April 14, 2004

Writers just want to be humiliated

Over and over. So says Robin Robertson, quoted by Dinitia Smith in a NYT review of his new book "Mortification: Writers' Stories of their Public Shame". Ms. Smith publically trips over her own shoelaces in the third paragraph, where she writes:

The idea for the book came to him after he escorted authors to readings and finding "a sign saying 'Reading Canceled,' or three chairs occupied by people released from mental institutions and not thought to be violent," he said. [emphasis added]

There's a simple and common mistake here -- the highlighted "and" connects the tensed verb "escorted" and the gerund "finding". This seems like an error to me, and it's not a difference in dialect or degree of formality, as far as I can tell. It's just a mistake. I bet that the author would call it a mistake if asked about it.

I speculate that she originally wrote "after escorting ... and finding ...", or "after he escorted ... and found ...", and then changed one of the verbs without changing the other. Or maybe some distracted editor was responsible.

But I didn't write this to one-up Ms. Smith, or even to underline the irony of the error's context. I'd like to use this example to suggest an irony of a different order: an argument for rationalist epistemology of language from the richness of the stimulus.

The usual (poverty of the stimulus) line-up is something like this:

Rationalist: Much grammatical knowledge must be innate, or at least somehow independent of experience, because linguistic experience is simply too impoverished to define the emergent patterns.

Empiricist: No, if you look at the facts, you can see that linguistic experience is much richer than you thought. The patterns that are learned are plausibly the result of inductive inference applied to documented experience.

See this article by Geoff Pullum for a fuller discussion. As Geoff points out, the poverty-of-the-stimulus case depends on "hyper-learning" -- cases where things are "learned" about structures that have never been seen.

The factual argument, to the extent that there is one, then comes down to the question of whether a language-learner's experience includes certain kinds of examples or not. And generally the rationalists are rooting for less stuff in the data, while the empiricists are rooting for more.

But this argument has an inverse form, it seems to me: "hypo-learning", where learners ignore commonly-encountered structures. As common sense tells us (and construction grammarians variously document and explain), people are ready and willing to learn all sorts of crufty constructions, from "what's X doing Y" to "all your X are belong to us". And as Smith's little editing error exemplifies, conjunctions of incompatible constituents are pretty common. So why don't we all learn a construction consisting of a tensed VP conjoined with a gerundive VP?

I don't know -- but here's a speculation. People are so constituted as to prefer their grammars to be "coherent", in some a priori sense. Conjunctions of tensed and gerundive VPs are incoherent enough that we are reluctant to learn a grammar that licenses them. When we occasionally encounter examples of such structures, we tend to ignore them or write them off as mistakes.

When we go to look at the facts, this argument reverses the traditional allegiances: the empiricists should want "incoherent" conjunctions to be as rare as possible, so that learning requires no special expectations about "coherence"; the rationalists should want the opposite.

Posted by Mark Liberman at April 14, 2004 02:54 PM