also be reinterpreted as a dialectic between the potential or virtual aspects of the language (what can be done with it) and the actual or realized aspects (what is in fact done in real discourse).²⁸ Saussure and Chomsky, who strongly insisted that linguistic theory concern itself only with the potential (‘langue’, ‘competence’), suggested that the actual (‘parole’ or ‘langage’, ‘performance’) was unsystematic (2.19f; 7.12). Bloomfield's postulates of ‘infinite’ variation of circumstance and universal ‘innovation’ of meaning, with ‘every person using speech-forms in a unique way’ (4.14, 16, 31 75, 78), carried a similar suggestion. But all this would be paradoxical: a virtual system could not persist if it were frequently realized in non-systematic ways, because the realizations offer the only tangible evidence that a system is indeed being deployed (2.20; 8.61; 12.67, 77, 83). Thus, the countertrend has been to see the use, the actual realized discourse, as a system in its own right, which Pike, Hjelmselv, Firth, Halliday, van Dijk and Kintsch, and Hartmann did (cf. 5.7f; 6.34-37, 45, 6¹¹; 6¹²; 8.43ff, 49f, 52, 65, 70, 76, 80; 9.22, 24, 26, 41, 55, 102, 109; 11.19, 23, 32, 56, 86; 12.47, 55f). Most recently, computational and cognitive approaches have undertaken to develop explicit models for the systematic quality of actualization processes, whether simulated or human (11.13, 16, 21f, 26f, 29, 34, 74f, 78) (cf. Winograd 1972, 1983). Despite the long-standing limitation of linguistics to the sentence (13.54), longer stretches of discourse are evidently not unmanageably higher in complexity; on the contrary, sentences in context are easier to process, whereas isolated ones will seem more indeterminate or ‘ambiguous’ (cf. 5.56f; 7.14, 61, 82; 9.16; 11.2f, 86, 91; 12.43). Similarly, ‘context-free grammars’ may look clearer and easier to write but prove more awkward and effortful to apply to realistic samples (cf. 7.48, 73f; 11.40).

13.40 Bloomfield's ‘innovation’ postulate and Chomsky's ‘uniqueness’ argument (4.16, 61; 7.90f) signal a pervasive discomfort about language being open and flexible, allowing even modifications of the system itself. To minimize the issue, linguists typically preferred ‘clear cases’ that were either obviously acceptable or totally bizarre, and discounted the effects of ‘farfetched’ contexts (4.67; 7.21, 41f, 58). Rules and formalisms were neatly constructed by postponing the stage of diminishing returns where we move beyond the core of clear cases and structures into areas where the controls are more variable and may be due to factors other than language (cf. 8.43, 52; 9.2, 40; 11.3; 13.43f). Distressingly, native speakers have proven unskilled in deciding what sentences or utterances do or do not belong to their language or ‘grammar’, no doubt because this decision seldom arises in real discourse. What is or is not produced or accepted as an utterance depends on the participants’ ‘intentions’, a factor whose relevance for linguistic theory was in dispute (cf. 2.20, 80, 3.15, and 8.63 vs. 5.65, 11.11, 10²¹; and 12.10). It also depends on paramteres of ‘style’, another disputed factor (cf. 3.69; 5.82; 6.52; 7.41, 53; 8.83; 9.102; 11.57). For these reasons, no secure empirical basis is likely to be found for a ‘purely formal’ account of ‘degrees of grammaticalness’ (cf. 7.42; 9.102) -- an ominous prospect for theories which rest on a firm opposition between ‘grammatical and ‘ungrammatical’ (cf. 7.36, 41f; 13.59).

13.41 Yet linguistics cannot indefinitely ignore creative uses of language not foreseen by the system (cf. 7.44; 12.35, 56, 58). A conspicuous instance is modernist poetry, which violates the conventions both of ordinary discourse (including grammar) and of traditional poetry, yet can be appreciated by focusing on relations among events and choices within the newly emerging pattern (Beaugrande 1979, 1986a). But more modest examples of creativity can be found in much ‘ordinary’ discourse, where unusual but appropriate usages are readily produced and accepted (cf. 9.42; Beaugrande & Dressler 1981). Such creativity can hardly be accounted for in terms of the ‘recursive processes’ developed in ‘mathematics’ (which only repeat embedded structures), nor of any purely ‘syntactic component’ (which only reshuffles formal sequences) (cf. 7.44, 67; 13.32).

13.42 The dialectic described between the knowledge and the use of language has a useful analogy in the dialectic between ‘theory’ and ‘data’ in linguistics, again with each side ‘controlling’ the other by limiting its indeterminacy (Fig. 13.2) (cf. Beaugrande 1987a; Yates & Beaugrande 1990).

-- INSERT FIGURE 13.2 HERE --

On the one hand, the ‘theory’ of language controls what items, relations, and significances linguists will assign to any instance of language data. On the other hand, the ‘data’ control what theoretical constructs are likely to be postulated, especially when the language under investigation has a markedly different organization than do the languages for which the theory had been designed. The organization of familiar languages tends to pervade the theory in less noticeable ways, whence the urgent need for defamiliarization (cf. 2.32; 3.5, 50; 8.14; 13.7).

13.43 This dialectic entails complex dilemmas because the determination may be too low or too high. Theory is underdetermined by actual data in that (a) several theories are usually possible for the same data; and (b) the data sample can never be complete (2.12; 4.67; 7.23, 43). Linguists must rely on their intuitions about when a ‘critical mass’ is attained such that the data sample can be judged sufficient and representative for a theoretical account of a ‘whole system’, and high-level frameworks can be applied to detailed analysis (cf. 3.4, 6; 4.16, 23, 29, 67, 78; 5.2, 37f; 6.20; 8.44, 65, 70, 76; 9.19, 26; 12.94). Ideally, the theory itself might supply explicit criteria for such a judgment, but in practice we have been content so far with approximations. Theory is also overdetermined in that (a) it sets up criteria and categories with standards for rigour and formality to which at least some data do not conform; and (b) the native speakers producing the data never have a full consciousnesss of its theoretical organization. Reciprocally, data are overdetermined in that (a) their occurrence always involves at least some circumstances that are merely accidental but necessary, such as exact time and place of utterance; and (b) other factors, such as speakers’ personality traits or emotional states, control the data besides the relation to an underlying language system. And data are underdetermined in the sense that (a) collected data are finite, but the data that could belong to language are infinite; and (b) specific choices are often significant in respect to others that were not made, but could have been, according to varying degrees of probability or ‘markedness’.²⁹ Due to the diverse pressures of under- and overdetermination, linguistic theory remains uncomfortably compelled both to enrich and to rarefy its theories and its data. We enrich by constructing theoretical categories too complex to be explicitly taught or learned by ordinary speakers, and rarefy by classifying large numbers of distinct data events as being, for our purposes, the ‘same’ or ‘different’. And we also enrich our data by adding ‘underlying’ organization and formality, and rarefy by discounting ‘superficial’ organization and fuzziness, often without explicit criteria for deciding what to add or detract and where to start or stop.

13.44 The complementary dialectic between theory and data strongly recommends a concerted interaction between inductive ‘models of data’ abstracted from empirical instantiations, and deductive ‘models on theory’ specifying the theory under given conditions (Yates 1986) (cf. 11.17; 12.8; 13.19, 31). Such an interaction plainly occurs during the use and comprehension of language: inductively taking into account the elements and structures (e.g., words and phrases) we judge to have been selected (‘bottom-up’ processing), while deductively constructing and testing hypotheses about what is being said or will be said, and what it probably means for us (‘top-down’ processing).³⁰ Like acquisition, comprehension attains a ‘critical mass’ whereby the discourse can be understood without major revisions (13.38), but this usually occurs so readily and rapidly that little is known about the inductive and deductive operations involved. In all probability, the knowledge being applied extends well beyond language (cf. 11.15; 12.10, 32, 36; 13.40).

13.45 Surprisingly, however, linguistic theory has tended to argue for just one outlook at the expense of the other, notably Bloomfield for induction, Hjelmslev and Chomsky for deduction (4.7, 76; 6.16f; 7.5ff, 25, 30, 34). These imbalances created predictable blind spots and vagaries in both argument and method. The inductivists’ heavy reliance on ‘observation’ of ‘manifest activity’ entrained them in a potential explosion of data for the ‘infinite’ variety of situations, while their theories remained parsimonous (e.g. based on ‘constituency’ or ‘minimal units’) (cf. 4.8, 13f, 31, 61; 5.19, 25, 28, 38, 52, 80f, 85; 8.42; 13.26). The deductivists’ reliance on ‘intuition’ left them uncertain about how data can be gathered and matched against theoretical constructions, which became luxuriant and highly technical (6.25, 59; 7.81). Moreover, the match between theory and data was often prematurely built right into the deductive terminology, e.g., by ‘using the term “grammar”‘ both for ‘the native speaker's internally represented “theory of his language”‘ and for ‘the linguist's account of this’ (7.15; cf. 7.28, 78). Applying the same terms to a set of events and to one's analysis of it stems from a long non-operational and non-empirical tradition (cf. 12.70, 77), which ultimately must be replaced by detailed demonstrations that the two indeed do match (13.57).

13.46 Although it is still far from settled what the relationships between language and linguistics is or should be, we can imagine at least five scenarios (Fig. 13.3) (Beaugrande 1987b).

-- INSERT FIGURE 13.3 HERE --

(1) Language contains linguistics (Fig. 13.3a): the activity of ‘doing linguistics’ is just one more instance of language being used, not essentially different in kind from other instances. (2) Linguistics contains language (Fig. 13.3b): the activity of ‘doing linguistics’ has language as one domain within its larger, more abstract study of the general formal, combinatorial, and organizational properties of sign systems. (3) Linguistics and language overlap, but neither contains the other (Fig. 13.3c): the two domains share some aspects, but neither can be fully subsumed by the other. Linguistics studies language in relation to other aspects, such as social organization; and yet linguistics never gets the entirety of language into its scope of vision. (4) Linguistics disturbs language (Fig. 13.3d): the activity of ‘doing linguistics’ suspends the normal operation or function of language in order to scrutinize, generalize, objectify, formalize, and so on, perhaps in the way that ‘doing biology’ entails starving, injuring, or killing living organisms. (5) Linguistics is independent of language (Fig. 13.3e): the activity of ‘doing linguistics’ is independent from language, perhaps in the way that ‘doing biology’ is separate from the coding and decoding of enzymes.

13.47 These five scenarios form a rough continuum between two extremes: complete mutual containment at one end versus complete independence at the other. The extreme scenarios are virtually impossible to maintain in an absolute sense, and none of our theorists does assert that doing linguistics is just a typical use of language, or that linguistics and language are fully independent. Nor did they, aside from Hjelmslev and Hartmann (6.10; 12.6), welcome the idea that linguistics must embrace the whole of language plus other sign systems, though ‘semiology’ or ‘semiotics’ might (2.8f; 6.50-56; 9.110; 11.43). The theorists usually opted for some version of the ‘overlap’ scenario, pushed in one direction or the other. Interactive approaches that acknowledged the role of the linguist as language understander and communicative participant (Sapir, Pike, Firth, Hartmann) tended toward the ‘containment’ scenario, whereas formalizing, logic-based approaches that discounted the linguist's role (Hjelmslev, Chomsky) tended toward the ‘independence’ scenario. The ‘disturbance’ scenario has been recognized by both fieldworkers like Pike and experimentalists like van Dijk and Kintsch, who air the prospect that their investigations can be intrusive on normal language operations (cf. 5.13; 11.92, 10⁴). Conversely, the generativists ironically implied that language disturbs linguistics due to ‘degenerate performance’, ‘irrelevant’ and ‘unrevealing surface structure’, and so on (7.24, 62, 82, 84; 9.5; 11.100).

13.48 The ‘overlap’ scenario offers the best framework to explore how the investigation of language is carried on to a great degree both in and by means of language (cf. 1.8f; 5.23; 6.55, 58; 8.33-36, 39; 9.27; 12.12f; 13.50). Reciprocally, we can inquire how the traits of language may parallel those of a science, as Hartmann has done, e.g.: ‘language science’ ‘shares the borderline status of language, compelled to run on multiple tracks, combine standpoints and results, and adopt a general outlook’ (12.12; cf. 12.40, 44, 46, 48, 69, 87, 93, 98). The conspicuous specialized terminologies in linguistics indicate a widespread ambition to create a ‘metalanguage’ whose application and force are not of the same order as ordinary language. This ambition falls under the general strategies for founding a ‘discipline’ or ‘science’ by setting it apart from the pre-scientific practice it proposes to describe, and supplying theoretical concepts and terms to supplant ‘pre-theoretical’ ones. But the new terms must both reliably relate to their definienda and form a coherent system among themselves, and so far such demands have been only provisionally engaged.³¹

13.49 Moreover, our theories must in some way take into account the presumed language knowledge of ordinary ‘speakers’.³² Some of our theorists have cautioned that such speakers would either not make any analysis or would make an inadequate one.³³ But most theorists compromised by arguing that their account addresses knowledge which speakers do have but of which they are not ‘conscious’; the linguist approaches the same data, but with higher ‘awareness’.³⁴ This argument puts the linguists in the awkward stance of claiming powers of reasoning and insight not open to normal people, much the same stance criticized for traditional grammarians (4.5; 8.7f). The best grounds for such claims would be results showing that one's insights generate predictions confirmed by empirical tests. Other solid grounds would be a thorough knowledge of numerous languages with extremely diverse organizations (2.10). To base the claims only on having a theory or formalism is less compelling, since the theory increases ‘the danger’ of ‘hearing one's own thoughts’ (12.38), especially when formalisms receive more attention than the data do, and when having a notation is equated with having a ‘theory’ (13.25, 49).

13.50 In any event, attempts to create a completely independent ‘theoretical’ apparatus without any grounding in ordinary language are unlikely to succeed. The normal business of any discipline, including the bulk of theoretical argumentation and practical demonstration, must be conducted in discourse, and no discourse, however many formalisms we deploy, can be fully separated from ordinary language. An artificial formal language can be a revealing construct only if we pay close attention to how it is made and used, and do not allow it to take on a life of its own by dictating to us what we can label and classify or by erasing the vital characteristics of natural data (cf. 8.31; 9.5). For example, the multiple structurings of a clause as a syntactic pattern, a transitivity configuration, and an information slope were often collapsed into a single one because most formalisms are syntactically oriented (cf. 5.40; 9.46, 55, 75, 109). Evidently, formal ‘algorithms’ are a two-edged sword for both discovering and concealing potential data (cf. 5.62, 86; 9.110; 11.14).

13.51 In recent linguistics, the balance between theory and data has been weighted by a large number of alternative non-language representations, such as ‘symbols’, ‘bracketings’, ‘trees’, ‘matrices’, and so on.³⁵ Such formalisms suggest generality, since each representation can ‘stand for’ numerous possible ‘realizations’ in language material. However, the further a representational mode is indeed removed from language, the more fresh problems can arise in providing reliable, intersubjective methods for translating between actual data and formal representation, and the more quarrels can come up about results, thus imperilling generality. The problems have been sidestepped somewhat by allowing actual words to appear as well in symbol strings or trees, and by treating actual sentences as if they were the underlying structures themselves (13.30; 7.80). A truly exhaustive conversion of language examples into formal representation would soon become explosive and opaque (for illustrations, see the structuralist approach of Koch 1971, or the generative one of van Dijk, Ihwe, Petofi, & Rieser 1972). A more workable and pragmatic approach would be to select a fairly language-like representation for ranking and measuring data (such as a proposition structure), which different researchers can apply with reliable agreement (cf. 11.43).

13.52 A disquieting tradeoff seems to be at work here. ‘General linguistics’ naturally wants to construct a metalanguage that does not inherit the same degrees of complexity, variety, and indeterminacy characteristic of ordinary language, but instead meets the aspirations of science for simplicity, unity, and determinacy. Yet an unduly forceful attempt to squeeze the complexity, variety, and indeterminacy out of the theory and its metalanguage simply tends to relocate it all in the relation between the theory and its domain of data (Fig. 13.4).

-- INSERT FIGURE 13.4 HERE --

The more we strive for rigour in theory, the harder it may become to decide exactly how the theory relates to natural language data. This would strongly apply to a fully formalized theory with a completely new and precise metalanguage, as attempted most radically by Hjelmslev, who also provides the least data for illustration. Chomsky hoped that formality could be combined with simplicity, and his ‘grammar’ was claimed to meet both standards; but his theory grew steadily more complex anyway, despite his tactic of moving complexities out of the grammar into the lexicon (cf. 7.36f, 50, 70, 73). A ‘realistic’ approach, in contrast, acknowledges the great variety of possible discourse events, which even controlled experiments cannot eliminate, but also the effectiveness of discourse strategies (not strict `rules’ or ‘algorithms’) for managing complexity and imposing determinacy on many kinds of ‘input’ and ‘information’ (11.6, 14).

13.53 A promising solution, and one finally gaining ground in science (cf. Bohm 1980; Prigogine & Stengers 1984; Bohm & Peat 1987; Gleik 1987), is to seek order in apparently accidental or uncontrollably complex domains by revising our conventional conceptions of ‘science’ and its clean-cut divisions into areas and disciplines with well-fenced problems. In linguistics, this strategy would mean that new models with an interdisciplinary basis can assist the treatment of more data which are ‘realistic’ in quality, large in scale, sensitive to context and goals, and controlled also by non-language aspects. Among the ‘strictly linguistic’ approaches to date, Halliday's has pursued this direction the furthest, and perhaps as far as could be done while still maintaining an applicability to the whole of English, and maybe to other languages as well (9.25). Wider contexts have been addressed heuristically from both the outside of the clause or sentence and the outside of linguistics proper by van Dijk and Kintsch, using knowledge structures (‘frames’, ‘schemas’, etc.) as controls on text processing (13.14, 24, 34). No claims were made here that the synthesis and analysis of language sequences entails an exhaustive segmental or classificatory treatment on each separate level; instead, people are presumed to use a mixture of cues and clues from all levels and to be content with a fuzzy, but reasonably adequate result (cf. 2.33; 5.34f, 53; 7.46; 8²³; 9.34; 11.15, 35; 12.82; 13.27, 57). It has indeed been argued that where appropriate, people circumvent syntactic analysis altogether (cf. 9.39; 11.4, 34; 13.38), a prospect linguists might not relish.

13.54 Still, the leit-motiv in linguistics until Halliday has been the accentuation of form over content and function. Categories, distinctions, and meanings have usually gained recognition only if they had some formal correlate.³⁶ For the structuralists, the forms had to be discoverable in the data; for the generativists, discoverable form was less crucial than underlying form (13.28ff). No doubt the notion of underlying form was attractive precisely because the manifest substrate isn't ‘formal’ enough to enable structural descriptions at the desired level of rigour and abstraction, and because one-to-one ratios between form and content or form and function are not predominant (cf. 3.16, 32; 4.17, 26, 50; 5.48, 64; 6.27; 9.39; 12.14). Also, underlying form provided an ideal level for supposing that manifest forms, like words, phrases, or whole clauses and sentences, are mutually derivable from or convertible into each other within the same system, rather than say, over long historical developments (cf. 3.26, 34, 39; 4.65; 5.54, 56; 7.52; 8.56; 9.75, 81f, 101). Thus, the ‘sentence’ could be treated as the essential unit, despite its uncertain empirical status, by supposing that other units can be made into or out of sentences (cf. 5.30, 56; 7.51ff; 9.5, 34, 82; 11.33f, 85f). The prospect of the sentence not being the basic unit was raised by Saussure, Firth, Halliday, and van Dijk and Kintsch (2.61; 8.55; 9.82; 11.2f, 16), but they all (except Saussure) worked with it anyway.

13.55 And yet this very striving for abstraction and formality entailed the tradeoff I described between a gain and a loss of control. The same contextual factors carefully reasoned out of the theory can be the actual means whereby people control their own operations when they use language (cf. 2.85; 5.57; 11.3, 92). An artificial vacuum was created when whatever did not seem properly abstract and formal was deemed ‘non-linguistic’ (13.22). What was left seemed extremely rarefied, and an increasingly large part of the discovery, preparation, and analysis of the data had to be done behind the scenes (13.36). The famous concept of ‘transformation’, originally introduced to gain control by compacting many structures into a few, also had the opposite effect by proliferating undesirable structures (cf. 7.50ff; Woods 1970).

13.56 For the same reason, the issue of how to choose between alternative ‘grammars’, now rated as ‘theories’ (7.8f, 37-40, 92), became more convoluted as the linkage to data became more mediated and abstract. The structuralists had typically made their ‘grammars’ out in the field and saw no special motive to justify them on any other grounds than the heuristic derivation (13.38). Chomsky correctly foresaw that his new approach could lead to a substantial number of competing ‘grammars’ for one and the same language, which would not have been such an imminent scenario for practising fieldworkers. He deployed this new plurality to develop complex arguments about criteria for choosing a grammar without regard to how it might have been constructed, and asserted that ‘evaluation’ and ‘decision’ should take precedence over ‘discovery’ (cf. 7.8). The structuralists were totally unprepared for such debates; they could only point to the hundreds of grammars they had in fact constructed, an achievement Chomsky refused to recognize (5.2, 89; 7.5, 87).

13.57 I see no principled abstract solution to these long-standing controversies over how to balance theory and data, knowledge and use, potential and actual, and so on. Our problems may after all have been aggravated by premature aspirations to find and enforce such solutions. A concrete empirical analysis and testing of ordinary language communication seems to me the only recourse for a multitude of issues which cannot be resolved by introspection or intellectual judgment; the greater the body of findings, the easier it will be to rate competing theories (or ‘grammars’) by criteria such as ‘psychological’ and ‘social reality’.³⁷ For example, it seems intuitively plausible, and congenial for linguistics as well, that when an utterance is produced and then comprehended, the same operations run first in one direction and then in the reverse: the producer goes from deeper (or ‘higher’) levels to the ‘surface structure’ (or ‘lower’ levels), and the comprehender goes back again, both working thoroughly and independently through each level in terms of its own proper constituents (cf. 7.83; 12.47). But empirical research proves that this appealingly reversible scheme does not fit the actual operations people perform: the two process groups run partly in parallel, and various levels are consulted throughout (cf. 5.32, 34f; 11.15, 81; 13.53). Evidently, ‘the strategy types of the largest scope’ are the ‘most fundamental to understanding’ ‘language’ (11.32; cf. 5.19). Leaving them aside exaggerates the impression that the relation between signifier and signified is ‘arbitrary’ (cf. 2.85; 13.27).

13.58 I would have to agree with Halliday that ‘a theory being a means of action’, we must consider what ‘action’ we ‘want to take’ ‘involving’ ‘language’ so we will know what is ‘relevant’ and ‘interesting’ for ‘the investigation or the task at hand’ (9.1; cf. 1.11; 11.6; 12.4; 13.31). Until we determine our goals, we do not have adequate criteria or controls either for designing a theory or for selecting a set of manifestations as evidence of a general system, pattern, or consensus within the language. Though our theorists may argue for a disinterested ‘objective’ viewpoint, the ‘subjective’ aspects cannot be fully eliminated but at best negotiated toward an intersubjective viewpoint -- just what language itself is so well-suited for doing (12.12; 13.24).³⁸ If ‘other sciences work with objects that are given in advance’ while in ‘linguistics’ ‘the viewpoint’ ‘creates the object’ (2.9), then we need to probe the commonalities and divergences of our viewpoints.

13.59 Inevitably, an intersubjective grounding cannot be fully stable. The approaches built close to large amounts of data, such as Sapir's, Pike's, Halliday's, and van Dijk and Kintsch's, can expect to encounter substantial ‘indeterminacy’ and ‘fuzziness’ in categories, concepts, and boundaries.³⁹ But apart from Pike's notion of ‘waves’ (5.31, 87), few provisions were made for representing this factor in an explicit part of the theory, and ‘probabilistic’ models and ‘statistical’ methods were often rejected with questionable arguments (cf. 4.27, 77; 6.62; 7.90f, 7³⁰; 7⁴²; 8.31; 10⁹; 12.35). Many linguistic theorists seemed to believe it devolved upon them to postulate and reconstruct some definitive determinate order or taxonomy. This aspiration may pass in phonology, where ‘opposition’ is the major factor, but makes superhuman demands in semantics, where ‘determination’ must be performed from case to case (12.54), where no set of ultimate ‘minimal units’ can be found, and where the idea of meaning as a sum of parts is ultimately unworkable.⁴⁰ The abstraction away from ordinary contextual controls magnifies the task still more, as does the demand to write a rule system explicitly excluding all disallowed utterances or sentences (cf. 7.41; 13.40). Many idiosyncrasies within usage are probably specific to a small area or even a single locution; but the proposals to pack them all into the ‘dictionary’ or ‘lexicon’ is not helpful as long as the organization of such a lexicon has not been explained in detail (cf. 2.29, 2²¹; 4.46, 48f, 52, 59; 6.48; 7.70f; 11.3). And even the most comprehensive lexicon could not be expected to store all the information and instructions for all possible uses of the entries. Thus, both lexicon and grammar must be designed to allow for fuzziness, indeterminacy, and probability (13.7, 54).

13.60 The ambition for a ‘general linguistics’ may encourage the notion that a prior statement of goals is not necessary, perhaps even restrictive. In consequence, when Hjelmslev calls for a linguistic theory to cover not merely all existing languages but all possible ones, the only goal he raises is to establish ‘linguistics’ as the ‘centre’ of ‘all science’ (6.10). Similarly, when Chomsky says ‘the main task of linguistic theory must be to develop an account of linguistic universals’, he offers no goal except ‘explanatory adequacy’ relating ‘an explanation of the intuition of the native speaker’ to ‘an empirical hypothesis about the innate predisposition of the child’, and thereby conveniently excluding demands for application by showing that ‘one cannot really teach language’ (7.19, 23, 32). Perhaps the ambitions to make linguistics the very model of theory (13.21) filled the place of more tangible goals, especially for the more formalized theories, which are inherently hard to apply to social or educational needs (cf. 9.2).

13.61 For van Dijk and Kintsch, ‘the main criterion for the success’ of a ‘general theory’ lies in its power to ‘derive fruitful situation- and task-specific models’ and ‘experimental tests’ of ‘principles and implications’ (11.101). This is still a self-directed goal: using theory to keep the usual methods moving along. However, they go on to suggest that ‘our knowledge’ about ‘comprehension processes’, derived from such ‘experiments’, enables better ‘control’ ‘over subject-text interaction’ (11.102). Wide prospects for applying empirically anchored theories to communication are indicated here, such as methods to make reading and learning more efficient, or to teach native or foreign languages, including ‘restricted’ ones (cf. 4.85-88; 5.89; 8.9f, 12, 14, 65, 89; 9.111; 11.94f). The more direct and detailed our applications, the broader our empirical base has to be.

13.62 Whatever their declared motives, our theorists certainly had expansive moods and evoked vast panoramas. For Saussure and Sapir, ‘linguistic questions interest all who work with texts’ (2.87), including the ‘outside public (3.2). For Bloomfield, ‘the study of language may help us toward the understanding and control of human events’ (4.88). Pike hoped for ‘a theory, a set of terms, and an analytical procedure’ to make ‘intelligible’ not merely ‘language behaviour and overt physical activity’ but `all human overt and covert activity’, ‘all psychological processes’, all ‘responses to sensations, all of thinking and feeling’ (5.89). Hjelmslev saw ‘linguistic theory’ ‘reaching its prescribed goal’ by ‘recognizing’ ‘man and human society behind language, and all man's sphere of knowledge through language’ (6.64). For Firth, ‘general linguistic theory’ should undertake ‘a serial contextualization of our facts, context within context, each one a function’ ‘of a bigger context, and all contexts finding a place’ in ‘the context of culture’ (8.91).⁴¹

13.63 Whether linguistics can attain such scope and significance remains to be seen. We must concede, without much surprise, that the outcome to date falls far short of these panoramas. As is typical for would-be founders, our theorists themselves often stress the incomplete or provisional character of their models.⁴² And in principle, the ‘image’ of ‘language’ ‘in linguistics’ remains ‘incomplete’ because it ‘must allow the right set or combination of features’ to ‘emerge from the progress of research and insight’ ‘by means of successive correction through new findings’ (12.39).

13.64 Still, linguistics has already been influential in language education and language policy-making, as well as psychology and social science. And computation and cognitive research show linguistic contributions being put to work in broad new ways, notably under the interdisciplinary aegis of ‘cognitive science’ (11.1, 5, 102). It might therefore not be unduly sanguine to expect further uses and influences in the future. Conditions will be most favourable if we can keep the richness of our past firmly in mind and view our total achievements within a concerted assembly, not within a set of rotating and disputatious fragments. With so much still to discover, we must strive to remain aware of all our options. If this book can aid such an awareness, I would be most content.

NOTES

¹ In this section I use the abbreviations listed at the front of the book or, where relevant, cross-references to particular paragraphs in the preceding chapters. Bulky lists of cross-references are given in footnotes. Individual terms can also be traced through the Index.

² A revealing contrast to surveys like mine can be found by seeing how recent Kuhnian textbooks invest steadily greater effort and hyperbole in making the development of linguistics fit one ‘normal’ paradigm. Newmeyer's (1980) Linguistic Theory in America makes only the merest mention of Sapir, Bloomfield, and Pike, and directly equates both ‘theory’ and ‘science’ with Chomsky's standard model, touted on the jacket as ‘the world's principal linguistic theory’. For Newmeyer, ‘no viable alternative exists’, and ‘the vast majority’ of ‘linguists’ ‘who take theory seriously acknowledge (explicitly or implicitly) their adoption of Chomsky's view of language’ (1980:249f). At the conclusion, we are asked to believe that ‘on the basis of [Chomsky's] idealization, more has been learned about the nature of language in the last 25 years than in the previous 2500’ -- calling to mind Firth's words: ‘to dismiss two thousand years of linguistic study in Asia and Europe’ ‘is just plain stupid’ (P1 139). But read carefully, Newmeyer's book reveals, against the author’s will, how heavily his acclaimed ‘revolution’ relied not on insights into language but on charisma and polemics (cf. 7.2).

³ Compare 2.3ff, 10, 36, 39; 3.2; 4.3; 5.2f, 5, 37, 62f; 6.3, 26, 49; 7.2-5, 7, 37ff, 62; 8.2, 4, 6ff, 10, 17, 19f, 25, 46, 87; 9.2-6; 11.2f, 34, 40.

⁴ The more usual disaffection with literary language and its study (2.24, 3.4; 4.41; 6.4) had its own political motives: the largest body of language study had been expended on literature and, for a time at least, many linguists had to occupy uneasy positions in departments of literature (cf. 5.56).

⁵ What might be ‘transformations’ or ‘transforms’ in a formal grammar appear in Halliday's grammar as ‘variants’, ‘close parallels’, ‘different analyses’, ‘interpretations’, ‘expansions’, and so on (cf. IF 61, 93, 225, 223, 165, 175). Whereas Chomsky's grammar sticks so close to the wording that it cannot handle the ‘relation’ between ‘“I liked the play”‘ vs ‘“The play pleased me”‘ (AT 162), Halliday's relations are flexible enough to link ‘pairs of clauses’ that are ‘representations of same state of affairs’ such as ‘“Mary liked the gift”‘ vs ‘“the gift pleased Mary”‘, which are not merely ‘active’ vs ‘passive’ (IF 107) (9.62). Still, Halliday concedes that ‘we can hardly explain’ a ‘clause by saying that it is doing duty as a replacement for’ another (cf. 13.54).

⁶ Especially in the eyes of other disciplines. I have heard this view expressed in conversation both by psychologists like Walter Kintsch and by computer scientists like Robert F. Simmons.

⁷ On English as a model, see 3.28, 53, 3¹⁴, ²⁰; 4.5, 27, 56ff, 68, 70, 4⁹, ²²; 5.54, 73; 7.5, 18, 41, 61, 66, 79, 81, 7³⁹; 8.12, 15; 9.24-27, 42, 52f, 58f, 61, 63f, 66ff, 82f, 87f, 91, 107, 9¹³, 9¹⁶, 9²⁰, 9³⁰, 9³³, 9⁴¹; 11.64, 67. On Latin, see 2.5; 3.27, 50, 53; 5.24; 6.5; 8.5, 8; 9.25; 12.20f.

⁸ And Sapir blamed the ‘social sciences’ for ‘the evolutionary prejudice’ that has been ‘the most powerful deterrent of all to clear thinking’ (3.2).

⁹ Pike was more even-handed in considering psychology (5.23, 27, 42, 89), probably because by that time ‘psycholinguistics’ had attained some success (cf. 5.10, 31; 13.19).

¹⁰ Van Dijk's recent work has now taken on a primarily sociological orientation, addressing such issues as ‘prejudice’ and ‘racism’ (e.g. van Dijk 1983, 1984, 1987).

¹¹ Compare 4.7, 5.86, 5¹; 6.56, 64; 8⁷, 8¹⁴; 11.40). For political motives, Chomsky denounced positivism (7.10) despite relying on logic more than any other linguist.

¹² Compare 3.23, 37; 4.4f, 4²; 5.73, 86, 89, 5²⁸; 8.5, 17, 31, 55; 9.3, 5, 59; 11.14, 36, 41, 10²⁷; 12.36f.

¹³ However, Pike acknowledged formal ‘approaches to language and logic’ and the ‘contribution of transformational grammar’ (LB 496f). He even stated a ‘feeling that tagmemics and transformational grammar should ultimately merge in the mainstream of linguistics’. He argued for instance that ‘transformation tests underlie our early methodology’, and deployed ‘transforms’ for describing structures (LB 223n, 67, 441, 459). Yet he was ‘unable to match’ Chomsky's (1961a:14) ‘disclaimer’ of having no ‘model of speaker or hearer’ ‘against the current work of some transformationalists’, e.g. when Lees (1959:1) says ‘speakers, both in producing and understanding sentences, make use of the “generation”‘ ‘by the grammar’ (LB 281, 495) (cf. 7.83). Pike also ‘believes’ ‘the dream of mechanical generation of all and only the well-formed linguistic units is hopeless’, partly because there are ‘linguistic units’ ‘beyond the sentence’, and partly because some texts, such as ‘poems’, ‘deliberately exploit departures from well-formedness’ (LB 494) (13.41). He agrees with Chomsky that ‘mechanical discovery procedures’ are not ‘possible’, but ‘vigorously rejects’ the dismissal of ‘“practical discovery procedures”‘ on those grounds (LB 225n, 492f; cf. 7.7ff; SS 56).

¹⁴ For a wide range of views about what is or is not ‘universal’, see 2.10; 3.67; 4.4, 72; 5.44, 84, 5²; 6.5, 10, 34, 44; 7.1-4, 19f, 27, 31, 77; 8.19, 27, 60, 86; 9.3, 18, 25, 47, 60; 11.63; 12.94. Many of these passages suggest that the ‘universal’ aspect is not ‘in’ language, and certainly not in its grammar; presumably, a language can have a grammar at all only as a self-contained system (cf. 2.10; 8.60; 9.19). What seemed ‘universal’ to Chomsky was merely due to the worldwide uniformity of the articulatory apparatus and his own familiar grammatical notions.

¹⁵ I can make nothing of Firth's claim that it should be ‘easier’ for ‘linguists’ to ‘acquire sufficient psychology and sociology’ than for ‘a psychologist or sociologist to acquire the necessary linguistic technique’ (8.16). The remark is unusually glib, even for Firth, since the other disciplines had far more substance at that time than linguistics did, at least in Britain.

¹⁶ Compare 2.32; 3.1, 10; 4.10; 5.7f, 69, 84; 6.9, 54; 7.10, 30f, 35; 8.10, 14, 28; 9.1f, 7, 14, 27, 38, 111f; 11.1, 5, 43; 12.6, 9. Compare also Bloomfield's sibylline remark that the ‘features’ ‘appearing in every language’ may ‘exist’ as ‘realities either of physics or of human psychology’ (4.71).

¹⁷ See 2.7, 9, 19; 3.12, 15; 5.8, 68; 6.8, 10, 29, 32f, 49; 9.23, 39, 59, 72; 11.22; 12.14; 13.55.

¹⁸ Compare 2.15, 65; 3.23, 46; 4.24, 27, 55, 70; 5.68, 5¹⁰, 5¹³, 5³⁴; 6.12, 29f, 54; 7.69, 71; 8.43; 9.14f, 27, 44, 60, 112; 11.8, 10, 83; 12.13, 18, 60-64, 76, 89.

¹⁹ See 2.28f; 5.7f; 6.12, 15, 31; 7.57, 69, 71; 9.60f, 68, 77; 11.10, 17, 20, 23f, 45, 51, 53.

²⁰ On ‘constituents’ and ‘minimal distinctive units’ see 2.17; 3.33; 4.33, 45, 48, 52, 59-62, 64f; 5.21, 28, 34, 51f, 58, 62, 5⁵; 6.23f, 39, 42; 7.6, 30, 36f, 40, 45, 49f, 68ff, 81; 8.31, 36; 9.33f, 75, 9¹⁵, 9¹⁶; 11.32, 10¹⁸.

²¹ The oddest offshoot of this derivation was the notion, quite fashionable for a time, that ‘thought’ should be viewed as silent articulation and studied through the speech muscles (3.10, 3⁷; 4.9; 5.39; 8.22, 8¹⁷). Even if the two operations reliably concur, which they do not (8¹⁷), we could only detect when a person is thinking, but not what or why.

²² On this kind of ‘arbitrariness’, see 2.59, 85; 3.13; 4.27, 49, 82; 5.17, 22, 30, 51, 60, 87; 6.15, 18, 24, 29, 31f, 60; 9.32, 35f; 11.41, 43; 12.64, 76; 13.57. An intriguing case is the intuitive disposition to see the central category of a grammar in the noun like Sapir and Halliday seemed to do (3.36; 9.81, 9⁴⁰), or in the verb like Firth seemed to do (cf. 8.61) Of course, the organization of a given language may encourage s