The sublexical learner
Collaborator: Michael Becker
The sublexical learner is an implementation of the sublexical grammar architecture proposed by Becker & Gouskova (2012). Essentially, it uses a list of word pairs (e.g. singular and plural noun pairs) to learn a set of probabilistic constraint-based grammars which encode the possible mappings between paired words, as well as using these grammars to generalize the learned patterns to novel ("wug") words. We show that such an architecture is able to very accurately predict judgments in novel word experiments. Our implementation is also able to encode a wide variety of morphological patterns, including non-concatenative morphology such as umlaut and metathesis.
Papers and presentations
(forthcoming) Allen, Blake and Becker, Michael. Learning alternations from surface forms with sublexical phonology. Ms. University of British Columbia and Stony Brook University.
Allen, Blake, and Becker, Michael. Learning alternations from surface forms with sublexial phonology. Poster presentation. LSA 2015 Annual Meeting. Portland, Oregon: January 9, 2015.
Allen, Blake, and Becker, Michael. Learning alternations from surface forms with sublexial phonology. Paper presentation. Phonology 2014. Cambridge, Massachusetts: September 19, 2014.