We just had a fun week grilling Charles Yang over his book, The Price of Linguistic Productivity. We discussed the mechanisms and principles in his framework, and how to apply to them to declension class and gender features, and to V2 and verb raising, how to formulate rules, the difference between contexts of applications and lists, and a slew of other topics. Here are some conclusions, as suggested by Charles.
1. Everybody should read Aspects.
2. Linguists should avoid overfitting the data: not every linguistic pattern is real, or worth theorizing over.
3. Productivity is categorical, not gradient.
4. There is no shame in achieving good data coverage.
5. A theory of competence needn’t transparently yield a theory of performance — but it would be irresponsible to insist that it cannot.
6. Third factor considerations require direct empirical motivations, not conceptual ones.
7. Some linguistic principles cannot be regressed out of data.
8. “Analogy” should be purged from linguistics.
9. Indirect negative evidence is wrong; Bayesian packaging doesn’t make it better.
10. Inductive/abductive learning of language does not mean that there are no a priori UG hypotheses (i.e., “parameters”).