Week 1 [2023-04-05 Wed]

We discussed the importance of a theory of how to connect our theory of grammar to actual data.

We hear again and again that linguistics is an empirical science. We hear it said that the goals of linguistic theory are to achieve descriptive adequacy, explanatory adequacy, and even beyond (what we might call evolutionary adequacy). These mean that we want to correctly account for linguistic data, explain how people learn language, and explain how language evolvolved in Humans. In particular, we want to correctly account for linguistic data. But…what data exactly? And…how?

As our theories are (nominally) built up over acceptability judgements, one might be forgiven for thinking we want to be able to account (at least) for this kind of data. One natural idea is to take the derivable/underivable distinction made by our grammars as the foundation for a binary distinction in acceptability judgements. That acceptability judgements are not in fact binary would then be the provenance of some other theory—the explanatory role of grammar would lie in attaching a single bit of information to each string of words, which could then play some role in the determination of actual behaviour. We could hope that the explanatory role of grammar was strong enough to be able to detect a difference between derivable and underivable sentences even without a spelled out theory of how exactly acceptability judgements emerge.

This is a totally reasonable thing to think, however, Chomsky wants to take a different tack (MP, p 13):

“In the earliest work in generative grammar, it was assumed that Jones’s language generates an SD for each of the permissible phonetic forms for human language, a set to be specified by UG. Thus, Jones’s language assigns a [structure] to [ungrammatical sentences and expressions of Swahili, Hungarian, etc]”

Thus for Chomsky, there is no explanatory role for the derivable/underivable distinction to play, because every string of words is derivable. If we adopt this view, we will have to find something else to hang our explanation of acceptability on. The commonly pursued idea is to look for particular structural configurations, such that having a structure with that kind of structural configuration leads to unacceptability. An example of this is moving too far, or over an intervener, etc. If we adopt this view, we will need to specify what happens when multiple configurations occur simultaneously.

My impression is that practicing linguists work generally with a mixture of both theories in the back of their minds: that is, they make use of a derivable/underivable distinction, but also look to structural configurations to refine this picture. Of course, this is always ‘in the back of their minds’—linguists just do not try very hard to specify the connection between their theories and grammaticality judgements.

We discussed work by Shalom Lappins group on relating acceptability to probability (see the readings tab).