Multiple Grammars and MOGUL

Optionality is a central phenomenon in second language acquisition (SLA), for which any adequate theory must account. Amaral and Roeper (this issue; henceforth A&R) offer an appealing approach to it, using Roeper’s Multiple Grammars Theory, which was created with first language in mind but which extends very naturally to SLA. They include in their article some fairly extensive discussion of the relation between their approach and that taken to optionality in the MOGUL framework (see especially Sharwood Smith and Truscott, in press; Truscott, 2006; Truscott and Sharwood Smith, 2004), suggesting that the two are largely compatible and that each can potentially help the other. I agree. The two approaches are presented in very different terms, so some translation is needed, and genuine differences exist, but these factors should not be allowed to hide the underlying similarity. In my comments here I want to consider the similarities and the differences and, in the process, to clarify some aspects of MOGUL and the approach that it takes to optionality.

The approaches are similar, first, in their recognition of a central phenomenon that has not received anything like the attention it deserves. Optionality is so pervasive and so salient that it simply must be accommodated in theories of language acquisition, first and second, as well as in linguistic theories. We approach the problem from somewhat different directions, but the difference may be more a matter of complementarity than of conflict. They are interested in developing a theory of representation that allows for optionality. We develop the MOGUL framework in such a way that any specific theory developed within it will include that crucial element. The framework is open to a variety of specific theories of representation, optionality following from the nature of processing and acquisition within its architecture rather than from any such theory.

This is not to say that MOGUL has nothing to say regarding representation, just that it stays at a relatively high level of abstraction in this regard and that optionality can be derived from the framework with only limited commitment to particular ideas on the character of representations. Nor is it a suggestion that representation is unimportant. As A&R note, we need to explain representation, development, and processing/use, together. None is any less important than the others. So it is good to have accounts that emphasize different aspects.

The two approaches are also complementary in another respect. While both deal with first language (L1) and second language (L2) phenomena and seek a unified explanation, MG has been primarily concerned with optionality in first language acquisition while MOGUL has focused on the second language side. A central tenet of MOGUL is that the same processes operate in L1 and L2 acquisition; there is no loss of UG ‘access’ in SLA, no impairments or deficits, just language development occurring in the presence of an already-established language system. Our work on optionality has naturally focused on the way that new L2 items compete with established L1 items, but the logic of the framework dictates that optionality should appear more generally.

An important feature of A&R’s approach is that once a rule has been established it is never deleted. This idea, which sets up optionality, is shared with MOGUL. It applies to two distinct types of case. First, a particular item in the grammar might be avoided in most contexts but useful in some particular context (explaining its initial establishment). The development of a strong alternative to it will not lead to its removal from the grammar but rather limit the range of its application. Second, an item that is established during processing might later turn out to be entirely inappropriate for the language: mistakes in learning are a natural consequence of noisy data, for one thing. In such cases the item will gradually become irrelevant as its resting activation level declines as a result of disuse and the levels of more suitable alternatives increase with use. But there is no mechanism for removing items. Such a mechanism would be an unnecessary and unwelcome complication. Mistakes made during learning are not eliminated, just made increasingly irrelevant by further processing experience.

Similarities between the two approaches can also be seen in A&R’s favorable comments on Schwartz and Sprouse’s (1996) Full Transfer Full Access theory, as well as the reservations they express about the common idea in this and other SLA theories that learning involves restructuring L1 grammars or resetting L1 parameter values; they suggest instead that we should think of the process as adding options. These comments are all very compatible with MOGUL thinking. We have previously shown that Full Transfer Full Access fits well with our framework and described in some detail exactly how it translates into MOGUL terms (Sharwood Smith and Truscott, 2006). The rejection of restructuring/resetting views of second language learning is not something we have explicitly addressed but is implicit in MOGUL; I am happy to see it explicitly presented.

Probably the most important difference between MG and MOGUL, and the area where further clarification is needed most, is our use of the concept of activation, which does not have a place in Multiple Grammars Theory. We use the term in essentially the manner that is standard in psycholinguistic research: to refer to the availability of a representation for use by a processor. Processors seek to use whatever representations are currently most active. Representations that have been used extensively in past processing have elevated resting levels as a result of this use, giving them a head start toward reaching the current levels that will get them included in current processing. Whether a given representation reaches such a level at a given time depends partly on its resting level and partly on the stimulation it is receiving. This stimulation comes from any other activity in the system, including perceptual representations of input, conceptual representations of a message to be expressed, affective representations of emotions that are associated with current activity, and representations (primarily conceptual) of context, both internal and external.

Activation thus offers a unification of the many factors that affect use of linguistic elements, directly in processing terms. So I would quarrel with A&R’s discussion of alternatives to an activation account of optionality. Most, if not all, are best seen as factors influencing activation level rather than as alternatives to it.

It should be emphasized that activation is not something invented to explain optionality. It is a fundamental concept in psycholinguistics and well beyond. Generative linguistics has not had reason to deal with it because the theorizing has generally abstracted away from questions of language use and the details of how development occurs. As long as the focus is on representation and abstract learnability, the work can go on just fine without any appeal to activation. But if we seek a more encompassing account of language then a notion like activation becomes essential. Another aspect of a more encompassing account should be an explanation for the gradual, quantitative character of development, also a central feature of language learning that must sooner or later be accommodated in any theory. This again is something readily handled by the incorporation of activation level, as we have shown in various places.

A general virtue of using activation level is that it allows connections to be made with work on language processing. That work can then contribute to the study of acquisition. The same can be said, somewhat more cautiously, regarding neural research and theory. Activation is a fundamental concept there, and its uses show interesting parallels with cognitive uses. The concepts are not identical, and some caution is required in efforts to link them, but activation nonetheless shows promise as a potential connecting point between the two kinds of theories; this is one more reason why its incorporation in accounts of second language acquisition is valuable.

A related point of contrast between MG and MOGUL appears in the way the former uses the features productive, lexical, and contextual to explain the various possibilities for pro-drop. I have always been skeptical about the idea that learners make categorical decisions (see Truscott and Wexler, 1989), deciding that one possible value is right for their grammar (or sub-grammar) and the other wrong. The main problem is that it is very difficult to reconcile this type of learning with the noisy nature of the data that learners encounter, including dialect/ register variation and genuine exceptions to the value to be learned, but also sloppy speech, literal noise, distraction, and misanalysis of input based simply on the imperfect state of the grammar at the time. The hypothesis of multiple grammars goes some way toward resolving this conflict, but I do not see how it can do more than reduce the extent of the problem.

I suggested some time ago (Truscott, 1998) that this problem might be handled through a synthesis of UG and instance theory, in which parameter values are interpreted as large sets of accumulated instances, each stored in terms of the parameter (+ or – pro for example) and relevant contextual features. In MOGUL, activation level is, again, at the heart of the explanation. The observed dominance of one of the possible options is to be derived mainly from resting activation level, which is a function of past experience. The derivation has to be based, at least in principle, not on how often the option is instantiated in the primary linguistic data but rather on how often it has been processed – what we have called its internal frequency. In such an approach no decision is made about productivity or dominance. The resting activation levels of the options change as a function of their frequency of use, and future use is determined to a large extent by these resting levels. This is only a partial sketch of the MOGUL approach, which also must include lexical and contextual factors, which receive a related but somewhat more complex explanation. A note of caution is required here, also, as it would be bold, at best, to say we have conclusively demonstrated that this kind of explanation is adequate; we must leave open the possibility that an explanation in terms of more abstract learning is needed.

The approach that A&R describe has considerable promise for advancing the understanding of some key phenomena in first and second language knowledge, learning, and use. There are a number of ways in which their approach differs from that taken in the MOGUL framework, but the similarities and shared goals are greater than the contrasts and the two approaches are in important respects complementary. I believe much can be gained, as they have suggested, by dialogue and cooperation.