Ontology-based question analysis in a multilingual environment: the MOSES case study

Question Answering (QA) systems (as QA track of the Text Retrieval conference (TREC-QA) competitions (Voorhees 2001) ), are able both to understand questions in natural language and to produce answers in the form of selected paragraphs extracted from very large collections of text. Generally, they are open-domain systems, and do not rely on specialised conceptual knowledge, while using a mixture of statistical techniques and shallow linguistic analysis. Ontological Question Answering systems, e.g. (Woods et al. 1972, Zajac 2000) attack the problem by using an internal unambiguous knowledge representation. As any knowledge intensive application, ontological QA systems have as intrinsic limitation the small scale of the underlying syntactic-semantic models of natural language. While limitations are well-known, we are still questioning if any improvement has occurred since the development of LUNAR, the first ontological QA system. Several important facts have emerged that could influence related research approaches: a growing availability of lexical knowledge bases that model and structure words: WordNet (Miller 1995) and EuroWordNet (Vossen 1998) among others; some open-domain QA systems have proven the usefulness of these resources (e.g. Harabagiu et al. 2001); the vision of a Web populated by “ontologically” tagged documents which the semantic Web initiative has promoted; this would require a world-wide collaborative work for building interrelated “conceptualisations” of domain specific knowledge; the trend in building shallow, modular, and robust natural language processing systems (Abney 1996, Hobbs et al. 1996, Ait-Moktar&Chanod 1997, Basili&Zanzotto 2002) which is making them appealing in the context of ontological QA systems, both for text interpretation (Andreasen et al. 2002) and for database access (Popescu et al. 2003). In such a new fascinating context, we are investigating a novel approach to ontology-based QA in which users ask questions in natural language to knowledge bases of facts extracted from a federation of Web sites and organised in topic map repositories (Garshol 2003). Our approach is investigated in the context of the EU project MOSES,