Ling F: Learnability and Linguistic Theory
Jon Rawski (San Jose State University)
This seminar covers learnability, aka the theoretical aspects of language acquisition. What makes a grammar learnable (or not!), under what conditions, and from what kinds of evidence?. We will consider both inductive and abductive approaches to learning. We will overview central paradigms in learning theory like identification in the limit, exact learning, probably approximately correct identification, and their numerous variants dealing with statistical criteria, constraints on hypotheses, generalization errors, evidence types, convergence properties, and feasibility. We will relate these paradigms to linguistic issues like poverty of the stimulus, structure vs statistics, domain-general vs domain-specific learners, the subset problem, the "too-many-grammars" problem, typological coverage vs learnability, and others. We will discuss positive and negative learnability results for various classes of linguistic phenomena in syntax, phonology, morphology, etc. as well as design and interpretation of grammar learning experiments in psycholinguistics and NLP.