Multilinguization and Personalization of NL-based Systems

Linguistic porting of content management services processing spontaneous utterances in natural language has become important. In most situations, such utterances are noisy, but are constrained by the situation, thus constituting a restricted sublangage. In previous papers, we have presented three methods to port such systems to other languages. In this paper, we study how to also personalize them by making them capable of automatic perception adaptation, using fuzzy evaluation functions. We have reengineered IMRS, a music retrieval NL-based system, to implement that idea, and ported it to French, English and Arabic using an enhanced version of our external porting method, building a unique content extractor for these three languages. More than 30 persons participated in a preliminary on-line qualitative evaluation of the system.