When: Friday, December 12, 3:00 p.m.

Where: 4623 Wean Hall

Curtis Huttenhower, LTI MS student

LTI MS Thesis Defense

Abstract:
Open domain question answering is slowly broadening its horizons, expanding from simple factoid questions to encompass broader and more complex queries. Recent areas of interest include scenario-based question answering, incorporating existing domain knowledge during question analysis, various types of inference, and intelligently processing subquestions. FLOOD (standing for Fluent, Linguistic, Object-Oriented, Dynamic) provides a planning platform within which these problems may be explored. It consists of three parts: A text processor which integrates existing tools to provide an end-to-end semantic parse. This begins with simple sentence breaking and tokenization and ends with theta-role filling. The text processor is organized as a separate module applicable to other tasks outside of FLOOD. The FLOOD planning platform, an environment for authoring and utilizing simple linguistic planning domains and algorithms. The platform handles routine tasks such as domain and state space management while leaving specifics up to individual reasoning modules; the primary goals of the platform are to provide a convenient and flexible interface for reasoner and domain development. A reasoner for question answering operating within the context of the FLOOD platform. The current implementation is based on the FLECS algorithm (Veloso and Stone, 1995) and performs simple inference and subquerying tasks based on the output of the text processor.

<< Back