Research

My research focuses on statistical models of natural language processing and acquisition, with an emphasis on joint inference, unsupervised learning and statistical relational learning. My primary interest stems from a fascination with how children are able to, with little exception, effortlessly acquire their first language. I develop models that explore this problem by examining how linguistic levels (morphology, phonology, etc.) interact, and how their relational structures can be leveraged to improve learning in the absence of labeled training instances and negative examples. Additionally, these models test different assumptions, biases, and learning mechanisms in an attempt to determine which best approximates human language learning, and which biases, if any, are linguistically universal.

Recent Publications

David Mimno, Hanna Wallach, Jason Naradowsky, David Smith and Andrew McCallum. "Polylingual Topic Models." To appear in Proceedings of the 2009 EMNLP Conference, Singapore, 2009.[pdf]

Jason Naradowsky and Sharon Goldwater. "Improving Morphology Induction by Learning Spelling Rules." To appear in Proceedings of IJCAI, Pasadena, California, 2009.[pdf]
Slides[keynote][pdf]

David Mimno, Hanna Wallach, Limin Yao, Jason Naradowsky and Andrew McCallum. "Polylingual Topic Models." Presented at the Learning Workshop (Snowbird), Clearwater, Florida, 2009.

Contact Details

Department of Computer Science
University of Massachusetts Amherst
140 Governors Drive
Amherst, MA 01003