Response to Seuren

Pieter Seuren is quite right in pointing out that an adequate theory of linguistic performance must account not only for language comprehension but also for language production, and that my discussion note (Jackendoff 2011) had more to say about the former than the latter. I also agree with him that language production probably makes more stringent demands on the grammar than comprehension, in that it requires resolution of many fine grammatical details that comprehension can often gloss over.

However, I have by no means ignored the issue of language production. To be sure, the discussion note had little to say about language processing. Its focus was on comparing the formal and conceptual bases of the minimalist program/biolinguistics with those of my own parallel architecture, and language processing was only one of many issues competing for a limited space. But immediately following the passage Seuren quotes, the reader is referred to chapter 7 of Foundations of language (Jackendoff 2002) for a more extensive treatment of language processing from the perspective of the parallel architecture. This chapter gives language production equal billing with language comprehension, dealing with such issues as lexical access in production, speech errors, and syntactic priming. In particular, there is a fairly detailed discussion of the degree to which the production theories of Levelt 1999 and Roelofs 1997 can be incorporated naturally into the architecture. The chapter acknowledges that linguistic theory cannot predict all properties of language processing, and it explores what aspects of language comprehension and production might be predicted from the architecture, such as relative timing of various subprocesses, and what aspects must be determined experimentally, such as absolute timing and brain localization.

I further agree with Seuren that at the moment little can be said about the relation between the theory of grammar and the neural processes that support the use of language. We have no idea how neurons instantiate something as simple as a speech sound, let alone prosodic contours, syntactic structure, or meaning. Hence, as Seuren says, the best strategy for linguistic theory is to try to formulate approaches that are consonant with what is known about the brain and the rest of the mind—what he calls ‘nonhardware realism’—in the hope of eventually achieving compatibility with theories of neural computation. Indeed, this was one of the major points of my discussion note: the parallel architecture has such aspirations and follows through on them, while the minimalist program/biolinguistics for the most part does not, seeking rather a sort of ‘perfection’ in the language faculty that is not at all characteristic of the way the brain operates in other respects.

However, the argument for nonhardware realism cuts both ways. Whatever our best efforts to bring linguistic theory into line with neural computation, language may still demand a kind of computation that cannot be accommodated by our current understanding of what neurons do. And then the question arises of which of the two must yield. This was the issue at the root of the ‘words vs. rules’ controversy of the 1980s and 1990s, between the connectionists (e.g. Rumelhart & McClelland 1986), who claimed that neural computation cannot and need not accommodate free combinatoriality in language, and various linguists and psycholinguists (e.g. Pinker 1999), who claimed that the facts of language and of language processing demand it. Section 3.5 of Foundations of language, entitled ‘Four challenges for cognitive neuroscience’, elaborates on the problems that linguistic combinatoriality poses for theories of neural computation—whatever one’s theory of language. It shows furthermore that these problems are not peculiar to language: counterparts occur in visual memory and visual processing as well. But linguistic theory per se cannot go much beyond this point. It can only be hoped that a more complete integration of the theories can be sought through meaningful engagement with open-minded cognitive neuroscientists.