UCLA

This dissertation aims to bridge the gap between theoretical and computational research on phonotactic learning and infant language acquisition, focusing on the morphophonological interface. Two acquisition findings are relevant. First, sensitivity to phonotactics emerges around the same time as knowledge of common affixes, suggesting that early stages of phonotactic learning are based on input with minimally partial morphological information. Second, morphological and phonotactic learning appear to mutually influence each other; early phonotactic knowledge can be utilized by infants to discover morphological boundaries. These developmental facts have fundamental implications for any account of the productivity and learnability of phonological patterns sensitive to morphological structure. This dissertation focuses on two such patterns, morpheme-bounded phonotactics and morphologically-derived environment effects (MDEEs). I present novel data from two empirical studies and use the results to motivate a developmentally-plausible model of morphophonological learning, implemented computationally in the Iterative MorphoPhonological (IMP) learner. The first empirical study focuses on a morpheme-bounded pattern, Arabic OCP. Even though the lexicon data shows a weak tendency for OCP to apply regardless of morphology, a nonce word judgment experiment demonstrates that Arabic speakers learn OCP as strictly morpheme-bounded. I show that existing models which learn phonotactics from a fully parsed lexicon or unparsed data fail to account for this discrepancy between the lexicon and speaker intuitions. I then show that IMP does successfully model Arabic speakers’ intuitions. Specifically, in utilizing emerging phonotactic knowledge to discover morpheme boundaries, the learner strengthens the morpheme-boundedness of OCP. In the second study I compare the laboratory learning of morpheme-bounded patterns to MDEEs, manipulating how much morphological information is available to learners. The results suggest an advantage for learning morpheme-bounded patterns even with minimal morphological information, akin to what is available to young infant learners. In contrast, MDEEs can only be learned successfully with maximal morphological information which is unusual in acquisition. Existing accounts which attribute the underlearning of MDEEs to a bias for morphologically general constraints are unable to account for these differences in learnability between the two patterns, but they emerge as a natural consequence of iterative morphophonological learning.