language as a cultural emergent property
The empirical, the actual and the real
The actual
This, then, is what we intend to convey when we refer to what is 'real' about language, while we trust that we have already made it clear what we mean by 'the empirical' in relation to language- those aspects of language which are capable of perception by the human senses. So do we need to introduce a third domain of reality? We think so. Bhaskar's claim allows us to distinguish between what causes things to happen in the world, which we often cannot observe (the real), and the instances of experience of which we are immediately aware (the empirical). The intervening domain, the actual, is that part of reality- events- which actually happen, as distinct from all those things which might have happened but did not. The reasons why the real does not inevitably coincide with the actual are attributable to the fact that the world is an 'open system', and the social world, in particular, 'is not a mechanism with fixed, indispensable parts and determinate relations between parts, pre-set preferred states and pre-programmed homeostatic mechanisms' (Archer 1995: 165). Outside of the laboratory, even the 'governing laws' of the physical sciences are often overridden by contingent variations: this is the very reason why scientists are obliged to exercise control in their experiments, and to contrive to make certain things happen which might never occur without their interventions. In the social world, the distinctive properties and powers of human beings, and of society itself, make it inevitable that underlying realities will not necessarily materialize into actual events.
Thus by 'the actual' in this context, we mean those linguistic phenomena which 'actualize' the powers and properties of language, some of which may not become manifest at the level of the empirical.
Our first example of this level of reality is, somewhat controversially, the different languages of the human population as it exists across the globe and throughout time. For Chomsky, languages in this sense are sociopolitical constructs, and the differences between them are of less interest to the scientist than their underlying similarities. It is customary for linguists to point out that lay classifications of languages, while adequate for everyday use, bypass a problem which is much more acute for the scientist, which is that the boundaries of varieties are notoriously difficult to establish. A typical example is that although 'Dutch' and 'German' are thought of as two distinct languages, 'the variety of German spoken near the Dutch border can be understood by speakers of Dutch who live nearby, though not by speakers of German who live in more remote areas' (Chomsky 1988: 37). The 'commonsense' conception of languages- what the lay speaker thinks of as 'French', 'Russian' or 'Swahili', for example- is rejected by linguists in this tradition partly because it is so difficult to specify precisely what 'knowledge of a language in this sense could mean. Yet the same theoretical approach proceeds entirely as though this problem did not exist. Thus Chomsky writes:
The grammar of English will generate, for each sentence, a deep structure, and will contain rules showing how this deep structure is related to a surface structure.... A person who has acquired knowledge of English has internalized these rules. (p.l06)
Statements such as these seem to us to rely on an independent existence for an entity which can be labelled 'English'- even if for this aspect of the linguistic enterprise the primary focus is on its under lying rather than its surface structure. Lightfoot (1991: 162) maintains that when linguists use an expression such as 'a change in the grammar of Language X', they are using 'a convenient fiction permitting the statement of certain generalizations and ignoring certain types of variation' (cited in Hopper and Traugott 1993: 36), but this approach seems to sidestep a paradox in the argument. For it is inconsistent to claim, on the one hand, that 'Japanese' or 'English' are simply 'convenient fictions' while yet appealing to the native speaker of 'Japanese' or 'English' for grammaticality judgements to aid in the process of identifying features of the underlying grammar.
Reliance on what speakers themselves believe when they say 'I speak Language X' is undoubtedly fraught with difficulties, some of them analogous to the problems of relying on respondents' self-classifications in respect of 'ethnic identity'(ef. Carr 1990). Nevertheless, for language learners it is the differences between languages that are the most salient, in the sense that what is to be learned, the 'target language', differs from what is known, the 'mother tongue'. In this context, the motivation to learn to speak 'English' or 'Chinese' is much more relevant, practically, than the concept of universal grammar. Learners enroll, and teachers are employed to teach, in 'French classes' or at a 'School of English'; learners' progress towards becoming speakers of these languages is assessed and may be formally certificated. Economic and social opportunities, on the one hand, or barriers, on the other, may well follow from the results obtained in these language tests: is everyone involved in these processes deluded to believe that different 'languages', in this specific sense, are 'real'?
The approach we are putting forward allows for a middle way between confining the concept of a language variety to something which exists exclusively inside each speaker's head, and granting a fully discrete and separate status to different languages. It suggests that it is the interaction between human practices and linguistic properties which may account over time for the separation of some varieties and the merging of others. These human practices include both the use of a variety for communication (where the emphasis is on agency) and sociopolitical regulation (with an emphasis on structure). Complexity theory provides a means of visualizing the process of such language change, as Nettle explains:
Consider a many-dimensional space with every dimension representing a different linguistic parameter. Thus, one dimension might be the size of the phonological inventory, another, some aspect of constituent order, another the level of morphological complexity, and so on. Different languages can be seen as occupying points in this space. Each point in the space has a fitness value, representing the extent to which a language at that point in the space will be effective at being learned and used by speakers. Underfunctional selection, languages should be expected to evolve towards the points with the highest fitness values.... [However,) in such a complex space, there is unlikely to be just one optimum position to which one can smoothly ascend from anywhere in the space. Instead there may be a number of positions that are locally optimal and onto which languages may settle.... [S]mall random changes, arising from random variations or social selection, are sufficient to push different languages off into different parts of the state space. (Nettle 1999: 34)
Note that there is no need to invoke any teleological argument here (d. Lass 1980). Languages are not organisms, and have no genetic imperative to 'seek' survival, despite the fact that some chains of changes observed by historical linguists give the appearance of a language, for example, '''trying'' to keep its phonemes apart' (ibid.: 51). Furthermore, our position here is emphatically not one which seeks to reify language varieties. They are not natural kinds, their boundaries are porous, their codification is indeed often a matter of politics rather than linguistics; nevertheless languages as actualized variants do have an existence as cultural emergent properties. '[T]he existence of blurred boundaries and transitional varieties between languages', as Nettle (ibid.: 65) observes, 'does not undermine the identification of the languages themselves. It merely means that we cannot place the boundaries precisely in those cases.' On our account, then, language varieties- 'languages'- are examples of the domain of the actual. The issue of naming and accurately classifying varieties is a different matter from recognizing and acknowledging their existence.
If movements of human groups across the globe had taken slightly different courses from those they actually did in history, if natural disasters, diseases or technological developments in weaponry had obliterated speakers of this language rather than that, if the development of cultural institutions had occurred here rather than there, then the actual languages in existence might look rather different from the collection we have now. In our account, all actual languages are connected in meaningful and significant ways to the generic human capacity for language, and both of these domains are similarly connected to the empirical products of the language capacity and actual language varieties, that is, utterances and texts. Once again, however, a concept of emergence is necessary to demonstrate the nature of such connections.
Our second example of what we mean by 'the actual' domain of 'real language' takes us back to the question of attested linguistic products, and the implications of developments in corpus linguistics. Our key claim here is that one function of a corpus is to make manifest the intersubjective properties of language.
As we have seen, for some linguists, the 'corpus' is of little interest. (Note that by 'corpus' is often meant any collection of attested utterances or written sentences, and not necessarily the macro, electronic corpora) As Halliday (1991:30) notes, Chomsky's theory of competence and performance drove 'a massive wedge between the system and the instance, making it impossible by definition that analysis of actual texts could play any part in explaining the grammar of a language'. As should be clear by now, our own interpretation of realism is quite different from that which disregards empirical data, or language behavior, and for us corpora have an important ontological significance. The conceptualization of corpora as evidence for language at the level of the actual is suggested by Beaugrande (1997), who posits language as a complex system, that '... runs on a dialectic between the virtual control that specifies constraints and the actual control that manifests constraints' (p.127).
However, of at least as much importance in the present discussion as theoretical linguists' objections to corpora is the criticism from within applied linguistics of their pedagogic role, and probably the best-known critique from this position comes from Widdowson. (See also the debate between Owen 1993 and Francis and Sinclair 1994.) Widdowson's criticisms are important, because he provides a persuasive account of why applied linguistics finds so little support for its enterprise in theoretical linguistics. This is because the pedagogic purpose of applied linguistics inevitably requires an understanding of language not limited to '... linguistics applied [which] is, in effect, misapplied linguistics' (Widdowson 2000: 6). His objection to corpus studies is that, while they may well be valuable, surprising and important, they are contributions to descriptions of language, rather than to language learning. As a description of language, a corpus is a 'static abstraction', underpinning a 'linguistics of the attested [which] is just as partial as the linguistics of the possible' (ibid.: 7). The corpus does not, of course- in itself- instruct the language teacher as to how language can best be learned, but it is important to note that many applied linguists are demonstrating what the possibilities seem to be. (See, for example, Burnard and McEnery 2000; Hunston 1995, 2002; McEnery and Wilson 1996; Wichmann et a1. 1997). Widdowson is skeptical of its pedagogic value partly because corpus data, especially when presented as the recurring patterns illustrated in concordance lines, take communication out of its ethnographic context. In direct contrast to the objection raised by many theoretical linguists, that the attested language behavior collected in a corpus is too tightly connected to the particular, Widdowson's objection is that it is not tightly connected enough. Corpus studies, he says, show 'only ... the textual traces of the processes whereby meaning is achieved' (2000: 7). In our view, empirical 'traces' are critically important in understanding both social and linguistic processes, and while we would agree with Widdowson that pedagogical issues are not resolved merely by having access to a corpus from which to draw examples of language use, we find in corpora, and the traces they reveal, some important facts about the nature of language.
As a body of attested language behavior, the corpus points both towards the unobservable but nevertheless real structures underlying that behavior. and towards the equally real, but not exhaustive, empirical experience of language in use which we can observe everywhere around us. The most significant feature of a corpus, for the present discussion, is its demonstration of the intersubjective properties of language. Recent developments in corpus semantics (Teubert 1999; Stubbs 2001a) have added to the evidence on which this orientation towards language is based. As Stubbs puts it, '[c]orpus linguists think they have identified a layer of order in these data where none was previously suspected' (Stubbs 2001b: 169), and one of the findings from within this relatively newly visible 'layer' is that 'repeated instances of collocations across a corpus show that meanings are not personal and idiosyncratic, but widely shared' (ibid.: 168).
It is probably worth noting at this point two other features of the corpus linguistic enterprise which are consistent with the realist ontology we are outlining, and the epistemological and methodological implications which we develop elsewhere. The first point is a general one which recurs throughout the literature on corpus linguistics, and it is the recognition that the traditional division of language into syntax and lexis is threatened by corpus findings. As Sinclair (1991: 108) expresses it:
The model of a highly generalized formal syntax, with slots into which fall neat lists of words, is suitable only in rare uses and specialized texts. By far the majority of text is made of the occurrence of common words in common patterns, or in slight variations of those common patterns.
Halliday takes an even stronger position: 'I have always seen lexico-grammar as a unified phenomenon, a single level of "wording", of which lexis is the "most delicate" resolution' (Halliday 1991: 32). Finally, Stubbs, again, suggests how the newly revised distribution of labor between lexis and syntax points to the intersubjective status of language:
If all words have predictable collocates, then lexical meaning is distributed over several words, the meaning of an individual word is influenced by these collocates, and the balance between propositional and connotational meaning shifts towards the latter. Many more words have evaluative connotations and discourse prosodies than previously recognized. (Stubbs 2001a: 242-3)
We shall investigate these propositions in more detail, with reference to the concept of emergence.
The second point, about method in corpus work, is made repeatedly by Stubbs, who, like Brumfit, engages with the responsibility of the language analyst to be in some sense 'objective'. Stubbs avoids quite explicitly any claims to naIve objectivity which would merely rely On the computer to bypass human interpretation. He does, however, claim that 'computers make it more difficult to overlook inconvenient instances' (1996: 154), and sets some store by the 'publicly accessible observational data' (2001b: 163) used in corpus studies, whose statements 'can be replicated and corroborated- or of course questioned and refuted- on independent data' (ibid.). Issues of method in applied linguistic research, and of the claims to a modest epistemic authority which we believe realism makes possible.
In concluding this discussion of different orders of the reality of language, we should point out that there is no rigid distinction to be drawn between the domains of the real, the actual and the empirical. Corpus studies provide a good illustration of the way in which certain features of language which were once not empirically observable migrate into the empirical domain as technology develops. Hence Stubbs' analogy between the technical resources of computing hard ware, concordance software and large corpora, in language description, and the telescope or microscope in descriptions of the physical world (1996: 231). Meanwhile, the development of new language varieties (such as creoles), on the one hand, and the 'death' of others, on the other, represent actualizations of the human capacity for language which may be present in one historical period and absent in another. The processes involved, we would claim, are emergent- from properties of language, properties of social structures and properties of human agency- and it is to the topic of emergence that we now turn.
