Robust Question Answering

A Question Answering (QA) system should provide a short and precise answer to a question in natural language, by searching a large knowledge base consisting of natural language text. The sources of the knowledge base are widely available, for written natural language text is a preferential form of human communication. The information ranges from the more traditional edited texts, for example encyclopaedias or newspaper articles, to text obtained by modern automatic processes, as automatic speech recognizers. The work developed in the present thesis focuses on the Portuguese language and open domain question answering, meaning that neither the questions nor the texts are restricted to a specific area, and it aims to address both types of written text. Since information retrieval is essential for a QA system, a careful analysis of the current state-of-the-art in information retrieval and question answering components was conducted. A complete, efficient and robust question answering system is developed in this thesis, consisting of new modules for information retrieval and question answering, that is competitive with current QA systems. The system was evaluated at the Portuguese monolingual task of QA@CLEF 2008 and achieved the 3rd place in 6 Portuguese participants and 5th place among the 21 participants of 11 languages. The system was also tested in Question Answering over Speech Transcripts (QAST), but outside the official evaluation QAST of QA@CLEF, since Portuguese was not among the available languages for this task. For that reason, an entire test environment consisting of a corpus of transcribed broadcast news and a matching question set was built in the scope of this work, so that experiments could be made. The system proved to be robust in the presence of automatically transcribed data, with results in line with the best reported at QAST.