The LIA at QA@CLEF - 2006
Laurent Gillard, Laurianne Sitbon, Eric Blaudez, Patrice Bellot, Marc El-Bèze · Cross-Language Evaluation Forum · 2006
This article presents the first participation of the Laboratoire Informatique d’Avignon (LIA) to the Cross Language Evaluation Forum (CLEF). LIA participated to the monolingual Question Answering (QA) track dedicated to French language, and to the crosslingual English to French QA track. Two runs for each track were submitted. English questions were first translated and then answered by using the monolingual French system. This QA System (QAS) already participated to the French Technolangue QA campaign (EQueR) but some improvements needed to be evaluated: definition questions answering module; or were developed specifically for CLEF: integration of Lucene search engine, and re-ranking based on redundancy for factoid answer candidates. The CLEF-QA provided an opportunity to evaluate these modules. Also, English conversion of the QAS was started, even if, only the Question Analysis module was adapted to English this year. The generic factoid QAS is based on an extraction of answer candidates in the form of Named Entities (NE) by using keywords density measures. Few factoid answers were also provided by a knowledge base module. Lastly, definition questions were answered by a module based upon detection of frequent appositive informational nuggets appearing near the focus to define. The system obtained reasonable results in all runs except for Temporal and List questions that were not recognized as such and wrongly handled as simple factoid one.