Syllables, Morphemes and Bayesian Computational Models of Acquiring a Word Grammar
Ça ri Çoltekin, Cem Bozşahin · eScholarship (California Digital Library) · 2007
We report a computational study on the CHILDES database for learning a word grammar of Turkish nouns.The syllable-based model converges to a morpheme-based model in terms of overlaps in the set of lexical hypotheses.Morphology is a hidden variable in all models, and the search problem for hypotheses is narrowed down by a probabilistic conception of universal grammar à la Combinatory Categorial Grammar.The convergence of the syllable model suggests that morphemehood can be an emergent computational property.