Open AccessUsing probability distributions to account for recognition of canonical and reduced word forms
Author(s) -
Meghan Clayards
Publication year - 2010
Publication title -
lsa annual meeting extended abstracts
Language(s) - English
Resource type - Journals
ISSN - 2377-3367
DOI - 10.3765/exabs.v0i0.529
Subject(s) - word (group theory) , canonical form , inference , non canonical , computer science , natural language processing , artificial intelligence , distribution (mathematics) , mathematics , speech recognition , linguistics , pure mathematics , mathematical analysis , philosophy , geometry , biology , microbiology and biotechnology
The frequency of a word form influences how efficiently it is processed, but canonical forms often show an advantage over reduced forms even when the reduced form is more frequent. This paper addresses this paradox by considering a model in which representations of lexical items consist of a distribution over forms. Optimal inference given these distributions accounts for item based differences in recognition of phonological variants and canonical form advantage.