Colloquium Talk – Lily Xu

A case for iterative morphophonological learning: the morpheme-boundedness of Arabic OCP

Acquisition research shows that the timelines of early phonotactic learning and morphological learning largely overlap. This conflicts with a common simplification in computational modeling of phonological learning, where phonotactic generalizations are learned from input with complete morphological information. I demonstrate a case of discrepancy between lexicon regularities and native speaker intuitions wherein models with this simplification fail to match human behavior. In Arabic, OCP restrictions disfavor sequences of homorganic consonants within roots. Even though these restrictions exhibit “leakage effects” in the lexicon, i.e., also apply with weak strength across morpheme boundaries, nonce word judgments show that Arabic speakers learn them as strictly morpheme-bounded. Instead, I propose the iterative morphophonological (IMP) learner to account for this result. This learner takes a developmentally-plausible approach to phonotactic learning, where 1) phonotactic restrictions are extracted from protolexical forms with partial, inaccurate morphological information, and 2) the developing phonotactic grammars inform the discovery of morphological boundaries in turn. As such, the tendency to learn strictly morpheme-bounded restrictions despite being exposed to evidence of leakage emerges as a result of phonotactic-based morphological parsing.

Location: Boelter Hall 5419