Knowledge Acquisition forCoreference ResolutionDissertationzur Erlangung des akademischen Gradeseines Doktors der Philosophieder Philosophischen Fakult atender Universit at des Saarlandesvorgelegt von Olga Uryupinaaus MoskauSaarbruc ken, 2007Dekan: Prof. Dr. Ulrike DemskeBerichterstatter: Prof. Dr. Manfred PinkalDr. Mirella LapataTag der letzten Prufungsleistung: 1.6.2007AbstractThis thesis addresses the problem of statistical coreference resolution. The-oretical studies describe coreference as a complex linguistic phenomenon, af-fected by various di eren t factors. State-of-the-art statistical approaches, onthe contrary, rely on rather simple knowledge-poor modeling. This thesis aimsat bridging the gap between the theory and the practice.We use insights from linguistic to identify relevant linguistic param-eters of co-referring descriptions. We consider di eren t types of information,from the most shallow name-matching measures to deeper syntactic, semantic,and discourse knowledge. We empirically assess the validity of the investigatedtheoretic predictions for the corpus data. Our data-driven evaluation exper-iments con rm that various linguistic parameters, suggested by theoreticalstudies, interact with coreference and may therefore provide valuable infor-mation for resolution systems. At the same time, our study raises severalissues concerning the coverage of theoretic claims. It thus brings feedback tolinguistic theory.