What is a Language?

This essay is from my book Language in Mind and Language in Society (1987). In 1993 Noam Comsky characterised this work as the only attempt known to him to try to explain such notions as "community language" and "abstract language", and he has gone on to cite the book in later publications. The essay identifies five accounts of what a language is: a natural kind, a name (Nominalism), an abstract object (Platonism), a social fact (Sociologism); both a natural kind and a social fact (Dualism). The work of Chomsky, Bickerton, Katz, Labov, Saussure, Itkonen - among others - is discussed at length. The essay tries to establish a framework in which both the psychology of language and the sociology of language can be pursued, and the inter-relations between psychological and socal realities properly understood. It does this by trying to clarify the ontology of different levels of linguistic reality and the methods by which they may be appropriately studied.

Part 1. Introduction

Lay readers of any scientific literature must expect to find terms with which they are pre theoretically familiar, refined and redefined within that science. In itself, this should occasion no particular problem. Problems arise when a science both trades on pre theoretical uses of a term and is committed to arguing the theoretical inadequacy of the lay vocabulary. Further and different problems arise, of course, when there is no intra scientific consensus on the sense or reference to be given within the science to a term taken over from lay terminology. As a lay reader of linguistic literature, both types of problem seem to me rampant within linguistics, where the idea of `a language' is freeIy used in both pre theoretical and various theoretical senses.

Now I agree with Popper (1979, especially ch. 8) that belief in the power of verbal definitions is the death of both philosophy and science, and the successes of modern linguistics are an existence proof of the claim that science can and must go on although key terms and concepts exist within it in a state of undefined animation. This essay seeks to order some of the uses of `a language', but its concern is with real definitions definitions of the essence of a thing not with merely verbal definitions. it is thus intended as a contribution to ontological arguments about languages to arguments about their mode of existence. I see this kind of discussion as logically prior to discussion of the epistemology or methodology of linguistics, for the sorts of reasons developed within realist philosophy of science and outlined in my book Language in Mind and Language in Society (Pateman 1987). As I proceed to consider various concepts of a language I shall also endeavour to indicate the place which the individual speaker or speaking subject must be assigned in relation to them.

Some measure of the extent of confusion prevailing in linguistics - or at least for readers of linguistics - can be gauged by reading the opening chapter (entitled `What is a Language?') of a popular textbook of Chomskyan linguistics (Smith and Wilson 1979). Written by two distinguished linguists, the chapter in question both says that `A language is definable in terms of a set of rules' (ibid., p. 13) constituting a grammar (see Endnote 1), and that `Since every speaker will have heard a different set of utterances, it is not surprising that he comes to possess a slightly different grammar from those of people around him. Strictly speaking, then, we cannot talk of the grammar of English, but only of the grammars of individual speakers of English' (ibid., p. 26). Unless the reader sees that this allows for English to be definable in terms of a set of sets of rules, since a set of sets is also a set (a formal version of the informal strategy later adopted at ibid., p. 197 - see Endnote 2), the reader could be forgiven for wondering what `English' refers to in the above quotation, and in other places throughout the chapter. The most reasonable interpretation of what is going on is that a language has been given a definition which excludes English from the class of languages, though `English' continues to be used in some pre-theoretical (or alternative theoretical) sense to gesture at something unavoidably present but not theoretically allowed for (see Love 1981 for an extended discussion of this aspect of Smith and Wilson 1979).

In the rest of this essay, I will consider five answers to the question, `What is a Language?', three or four of which have some currency in contemporary linguistics and philosophy of linguistics, and defend a particular combination of answers as getting things right. It may be helpful to have the five answers listed in brief and labelled form at the outset. Here they are:

  • (I) A language is a natural kind. (NATURALISM)
  • (II) A language is an abstract object. (PLATONISM)
  • (III) A language is a name given to a set of objects (for example, a set of grammars, lects, or idiolects, characteristically taken to be properties of individual speakers) . (NOMINALISM) (See Endnote 3)
  • (IV) A language is a social fact, and that social fact is also a (or, in a stronger version, the only) linguistic fact. (SOCIOLOGISM)
  • (V) A language is a social fact, but that social fact is not a linguistic fact. (DUALISM, for want of a better word to indicate a view of reality as stratified and with at least `weak' emergent properties) .

By the end of the essay I will have argued that this last account (V) is true; that in addition linguistic facts (e.g., individual linguistic rules) are not necessarily social facts and, indeed, frequently are not contra position ( IV); and that the truth of (V) is compatible with the truth of some versions of (I). Positions II and III are rejected as satisfactory answers to the questions being posed.

The discussion which follows considers in relation to each five positions, and some variants and elaborations of them: (1) arguments for and against them; (2) compatibilities and incompatibilities between them; (3) exemplifications of them in the literature of linguistics and philosophy of linguistics; (4) the place each position explicitly or implicitly assigns to the individual speaker. One particular aim of my discussion is to outline an ontology which would permit and require both a psychology and a sociology of language, related in a particular way, and in this connection versions of positions (I) and (V) become focal for the discussion and lead into the positions developed elsewhere in myLanguage in Mind and Language in Society.

Part 2. Languages as Natural Kinds

At first approach it might appear that currently influential doctrines of natural kinds, as developed in the work of Saul Kripke and Hilary Putnam (Kripke 1972; Putnam 1975) are instantiated within linguistics only in the discredited and obsolete biologistic (`family tree') theories of the last century: those of August Schleicher, for instance (see Tort 1980). Even if tigers and lemons have essential properties, replicated (with variation) in the processes of natural reproduction, surely no one believes that English or Hopi could be natural kinds which replicate themselves in purely natural processes, which have essential properties, and which change only through some (natural) selection of variations. Cast into sociobiological terms, we could talk of memetic rather than genetic replication (Dawkins 1976) and look upon individual speaking subjects as providing the supports or interactors (see Harre 1981) through which replications can take place . But surely such talk is no more than a metaphor or an analogical parasite on properly social (non natural) models of cultural transmission!

The knock down argument against the possibility thatlanguages are natural kinds is surely and simply this: that tiger cubs brought up among humans become tigers, not humans, whereas Vietnamese orphans brought up in England speak English, not Vietnamese, and - if English is their first or early second language - without so much as a hint of Vietnamese in their English.

But this is not the end of the matter. For it does not follow from the fact that a language - one language on its own - is not a natural kind that languages as a class are not a natural kind. In other words, it does not follow from the fact that English is not distinguishable from Vietnamese by essential, natural, replicable properties that languages are not distinguishable from other human or animal semiotic systems by essential, natural, and replicable properties. And this, of course, is exactly what Chomsky and fellow nativists claim, against those linguists and philosophers who have argued that languages can and do vary without definable limit (i.e., without it being possible to specify essential features or universal features which are other than accidental or contingent universals).

Now linguists and philosophers are frequently at pains to distinguish the concept of a language from the concept of language: Saussure does just this when he distinguishes la langue from la faculte de langage(Saussure 1983; and see Harris 1981, ch. 1, for a discussion of the distinction). But while I agree that the distinction is of fundamental importance, I do not believe that it is a confusion of the issues with which this chapter is concerned, or a distraction from them, to consider the nativist position and one possible development of it which may well re establish a place for languages as natural kinds.

For Chomsky, as is familiar and as I discuss elsewhere (Pateman 1985a), there are innate constraints on the class of languages possible for humans, more specifically, on the class of languages learnable as a first language under normal conditions of exposure and time available. It is a contingent fact that the constraints universal grammar among them are what they are, and in this way Chomsky's nativism differs from classical rationalism, which was interested in the necessary conditions of any possible experience.

The contingently given innate constraints on natural language acquisition, whatever they are, define the class of (possible) human languages - languages possible for humans - and are essential properties which distinguish language from other semiotic systems. It does not follow that possession of the innate characteristics is a necessary condition of acquisition of a human language, so it is possible that chimpanzees and global aphasics may acquire a human language they will employ other than the characteristic methods, under different conditions of time and exposure. Nor is it impossible for humans to learn languages which fall outside the class defined by the innate constraints, but they will do so using other cognitive faculties than the language faculty and under different conditions of time and exposure (see Chomsky 1975a, 1980 and notably 1979 for extensive discussion of these characteristically realist points; also the discussion in Pateman 1982a and 1985a).

The doctrine of universal grammar and of innate constraints on language acquisition, as developed by Chomsky, can be given a form (and in outline has been given such a form by Chomsky) in which it is possible to argue that in some sense not only language butlanguages are natural kinds. I do not require the truth of this possibility for anything which follows, but sketch it here because it is interesting and relevant to my present purpose which is the consideration of various real definitions of the concept of a language.

In Rules and Representations (1980), Chomsky points out that innate constraints on the class of languages can be invoked to account not only for similarities between languages (universals of language) but differences between them:

If the system of universal grammar is sufficiently rich, then limited evidence will suffice for the development of rich and complex systems in the mind, and a small change in parameters may lead to what appears to be a radical change in the resulting system. (Chomsky 1980, p. 66; see also Chomsky 1981, ch 1)

The parameters here are the input of utterances which the language learner receives. Now I see no reason to doubt Chomsky's claim, and technical work in learnability theory (Wexler and Culicover 1980; Baker and McCarthy 1981; Hornstein and Lightfoot 1981) is concerned with just such issues. But if Chomsky is right in the passage quoted, then it is also entirely possible that the grammars which universal grammara classifies as possible, and consequently the languages they define, should be - to put it loosely - clustered rather than dispersed within linguistic space. It may even be the case that there would be prototypical languages within those clusters which are particularly learnable (accessible). The type of work on implicational relations and more generally typological linguistics now often conducted in a non or anti- nativist framework (e.g., Comrie 1981) then appears as far from incompatible with it.

The creolist Derek Bickerton has transformed Chomsky's account of language as a natural kind into a theory of an ur language, a theory that there is a language (but only one) natural to the species, specified in what Bickerton calls the bioprogram (Bickerton 1981, 1984). Whereas universal grammar constrains the class of possible natural languages, the bioprogram is the outline of one language. Bickerton's theory is designed to account for the universal or near universal features of the grammars of young creoles that have been generally recognized by many workers in the field, and accounted for in various ways. Creoles are the creation of the children of parents whose native language is of little or no use to them in their community (e.g., as slaves or plantation labourers), and who use a pidgin for communicative purposes. Whereas the pidgin has features clearly derived from its substrate and superstrate languages for example, syntactic features of the substrate with the lexicon of the superstrate the creole has syntactic features not in the substrate, nor in the superstrate, but appearing in other creoles (whose speakers have no contact with speakers of the creole in question). Creolization is very close to an experimentum crucis for theories of the innate aspects of language, and Bickerton argues that the experiment supports the idea of one ur or proto language, to which creoles are very close, but from which more developed "cultural" languages diverge to a greater or lesser degree. This last idea is not foreign to Chomsky, who writes of the "actual systems called 'languages' " that they may be " 'impure' in the sense that they incorporate elements derived by faculties other than the language faculty and hence depart in certain respects from principles of universal grammar" (Chomsky 1980, p. 28). Bickerton 1984 explicitly compares his notion of bioprogram with Chomsky's more recent ideas (Chomsky 1981) on core grammar. Bickerton comments that his bioprogram, in comparison with Chomsky's core grammar, constitutes 'a kind of "inner core grammar" from which more complex and varied grammars have evolved and which may bear a relationship to core grammar similar to the one that core grammar bears to the marked periphery' (1984, p. 188). It does not matter for present purposes whether Chomsky's or Bickerton's or some other yet-to-be thought-of account is nearer the truth, or indeed if any of them are. All that matters is that it makes sense to sustain some basically 'natural kinds' approach in the study of language, as a specific kind of naturalism. This seems to me to yield a coherent and plausible research programme.

Finally, a brief word about the relation of the individual (the speaking subject) to languages, construed as natural kinds. The individual speaker stands to the language as token to type, or phenotype to genotype. The token (phenotype) may well possess properties which do not belong to the type (genotype) and is the source of variation, which will occur during replication or reproduction. The speaking subject as the support or interactor through which replication occurs may be more or less powerful in acting self consciously (reflexively) on the process of replication: this is a topic I take up in Pateman 1985a.

Part 3. Languages as Abstract Objects

For many linguists, Chomsky ties linguistics too closely to other sciences - notably psychology and neurology and also to high theory with claims which are not open to any obvious falsification, or which are too easily reinterpreted to cope with what some have taken to be obvious falsifications the linguistic abilities of apes, for instance. On the other hand, he has furnished for those same linguists a clear understanding and powerful models of what it is to write a grammar, the central descriptive activity of linguists. One resolution of the tension this state of affairs creates is to recast the object of linguistic investigation in such a manner that the writing of grammars can go on in a serious way without being dogged by commitments to `psychological reality'. In his book Language and Other Abstract Objects (Katz 1981; see also Katz 1985, Langendoen and Postal 1984) Jerrold Katz recasts the object of linguistics in such a way as to free linguistics from relations with psychology and neurology.

Katz argues that sentences and languages, like numbers and implication relations, are Platonically real, abstract objects, knowable in corrigible intuition, and that linguistics is properly construed as a branch of mathematics (Katz 1981, p. 93) which studies the properties of such abstract objects. Abstract objects have no temporal or spatial properties but are as objective as physical objects, in the sense that no one person has a `special relation' to an abstract object, and are changeless and cohesive 'in the sense of having logically inseparable basic properties' (p. 186). Katz develops his argument in opposition to what he calls Chomsky's 'conceptualism' which 'imposes its own psychological constraints which require grammars, over and above getting the facts about the grammatical structure of the sentences right, to say something about the psychological states of speakers' (p. 52).

Katz employs what he calls 'general scientific methodology' and especially a simplicity criterion (Occam's Razor) as a 'neutral basis' (p. 64) from which to argue against Chomsky's conceptualism (pp. 63-73, pp. 232 8), and though this use of a simplicity criterion is contestable, it is more important for present purposes that adherence to it is inconsistent with Katz's ontological and epistemological arguments in the following way. In Chapter VI of his book Katz defends a Kantian theory of intuition as providing the appropriate epistemology for his ontology of abstract objects. 'Kant's account conceives of intuitive awareness, not as a causal effect of an external event, but as the effect of an internal construction' (p. 202). In other words, intuition is 'the inner construction of concepts of abstract objects in concreto' (p. 212). Particularized to language, this yields a characterization of actual intuitions 'as the result of projections from an a priori enumeration of systems of competence which corrects and depsychologises complexes of grammatical information and casts them into concrete form as concepts of sentences modelled on the notion of an abstract object' (p. 206). Now just as Katz distinguishes in his critique of conceptualism between knowledge of language and the language of which it is knowledge, so here we must distinguish between concepts of abstract objects and abstract objects themselves. And on a Kantian epistemology, the only kind effectively available to Katz, the latter (noumenal) objects remain unknowable things in themselves, and our knowledge is confined to the phenomenal world of concepts of objects. This being so, the noumenal objects can make no difference to our theories, and it follows that Katz ought to wield Occam's Razor to dispose of these abstract objects, though the whole purpose of his book is to convince us of their existence . Endnote 4

It is difficult, though not impossible, for Katz to salvage his theory from this contradiction. Should he abandon the simplicity criterion, he loses the essential instrument of his critique of Chomsky's conceptualism. Should he abandon abstract objects, he abandons the book under discussion. He would then be saying that intuition gives access not to abstract objects, but to the structures and mechanisms which comprise our linguistic competence, a position indistinguishable from that of Chomsky. Should he seek an alternative to his Kantian theory of intuition he will straightway re encounter the problem which motivated his Kantianism in the first place, namely, the problem of how we can have any knowledge of abstract objects in the absence of a causal relation between them and us: a problem much discussed in Platonist philosophy of mathematics (see, e.g., Steiner 1975). This third strategy is the obvious one for Katz to develop, and Katz may have more to say about intuition at a later date, since he indicates that he is working on a book which he refers to as Sciences of the Intuition (Katz in preparation). One of Katz's more remarkable claims is that English is an abstract object, of which speakers of English have partial and imperfect knowledge, and 'English' a proper concept of linguistic science (ch. III). Katz regards as 'astounding' (Katz 1981, p. 80) and, evidently, absurd Chomsky's claim (which I discuss below) that 'such notions as "the English language" are not linguistically definable, but are rather sociopolitical in nature' (ibid., p. 79). And he goes on, 'Empirical evidence shows no more than that speakers have internalised some principles [of language TP], but for the internalised principles to be counted as knowledge of the language [e.g. English TP] they must be related in the proper way [truth conditional correspondence -TP to the language' (ibid., p. 83). These quotations are inserted here to indicate just how thoroughgoing is Katz's Platonism: his abstract objects include ones which really are 'out there' awaiting discovery.

Aside from my argument that Katz's epistemology cannot yield knowledge of the abstract objects he postulates, is there anything which can be said more directly for or against the existence of linguistic abstract objects? If it could be shown that there are no abstract objects, then there would be no linguistic abstract objects, but from an argument to show that there are no linguistic abstract objects it does not follow that there are no abstract objects of other kinds, for example, mathematical objects or logical relations. Because I do not wish to enter into philosophy of mathematics or logic, I confine myself to the question of the existence of linguistic abstract objects.

The main argument I wish to use here against Katz's claim for the existence of linguistic abstract objects, including languages (as the quotations above show Katz is quite clear that he regards English as an abstract object; see also ibid., pp. 77 83), is that the propertiesKatz atributes to these objects are arguably possessed by objects of a much less contentious character than thoroughgoing Platonist abstract objects. For Katz, an abstract object is objective but does not have temporal or spatial properties, and its objectivity is established by the fact that individuals do not have a special relation (privileged access) to it. In addition, abstract objects are changeless and cohesive. (Katz's epistemological claim about these objects is that they are knowable in corrigible intuition. I shall say something about how his epistemological difficulties are soluble when different objects are substituted for abstract objects.) The properties Katz assigns to abstract objects appear all to be possessed by the kind of conventions of mutual knowledge or belief which Esa Itkonen argues are constitutive of linguistic rules (Itkonen 1978a; not cited in Katz 1981 but see endnote 5) following a familiar line of post Wittgensteinian argument and using the theories of convention and mutual knowledge, the most important of which - David Lewis's I discuss at length in Pateman 1982c. Thus, conventions of mutual knowledge (public rule systems) are not objects in space or time because claims about them are not falsifiable by empirically occurring (spatio temporal) counter instances. A claim about a rule differs from a claim about a regularity in just this kind of way. That such objects are 'cohesive' requires only the granting of a general structuralist point it is not necessary for an object to be abstract for it to be an object ou tout se tient , as Meillet put it. It may seem less obvious that such objects are changeless that, as Dummett puts it, 'An abstract object can neither be the cause nor the subject of change' (Dummett 1973, pp 491-92) but this is, in fact, a corollary of their non existence in space and time. Rule systems do not change, though in use one rule system may replace another. It then appears that the system itself has changed simply because both the previous and the present system are called by the same name.

As for the relation of the individual to systems of mutual knowledge, this is not privileged but only because all participants are equally privileged: being parties to the system individuals speak for all when they speak for themselves. This idea has been explored at length by philosophers such as Stanley Cavell (Cavell 1958), with whom Katz has crossed arguments in the past (see Fodor and Katz 1963), and has even been offered as a rational reconstruction of Platonic anamnesis: philosophical discoveries are acts of remembering what we learnt on our mother's knee and have forgotten ever learning (Hare 1960; this item and the preceding two, together with other pertinent contributions are anthologized in Lyas 1971). Though some of the contributors to this discussion (e.g. Cavell) have taken the view that intuitions about our mutual knowledge are incorrigible, both Katz and Itkonen are agreed that our intuitions, about what for Katz are abstract objects and for Itkonen rules known a theoretically with certainty, are fallible and corrigible, since they are always subject to interference in the process of their articulation.

The upshot of this discussion, then, is that Katz defines abstract objects in terms of characteristics possessed by at least one other object, conventions of mutual knowledge, and so does not make out a distinctive case for believing in abstract objects as other thanconventions of mutual knowledge. These latter raise fewer ontological and epistemological problems than the Platonist idea of abstract objects, and so on grounds of simplicity might appeal to, and indeed ought to be preferred by Katz.

In conclusion of this discussion it is worth remarking that there is no doubt an attraction in the idea of abstract objects which stems from confusing the practice ofabstraction in linguistics with the study of abstract objects in Katz's sense. Thus, one might consider properties of sentences generated by a grammar in abstraction from any consideration of the existence of the grammar or the sentences other than within the theory. In one sense, what is then being studied is an abstract object. But not in Katz's sense. For as the passages from Katz cited above indicate, he is committed to a view of abstract objects in which they form the reality to which true statements or propositions of linguistics correspond. But the interest of the formal or mathematical grammarian is entirely in questions ofcoherence, such that no reference to extra theoretic entities is required. Formal or mathematical grammarians are not in the business of makingexistential claims.

As for the relation of the individual to Katzian languages, it is, as indicated, a relationship between the Knower and the Known, where the Knower is liable to error. If nothing else, this doctrine will gladden the hearts of hard pressed prescriptivists, who can now believe that English has a Platonic existence and that they are its Guardians.

That this is the implication of Katz 1981 is clear from Bever 1982. Bever, contributing to a state of the art volume on language acquisition, contrasts the view that the essence of language has a cause, which is our biology or psychology, with the view that the essence of language is uncaused that it is an abstract form, which humans may discover but do not cause (e.g., p. 433). It is not language which evolved, but humans who evolved to a point at which they could learn it and had a motive to learn it (p. 436). The radical upshot is indicated by the following quotation: 'The essence of language has no cause [according to "the linguistic realist" TP]. The child's problem is to apply general intelligence or special capacities to construct a representation of language sufficient to serve human purposes. It is possible that children never acquire a real language at all, just as they do not acquire a consistent logical system [for "children" one can equally read "adults" TP]. But just as everyday human reason proceeds well enough with faulty logic, human communication could proceed with a faulty language' (p. 443). As this quotation indicates, Bever motivates the idea of language as an uncaused, abstract form by a comparison with logic, and elsewhere with arithmetic, and his approach leaves it open that the essence of language or logic might actually be an unknowable mystery if our construction does not allow us to discover it. In Kantian terms, Bever has a concept of language, logic, and mathematics in which their objects belong to the noumenal world.

Part 4. Languages as Names

Set against Naturalism and Platonism, the view that languages are names might appear simple and uncontroversial in the extreme. It is also a view which attracts linguists whose focus of interest is the individual idiolect or mentally represented grammars, and all linguists in so far as they see that no two speakers are ever linguistic identical twins. Even sociolinguists, who might seem to have a professional interest in linguistic facts or facts about language not reducible to facts about individuals can be found to express a scepticism about the reality of languages or dialects as other than names. Thus Richard Hudson writes in his textbook of sociolinguistics that, 'Varieties [which he uses as generic for languages, dialects, etc. TP] do not exist except as informal ways of talking about collections of linguistic items which are roughly similar in their social distributions' (Hudson Hudson 1980, p. 55). Here the use of the idea of social distribution is an attempt, one might well say, to save the idea of a specifically socio linguistic inquiry. In other kinds of linguistic inquiry, other bases of similarity may be suggested for grouping together linguistic items. Thus, in their textbook, Smith and Wilson have in mind formal similarities between items when they write that:

Strictly speaking . . . both dialect and language [in the sense of 'a language' TP] are relative terms, based on grouping together speakers with broadly similar grammars. (Smith and Wilson 1979, p. 197)

This idea of grouping lends itself to at least formal refinement in the idea that a language can be defined in terms of some elaborate set-theoretic operation on mentally represented grammars or idiolects: Wunderlich describes an elaborate version of such a theory due to Kanngieser (Wunderlich 1979, pp. 339 46).

It is difficult to see why anyone should bother with such elaborations, since the language so painfully constructed by a set-theoretic operation (and, of course, the construction is purely programmatic) has no theoretically interesting properties which are not properties of the idiolects or grammars on which the operation is performed. The only possible interest I can see lies in the possibility of constructing languages, dialects, etc., using different principles of grouping and comparing results. Thus, one could hope to show (for example) how mutual intelligibility is partly dependent on speakers' and hearers' attitudes by holding differences in formal properties of grammars constant and showing variability in the extent to which speakers and hearers claim to understand each other. This would be more elaborate than the kind of work which points out that if you are a Dane well-disposed towards Swedes, you will understand more of what they are saying than if you are not.

It may be that some nominalists have thought that it is possible to reconstruct by set theoretic operations a language the extension of which is somehow known in advance for instance, knowing who are the (native?) speakers of English, one would then seek a principle of grouping which would keep English speakers together and separate them from all other speakers. Trivially, there is undoubtedly such a principle, since there is no limit on the complexity or silliness of permissible operations for mapping one domain into another. But there is absolutely no reason to suppose that an operation which keeps English speakers together now would be of any use in grouping English speakers in the past or future, or in grouping speakers of other languages. Any such operation is devoid of interest; what is interesting is how it is possible to know in advance who are all and only the speakers of English. I suggest one essential element of any plausible answer to this problem below, involving reference to howspeakers define the language they speak.

It is indeed difficult to explain the attraction and use of essentially nominalist answers to the question, What is a Language?, in contemporary linguistic literature except as indicating a tacit assump tion about a proper division of labour between linguists and sociologists, or else as showing the impact of a particular philosophy of science within linguistics. I should like to say a bit about each of these possibilities.

In relation to the first, a linguist could say that not all facts about language are linguistic facts, and that though linguists' nominalist definitions of languages do, indeed, show languages to be devoid of linguistic interest, it does not mean that they are devoid of sociological interest. Language has an 'external' as well as an 'internal' reality, and though there is nothing in a name as far as the linguist is concerned, there may be a lot in it from a sociological standpoint. This point seems to me to be on the right lines, but as I indicate below, the sociologist may decline to take over the linguists' nominalism as adequately expressing the reality (the ontology) of languages considered as social facts. People do indeed fight over names, but there may be more to languages than their names. What more there is is the subject of parts 5 and 6.

In terms of the impact of the philosophy of science on linguistics, it seems that the nominalist solution (or dissolution) of the problem of what languages are shows the impact of the ideas of positivism in general, and ofindividualism and reductivism specifically. Nominalist approaches are positivistic when they treat the object of their investigation as a fact definable and describable purely 'externally' without reference to a hermeneutic moment in which speakers define the fact or object for themselves. The approach is individualistic in assigning all theoretically interesting properties to properties of individual speaking subjects, and reductionistic in denying any emergent properties to language that is, in denying languages any causal powers. I point out these features here, and in part 6 outline a view of languages which is non reductionist and incorporates recognition of the hermeneutic moment, but which in one sense remains individualistic. In part 5, I consider some conceptions of language which are more or less clearly non individualistic.

As for the individual's relation to a language in nominalism, there is simply no relation when the name is a linguist's construct from what other speakers do. And where the name is taken by the linguist from speakers, and used to demarcate the set of grammars or idiolects requiring reconstruction as a language, what the name then means to speakers is outside the scope of nominalist enquiry it has to be considered hermeneutically, as it is in part 6 below.

Part 5. Languages as Social Facts which are Linguistic Facts

Nominalism has always had a strong implantation in the social sciences, at least from the time of Hobbes. But it has never gone unchallenged, and within linguistics there have always been powerful theoretical alternatives to it, which are simultaneously alternatives to Naturalism and Platonism. Thus, any theory which says that a language is a social fact and that that social fact is a linguistic fact, is incompatible with Naturalism, Platonism, and Nominalism.

By a linguistic fact, I mean (in the present discussion) the kind of fact for which a grammar or a rule of grammar can be written. It is less straightforward to say what a social fact is it has been a matter of sociological controversy from the time of Durkheim, and it is too easily defined in ways which make it both indistinguishable from an abstract object and totally implausible. Durkheim was very conscious of this problem, and protested against the reified versions of his theory attributed to him. Yet there remains the problem of formulating his theory in such a way that it does not lead to the positions generally attributed to him. For present purposes, and as a first approximation, I will characterize a social fact as a fact which pre exists any individual considereddistributively but does not pre exist every individual considered collectivel. All kinds of familiar institutions are social facts in these terms churches, schools, legal systems, and so on. The relation of the individual to the institution is then one of participation and membership, and the entry procedure one of socialization, initiation, etc.

Now in these senses of linguistic fact and social fact it has seemed obvious to many linguists (and to allsociologists) that languages are social facts, and that those social facts are linguistic facts. I shall call facts which are purported to be both linguistic and socialsocial-linguistic facts. The proper object of linguistic description is then a social fact (or institution) which Saussure called la langue, and which a contemporary linguist calls the grammar of a speech community (Labov 1972). I shall say a little about both Saussure and Labov, and Labov's presentation of Saussure in his formulation of the so called `Saussurean paradox', at the same time recognizing that there is a great deal more which can be and has been said relevant to my purposes.(See endnote 8)

Saussure states in fairly clear form what, I suppose, is also the commonsense view of languages. They are structured, closed systems, self contained wholes in 'which all the parts can and must be considered synchronically interdependent' (Saussure 1983, p. 86). They exist as social institutions, but eluding both individual and social control (pp. 15 -16). The social character of languages is not fortuitous: the social nature of a language `is one of its internal characteristics' (p. 77). Individual speakers participate in a language which they passively register (p. 14), but 'the language is never complete in any single individual, but exists perfectly only in the collectivity' (p. 13). Such an approach is reasonably called a Durkheimian one, though Whitney is the writer to whom Saussure explicitly refers in connection with the idea of languages as social facts.

Saussure's difficulties with this account of languages stem from his recognition of linguistic variability. He seeks to incorporate recognition of this fact by saying that 'All the individuals linguistically linked will establish among themselves a kind of mean [une sorte de moyenne]: all of them will reproduce doubtless not exactly, but approximately the same signs linked to the same concepts' (p. 13). But this simply does not fit with the idea of a language as self contained, homogeneous, and passively assimilated; nor is the idea of a `kind of mean' the same idea as the idea of a language of which each speaker knows only parts, as I know only some of the words in the Oxford English Dictionary (cf. Saussure p. 13 for this `Storehouse' aspect of language).

For example, Saussure says that 'If two people from different regions of France say se facher (`become angry') and se focher respectively, the difference is very minor in comparison with the grammatical facts which allow us to recognise in these two different forms one and the same linguistic unit' (p. 180). But the nature of this `very minor' difference can be construed in at least two different ways, one of which is consistent with Saussure's conception of a language, but implausible, the other inconsistent but plausible. The implausible view is that the underlying phonology of the two forms (the signifiers) is more abstract than either of the two realizations, so that the difference counts only as variation in the token (the phonetic item) not the type (the phonological item). This view would preserve the notion of a single language (langue) shared by speakers of these two variants at the price of a very abstract conception of phonology. The plausible view is that the two forms are realizations of two different phonological items /se facher/ and /sefocher/, where the difference in phonological type shows up in phonetic realization. The only sense in which they are different forms of one and the same linguistic unit is that they both mean the same - the signifieds remain constant though the signifiers vary. But identity of meaning does not give Saussure what he needs - this is obvious if one considers that red androuge have the same meaning, yet no one supposes they belong to the same language and are no more than different forms of the same linguistic unit.

Much more plausible, but inconsistent with the positions he has previously sought to defend, is Saussure's subsequent statement that 'The boundaries held to separate two languages can only be conventional ones' (p. 202), simply because of the continuously changing character of language as distributed in space.

This remark suggests the view that though it is possible to believe that a language may be no more than a name (`German', `Dutch', etc.) the facts named do and must exist as social linguistic facts in the way Saussure is driving at, perhaps at a level no higher than that represented by individual grammatical rules. This is certainly not implausible. Nor is it implausible to suppose that the boundary between the purely nominal aspects of a language and the social-linguistic reality of its rules could be a shifting one. Standardization is precisely about making a language more than a name for a set of heterogeneous objects by making those objects more homogeneous.

The key theoretical issue is whether at the very lowest level, grammatical rules must exist as social linguistic facts (as, one might say, duolectal rather than idiolecta). I discuss this issue below in connection with the work of Esa Itkonen, and again elsewhere on this website in the essay "Wittgensteinians and Chomskyans".

For the moment, I turn to the work of a linguist whose work seeks both to save the idea of a language as a social linguistic fact, and to reconcile that with the reality of linguistic variability which made Saussure write about a 'kind of mean' and 'conventional' boundaries. At least in his early work, William Labov is aiming to give an account of languages existing at the level of a social linguistic fact which nonetheless recognizes linguistic variability.

Thus, in his seminal paper on contraction and deletion of the copula in Black English Vernacular (BEV) (Labov 1969; revised version in Labov 1977, ch. 3 to which I refer) Labov declares that, `The aim of linguistic analysis is to describe the regular patterns of the speech community, rather than the idiosyncrasies of any given individual' (Labov 1977, p. 95). Of course, the expression `regular patterns' needs to be cashed for a proper concept of a social linguistic fact, and indeed it is. For Labov proceeds to construct a 17 step rule (a variable rule) for the contraction and deletion of the copula in BEV which not only accounts in a statistical sense for the use of the copula in the speech community studied, but about which he makes strong ontological claims, as follows: `We are in no way dealing with statistical statements or approximations to some ideal or true grammar. We are dealing with a set of quantitative relations which are the form of the grammar itself' (p. I 25). So the `regular patterns' are the `form of the grammar' (in the existential sense of what exists in the community rather than in the linguist's grammatical reconstruction of what happens).

The ontological claim is motivated and justified in the following passage:

The construction of complete grammars for `idiolects' even one's own [the practice Labov believes of both Saussure and Chomsky TP], is a fruitless and unrewarding task; we now know enough about language in its social context to realise that the grammar of the speech community is more regular and systematic than the behaviour of any one individual. Unless the individual speech pattern is studied within the overall system of the community, it will appear as a mosaic of unaccountable and sporadic variation. (p124)

I find Labov's linguistic ontology totally implausible, for reasons I shall now indicate. Labov's methodology is to collect speech data from individuals, subject variation in the data (e.g. phonetic realization of a phoneme, most famously /r/) to statistical analysis to establish linguistic and objectivistically defined social correlates of the variation, and then write variable rules which will generate the appropriate variant for any linguistic or social context. Why the variable rule should be regarded as other than an artefact of the methodology a theoretical fiction is completely unexplained. Labov does seek to avoid the Platonism or vulgar Durkheimianism of making the variable rule a property of the community independent of individuals collectively considered by attributing to individuals distributively a mentally represented grammar the rules of which are isomorphic with the community grammar: `It is evident that rules I 17 [for contraction and deletion of the copula in BEV TP] . . . are a part of the speaker's knowledge of the language' (ibid., p. 125). But there is no basis for this claim. The performance of individual speakers (lost in the statistical processing) does not have to be and as a matter of fact (Bickerton 1971) is not isomorphic with the variable rule, and short of claiming that all non isomorphic variation around the variable rule is the result of performance factors (fatigue, slips of the tongue, and so on), there is no reason to suppose that `the speaker's knowledge of the language' is other than isomorphic with their performance rather than the community grammar. In addition, the isomorphism which does exist can be accounted for without postulating a community grammar as really existing (see, e.g., Bickerton 1971,1975)

It may seem that Labov has an argument in the last sentence of the passage quoted above from his page 124. There are two questions here. Is it true that the individual speech pattern is unaccountable except in a larger context? And, does that demonstrate the real existence of a community grammar? Before answering it may be helpful to add into the discussion a passage from Labov's major paper on linguistic methodology (Labov 1972) where he writes:

It seems clear that a great many disputes in generative grammar do revolve about an area of idiosyncratic behaviour where the social compact has disappeared. For rare sentence types, it is only natural that each individual should have to solve the problem for himself; insofar as he can do this by extending his current roles in a predictable manner, we are dealing with langue [which for Labov is aquantitative phenomenon TP]; insofar as individuals diverge without any observable pattern, we are dealing with true idiolects. The very concept ofidiolec, of course, represents a defeat for the Saussurian notion of langue as the general possession of the speech community. Our general aim is to write the grammar of that speech community with all of its internal variation, style shifting, change in progress .. . when the data begins to fragment into unpatterned idiosyncrasies . . . then linguistics comes to a stop. (Labov 1972, p. 108)

It seems to me that Labov commits two errors in this passage and in the previous one: (1) he fails to distinguish unaccountability in practice from unaccountability in principle, and effectively infers the latter from what may only be cases of the former; (2) he offers a false choice between statistical predictability and true randomness, when what is at issue is whether there is a use for the idea of rule governed idiolects non isomorphic with la langue. It is error (2) which helps us answer our two questions. Clearly, if you believe that linguistic regularity is statistical, then you need more than one instance of a form to establish what the regularity is of which it is an instance. But this does not establish the existence of a community grammar, only of regularities of speech patterns and a trivial methodological limit on their study. And if there was truly random language use, however `creative' that might seem to some, there would be nothing to say about it, either in the explanatory or predictive mode. But to suppose these are the only options is to beg the question. Further, from the need to study more than one token to establish the type, it does not follow that the tokens must be culled from different individuals. There is no objection in principle (at least in the context Labov aims to establish) to a statistical approach to the study of idiolects. Put another way, the idea that the individual speech pattern is a `mosaic of unaccountable and sporadic variation' (Labov 1977, p. 124) is inconsistent with the idea that variable rules `are a part of the speaker's knowledge of the language' (p. 125). In short, none of Labov's anti Chomskyan argument establishes a case for the reality of community grammars, or any other kind of `superlanguage' (cf. Chomsky 1980, pp. 117 18; and the work of Romaine 1981, 1982).

In conclusion of my discussion of Labov, let me just remark on the paradox Labov has described and made famous under the name `Saussure's paradox', since it bears on discussion in part 6 below and in Pateman 1985a. Labov formulates the Saussurean paradox as follows:

Saussure argues that langue is a social fact, knowledge possessed by practically every member of the speech community. It follows that one can find out about langue by questioning one or two speakers of the language even oneself. On the other hand,parole reveals individual differences among speakers that can be examined only in the field by a kind of sociological survey. Thus the social aspect of language can be studied in the privacy of one's own office, while the individual aspect would require social research in the heart of the speech community. (Labov 1978, p. 267)

It is far from clear that, even granting the accuracy of Labov's rendering of Saussure, there is any paradox here. For if la langue should be, for example, a system of mutual knowledge conventions, it would indeed be the case that one speaker could speak corrigibly for an indefinite number of other speakers, with reference to questions about la langue. If parole was the area where individual differences came into play, then indeed it would require fieldwork study. That is not a paradoxical view; it is just that it is mistaken to equate parolewith individual differences, as Bakhtin/Volosinov soon pointed out in their 1929 book (see Volosinov 1973). And in so far as there should be pragmatic competences which shared with languea mutual knowledge character, then individual speakers would be authoritative for others in that domain in the self same way as in the domain of langue. It seems to me that `Saussure's paradox' is little more than an index of Labov's naive conceptions of scientific activity, which in his most famous works is essentially exhausted by his preoccupation with establishing statistical correlations an emphasis which, from a realist standpoint, is totally misplaced.

This concludes what I want to say about Saussure and Labov. I turn now to consider at rather greater length the work of a philosopher of linguistics, Esa Itkonen, whoseGrammatical Theory and Metascience (Itkonen 1978a) defends the view that the core of linguistic facts must be social facts, whatever else they are, and even if there are linguistic facts which are not social facts. Itkonen is concerned both with the ontology of individual linguistic rules and with the ontology of languages, which he thinks of as sets or systems of rules to which names of languages refer. His work is critically relevant here for two main reasons. First, for his attempt to ground philosophically the claim that linguistic facts are centrally social facts, a claim I wish to dispute in section 6 below and elsewhere (notably in "Wittgensteinians and Chomskyans" on this website). Secondly, because my own solution in Part 6 below to the ontology of languages is, essentially, a transposition of Itkonen's account of linguistic rules as social facts onto languages, which I conclude are social facts without being linguistic facts. My discussion here consequently prepares the ground for material still to be presented. My presentation of Itkonen's position in the next few paragraphs has been silently corrected in the light of Itkonen's critique of the article on which this essay is based (Pateman 1983e, Itkonen 1984). There is a direct response to points in the critique with which I disagree in part 6 below.

Itkonen uses `the Wittgensteinian approach to prove the primarily social nature of knowledge and language' (Itkonen 1978a, p. 95), and develops much more thoroughly an approach which is also to be found in Apel 1979 and Habermas I970 and I979a. The conclusion at which he arrives is that, whatever else it is, the essence of language is constituted by conventions in relations between people rather than as rules which are properties of people (e.g. as mentally represented grammars), though he generally uses the word `rules' for what I have called `conventions', partly because he argues against D. K. Lewis (see Lewis 1969 and Pateman 1982c) that the aspect of normativity in rules is ineliminable. However, Itkonen's position is very similar to Lewis's in his conclusion that a language `is a set of rules existing at the level of common knowledge, and grammar is a (theoretical) description of these rules, or of this knowledge' (Itkonen 1978a, p. 136), where common knowledge is defined in terms of the satisfaction of three conditions: for x to be an object of common knowledge, it must be the case that,

`(1) A knows1 x; (2) A knows1 that B knows2 x; (3) A knows3 that B knows2 that A knows1 x' (p. 123; compare Lewis 1969 and Schiffer 1973).

Two points before I continue with the argument. First, in the quotation above from page 123, it is unspecified whether `A' or `B' range over individuals or classes of individuals. The point relates to the question of how it is that a speaker defines his or her co linguals, with whom that speaker enters into common knowledge relations. Is it that I know that my mother knows that the definite article precedes the noun? Or is it that I know that my consociates know that the definite article precedes the noun? (But then what of multilingual communities?) Or is it that I know that other speakers of English know that the definite article precedes the noun? (And if so, why is such an account not viciously circular?) Clearly, Itkonen thinks that `A' and `B' range over classes rather than individuals, but there seems to me a considerable, and interesting problem, in specifying what those classes are. My own arguments in part 6 should contribute to a solution to this problem.

Secondly, though Itkonen thinks that it is necessary for some linguistic rules to be common knowledge in a speech community, he does not think it necessary for all to be. This allows him to acknowledge the existence of idiolectal linguistic variation and the possibility of linguistic change (pp. 151-154) within a common knowledge framework. Where there is variation and change, linguistic rules cease to be known with absolute a theoretical certainty, but that this can sometimes happen does not mean it can always happen (compare Pateman 1982c and Wittgenstein 1958, para. 345). Or, as Itkonen puts it:

Uncertainty [about what a convention is, or whether there is a convention TP] presupposes certainty; rules which hold only approximately presuppose the existence of rules which hold absolutely . . . the uncertainty typical of linguistic change always obtains within the limits of absolute certainty: the uncertainty about the correctness vs. incorrectness of the two entities, A and B, of which the one is in the process of supplanting the other, is to be contrasted with the absolute certainty with which we know that all other (possible) entities D, C etc. are incorrect. This is why linguistic change, or linguistic variation, relativizes only to a quite limited extent the general thesis about absolute atheoretical knowledge of language. (Itkonen 1978a, p. 152; See also Endnote 9)

Of course, this does not tell us to how often what can sometimes happen can actually happen, and neither Itkonen nor Wittgenstein, who supplies the argumentation in the passage quoted above, has a catastrophe theory which would allow him to answer this question - that is, a theory which would indicate the point beyond which further uncertainty would cause a breakdown of the language system and communication by means of it. Though the question is, perhaps, uninteresting to the philosopher, it might be of interest to the linguist asked to accept that an unspecified proportion of linguistic rules must, at any one time, be constituted in common knowledge.

Leaving this aside, I turn at last to the argument for the view that linguistic rules have an essential common knowledge character, holding between at least two people in the central case (as duolectalrather thanidiolectal rules).(See endnote 10). Itkonen's argument (ibid., ch. 4. especially pp. 109 I3) is a Wittgensteinian one: an inner process stands in need of outward criteria (logical, but non reductive, behaviourism). For someone to be said to follow a rule, it must be possible for them to make a mistake and be corrected. This can only occur in interaction or possible interaction, since one's own memory cannot serve as the criterion for determining whether one has, on some occasion, followed a rule correctly. So if no one disputes that language involves rules, linguistic rules must be public rather than private objects, and the character of these public objects is elucidated in the common knowledge i.e. social object analysis. Itkonen regards Wittgenstein's private language argument as simultaneously a refutation of the philosophy and methodology of Chomskyan linguistics: transformational grammar, he says, `maintains a conception of language which is demonstrably equivalent to the private language conception' (ibid., p. 133).

I comment on private language arguments again in "Wittgensteinians and Chomskyans" on this website. Here I offer one challenge to Itkonen's use of private language arguments which is particularly relevant to the concerns of this work, and to the position I advance in Pateman 1982c (which may in due course also appear on this website). It draws on a body of scientific work which, it seems to me, Itkonen either has to regard as a priori false and misguided or else, if he accepts the work as true and interesting, it can only be at the price of showing the fundamental irrelevance of Wittgensteinian private language arguments to any actual problems of linguistic theory and methodology. (The text of the remainder of this section is unaltered in content from the original article on which Itkonen commented in Itkonen 1984; a few stylistic improvements have been made).

Itkonen rightly insists that, `Describing a game is different from describing how it is learned' (ibid., p. 182). An account of the nature of a game, however, must be compatible with that game's being learned, and must not be excluded or made redundant by a true theory of learning. It seems to me that in this context, claims for the necessarily common knowledge character of linguistic rules are falsified by the form which it appears any true theory of learning or acquisition (not to mention growth) must take, and by phenomena of linguisticinventivenes.

Thus, consider the learning of language. Suppose that a child is born into a community whose language is indeed constituted as a set or system of mutually known rules, conventions to which existing speakers are party. The child who enters this community does not thereby become party to its linguistic conventions. To do that the child must begin (on any but the strongest nativist view of language acquisition) by making falsifiable abductions about the linguistic conventions (rules) which obtain in its community, and which generate the speech output the child receives as input. On this basis, the child can begin to speak and communicate successfully. But that is to say that falsifiable knowledge is sufficient for the child to communicate, and that being a party to conventions is unnecessary to its doing so. No doubt the child insensibly becomes a party to conventions, if conventions define the language of its community, but that it does so appears on this view a contingent matter. If it is argued that the child necessarily makes assumptions about what its auditors know and that they necessarily make assumptions about what it knows, so that a fresh set of public rules are constituted, that seems to me to be either whistling in the dark, or else to confuse what it is for something to be a linguistic rule with what it is for something to be uttered with communicative (Gricean) intent.

More radically, the child abduced its rules not from the rules held in common by its interlocutors, but from theoutput of those rules. It made no use of their common knowledge character, and could have succeeded just as well had its linguistic input been derived from the output of a regular rather than a rule governed device: there would be no outward criteria to show the difference in the underlying generative mechanism, about which the child has no need to speculate.

More radically still, it seems quite well established that children do not simply abduce or reinvent existing rules in the process of their linguistic development; they quite clearly invent new rules, the output of which becomes input to the speech of the community to which they belong. Though the outward criteria for ascription of these rules is present, the collectively shared character of these rules is missing: if they are shared at all, it is distributively (in virtue of a shared natur) rather thancollectively (in virtue of shared membership of a linguistic community). This distinction between something shared distributively and something shared collectively is crucial, but sometimes overlooked. There are visual illusions to which we all respond in the same way (distributively) but no one supposes that this is because there is some kind of convention or collective rule to fall for the illusion in the same kind of way. I shall give three examples from language in which the distributive/collective distinction is critical, then consider their implication for Itkonen's arguments.

First, consider the first generation creolization of pidgin languages. In this case, it appears to be empirically false that there is a pre-existing language to be acquired by children. Though `every existing theory of acquisition is based on the assumption that there is always and everywhere an adequate language to be acquired' (Bickertor I98I, p. 5), this is not true for the creolizing child: `Everywhere else in the world it goes without saying that the parent knows more languagc than the child; here, if the child is to have an adequate language, he must speedily outstrip the knowledge of the parent' (ibid., p. 5). Bickerton drives home the point prior to introducing his empirical evidence:

The act of `expanding' the antecedent pidgin, which each first creole generation has to undertake, involves, among other things, acquiring new rules of syntax. In the conventional wisdom, children are supposed to derive rules by processing input (with or without the help of some specific language learning device); in this way, they arrive at a rule system similar to, if not identical with, that of their elders. If this were all children could do, then they would simply learn the pidgin, and there would be no significant gap between the generations. In Hawaii, at least, we have empirical proof that this did not happen that the first creole generation produced rules for which there was no evidence in the previous generation's speech. (Bickerton, 1981, p. 6)

Bickerton attributes the apparent inventiveness of the creolizing chill to an innate `bioprogram' which allows the child to go beyond not merely the information given, but the range of socially available rules. Bickerton generalizes his argument to the normal first language acquisition situation:

The vast mass of human children are not growing up in even a partial linguistic vacuum. There will be a ready made language which their elders will be determined that they should learn. Thus, almost (but not quite) from the earliest stage, the evolving bioprogram [for language TP] will interact with the target language. Sometimes features in the bioprogram will be very similar to features in the target language, in which case we will find extremely rapid, early, and apparently effortless learning. Sometimes the target language will have evolved away from the bioprogram, to a greater or lesser extent, and in these cases we will expect to find, common or even systematic `errors' which, in orthodox learning theory, will be attributed to `incorrect hypotheses' formed by the child, but which, I shall claim, are simply the result of the child's ignoring (because he is not ready for it) the data presented by speakers of the target language and following out instead the instructions of his bioprogram. (Bickerton 1981, p. I35)

By way of second example, the account Bickerton gives in the preceding quotation of first language acquisition also appears to hold up well for second-language acquisition, where there are now a large number of studies supporting the view that the language learner's `interlanguage' possesses many features not derivable either from the learner's first language or the language being aimed at, but which are spontaneously generated. A leading researcher in this field expresses the position as follows:

The claim I am making is . . . that in the absence of attempts to control the learning process, as in the classroom, the sequence of development of the interlanguage syntax of learners, whatever their mother tongue and whatever the target language, will show general overall similarities particularly in the early stages. This claim implies that there is a property of the human mind which determines the way language learners process the data of language to which they are exposed, whatever the superficially different properties of the data may be. It is the object of interlanguage studies to discover what these processes are and what the `natural' sequence of development is. (Corder 1981, p. 72 and see endnote 11)

My third and and perhaps most striking example concerns deaf children. There is a body of evidence, forcefully presented in Feldman, Goldin Meadow, and Gleitman 1978, which, if true, shows that isolated deaf children, whose hearing parents who do not use sign language, nonetheless spontaneously develop signing systems as means of communicating with their hearing interactants. Here are two quotations, one setting the scene, and the other bearing directly on the argument with Itkonen:

All of the subjects we studied developed the ability to communicate with manual gestures. All began, like hearing children, simply by pointing to the people and things around them, engaging their listeners by eye contact (that is, they did not point if no one else was looking). These points continued to bear a heavy burden in the subjects' communications throughout our observation. In comparison, gesturing tends to decline for hearing children as speech emerges. As our subjects grew older, two striking changes in their communicative attempts took place. They began to use the points in combination, in ways that seemed clearly intended to convey semantic relations between the referents of the points. And they began to invent motor-iconic gestures that seemed to specify predicates of various kinds. These gestures, which we have called characterizing signs, soon came to be used in combination with the points in structured ways. Such extensive elaboration of the gestural mode is not typical for hearing children. As observers, we came inescapably to understand these signs and sign combinations as communicative acts as comments about the world, queries, requests, and demands addressed to the listener. All in all, it is impossible to observe these subjects and deny seriously that they achieved considerable communicative skill in a gestural mode: each subject must be credited with an idiosyncratic `home sign' system that puts him into mental contact with those around him. (Feldman et al. 1978, p. 375)

and again:

Our subjects have no richly elaborated language system to learn; they must create, as well as acquire, a means of communication . . . in so far as an actively teaching individual is implicated in the acquisition process . . . one is lacking in this situation: though the caretakers are cognitively mature they are no more fluent than the children in the manual system. Finally, and probably most important, there is no clearly co operating communicative partner. The adults around our subjects do not accept the manual communicative mode in ways that would seem to be conducive to joint, active, attempts to specify and enrich the system. No conventions for expression seem to be adopted by mutual agreement. If the child makes a gesture, we do not observe the mother gesturing it in return. Even to the extent that the adults do gesture, they are quite inept, dependent on props, and often unaware that they are being expressive. If humans have the capacity to create communication systems on their own, this hardly seems the ideal situation in which to try to do so. (Feldman et al. 1978, pp. 405 6).

The question now to be considered is what bearing the kind of material presented in the preceding three examples has on Itkonen's arguments. I am putting forward the three examples as falsifications of the claim that linguistic rules are centrally constituted in common knowledge conventions. Itkonen might make any of the following four counter claims:

  • (1) as described, the rules of creoles, interlanguages, and signing systems are not clearly private rules; so they do not falsify the claims at issue
  • (2) as described, the rules are indeed private rules, but are peripheral to the core of language and so can be accepted as things which sometimes occur
  • (3) what has been described are not rules, but (say) processes or regularities and so do not bear on the question at issue
  • (4) as described, these are indeed private rules, and extensive and systematic ones, but they belong to a different level of reality than that with which the private language arguments are concerned, and hence do not falsify the claims made on the basis of private language arguments.

Let me consider each of these counter-claims in turn.

(1) Clearly, we are able to study the output of creoles, interlanguages, and spontaneous signing systems, and in the first two cases, at least, call upon speakers' intuitions in formulating the rules of their systems. Again, speakers of these languages can make themselves understood by nonspeakers: this is especially striking in the third case. So if there are rules here, they are not radically private: they are constructible and reconstructible by others than their users. But that is not to say the rules are common knowledge: they are strictly idiolectal, and, if shared, they are shared distributively, not collectively. I cannot see that the intellegibility of the users of these rules ipso facto shows that those who understand them are therefore parties to a convention of mutual knowledge with their users. If it did, it would obliterate any distinction between creoles, interlanguages, and `home sign' systems, on the one hand, and standard languages, on the other, and it would remain to specify the differences. In sum, if the rules of creoles, interlanguages, and `home sign' systems are not private in the relevant sense, then that sense is surely irrelevant to linguistics. But if they are not private in the relevant sense, it does not follow that they are commonly known rules: there is a difference between there being outward criteria, that there are rules and what the rules are, and those rules being common knowledge.

(2) It could be accepted that the rules considered above are private, in the sense that they are rules which are or could be followed by one speaker only, yet doubted that these were central cases. Using Wittgenstein's version of the fallacy of composition argument, it could then be said that here we have cases which can sometimes happen, but not always happen. Against this, the creole situation is clearly one where the language of a whole community is transformed, effectively in a single generation, and that surely undercuts the argument.

(3) Itkonen defines rules partly in terms of normativity and social control. In so far as the output of creoles, interlanguages, and `home sign' systems is not normative for anyone, including their users, it might be claimed that the output is the result of processes not properly described as rule governed (guided).(See endnote 12). But since no one is going to deny that the processes and their output is linguistic in character, denying creole, interlanguage and `home sign' processes the status of rules only serves to sever the connection between language and rules. That is something unacceptable to most linguists and philosophers, and, more pertinently, would deprive Itkonen's philosophical analysis of the general application he wants for it in the domain of linguistics.

(4) The last mentioned problem is also raised by counter claim 4. It could be claimed, with justice, that when Bickerton, Corder, and Feldman et al. talk about rules or grammars, they are talking about objects or processes they believe to be mentally represented but probably non introspectible and constituted in part by innate cognitive structures. Itkonen could then say that this is not what he is talking about, which is the line he takes in response to Fodor 1976's defence of private language:

Fodor's putative defence of private language rests on his identification of `private language' with `unconscious, largely innate cognitive structure'. Yet there is nothing to justify such an identification. Just as well one might call the internal structures of individual atoms their `private languages'. (Itkonen 1978a, p. 320, n. 55; cf. Fodor 1976, p. 74, n. 15)

Itkonen now has two problems. First, if he agrees that there are innate and other cognitive structures which can legitimately be talked about and theorized (e.g. in psycholinguistics), then he has to tell us which concepts may legitimately be used in this enterprise, currently dominated by a computational paradigm in which concepts of `rules' and its cognates (such as `procedure') are central. Secondly, there follows on from this the problem for him of showing that he has not simply theorized the domain of a discipline parallel to psycholinguistics - namely, sociolinguistics but has indeed characterized the nature of language as the object of some third, and disjoint, study namely, autonomous linguistics. Itkonen sees himself in Itkonen 1978a as having theorized the domain of autonomous linguistics (see endnote 13). His views on nonautonomous linguistics (including under that both psycholinguistics and sociolinguistics) are set out in Itkonen 1983. My own view is that Itkonen 1978a is best reconstructed as a contribution to the study of the social side of language. But even if Itkonen admitted my empirical material as showing the operation of private rules, but argued for its non falsifying character by consigning it to a different field of study, the upshot would be to make the private language arguments irrelevant to that field of study. It could not then also be the case that (to repeat a quotation), transformational grammar `maintains a conception of language which is demonstrably equivalent to the private language conception' (Itkonen 1978a, p. 113), granted identity between the character of the objects Chomsky is interested in, and those Bickerton, Corder, and Feldman et al. (see endnote 14) are concerned with. Furthermore, whether Itkonen has succeeded in theorizing the domain of autonomous linguistics or only of a part of social linguistics, it would seem that the material I have presented is sufficient to establish the claim that there are linguistic facts which are not social facts, and hence to undercut any thoroughgoing sociologism. As I have constructed it, linguistic facts which are not social facts are psychological (mental) facts, rather than abstract facts, though at least some of those psychological facts are natural facts. To round out the picture, and in particular to give proper recognition to that hermeneutic moment of language and linguistics which Itkonen is one of the very few to theorize, I now want to look at the other side of the coin: for indeedthere are facts about language which are social facts, but not linguistic facts.

Part 6. Languages as Social Facts which are not Linguistic Facts

I take it that the view that languages are social facts, but not linguistic facts, is expressed by Chomsky when he says in a statement previously cited that, `Such notions as "the English language" are not linguistically definable, but are rather socio political in nature' (in Piattelli-Palmarini 1980, p. 313). This view arises not only from a theoretical preference for writing grammars for idiolects or, more precisely, grammars for mentally represented objects but from positive doubts about the linguistic definability of `German', `Dutch', etc. (see e.g., Chomsky 1980, pp. 217 18). By saying that languages are sociopolitical facts, however, Chomsky allows us to avoid resort to the kind of nominalist position already discussed, while leaving it open what sort of sociopolitical fact a language like English is.

In this section, I will outline an account of languages as sociopolitical facts in a way which is consistent with the way Chomsky treats linguistic facts, and consider some advantages of the approach and the possibility of combining it with two other approaches already considered - first, the naturalism discussed in Part 2 above; and, second, an account which takes at least some linguistic facts in sensu diviso (individual linguistic rules) to be social facts.

The basic idea is this: a language is an (intentional) object of (mutual) belief, appropriately studied hermeneutically within a sociology of language. (See also on this website the essay, ",What is English if not a Language? ")

What is involved in this idea and what its consequences are can begin to be appreciated if, for a moment, the standpoint of an anthropologist studying an alien culture is adopted. The anthropologist does not wish to prejudge questions of truth and falsity, and hence records what native informants say without prejudice to questions of truth. So if the anthropologist asks a (male) informant S `What language do you speak?' and is informed `English', it is proper to record this information in form (2) not form (1):

  • (1) S speaks English;
  • (2) S believes he speaks English.

If it is granted that there is indeed something linguistically characterizable call it a languageithat S speaks (see endnote 15), then (2) can be expanded to yield its logical form (3):

  • (3) S believes, of the languagei that he speaks, that it is English.

In philosophical terms, `English' appears in (3) as the intentional object of a speaker's belief, such that the intentional occurrence of `English' is referentially opaque: from the fact that some S believes he speaks English, it does not follow that he believes he speaksEngelska, since he may not know that `Engelska' is the Swedish word for `English'. In contrast, `the languagei' appears extensionally and transparently in (3), so that if this language can be designated by a name such as, say, `Jamaican creole', then we can substitute `Jamaican creole' or `creole de Jamaique' for `the languagei' in (3) without affecting the truth value of (3). It just is the nature of extensional, transparent occurrences of a term that they can be substituted for salva veritate in this kind of way. This is nothing to do with linguistics, just with bog standard treatments of logical form.

The illustration indicates how the languages S speaks and S's beliefs about the language he speaks can apparently vary independently of each other. If this is indeed so, then linguists' problems with the inherent variability of language simply dissolve: the form (3) allows for the possibility that a set of different S's may believe they speak the same language, but at the same time allows for the possibility that the languagei they speak may be different for each one of them. The differences may be minor or major - there are no a priori constraints on how big or small the differences can be.

This analysis may be accepted , yet one may still feel that there must be some connection indeed, an intimate connection between the language S speaks and his belief as to what language it is, for otherwise we should sometimes find cases truly represented by something like (4), which is something we do not actually find.

  • (4) S believes, of the Germani he speaks, that it is English.

Two comments in response. First, it is by no means clear that we do not find cases like (4).

First, to argue from an example, it is commonly said that some children of West Indian origin have encountered problems in the British educational system because they think they are speakers of English but their teachers do not. Their teachers think of them as speakers of creoles with insufficient resemblance to, or mutual intelligibility with, English to be counted as English (see Edwards 1979).

Second, if you consider how, developmentally, speakers might acquire beliefs about which language they are native speakers of, it is intuitively obvious that they do not acquire them by inspecting the languagei they speak or by matching it with a named language. Rather, they are told which language they speak in the context, for example, of overhearing speakers of a foreign language for the first time (`They are speaking French. We speak English'). The belief is acquired on authority and, consequently, is a mutual belief: I believe I am a speaker of English because you believe I am, and I believe you believe you are a speaker of English, and so on. (It is not, however, necessary that such a belief be acquired at all: see endnote 16 for the argument and evidence).

Let me expand on this second comment. S's belief in his English speakerhood cannot be based on inspecting the languagei he speaks in so far as this takes the form of a mentally represented grammar, since this is, according to Chomskyan hypothesis, simply not available to introspective inspection. Nor can it be based on observation of the output of a mentally represented grammar, the sentences S uses in his utterances. For these do not come labelled as `sentences of English' nor are they painted red, white, and blue. Again, in the normal course of events, S will not acquire the belief that he is a speaker of English by matching the output of his mentally represented grammar with the output of a named grammar of English (but see endnote 16 for possible exceptions) though S might well conclude that he is a poor speaker of English by matching his output with the output of a prescriptive grammar of English (more on this theme below).

In any case, though individuals distributively considered (i.e., each considered in turn, facts about the others being held constant) could acquire their beliefs about their English speakerhood by a process of matching, that is only because English has already been defined. For all individuals in a speech community considered collectively, such could not be the case. Rather, when we consider individuals collectively it is clear that it is they who can define what English is but only consequent upon (and subsequent to) believing of themselves and each other that they are speakers of English. That is to say, the reality of (the) English (language) is constituted as a sociopolitical fact through its appearance as the intentional object of speakers' mutual beliefs. (See endnote 17). It is its place in a package of mutual beliefs which sustains any individual speaker's belief in his or her English speakerhood. (See endnote 18)

More formally:

  • (A) S believes, of the languagei S speaks, that it is English; and
  • (B) S1 believes, of the languagei S1 speaks, that it is English, if and only if
  • (i) S believes S1 is a speaker of English; 
  • (2) S1 believes S is a speaker of English;
  • (3) S believes S1 believes S is a speaker of English;
  • (4) S1 believes S believes S1 is a speaker of English.

In this analysis, S and S1 are being used as names of individuals. This can only be a provisional approach: clearly I have beliefs about co-linguals whose names I do not know, so I must identify them in some othere way.

Now, whereas in one sense it follows that S and S1 arespeakers of English in the sense that all and only those who believe themselves to be chic are chic in another sense, it does not. In particular, from the mutual belief in their English speakerhood it does not follow that the languagesi (the linguistically characterizable languages or mentally represented grammars) of S and S1 are type identical or even isomorphic (in the way Labov proposes and as I discussed in the previous section). This brings us back to the distinction made earlier between the extensional and intentional occurrences of the concept of language between the language-subscript-i of which one has knowledge and the beliefs that one has about it - that it is, for example, the language English.

It must be added at this point that speakers clearly do not have just one metalinguistic belief, that they are speakers of English (or whatever language they think they are speakers of). At the very least, this belief will itself be connected to other explicit or implicit beliefs. It may be that a set of speakers uses mutual intelligibility as a basis for justifying their beliefs about their co-lingual status. It may be that they define their language in terms of a small number of stereotypical or distinctive grammatical features. It may be that they operate a division of intellectual labour in which grammarians are left to define their language by writing a grammar of it. In this connection, the point I wish to stress is that in the last case mentioned (which is the case for speakers of standard languages), speakers will come to acquire beliefs about individual linguistic rules of their language (in sensu diviso) from their grammarians and their teachers and that these rules will thus become objects of mutual belief or common knowledge in the way Itkonen envisages as the standard case. But they become common knowledge consequent and subsequent to speakers holding beliefs about their language in sensu composito (as being a big bundle called English, in the example I am using). Furthermore, in so far as speakers have both a belief that they share a language and have non isomorphic mentally represented grammars (or languagesi), then defining the individual rules of their language (in sensu diviso) will necessarily involve choice, controversy, and prescription since they do not start out from the same internally represented system. Inevitably, there will emerge a politics of the language bounded by whatever starting beliefs there are about common membership in a language community. At this level (and at this level of analysis alone) the kind of sociological analysis offered, for example, by Bourdieu (1982) becomes appropriate.

In more general terms, I see at least five advantages in approaching languages, in the first instance, as the intentional objects of mutual belief, where these beliefs are held about the object consdiered as a whole in sensu composito rather than considered rule by rule in sensu diviso.

The first advantage I see is that it allows for (and explicates) the possibility that speakers who agree that they are all speakers of English (who mutually believe each other to be speakers of English) can disagreeabout what English is (what is grammatical in English, etc.), and this is a perfectly genuine disagreement, which would be unintelligible if we only allowed individual knowledge of language or beliefs about individual rules into our ontology: on that ontology, there could not be disagreement, only difference. It is one task of the sociologist of language to explain the conflicts over language which take place between people. The present analysis elucidates one condition of possibility of one class of conflicts those conflicts between people who both believe they share an identity and define it differently.

A second advantage of the approach outlined is that since beliefs are things we can acquire, add to, and change, the approach allows for people to acquire, add to, and change their beliefs about what English is. This secures the conditions of possibility of the rationality of arguments about language, including prescriptivist arguments. Though, in the first instance, prescriptivists influence speakers' beliefs about language, it is possible through time that they have an influence on knowledge of language itself in the full-on Chomskyan sense. I present a detailed case for this actually happening in Pateman 1982c (which should in time also appear on this website)

A third advantage of the approach is that it not only specifies conditions of possibility for the operation of linguistic prescriptivism, but also provides a solution to the problem of the limits of the operation of prescriptivism in space and time. When people do not believe themselves to be speakers of some language or other, prescriptivists have no audience; and whatlanguage people believe themselves to speak determines which prescriptivists can influence them: French and Italian dialects may form a continuum across the Franco Italian border, but the influence of theAcademie Francaise does not pass the boundary beyond which speakers do not regard themselves as aspirant speakers of its version of French. This may be a geographical-political boundary, or a social one.

A fourth advantage of the approach, certainly bearing on a problem which has interested linguists, is that it handles the phenomena of hypercorrection with ease. To say that a speech form is hypercorrective is to say that the explanation of it must make necessary reference to some modification to their speech output which speakers make in virtue of false beliefs they hold about the language they speak or wish to speak. More precisely, though still quite informally, and using English as an example, hypercorrection occurs when speakers who believe themselves to be speakers of English believe that English (or `good English') possesses features which they can only realize in their speech by adaptive modification of their natural speech patterns (the output of their knowledge of language) and who modify their speech accordingly; but where other speakers (who the first group of speakers would regard as authoritative on English or models of English speakers) do not believe English to possess such features (and do not realize them in the relevant environments). This explication of hypercorrection successfully distinguishes it from, for example, speech forms which are the product of hypercorrection by previous generations (but which do not involvemodification to present speakers' speech), and, again for example, distinguishes hypercorrection from target modification of speech where the target aimed at actually exists. It also indicates quite clearly how a social fact can also be a political fact involving unequal distribution of resources permitting the exercise of domination. For in any given community of mutual believers (that is to say, in the present context, a community constituted by all and only those who believe themselves speakers of some language, L), it does not follow that all are equal with respect to their ability to articulate or impose a definition of what the language is, and whereas articulation will necessarily draw on linguistic cognitive resources, imposition can draw on resources which are not linguistic cognitive at all (material resources; deference to the materially powerful, etc.).

In general, it is a fifth advantage of the intentionalist approach to the reality of language which I have outlined that it allows an approach to questions not only about linguistic standardization but linguistic hegemony, such as concerned Gramsci (Lo Piparo 1979) and concern Bourdieu 1982, but without abolishing the distinction between what is often called the `external history' of language(s) and its/their `internal history', centrally because of the use of the distinction between knowledge of language and beliefs about language (what Gramsci distinguishes as grammatica immanente and norme: Gramsci 1975, p. 2342; see now Gramsci 1985, pp. 164-95)

In short, the intentionalist approach to languages (langues) is one which captures the hermeneutic moment of languages as non linguistic, social facts: the moment in which actors define their social and cultural world.

I wish to consider in the remainder of this essay four possible claims about languages as objects of belief and languages as natural kinds, as a preliminary to attempting (in chapter 4 of my Language in Mind and Language in Society: to appear on this website) a dynamic theory of the relation between, on the one hand, linguistic entities as psychological objects and, on the other, cultural practices on language. The four accounts I wish to consider, albeit briefly, are these; I list them and then discuss each in turn:

  • (1) Languages have no existence as natural kinds, only as cultural kinds (objects of mutual belief); 
  • (2) Languages exist only as natural kinds;
  • (3) Languages exist as natural and cultural kinds, and the relationship between them is one to one; 
  • (4) Languages exist as natural and cultural kinds, but cultural kinds of languages are discontinuous with or additional to their natural reality (there is no one to one relationship).

(1) Culturalism or sociologism is a main tendency within social theory, out of which, for example, have come anti psychiatric culturalist theories of schizophrenia, which hold that there is no more to schizophrenia than there is to witchcraft; both are no more, and no less, than socially constructed. The same view applied to the case of languages would mean that though, for example, standard languages could be studied as the products of cultural practices of standardization, there could not be any intermediate study between that study and the study of idiolects. Typological linguistics would, it seems to me, have to shup up shop because resting on unsatisfactory ontological doctrines.

(2) Central to twentieth century Marxism, notably in the work of Lukacs 1971, has been the critique of the naturalization of cultural phenomena, and this critique also inspires the anti psychiatric critique of the disease entity status of schizophrenia. It does, indeed, seem to me that there is an illegitimate, everyday reification of languages as natural kinds in which languages are thought to possess properties which make them (essentially) what they are, and which are or ought to be transparent to the understanding. Most talk of the corruption of languages proceeds from some such reification.

(3) As far as I know, and errors of diagnosis excepted, haemophilia and haemophiliacs stand in a one to one natural to cultural relationship, though, of course, how haemophiliacs are socially defined can vary without limit, within the bounds that haemophiliacs must be all and only people who have haemophilia. It is clearly not the case that the cultural kind, English, stands in such a one to one relationship with a natural kind, English: the whole point of the analysis in this essay has been to theorize a situation in which not all speakers of English speak the same language (there is always idiolectal and dialectal diversity). This does not, however, preclude the truth of (4):

(4) It could, indeed, be the case that languages have some natural mode of existence as one ur language (Bickerton's bioprogram, for example) or as prototype clusters or in terms of accessibility, or in some other manner yet unconceived of, but that these natural modes do not exist in a pure form and/or are not mapped and represented one to one in cultural definitions of languages. Chomsky writes of actual languages as `impure' (Chomsky 1975a, p. 28; see also Chomsky 1981, ch. 1), for instance. Analogously, it is possible to think that there is some disease entity, schizophrenia, but that this is manifested in `impure' forms, not least because of the unfortunate tendency of psychiatrists to diagnose as schizophrenic some schizophrenics together with lots of paranoids, depressives, rebels, and exasperated adolescents. With the progress of science we could hope to get things right. If anything like this is true in the case of languages, then there is space for a linguistics which is a science of enlightenment, pointing up and perhaps aiming to close the gap between natural kinds and cultural definitions of languages, though I do not want to advance this particular idea other than very speculatively.

This concludes my outline of an intentionalist approach to the ontology of languages, which both recognizes the hermeneutic moment in which reality is socially constructed and allows for (and this is its dualist aspect) the real existence of languages as natural kinds. The relationship between psycholinguistic processes and structures, and cultural practices on language, is a further question, and it is this question which I consider in the essay "From Nativism to Sociolinguistics" (to appear on this website; in the meantime to be found at chapter 4 of my book Language in Mind and Language in Society)


1) This follows Chomsky, as the following statement indicates: `I have argued that the grammar represented in the mind is a "real object", indeed that a person's language should be defined in terms of this grammar', and he goes on to make a reductionist claim which Smith and Wilson also follow (though Chomsky himself also has other, and better, ideas for defining languages than this one: see Chomsky forthcoming): `and that the vague everyday notion of language, if one wants to try to reconstruct it for some purpose, should be explained in terms of the real systems represented in the minds of individuals and similarities among these' (Chomsky 1980, p 120).

2) This was pointed out to me by a linguist, Margaret Deuchar. But that such a reading is not a natural one to make in the context in question is supported by the fact that another linguist, Nigel Love, has independently read ch. 1 to embody just the kind of problem I have noted, and has discussed it at some length (Love 1981).

3) In this case the Nominalism is both reductivist and psychologistic.

4) I should note at this point that Katz denies that he is or ever has been a Kantian (Katz, personal correspondence 1982), though I think he is a Kantian despite himself.

5) Katz 1985, p. 15 n., cites one of Itkonen's essays but does not discuss his work.

6) Itkonen's contribution on this question has provoked considerable discussion: see, notably, Dahl 1980 (an exchange with Itkonen) and Linell 1976.

7) Katz 1981 discusses this past debate in a footnote (n. 45, pp. 19 20).

8) See, inter atia, Bickerton 1971; D'Agostino 1979 (my response, Pateman 1980d); Doroszewski 1933; Gazdar 1977; Koerner 1973; Romaine 1980, 1981, 1982; Sampson 1980b, ch. 2; Sanders 1979; Washabaugh 1974; Harris 1985. When I write `Saussure' in this section, I mean the Bally/Sechehaye text of the Cours de linguistique generale (trans. as Saussure 1983).

9) Note that for Itkonen, though linguistic rules are known with absolute a theoretical certainty by those who participate in them (since there is no extra linguistic spatio-temporal reality to which such rules correspond), speakers' intuitions about the rules of which they have certain knowledge are fallible and corrigible, though even here, 'it is conceptually necessary that disagreement of this kind [conflicting intuitions TP] is the exception, and not the rule' (Itkonen 1978, p. 137).

10) Itkonen (personal communication 1982) thinks rules must hold minimally between three people (as trilectal rules), `because the refutation of private languages . . . is based on the notion of social control . . . which makes it possible to recognize errors as errors; social control is, roughly, majority opinion, and to have a majority, you must have at least three persons, preferably more.'

11) Corder goes on to consider parallels with first language acquisition (Corder 1981, p. 74) and elsewhere has related second language acquisition to creolization (see Corder and Roulet 1977).

12) Note added. In this paragraph I run together the idea of normativity as socially prescriptive and normativity as a standard of consistency. This mistake is picked up and corrected "Wittgensteinians and Chomskyans" (on this website)

13) For scepticism about the possibility or necessity of autonomous linguistics, see Linell 1980b.

14) Feldman et al. specifically acknowledge their indebtedness to Chomsky, `whose writings on the formal nature of language and on language acquisition quite obviously provide the general framework within which we thought about our subjects' communications' (Feldman et al. 1978, p. 413)

15) So that the i could mean `internalized' and `languagei' mean `internalized language' as characterised in Chomsky 1986, ch 2 and forthcoming.

16) Imagine a speaking child transported as a slave and isolated from other members its community later discovering its linguistic identity by matching its native speech with a linguist's grammar. This possibility is not altogether fanciful: some deaf people are just learning that they are users of British Sign Language, a fact of which they were previously ignorant, though, of course, it did not stop them signing. (I owe this information to Margaret Deuchar.)

17) There are several theories of intentional objects of belief. For my purposes I require only that (1) an object of belief may not exist as anything other than an object of belief (so that though people believe in witches or English, it does not follow that witches or English exist) and (2) that the objects in the that clauses of beliefs are referentially opaque. See Woodfield 1982 and Searle 1983 for current debates on intentional objects.

18) In the literature, `mutual belief, `mutual knowledge', and `common knowledge' are used interchangeably; though Lewis and Itkonen speak of `common knowledge', I use `mutual belief' here in order to get a stronger contrast between `knowledge of language' and `beliefs about language'.

This website version first published 2006. Revised from chapter 3 of Trevor Pateman, Language in Mind and Language in Society (Oxford University Press 1987). Click here for the Bibliography [to be added shortly] for this essay, which is in fact the Bibliography for the complete 1987 book.