Learning abstract underlying representations from distributional evidence
Ezer Rasin, Roni Katzir
May 2018 

Human learners have been argued to infer underlying representations (URs) that are sometimes different from their corresponding surface representation even without being forced to do so by an alternation. To account for such inferences, McCarthy (2005) proposed the Free Ride Principle (FRP), which allows the learner to extend nonidentical mappings in alternating forms to non-alternating forms. Recently, Rasin and Katzir (2016) proposed Minimum Description Length as an evaluation metric for Optimality Theory and showed how complete phonological grammars can be acquired from distributional evidence alone, including URs (that are sometimes different from their surface forms), constraint rankings, and the constraints themselves. The present work shows how Minimum Description Length supports the induction of abstract, nonidentical URs from distributional evidence alone, both in cases that have motivated the FRP and in cases that have been argued to involve nonidentical URs in the absence of supporting alternations (and are thus outside the scope of the FRP).

Format:	[ pdf ]
Reference:	lingbuzz/004121 (please use that when you cite this article)
Published in:	To appear in Proceedings of NELS 48
keywords:	learning, phonology, underlying representations, minimum description length, sanskrit, yimas, free ride, phonology
Downloaded:	230 times

 

[ edit this article | back to article list ]