James White at LEL Research Seminar

Posted on March 2, 2017 by

Next Tuesday, March 7,  we welcome James White at the LEL Research Seminar. His talk promises to be a must-attend for those interested in phonology, phonetics, and language acquisition. Here is the abstract:
Phonological learning is constrained by phonetic similarity: Experimental and computational evidence
James White
In this talk, I argue that phonological learning is constrained by a bias against large phonetic changes (e.g. Steriade 2001/2009). As evidence, I consider the problem of “saltatory” alternations, cases where alternations leap over phonetically intermediate sounds (e.g. [p] alternates with [β], but [b] does not alternate). Typological evidence suggests that saltatory systems are marked: they occur infrequently, and when they do arise,  they tend to be unstable. Furthermore, experimental evidence from adults and 12-month-old infants suggests that saltatory systems are dispreferred during learning. On the theoretical side, we must therefore account for the fact that saltatory alternations are dispreferred during learning, but ultimately learnable. I show that both observations can be captured in a probabilistic maximum entropy model (e.g. Goldwater & Johnson, 2003; Wilson, 2006) with two components: (1) constraints banning correspondence between individual segments (*MAP; Zuraw 2013), and (2) a soft bias that causes the learner to be more skeptical of alternations between dissimilar sounds during learning, in the spirit of Steriade’s P-map (2001/2009). The computationally implemented learning model closely matches human performance in artificial language learning experiments.

Please join us for the talk! As usual, everyone is invited to stick around afterwards to talk to our speaker over some wine and crisps.

Event details:

Location: Ellen Wilkinson Building, room A2.6

Date and time: Tuesday, March 7, 4.15pm-5.30pm

Posted in: Uncategorized