Variable competence models are often discussed with some enthusiasm by the staff responsible for teaching MA programmes in Applied Linguistics, although a quick trawl through journals shows that interest is dwindling. I suppose part of their attraction might have come from the fact that Rod Ellis is a keen advocate. When not involved in updating his book on SLA, a very poor reference work indeed, in my opinion, he spends time working on his version of variable competence, which is the best example I’ve ever come across of confusing data with phenomena, and of muddled thinking in general. I think it’s good for students to take a look at bad scholarship, in the hope that it will help them to develop a more critical approach to anything they read. My advice to all MA students is to seriously question everything you’re told.
As usual, we have Kevin Gregg, that unsurpassed guardian of rational thinking, to thank for drawing our attention to the profound weaknesses of this model. His article (Gregg, K. R. 1990: The Variable Competence Model of second language acquisition and why it isn’t. Applied Linguistics 11, 1. 364—83) provides a lucid and well-considered criticism of variable competence models. Here, I give my own summary, leaning heavily, as always, on Gregg. The text is an adaption from my book on Theory Construction in SLA
Variable competence models take a sociolinguistic approach to SLA, and abandon Chomsky’s distinction between competence and performance, viewing competence as variable, not homogenous. Chomsky’s “competence” refers to underlying linguistic (grammatical) knowledge, and performance refers to the actual day to day use of language, which is influenced by a variety of factors, including limitations of memory, stress, tiredness, and so on. Chomsky argues that while performance data is important, it is not the object of study (it is, precisely, the data): linguistic competence is the phenomenon that he wants to examine.
Tarone argues that “capability” underlies performance, and that this capability consists of heterogeneous “knowledge” which varies by speech style, which in turn is related to the task being undertaken by the learner, the interlocutor, and the topic. In other words there is no homogenous competence underlying performance but a variable competence or “capacity” which underlies specific instances of language performance.
According to Tarone (1983a, 1983b, 1984, 1988) L2 learners acquire a continuum of “styles” or rules for the L2, ranging from the most informal vernacular style to the most careful style. When speakers use the vernacular style they pay least attention to what they say, and as a result it will be the least target-like, but the most internally-consistent, the least variable, whereas the careful style is the most target-like, the least internally-consistent, the most variable. Tarone claims that the variability in the production of L2 learners is caused by style-shifting along the capability continuum, which in turn is linked to the amount of attention the learner pays to language form. Thus, according to Tarone, L2 learners have several interlanguages which they use for different types of situation.
Ellis, while not adopting the continuum of styles model, suggests that the L2 learner’s interlanguage grammar allows variable rules, whereas the target grammar does not. The observed variability of L2 learners’ performance is explained by learners at early stages having a non-variable representation for a grammatical property in the L2, but then successively noticing forms in the input which are in conflict with the original representation. The learner does not initially abandon the original representation, but adds the forms to the representation, so that he acquires more and more versions of the original rule. This leads to either free variation (where forms alternate in all environments at random) or systematic variation where one variant appears regularly in one linguistic context, and another variant in another context.
Tarone’s and Ellis’ models are examples of a certain type of general learning theory – learners acquire the L2 by general learning strategies such as noticing novel aspects of the language, hypothesis testing, etc. The novelty is the suggestion that L2 learners use different “styles” of the interlanguage in different contexts (Tarone), or that the way their interlanguage develops has the peculiarity of the sorts of variation Ellis describes.
Such models confront the immediate problem of assuming that the development of L2 interlanguages is random, i.e., they assume that the order in which new items enter a developing L2 grammar, and the subsequent stages of development of that grammar, are random. When the learner notices a discrepancy between the prediction of his interlanguage grammar and a sample of input, this new form enters his grammar and then gradually spreads, either along a continuum in Tarone’s case, or to linguistically or task-determined contexts in Ellis’ case. But which particular forms enter at any particular moment is random, an assumption that goes against a great deal of evidence suggesting that there is a predictable order in these stages of development.
Moreover, some areas of variability in SLA are more quickly resolved by L2 learners than others. The development of negation in English, for example, often starts with “no”, and following this stage, learners soon add other forms, such as “don’t”, “can’t”, “isn’t”, “won’t”, “didn’t”, and eventually seem to have internalised the auxiliary verb + n’t. In this case we observe the stage of free variation giving way to systematic variation, which then gives way to target-like invariance. Other aspects of the L2 grammar, however, appear to be persistently variable (where the target language is not). As Towell and Hawkins say: “Although both Tarone and Ellis “appear” to have a model for handling the variable production of L2 learners, and “appear” to be able to link variability to staged development, in essence, why particular L2 forms enter L2 grammars when they do, and how they spread through L2 grammars, are as mysterious as the notion of “acquisition” is in Krashen’s system” (Towell and Hawkins, 1994: 41).
Thus, the root of the problem of the variable competence model is that there are serious conceptual objections to it. Chomsky made the distinction between performance and competence because it fits his theory of language and first language acquisition. Competence is what we know about language, it is, in Chomsky’s theory, a well-defined phenomenon which is explained by appeal to an innate human capacity which the vast majority of human beings are born with. The job of linguists is to describe the rules that make up this linguistic competence and then test empirically the theory that all languages obey these rules. In order to test these rules, Chomsky says that we cannot just use performance data, since much of it cannot be counted as evidence for or against the theory.
In contrast to Chomsky, Tarone and Ellis argue for an underlying variable competence, but do not define that competence, or make it the source for generalisations or general explanations. It seems to be a theoretical “bolt-on” which sits uneasily with a clearly descriptive approach to what L2 learners do, particularly in classroom settings. While it is perfectly valid to collect data about L2 (classroom) performance, and to suggest patterns and tendencies in this data, the suggestion that this is evidence of an acquired competence seems not only unjustified but incongruous.
As Gregg (1992: 368) argues, Tarone and Ellis offer a description of language use and behaviour, which they confuse with an explanation of the acquisition of grammatical knowledge. If we abandon the idea of a homogenous underlying competence, Gregg says, then we are stuck at the surface level of the performance data: any research project can only deal with the data in terms of the particular situation it encounters, describing the conditions under which the experiment took place. The positing of any variable rule at work would need to be followed up by an endless number of further research projects looking at different situations in which the rule is said to operate, each of which is condemned to uniqueness, no generalisation about some underlying cause being possible.
At the centre of the variable competence model are variable rules. Gregg points out that Labov (1969) first proposed variable rules to account for the variation in the copula in Black English vernacular speech, i.e. it was a tool of sociolinguistic research used with native speakers whose L1 competence was fully-developed. Gregg suggests that such a tool cannot become a theoretical construct used in attempts to explain how people acquire linguistic knowledge. Labov attempted to explain under what conditions and in what environments variants of the copula were observed, and concluded that the variation was systematic. His variable rule amounts to “including a probabilistic element to the grammar of a natural language: where a normal rule states that F appears in environment E, … a variable rule says that F appears in E with probability P” (Gregg, 1990: 371). But while variable rules may serve well as a tool of analysis for sociolinguistics, statements of the type “in given circumstances a given person has a given probability of using a given variant” do not imply that any variable rule has been acquired.
In order to turn the idea of variable rules from an analytical tool into a theoretical construct, Tarone and Ellis would have to grant psychological reality to the variable rules (which in principle they seem to do, although no example of a variable rule is given) and then explain how these rules are internalised, so as to become part of the L2 learner’s grammatical knowledge of the target language (which they fail to do). Tarone suggests that “a new approach to grammar-writing might be needed, because grammars must be written to describe and model several language styles at a single point of time, each style containing both variable and categorical regularities” (Tarone 1985, cited in Gregg 1990: 373).
The variable competence model, according to Gregg, confuses descriptions of the varying use of forms with an explanation of the acquisition of linguistic knowledge. The forms (and their variations) which L2 learners produce are not, indeed cannot be, direct evidence of any underlying competence – or capacity. To quote Gregg again: “There is no rule for producing ellipted or non-ellipted sentences in English. Or rather, there is no such rule in the grammar, although there may very well be pragmatic rules controlling ellipses, just as their are Gricean rules, rules of politeness, etc.” (Gregg, 1990: 374).
By erasing the distinction between competence and performance “the variabilist is committed to the unprincipled collection of an uncontrolled mass of data” (Gregg 1990: 378).
Tarone, in a reply to Gregg, labels Gregg’s approach “rationalist”, and complains that “such scholars, perhaps motivated by “physics envy”, are trying to turn the study of language into an exact science” (Tarone, 1990: 395).
Ellis, in his response to Gregg, claims that there are two styles of SLA research: the theory-then-research style (Gregg’s), and the research-then-theory style (Ellis’), and that these two styles “are reflected in different theories of language, and different research methodologies.” (Ellis, 1990: 384) Ellis says “I remain strongly committed to ethnographic and descriptive studies of classroom learners. It follows that I must also remain a variabalist”(Ellis, 1990: 385).
Ellis goes on to assert that Gregg fails to recognise “the context-dependency of theory construction”, that performance data is the only data SLA researchers have at their disposal, and that the crux of the argument is whether “a mentalist (Chomsky), or functional (Halliday) explanation best fits the facts” (Ellis, 1990: 390).
The arguments of Ellis and Tarone are confused and circular; in the end what Ellis and Tarone are actually doing is gathering data without having properly formulated the problem they are trying to solve, i.e. without having defined the phenomenon they wish to explain.
Both Tarone and Ellis reply to Gregg’s criticisms by invoking some version of the incomensurability argument. Tarone, for example, accuses Gregg of “science envy” and of trying to insist that researchers treat the study of language as an exact science. Tarone’s “argument” is, of course, no reply whatsoever to Gregg’s criticisms. Neither does Ellis’ reply, quoted above, do any better. Ellis’ claims that there are two styles of SLA research: the theory-then-research style, and the research-then-theory style, each with its own different theory of language, and different research methodology, likewise fails to deal with Gregg’s criticisms of the model itself.
What, we need to ask Ellis, constitutes an “ethnographic, descriptive” approach to SLA theory construction? How does one go from studying the everyday rituals and practices of a particular group of second language learners through descriptions of their behaviour to a theory that offers a general explanation for some identified phenomenon concerning the behaviour of L2 learners?