Charles Yang
Yale University
What happens when rules run into exceptions?
According to one approach, exceptions are just exceptions, and that is that. In the famous case of English past tense (Pinker 1999), the exceptional irregular verbs are simply memorized by rote, in co-existence with the regular rule: add -d.
Another approach (Culicover 1999; Culicover & Jackendoff 2004) is even more pessimistic: it rejects the rules. After surveying a wide range of intriguing periphery constructions in the sense of Chomsky (1981), Culicover concludes that the core parametric system be abandoned as it cannot accommodate these idiosyncratic nuts.
At the heart of the matter is the old question of storage vs. computation. More succinctly, we need a model of learning that strikes a balance between rules with exceptions: it must know when the postulation of a rule, or a statement of generality like parameter values, is warranted even if it has a few exceptions. For example, children construct the -d rule in spite of over 100 irregulars, but they do not construct a productive rule that inflects verbs ending in -ow to -ew in past tense--along the lines of blow-blew, know-knew, grow-grew. In the latter case, apparently, they have seen too many exceptions to the ow->ew "rule" and decided that enough is enough.
"How much is enough" will be the concerns of this talk. We present a mathematical model of storage and computation, on the assumption that the lexicon is organized by a principle of economy, which minimizes the time complexity of lexical access and retrieval. The result is a criterion that tells the learner when a rule is justified. Its application in language learning, change, and evolution will be discussed.
If you are interested in
meeting with the speaker, please contact Kaori
Ozawa.
To go back to the Colloquium schedule, click here.