Bilingual Pronoun Resolution: Experiments in English and French

Anaphora resolution has been a subject of research in computational linguistics for more than 25 years. The interest it aroused was due to the importance that anaphoric phenomena play in the coherence and cohesiveness of natural language. A deep understanding of a text is impossible without knowledge about how individual concepts relate to each other; a shallow understanding of a text is often impeded without resolving anaphoric phenomena. In the larger context of anaphora resolution, pronoun resolution has benefited from a wide interest, with pronominal anaphora being one of the most frequent discourse phenomena. The problem has been approached in a variety of manners, in various languages. The research presented in this thesis approaches the problem of pronoun resolution in the context of multilingual NLP applications. In the global information society we are living in, fast access to information in the language of one’s choice is essential, and this access is facilitated by emerging multilingual NLP applications. As anaphora resolution is an integral part of many NLP applications, it necessarily has to be approached with the view of multilinguality in mind. The present project describes an original approach to bilingual anaphora resolution, for English and French. The method is based on combining hand crafted rules with automatic learning. Although only designed and tested for English and French, the approach has the potential to be extended to other languages. The steps preceding the practical development of the system consisted in an investigation of common and distinct features of English and French pronominal systems, while the evaluation of the system provided an interesting insight into specific problems relating to bilingual anaphora resolution.