Incremental learning of lexically specific morphophonology: an integrative approach

Abstract Acquisition and processing results indicate idiosyncratic, lexical knowledge interacts with productive, grammatical knowledge in systematic ways. Evidence from language processing demonstrates that higher frequency and less productive complex words are more likely to be retrieved holistically from the mental lexicon rather than being decomposed into constituents. Meanwhile, acquisition findings in both natural and artificial languages provide important evidence about the time course of learning and the role that lexical conditioning plays during development. Finally, work in experimental phonology demonstrates adult native speakers have productive knowledge of lexicalized morphophonological patterns, with learners extending lexical trends stochastically to novel forms. Although there have been significant computational advances in modeling the learning of such gradient grammatical generalizations, less attention has been devoted to modeling the joint learning of both the lexical and grammatical components of these systems, their interactions, and their connections to the processing and acquisition findings. This paper examines concrete predictions that two families of existing models of lexical and grammatical learning make across all three of these empirical domains. I show the models already align well with existing processing evidence on lexicalization effects, identify aspects of their predictions requiring further attention, and discuss implications for linguistic theory and computational modeling.