In this talk, I will cover two subtopics in my Ph.D. thesis - soft pattern matching for definitional question answering (QA) and fuzzy machine of dependency relations for factoid question asnwering. Both techniques are statistical methods applying to sentence retrieval for question answering.
(1) Soft pattern matching for definitional QA - we present two formal statistical matching models (i.e. soft matching models) for lexico-syntactic pattern matching in identifying definition sentences. Compared to current systems that are using rigid, hard matching rules to match definition sentences, our method are more flexible in dealing with language variations. The soft matching modes are based on bigram model and Profile Hidden Markov Model. They can be extended to other applications that use text pattern matching.
(2) Fuzzy matching of dependency relations for factoid QA - Most of current QA systems are using lexical-density based methods for passage retrieval. It often leads to false positives because they don't take into account relationship between matched question words. Previous studies attempted to address this problem by matching dependency relations between questions and answers. They used strict matching, which fails when semantically equivalent relationships are phrased differently. We propose fuzzy relation matching based on statistical translation models, which out performs lexical matching methods and methods that use strict matching of dependency relations.
Hang Cui is now a Ph.D. candidate in the Department of Computer Science, National University of Singapore. His thesis is about precise passage retrieval for question answering, under this supervision of Prof. Tat-Seng Chua and Prof. Min-Yen Kan. His research interests include information retrieval and natural language processing. He had been interning at Microsoft Research Asia and Goggle Inc., where he conducted research in web search.
This talk is part of the CLIP Colloquium Series, organized by Jimmy Lin (jimmylin -at- umd .dot. edu). For the complete schedule, please visit http://www.umiacs.umd.edu/research/CLIP/colloq/.