Abstracts

Most events on this list can be tied to a specific page somewhere on the internet.  List entries contain links to such pages through the "Details".   Occasionally events have no such page or the information on it is inadequate.  In such cases I attempt to cobble together a details page myself.



Naomi Feldman (U of Maryland) will be visiting MIT from 3/15-3/18 and giving two talks.

How phonetic learners should use their input
Time: Wednesday, 03/16/2016, 3:30-5:00 pm
Venue: 32D-461

Children have impressive statistical learning abilities. In phonetic category acquisition, for example, they are sensitive to the distributional properties of sounds in their input. However, knowing that children have statistical learning abilities is only a small part of understanding how they make use of their input during language acquisition. This work uses Bayesian models to examine three basic assumptions that go into statistical learning theories: the structure of learners' hypothesis space, the way in which input data are sampled, and the features of the input that learners attend to. Simulations show that although a na´ve view of statistical learning may not support robust phonetic category acquisition, there are several ways in which learners can potentially benefit by leveraging the rich statistical structure of their input.

Modeling language outside of the lab
Time: Friday, 03/18/2016, 3:30-5:00 pm
Venue: 32D-461

Speakers and listeners operate in complex linguistic environments. They extract phonetic information from highly variable speech signals and track the salience of entities in rich discourse contexts. However, little is known about the representations that support language use in these complex environments. In this talk, two cognitive models that were developed for laboratory settings are modified to operate over more naturalistic corpora. A rational speaker model is used to predict how entities are referred to in news articles, and a model of speech perception is trained and tested directly on speech recordings. In each case, simulation results show how cognitive modeling can be used to probe the way in which speakers and listeners represent the complexity of their linguistic environment.