XXTH-CENTURY THEORIES OF LANGUAGE : AN EPISTEMOLOGICAL DIAGNOSIS

This article is intended as a study in the methodology and epistemology of linguistics, a field which developed out of theoretical linguistics in the past thirty years. Methodology and epistemology (or philosophy)1 of linguistics can be subsumed under the general domain of 11 philosophical linguistics11 (cfr. Kasher Lappin 1977), which also includes a theory of meaning and reference, a theory of linguistic ( or, more generally semiotic) communication, and in some cases a formalization of linguistic subsystems. The specific contribution of methodology and epistemology of linguistics lies in the definition of the object of linguistics, in the determination and justification of its research techniques, in the appreciation of its results with respect to a broader field of investigation, in the reflection on the nature, status, and variability of approaches to language. 2

The history of general linguistics (which is still an ill-defined concept) since the beginning of this century shows that the basic notions -such as 11 language 11 , 11 grammar 11   ,   11 (linguistic) meaning 11   ,   11 (linguistic) structure 11 -underwent a radical change in intension and extension, and have been focused upon from different points of view, in variegated perspectives (cfr.Swiggers 1989).It would be presumptuous to analyse the various 11 transformations of linguistics 113 in this article; our aim here isto offer some ge-I am taking epistemology here in its "correlative" acceptation (viz."epistemology of -"; cfr. the use of epistemologie in French), and not in its "absolute" acceptation (epistemology as theory of knowledge), which has been the most common one in Anglo-Saxon philosophy.The advantage of taking "epistemology" in its correlative meaning is that it allows us not only to bridge the gap between English and French usage (see, e.g., Dominicy ed. 1991), but also to account for the fact that "epistemology of linguistics" is also about the theory of Iinguistic knowledge.The epistemologist's task is to define the kind of knowledge aimed at (and, ideally, attained) by (the application of) a particular theory within a particular domain, and to compare theories which are intended to cover an almost identical extensional field.
neral reflections on a few aspects of these transformations, serving as a prelirninary to a more comprehensive "historical epistemology" 4 of modem linguistics.
l. Sign-centered theories of linguistics.
The courses in general linguistics which Saussure taught at the University of Geneva between 1906 and 1911, and which provide the substance of the posthumously published Cours de linguistique generale edited by Charles Bally and Albert Sechehaye, can be seen as the unfolding project -which can be traced back to the early 1890s -of defining the field of general linguistics, and of formulating the principes generaux of the science of language.Saussure's endeavour cannot be isolated from its intellectual context, in which the discussion on the nature and classification of sciences played a crucial role.In 1898 Edmond Goblot, in his Essai sur la classification des sciences, remarked that the "moral sciences" had not yet fixed their object, nor their principles and methods.These sciences, aspiring to become "positive", show a tendency to depart from their empirical grounds, and to idealize their objects: in this they manifest what Goblot calls "un dualisme logique".Adrien Naville, in his Nouvelle classification des sciences (1901), divides the sciences into three kinds: theorematical, historical, and canonical.The first kind is concerned with possibilities (constrained by laws), the second with realities, the third with the notion of "good" (or "useful").Within the first kind Naville includes mathematics, physics (subsurning biology, mechanics, chernistry), and psychology (and sociology).The logically prior science within the theorematical sciences is nomology, the object of which is the notion of law 5 • Saussure's linguistic itinerary can be seen as a constant search for the elaboration of a nomology of linguistics, structured in an axiomatic way.linguistics involves a distinction between three distinct phases (a philosophically oriented approach of language categories; a historically oriented approach of language forms; a syntactic analysis of the "intrinsic reality" of language).At the same tirne Benveniste (1966: 16-17) outlined various possibilities of further development.For a stimulating overview of "changing perspectives" in modem linguistics, see Chomsky (1986a,b).For a penetrating analysis of phonological theories, see Anderson (1985).For globa! epistemological analyses, see Dougherty (1973) and Itkonen (1978Itkonen ( , 1984)).
Although Saussure arrived only very late (in May 1911) at the distinction between signifie and signifiant, he soon saw that the specific nature of linguistics -or of the fait linguistique -lies in the interplay between sound and sense, an interplay which is concretized, in a creative (and oppositive) way, in the linguistic sign: "On a discute pour savoir si la linguistique appartenait a !' ordre des sciences naturelles ou des sciences historiques.Elle n'appartient a aucun <les deux, mais a un compartiment <les sciences <qui, s' il n' existe pas, devrait exister sous le> nom de semiologie, c' est-a-dire science des signes ou etude de ce qui se produit lorsque l'homme essaie de signifier sa pensee au moyen d'une convention necessaire.<Parmi tous les systemes semiologiques> le systeme semiologique "langue" est le seul (avec l'ecriture) qui ait eu a affronter cette epreuve de se trouver en presence du Temps, qui ne se soit pas simplement fonde de voisin a voisin par mutuel consentement, mais aussi de pere en fils par imperative tradition et au hasard de ce qui arriverait en cette tradition, chose hors de cela inexperimentee <non connue ni decrite>.Sil' on veut, la linguistique est done une science psychologique en tant que semiologique, mais les psychologues n' ont jamais fait intervenir le 1EMPS dans leur semiologie.Ce fait qui est le premier qui puisse exciter l'interet du philosophe reste ignore des philosophes: aucun d'eux n'enseigne ce qui se passe dans la transmission d'une semiologie.Et ce <meme> fait accapare en revanche tellement l'attention des linguistes que ceux-ci en sont a croire <pour cela> que leur science est historique ou eminemment historique, n'etant rien d'autre que semiologique: par la completement comprise d'avance dans la psychologie, a condition que celle-ci voie de son cote qu'elle a dans la langue un objet s'etendant a travers le temps, et la fon;:ant de sortir absolument de ses speculations sur le signe momentane et l'idee momentanee" (Saussure [Engler ed.] 1968[Engler ed.] -1974: vol: vol.II, fragment 3342.1).
Saussure's choice of the linguistic sign as the axiomatically central concept of linguistics not only entailed the "semiological" status of linguistics, but it also implied the dichotomies elaborated upon in the Cours: the langue/parole distinction within the "facultas signatrix" (corresponding to the systematic insertion of signs within the social context of language, and to its expressive realization in individual speech); the synchrony/diachrony distinction (corresponding to the opposition between the status of elements used as termes d'un systeme, and the chronological, non-systemic, link between (parts of) systems); the distinction between the associative and syntagmatic axes (corresponding to static versus dynamic differentiality of signs) 6 .
Saussure's theory of language 7 is based on an epistemology which is neither instrumentalist nor realist: language is defined in terms of a phenomenological conceptualism, stressing the creative articulation of linguistic entities, fulfilling a social function.These entities are not sentences; the reason why Saussure discards sentences as a unit of analysis is not so much that they belong to parole, but that there is no property in 6 In fact, one can say that the associative axis concerns the caractere differentiel of the sign as a bifacial unity (signifiantlsignifie), whereas the syntagmatic axis concerns the oppositive nature of signs taken on their own.
common to them 8 • The linguistic sign is a unit of both langue and parole, and it invariably presents a signifiant/signifie structure 9 .Saussure's choice of the linguistic sign as the central notion of his language theory is paralleled in Vendryes's (1921) distinction betweenform and meaning, and in Sapir's (1921) distinction betweenform and content (cfr.Swiggers 1991a).Both authors, however, present us with an extension of Saussure's theory of language.Vendryes's distinction between form and meaning is not only relevant to the nature of the linguistic sign, but it also pervades the entire linguistic system, viewed as a complex distribution of relationships between form-units and meaning-units.A sentence results of the synthesis of elements which are the (analytical) expression of content-representations and of the relations between these.Sapir's contribution lies in the typological elaboration of the relationship between form and ( conceptual) content.Language, the symbolic expression of experience, manifests a relationship between matter and form: only the latter is relevant to the linguist, and more particularly insofar as linguistic form has no direct link with a particular function.The asymmetry between form and function explains why Sapir distinguishes between grammatical processes and grammatical concepts: the former correspond to a view "from form to function", the latter to a view "from function to form" (or even from function to function).Typology therefore has a double role in Sapir (1921): it shows the multiple relationship between form and function within one language, and it manifests the variance of form/function-coupling among languages.In Sapir's typological table (1921: 150-151) this corresponds to the basic distinction between the "capital" (conceptual) types oflanguages (Simple Pure-Relational, Complex Pure-Relational, Simple Mixed-Relational, Complex Mixed-Relational) and the process-subtypes (indicated by smaller letters: a, b, c, d) within one language ( cfr.Swiggers 1993).Vendryes (1921) and Sapir (1921) were not so much preoccupied with defining the field of general linguistics, nor with determining its axiomatically central units.In fact, 8 See Saussure (1916: 152): "Une theorie assez repandue pretend que les seules unites concretes sont les phrases: nous ne parlons que par les phrases, et apres coup nous en extrayons les mots.Mais d'abord jusqu'a quel point la phrase appartient-elle a la Iangue?Si elle releve de la parole, elle ne saurait passer pour l'unite linguistique.Admettons cependant que cette difficulte soit ecartee.Si nous nous representons l'ensemble des phrases susceptibles d'etre prononcees, Ieur caractere le plus frappant est de ne pas se ressembler du tout entre elles.Au premier abord on est tente d'assimiler l'immense diversite des phrases a la diversite non moins grande des individus qui composent une espece zoologique; mais c'est une illusion: chez les animaux d'une meme espece les caracteres communs sont bien plus importants que les differences qui les separent; entre les phrases, au contraire, c' est la diversite qui domine, et des qu' on cherche ce qui les relie toutes a travers cette diversite, on retrouve, sans l'avoir cherche, le mot avec ses caracteres grammaticaux, et I'on retombe dans les memes difficultes" (for the different versions in the students' notes, see Saussure [Engler ed.] 1968[Engler ed.] -1974: vol. 1, 240): vol. 1, 240).
their epistemological outlook was less "phenomenological" than Saussure's in that they identified some "real" properties of language structure as being the solid foundations for a theory of language.In addition to this, both authors differed from Saussure on the issue of the autonomy or non-autonomy of linguistics.

Level-oriented theories of linguistics.
A major breakthrough in the theory of language was achieved by the elaboration of a level-analysis of languages.The concept of level-analysis implied two major axioms: (a) units of description are defined with respect to a level of analysis (as such, two elements can be analysed as distinct on the phonemic level, but as non-distinct on the morphophonemic level); (b) the study of each level involves the application of a particular technique and notation.
Basically, these insights were on line with Saussure's idea that linguistic units are the results of analysis, and are not given, and with its implication that units are ultimately defined by the analytical determination of their oppositive nature 10 • But levelanalysis makes it possible to study language as a system of recursive combinations involving finite sets of elements: "The number of orders of phonemes in the morphemes and words of a language is a sub-multiple of the number of possible orders (".) The number of constructions in a language is a small sub-multiple of the number of forms ( ... ) Each position in a construction can be filled only by certain forms" (Bloomfield 1926: 157-158).
11 See Bloomfield (1933: 509): "The methods and results of linguistics, in spite of their modest scope, resemble those of natura!science, the domain in which science has been most successful.It is only a prospect, but not hopelessly remote, that the study of language may help us toward the understanding and control of humans events"; and Bloomfield (1939: 11): " ... widespread and deep-seated errors in supposedly scientific views of human behaviour rest upon ". failure to the tremendous selectivity of language to describe how language carries messages.The focus on selectivity led to discoveries of correlations between meanings and selections of forms.As Bloomfield pointed out, it frequently happens that when we do 10.ot content ourselves with the explanation "that something is due to meaning, we discover that it has a formular regularity or 'explanation'" (Harris 1970: 785).
Selectivity also is the basis of Bloomfield's theory of meaning (cfr.Bloomfield 1933: 139-157;Bloomfield 1939).Contrary to a widespread belief-based on unfarniliarity with Bloomfield's work -he wrote more about meaning then most linguists of his tirne.For Bloomfield, meaning is not just a function of a speech utterance; it is a relation between the speaker's stimulus, the utterance, and the hearer's recepti.onand response (Bloomfield 1939: 18)."The term 'meaning', which is used by all linguists, is necessarily inclusive, since it must embrace all aspects of semiosis that may be distinguished by a philosophical or logical analysis: relations, on various levels, of speech forms to other speech-forms, relations of speech-forms to non-verbal situations (objects, events, etc.), and relations, again on various levels, to the persons who are participating in the act of communication" (Bloomfield 1939: 18; see also Bloomfield 1933: 77).
Bloomfield's work -and that of some of his followers, e.g.Z.S. Harris and Ch.F. Hockett -has been crucially important in two ways: (a) it has shown the importance of refining techniques for linguistic description, and has thus led to a highly controllable account of how levels can be distinguished, and how units can be categorized within a homogeneous description of speech events; (b) it has shown that for a given data equivalent procedures, all of them uniform and consistent, can be set up.As a consequence, no ontological import should be attributed to the descriptive model: the structure of the description is not identical with the structure of the descriptum; at the most, there is a correlative homomorphy.
Bloomfield's theory of language is coupled with an instrumentalist epistemology of linguistics, and of science in general (cfr.Hi:ž: -Swiggers 1990).Bloomfield welcomed the physicalism and mechanism of Carnap and Neurath, their program for a unified language and method of science.A scientific description of language should be made in a language of science in general.Bloomfield was convinced of the unity of science, as a social and linguistic process, and this explains why he speaks of scientific practice, of the branches of science, and of the "great vocabulary of science" (Bloomfield 1939: 20, 49, 55).The discourse construed by the scientist is basically of two types.In the first type the correctness of a discourse is bound only by the verbal agreements on which it is based.This type of discourse, as far as it is limited to verbal activity, in fact does not constitute a science.This would be the case of.mathematics.distinguish between linguistic and nonlinguistic events, upon confusion of the biophysical and the biosocial aspects of language and its subsidiaries, and above ali, upon a habit of ignoring the linguistic parts of a sequence [ = stimulus, speech act, recepti on, response] and then calling upon metaphysical entities to bridge the gap".Bloomfield adopted Carnap's view that uninterpreted mathematics is without factual content; it is a result of agreements on the use of words.In mathematics certainty results from its nonfactuality.Applied mathematics is factual, but no more certain than the science to which it is applied."The invention and skilful manipulation of speechforms is nota science, buta skill, craft or art" (Bloomfield 1939: 56).The other type of scientific discourse involves, besides the verbal agreements, the biosocial agreements affecting the reception of the discourse by a community: Bloomfield refers to it as "publicity"."U nique personal or communal behavior figures in science as an object, which may be observed like any other; but it does not figure as a part of scientific procedure" (Bloomfield 1939: 47).The language of scientific procedure (as public activity) is a language stripped of the private feelings, attitudes, and connotations attached to linguistic forms by the participants: "each participant burns his own smoke" (Bloomfield 1939: 47).

Transformation-oriented theories of linguistics.
The instrumentalist view of language has been refined in work on phonology and morphology by Harris, Hockett, Nida, Swadesh, Voegelin; it has focused on problems of segmentation and classification, which -no linguist would disagree on this -remain essential to any type of linguistic work.The upper level of this descriptive work has mostly been that of morphemes; attempts at extending the taxonomy to syntagms can be found in Wells (1947), Harris (1951: chapter 18), Hockett (1958: chapters 17-31), and Pike (1967: chapters 11-14), but one cannot say that parallel to the procedures for establishing paradigms of morphemes there were procedures for establishing paradigms of phrasal relationships.Zellig Harris's work in the 1950s12 and Chomsky's transformational grammar aimed at offering this extension of the structuralist methodology to syntactic relationships; the implications of this was that syntactic analysis was autonomous (a view held by Chomsky up to the 1970s), and could be kept separate from semantics, as well as from prosodical and intonational phenomena.The autonomous conception of syntax has been severely criticized for its neglect of the semantic component; personally I regret as much 13 the discrirnination of the suprasegmental level, a fact which stili has to be deplored in the most recent versions of generative grammar.
The autonomy of syntax -abandoned in the Government and Binding Model 14was a strong claim, which had a positive and a negative consequence.The positive one was that it could be shown that relationships extend beyond linear sequences and substitutional slots; the negative one was that this gave way to highly tentative conclusions, which in my opinion are still in need of proof.The eagerness to formulate such conclusions should be seen in the light of Chomsky's interest in metatheoretical problems 15 , starting with discussions concerning the adequacy of grammars and of taxonomic techniques, and leading to claims concerning the nature of linguistic knowledge, the status of grammar, and the place of linguistics.Chomsky's passionate and patient search for a theory of language has not yet received a detailed epistemological analysis 16 , and one should bear in mind the fact that Chomsky feels that linguistics still has to undergo "something like a Copernican or Galilean revolution".The issues of the social impact of Chomsky's work, of its novelty, of its position with respect to older and contemporary theories, of its strategies (for a case study, see Swiggers 1995) should also be addressed in such an investigation, which cannot exclude the social history and the intellectual context of generative grammar.But what should be addressed in the first place are the following questions, which are essential to Chomsky's practice of linguistics, and to his conception of grammar and language; each of these offers ample matter for debate, because it seems to me that Chomsky has not yet offered a cogent argumentation on these issues.
( 1) Chomsky claims that linguistic competence, studied in depth, can be studied as a conceptually innate structure.While this position justifies the use of the introspective method, viz. the appeal to the ideal speaker's intuitions, it is not clear (a) what is precisely innate (what type of structure, or what type of principles), (b) why in such a view so much stress should be laid on the "empirical" nature of the investigation, (c) why grammatical rules are so diverse, and at times so undetermined (c is both a problem for universal grammar, and for the categorial content of any "private" grammar) 17 , (d) how an operational language concept can be extracted from the concept of I-Grammar.
transformational predecessors, but it is, on the other hand, a conceptually heterogeneous frame, in which for instance the theory of thematic roles is an oddity.As a general problem I see the opaque relationship between what Chomsky calls Logical Form and S-Structure.
(2) Chomsky has repeatedly insisted (recently, in Chomsky 1987) on the necessity of construing a grammar which has psychological reality.But the content of the concept of "psychological reality" has remained vague (for harsh criticism, see Derwing 1973); reference to the "modularity of mind" (see Podor 1983) may seem an easy wayout, but it is in fact a dead end: ( a) one can hardly deny that in a concrete discursive fragment (involving, e.g., cases of anaphorization, cataphorization, deletion of forms, switching of deictic elements, as well as phenomena such as presupposition, implication, ellipsis, etc.) the human mind produces and retrieves information synthesized in decodable structures: the question then is whether modularity is not an unrealistic ex- planans; (b) a modular grammar (involving X-bar theory, theta theory, binding theory, case theory, control theory, bounding theory) may be easier to "control" in terms of descriptive efficiency, but then we would like to know how the (language faculty in the) mind controls these modules, and what kind of creative control there can be.
(3) This brings us to a third problem.Chomsky has, since the early sixties, stressed the creative nature of language.He has however, never shown in what way linguistic creativity is different from (oris more than) analogical construing.Chomsky (1986a: 11-12) claims that analogy is nota sufficient explanation for "Plato's problem", but his argument is extremely weak: ( a) he does not consider the full range of analogical constructions (not to speak of the sets out of which analogies could be construed); (b) the type of proportional analogy he seems to use, involves a wrongly construed proportion (see Itkonen 1991); ( c) he does not show in what way analogy would be a deficient explanation.Of course, an appeal to analogy as an explanatory principle would urge a revision of (a') the notion of ideal speaker (correlated with grammaticality judgments), (b') the notion of intemalized grammar (coupled with the issues of modularity and of innate universal principles), (c') the notion of "rule" (correlated with Universal Grammar, and with grammaticality judgments) 1 8.
(4) Chomsky has, consequently, maintained his claims about the universality of grammatical principles (for a critical view, see Hagege 1976).Integrating cross-linguistic research, the Govemment and Binding model has set upa "parametric(al) syntax"; while this may be an interesting prototypological device, it cannot conceal the following problems: (a) it makes little sense to put up explanatory parameters if so much taxonomical work has still to be done (cfr.Gross 1979) 19 ; (b) a parametrical option should 18 In addition Chomsky leaves unmentioned the fact that structuralist accounts of analogy involve a not uninteresting recognition of linguistic creativity; see, e.g., Bloomfield (1933: 275-276): "A grammatical pattern (sentence-type, construction, or substitution) is often called an analogy.A regular analogy permits a speaker to utter speech-forms which he has not heard; we say that he utters them on the analogy of similar forms which he has heard ( ... ) The regular analogies of a language are habits of substitution ( ... ) When a speaker utters a complex form, we are in most cases unable to teli whether he has heard it before or has created it on the analogy of other forms".On the notion of "rule", and its link with explanatory adequacy, Quine's reflections are stili worth meditating (Quine 1972).
19 For a similar criticism see my remarks (Swiggers 1984) on Chomsky's account of control-structures, which are explained by reference to the "nature" of the head verb (see Chomsky 1982: 75).
be related to a structural invariant (what would be the invariant of the options concerning a fixed or a free word order ?);(c) any typology which starts from taldng the "familiar" case as the "normal" (or "configurational" .")one, is condemned to step on its own tail.

Conclusion.
This paper has focused on some aspects of linguistic theory in its relation to general epistemology and epistemology of linguistics.The changes which characterize XXth-century linguistics have affected the relationship between language and grammar (the latter being more and more conceived asa theory of language structure or even asa theory of mind about internalized language; cfr.Chomsky 1986a), as well as the relationship between linguistic description and reality (also reality of knowledge).There has been, in retrospect, much more continuity in methodology than in epistemology: much of Chomsky's earlier work (roughly up to the 1970s) has been, from the point of view of methodology, in continuity with Bloomfieldian (and more particularly Harris's) linguistics.But Chomsky's objective realism (cf.Blackburn 1984: 27-29), and his insistence on the psychological reality not of units but of structures (and their relationships), have led to a totally different view of both grammar and variance20 (distinct from sociolinguistic variation).
Chomsky's enterprise has gained in technicality; logical coherence is, however, only a prerequisite, but not a guarantee of adequacy.Today its strongest opponent is a cognitive approach (cfr.Lakoff 1986;Langacker 1986Langacker , 1987Langacker , 1990;;Rudzka-Ostyn ed. 1988) which rejects the modular conception of mind, and which focuses on (prototypical) semantic relationships (Chomsky's approach has always been a basically "syntacticist" one)."Internalized realism" vs. "externalized realism" seems, today, the basic controversy in linguistic theorizing, as well as in the theory of knowledge (see Burge 1986aBurge ,b, 1988;;Davidson 1987Davidson , 1991)): this should, however, not make us forget that much more work needs to be done on the level of description.And whether one subsumes this work under an instrumentalist epistemology or not, one thing is clear: language and its structure -the object of grammar -cannot be isolated from communication, and more specifically of information (see Harris's foundation-laying work: 1988Harris's foundation-laying work: , 1991)).The latter notion is essentially a structural one: it corresponds to both organizational pattems of language(-types ), and the individual ( and diachronically as well as socially variable) exploitation of these patterns, in function of communicative needs.In the light of this one feels that linguistic theory will have to elaborate an adequate, and more comprehensive, epistemology.It also seems that this epistemology should take a new look at analogy, as a linguistic and a cognitive process.
remarks on Chomsky's metatheoretical expansion (or, better, deepening) of Harrisian methodology: "In short, what Chomsky has done is to retain the scope of Iinguistic theory established by the Bloomfieldians and particularly in his case by Harrisforma!linguistic structure -and, while deepening the concept of structure itself, invest theory of such scope with utmost significance.Put in other, and quite appropriate terms, the relation between methodological form and substantive content in Chomsky's work is consistently interpretive ( ... ) Chomsky did not so much add theory to methodology, as find theory in methodology".