Host: Premkumar Devanbu
When: Thursday, October 27th at 3:10pm
Where: 1065 Kemper Hall
Automatic analysis of the structure and meaning of natural language is a fundamental problem in computational linguistics and an enabling capability for a wide range of language technology applications, such as machine translation, question answering and dialogue systems. I will present a framework for efficient parsing that deals effectively and efficiently with the exponential search space of possible analyses for natural language input. This involves a shift-reduce parsing algorithm and a novel approach for structured prediction that, unlike previous approaches, allows both efficient training and the use of best-first search during run-time. Using a simple feed-forward neural network as the learning component, this work outperforms previous approaches, including those based on more complex architectures for structured prediction.
Kenji Sagae is an assistant professor of Linguistics at UC Davis. Since completing his PhD at Carnegie Mellon University, he has been a research associate in Computer Science at the University of Tokyo, a research faculty member of the Computer Science department at the University of Southern California, a project leader at the USC Institute for Creative Technologies, and a co-founder of KITT.AI, a natural language processing start-up. His research interests are in computational linguistics and data-driven natural language processing, including the design of algorithms and learning approaches for linguistic analysis and language technology applications.
1131 Kemper Hall