But the challenge to the formalist approach in respect to validity is quite different. It is not tolerant and neighbourly at all, but a matter of competing claims for the same territory. It is not just an issue of delimitation but of definition, and proposes a functionalist one in opposition to a formalist one. The argument here is that it diminishes the very study of language to reduce it to abstract forms because to do so is to eliminate from consideration just about everything that is really significant about it and to make it hopelessly remote from people's actual experience. Language, the argument goes, is not essentially a static and well-defined cognitive construct but a mode of communication which is intrinsically dynamic and unstable. Its forms are of significance only so far as we can associate them with their communicative functions. On this account, the only valid linguistics is functional linguistics.
But, as was indicated in Chapter z, there are two senses in which linguistic forms can be said to be associated with functions, and therefore two ways of defining functional linguistics. Firstly, we can consider how the linguistic code has developed in response to the uses to which it is put. In this sense, functional linguistics is the study of how the formal properties of language are informed by the functions it serves, how it encodes perceptions of reality, ways of thinking, cultural values, and so on.
Secondly, we can think of the form-function association as a matter not of encoded meaning potential but of its actual realization in communication; and here we are concerned with the way language forms function pragmatically in different contexts of use. In this case formalist linguistics is challenged not because it defines the language code too narrowly without regard to the social factors which have formed it, but because it defines language only in reference to the code, without regard to how it is put to use in communication. The argument here is that linguistics should extend its scope to account not only for the knowledge of the internalized language of the code, or linguistic competence, but for the knowledge people have of how this is appropriately acted upon, or communicative competence.
These two senses of functional linguistics are frequently confused, and there has sometimes been a tendency to suppose that if you define the code in reference to the communicative functions that have influenced its formation over time, then it follows that you will automatically be accounting for the way in which the code functions in communication here and now. But to do this is to equate the semantic potential of the code with actual pragmatic realizations of it in communication.
Functional linguistics, in both senses, considers language as an essentially social phenomenon, designed for communication. There is no interest in what makes human language a species-specific endowment, in those universal features of language which might provide evidence of innateness which were described in Chapter 1. The concerns of functional linguistics are closer in this respect to the reality of language as people experience it, and it is therefore often seen as more likely than formal linguistics to be applicable to the problems of everyday life. Opponents might argue that this is only achieved at the expense of theoretical rigour. This raises the general question of how far relevance and accountability are valid considerations in linguistic enquiry, and this will be taken up again a little later. It also raises the question of what the source of linguistic data should be, and it is to this matter that we now turn.
The data of linguistics
There are, broadly speaking, three sources of linguistic data we can draw upon to infer facts about language. We can, to begin with, use introspection, appealing to our own intuitive competence as the data source. This is a tradition in linguistics of long standing, and essentially makes operational Saussure's concept of langue as common knowledge, imprinted in the mind like a book of which all members of the community have identical copies. So if linguists want data, as representative members of a language community they have only to consult the copy in their head. Most grammars and dictionaries until recent times have been based on this assumption that linguistic description can be drawn from the linguist's introspection. And it is not only linguistic competence which is accessible to introspection, but communicative competence as well, so the argument is that the conventions that define appropriate language use can also be drawn from the same intuitive source.
If, however, there is some reason to doubt the representative nature of such intuitive sampling, there is a second way of getting at data, namely by elicitation. In this case, you use other members of the community as informants, drawing on their intuitions. And again, this might be directed at obtaining the data of the code or its communicative use. Thus, you might ask informants whether a particular combination of linguistic elements are grammatically possible in their language, or what would be an appropriate expression given a particular context.
Introspection and elicitation can be used to establish both the formal properties of a language and how they typically function in use. But in both cases the data is abstract knowledge, and not actual behaviour. They reveal what people know about what they do but not what they actually do. If you want data of that kind, the data of performance rather than competence, you need to turn to observation.
The development of computer technology over recent years has made observation possible on a vast scale. Programs have been devised within corpus linguistics to collect and analyse large corpora of actually occurring language, both written and spoken, and this analysis reveals facts about the frequency and co-occurrence of lexical and grammatical items which are not intuitively accessible by introspection or elicitation.
It would seem on the face of it that this is a much more reliable source of data. It is surely better to find out what people actually do than depend on intuitions which are often uncertain and contradictory. Claims have indeed been made that these large-scale observations reveal patterns of attested usage which call for a complete revision of the existing categories of linguistic description, which are generally based on intuition and elicitation. Corpus linguistics, in dealing with actual behaviour, clearly has an affinity with functional linguistics in that it too claims to get closer to the facts of 'real' language.
There is no doubt that corpus analysis can reveal facts of usage, the data of actual linguistic performance, which throw doubt on the validity of any model of language based on the idea of a stable and well-defined system. The elaborate picture it presents is very different from the abstract painting proposed by the formal linguist. If language use is indeed a rule-governed activity, as is often said, the rules are not easy to discern in the detail. And it is also true that this detail is not accessible to introspection or elicitation. Even a limited corpus analysis can show patterns of occurrence of which language users, the very producers of the data, are unaware. Corpus linguistics transcends intuitive knowledge and in this respect can be seen as a valuable, and valid, corrective to unfounded abstraction: a case of description influencing theory for once, rather than the other way round.
But the claims of corpus linguistics can be questioned too. The facts of usage revealed by computer analysis, for example, carry no guarantee of absolute truth. The intuitions that people have about their language have their own validity as data. These conceptual constructs are also real, but the reality is of a different order.
One example of this is the way lexical knowledge (in some areas of vocabulary at least) seems to be organized semantically in terms of prototypes, and these cannot be observed, but only elicited. Thus, when a group of English-speaking informants were asked to give the first example that came to mind of a more inclusive category of things they showed a striking unanimity. The word 'bird' elicited 'robin' (rather than, say, 'chaffinch' or 'wren') and the word 'vegetable' elicited 'pea' (rather than, say, 'parnsip' or 'potato'). For these informants, then, a robin is the prototypical bird, a pea the prototypical vegetable. But this conceptual preference does not correspond with how frequently these words actually occur in a corpus. The same point can be made about grammatical structures. If English-speaking informants are asked to provide examples of a sentence, they are likely to come up with simple subject-verb-object (SVO) constructions ('The man opened the door'; 'John kissed Mary'). These, we might say, are prototypical English sentences. But they are unlikely to figure very frequently in a corpus of actual usage. Since people do not use simple sentences like this very often, they do not have much reality as observed data, but they may have a significant psychological reality nevertheless. They may be evidence of competence which is not reflected in the facts of performance.