Ever since the derivational theory of complexity (DTC) apparently bit the dust in the late 1960s, experimental psycholinguistics have been afflicted by a dualism at least as troublesome as the mind/brain dichotomy, namely, the grammar/parser distinction.¹

The idea that mentally represented grammar² is something fully dissociated from the human language processor is less than compelling, yet it has implicitly informed much of the last half century's psycholinguistics practice on both sides of the formalist–functionalist divide.

On the one hand, those who interpreted the apparent failure of the DTC to mean that generative grammar was not “psychologically real” proceeded to develop models of language processing that have typically neglected the fact that language users come to know highly abstruse, often dysfunctional, constraints on grammatical analysis, such as those on ellipsis, parasitic gaps, wh- movement, and so forth. In other words, notwithstanding their many virtues, these models neglected complex grammar. (Exemplary psycholinguistic work of this sort includes Marslen-Wilson & Tyler, 1980; Seidenberg & MacDonald, 1999; and Tyler, 1992.)

On the other hand, there are those who placed formal syntactic grammatical knowledge at the very center of their theory, and then proceeded to develop parsing theories to address just those behavioral contrasts about which the grammar itself had nothing to say, in particular, various instances of (grammatical) structural ambiguity. Here, the work of Janet Fodor, Lyn Frazier, and their respective colleagues is obviously central (see Frazier, 1987, for an overview). In this scheme of things, grammatical information served to exclude ungrammatical analyses whereas parsing theories were developed to pick out the preferred analyses from the remaining set (the “grammar proposes, parser disposes” doctrine). Ironically, this branch of psycholinguistics also often came to neglect grammar in developing parsing theories: it was so central as to be ignored.

At the risk of flippancy, the last 40 years of psycholinguistics has witnessed two competing “doughnut theories”: the JAM DOUGHNUT theory, with grammar the jam hidden inside, and the RING DOUGHNUT theory, with grammar the hole in the middle.³

CF's paper is a highly commendable example of work within the latter theory, where the boundary between grammar and parser is taken to be as self-evident as it is well-delineated. Were this not the case, it would be impossible as a matter of principle to infer that differences between native-speakers' and second language (L2) learners' performance in experimental tasks are due to differences in their processing systems rather than to differences in their grammatical competence, which I take to be CF's central claim. That CF have little difficulty in drawing the line between processing capacity and grammatical competence is especially evident in the following quotes:

The clear implication of these quotes is that the competence grammar and the processing system are distinct entities; moreover, that data from judgment tasks and proficiency tests provide a direct channel to (L2) competence, which allows one to infer complete or incomplete acquisition of grammar, and which is unmediated by the processing system. Neither of these implications is acceptable when spelled out, I think.

Notice that what is at stake here is not the issue of the autonomy of syntax (within the language processing system), nor the question of whether aspects of the language processing may be modality specific, nor whether on-line and off-line tasks engage different processing capacities, and certainly not the validity of the competence/performance distinction.⁴

These are all matters where I suspect CF and I would find broad agreement.

What is disputed is whether one can say that a particular piece of linguistic performance—whether it comes from traditional behavioral methods such as response latencies, or from more modern techniques such as event related potentials—is uniquely due to the grammar or to the processing system. Surely one cannot: grammatical competence is always mediated by the processing system (in virtue of being part of it); the ghost, if one exists, is in the machine, not somewhere else.

In conclusion, this perspective on the relationship between syntactic theory and language processing⁵

prevents me from sharing CF's main theoretical conclusions, even while applauding the empirical research presented here. The paper offers thorough and, to my mind, convincing arguments that adult L2 learners really “do (theoretically interesting) things differently,” and hopefully will inspire further work in this crucial area of comparative first language and L2 studies. However, the question of why native speakers and L2 learners do what they do will have to await a better theoretical doughnut recipe.