Percy Liang

: grad student at Berkeley → post-doc at Google → assistant professor at Stanford

Language: How do words evoke meaning? I work on methods that infer representations of meaning from sentences given limited supervision. What's particularly exciting to me is the interface between rich semantic representations (e.g., programs or logical forms) for capturing deep linguistic phenomena, and probabilistic modeling for allowing these representations to be learned from data. More generally, I'm interested in modeling both natural and programming languages, and exploring the semantic and pragmatic connections between the two.
Learning: How can we design computationally-efficient algorithms that learn from weak supervision? Towards computational efficiency, I work on approximate inference algorithms. Towards weak supervision, I work on methods that learn from partial labels and that can share statistical strength across multiple related learning problems. I'm also interested in analyzing learning algorithms, both theoretically and empirically.

Graduate fellowships: NSF, NDSEG, GAANN, Siebel Scholar
Research: Best student paper (ICML 2008)
Programming contests: 2nd place at 2002 ACM ICPC World Finals, silver medalist at IOI 2000
Music competitions (piano): Winner of KDFC Classical Star Search (2008, over-21 division), MIT Concerto Competition (2004), Phoenix Young Musicians Competition (2000)