Overgeneration

Bob Kennedy comments on my rant about Odden on ordering:

My take on the moral of the story is that the overgeneration argument should be put to bed.

This isn’t quite the moral I had in mind (but it may have been the one Odden had in mind). In any case, I’m glad Bob raised this issue. I think it’s a very interesting topic for discussion on this blog.

Bob’s post continues:

At a recent LSA Charles Reiss presented a derivational stress-assigning parser – it seemed that it allowed some really weird stress patterns to be learned, like a stress every four syllables. I asked about it, and he replied that it was possible, but no learner would ever encounter the evidence to generalize to that system. I’m totally satisfied with that answer; it’s the essential gist of Reiss’s research – but it opens up the same response to any potentially overgenerative system. So, an OT model of tone assignment can describe the hypothetical but unattested language (Kintupu) – but no learner would encounter such evidence, so Kintupu just would never evolve into its peculiar representational state.

The way that Bob has characterized it here, it sounds to me like Reiss is appealing to unattestedness as an explanation for unattestedness. If this is a fair characterization, it reinforces my suspicion that Reiss and others are (perhaps not consciously) simply employing a very convenient rhetorical strategy: when you have nothing substantive to say about an issue, deny that there’s any issue to say anything substantive about.

This strategy works very well in this case because we are indeed far from understanding the distinction between what one might call “weakly unattested” patterns (accidental gaps of attestation) and “strongly unattested” patterns (things we never expect to find). However, I think we all make this distinction in some way or another regardless of our theoretical and/or methodological persuasions, and we even have intelligent disagreements about where/how to draw the line.

I’ve read work by Reiss (often in collaboration with Mark Hale) where appeals are made to independent theories of language acquisition and/or language change to account for (certain kinds/examples of) unattestedness. Consider Reiss’s recent paper in The Linguistic Review (2003), the subtitle of which is “Attested and unattested patterns”. (Here’s the paper on Reiss’s website, and just in case it’s ever removed, here’s a copy.) The following extended quotation summarizes Reiss’s position on the matter (pp. 25-26, emphasis added):

One way to do so is to build the fact into UG […]. This strikes me as the wrong way to approach the issue, if it merely consists of restating the descriptive observation as a principle of grammar and not being open to explanations outside of the realm of grammar. In this particular case, it may be possible to derive the gap, from the relationship between language change and phonetics. Note that this approach is in no way incompatible with a nativist perspective–the nativist position is just that some (not necessarily all) non-trivial aspects of the language faculty are innate. […] [I]t is to be expected that attested patterns in the phonological systems of the world’s languages reflect only a subset of what is computationally possible for the human phonological capacity. In other words, all attested patterns must be generatable by the UG-given phonological capacity, but not all generatable patterns will arise, due to the nature of sound change and language acquisition. This point of view may be helpful in explaining why [some generatable patterns] are unattested.

The two “may be” parts of this quotation are, I think, significant; in my view, the ensuing discussion does not convincingly substantiate Reiss’s position. (Don’t take my word for it; read for yourself. If you read things differently than I do, please blog about it.)

It is clear from the above quotation that Reiss advocates all three of the following explanations for unattestedness, and it is also clear that he favors the first two.

  1. The theory of language acquisition.
  2. The theory of language change.
  3. The theory of UG / nativism.

And, from what Bob reports, Reiss appears to have given himself a backdoor in case all of these explanations fail to pan out: unattestedness itself.

In closing, I should note that I do think that Hale & Reiss are correct in pointing out how naive many phonologists can be about the issue of attestation. They suggest that it is not (always) the job of the phonology to account for the absence of strongly unattested patterns, but they appear to concede that it is the job of the phonologist to do so. It is thus particularly disappointing to see Hale & Reiss not really putting their money where their mouths are. (Perhaps they do so in one of their forthcoming co-authored books.) Hale often explicitly hides behind the fact that he’s not a phonologist; I saw him give a talk in 1997 and the title of the first section on the handout was “I am not a phonologist”. Reiss presumably does not have this excuse.