Computational Linguistics Colloquium

Tuesday, 23 June, 16:15
Conference Room, Building C7 4

Entailment-Based Question Answering in Restricted Domains

In this seminar I will present a methodology for question interpretation based on the use of a Textual Entailment (TE) Engine. The engine attempts to recognize all semantic relations expressed in the question by checking whether the question entails a textual pattern for a give relation, automatically extracted from a corpus of questions.
There are two expected benefits of this approach: (i) it allows the TE engine to be trained using a corpus of questions for the application domain; (ii) it provides a deep analysis of the question, which can be exploited both in the answer extraction and in the answer presentation phase. A relevant consequence is the potential for intensive use of data driven techniques, this way significantly reducing the cost for the development of the application, while still maintaining the fine-grained analysis of traditional knowledge-based systems.
The approach makes use of the EDITS (http://edits.fbk.eu/) textual entailment package, and has been successfully applied in the context of the European project QALL-ME (http://qallme.fbk.eu) in several demonstrators.

If you would like to meet with the speaker, please contact Rui Wang.