Thursday, April 09, 2015
12:45 PM
CBS Room 209
Percy Liang
Stanford University
Learning to Execute Natural Language
A natural language utterance can be thought of as encoding a program, whose execution yields its meaning. For example, "the tallest mountain" denotes a database query whose execution on a database produces "Mt. Everest." We present a framework for learning semantic parsers that maps utterances to programs, but without requiring any annotated programs. We first demonstrate this paradigm on a question answering task on Freebase. We then show how that the same framework can be extended to the more ambitious problem of querying semi-structured Wikipedia tables. We believe that our work provides a both a practical way to build natural language interfaces and an interesting perspective on language learning that links language with desired behavior.

BIO: Percy Liang is an assistant professor of Computer Science at Stanford University (B.S. from MIT, 2004; Ph.D. from UC Berkeley, 2011). His research focuses on methods for learning richly structured statistical models from limited supervision, most recently in the context of semantic parsing in natural language processing. He is the recipient of Sloan Research Fellowship (2015) and Microsoft Research Faculty Fellowship (2014). He won a best student paper at the International Conference on Machine Learning (ICML) in 2008, received the NSF, GAANN, and NDSEG fellowships, and is also a 2010 Siebel Scholar.