Abstract

CINDI_QA is the result of the CINDI group's efforts at tackling multilingual Question Answering. The goal of the system is to receive a French question and return the answer to that question in English. This thesis outlines the architecture of CINDI_QA and how it is integrated with several open-source tools. Google Translate is used to produce the English equivalent of the French question. The Link Parser is a semantic parser of English that identifies keywords such as nouns, verbs and adjectives. WordNet is a lexical analyzer that generates synonyms of those keywords. Finally, Lucene handles indexing and searching of the large data collection made up of Wikipedia static pages. In addition to these tools, CINDI_QA heavily relies on templates to better understand the question entered by the user. Six templates help the system answer questions about different subjects. The performance of CINDI_QA has been assessed by participating in the 2007 edition of the QA@CLEF competition, an annual conference focusing on multilingual QA, to which the CINDI group submitted two runs that ranked second and third out of eight candidates