An algorithm for pronominal anaphora resolution：（一种算法对代词的重复法解决）.pdfVIP

An Algorithm for Pronominal Anaphora Resolution. Shalom Lappin∗ Herbert J Leass† SOAS, University of London. Sietec, Systemtechnik This pap er p resents an algorithm f or identify ing the noun phrase antecedents of t hird person p ronouns and lexical anaphors (refl exives and recip rocals) . The algorithm applies to the sy ntactic rep resentations generated by McCord s Slot Grammar pa rser, and relies on salience measures derived f rom sy ntactic structure and a simp le dy namic model of attentional state. Like the pa rser, the algorithm is imp lemented in Prolog. The authors have tested it extensively on compu ter manual texts, and conducted a blind test on manual text containing 360 pronoun occurrences. The algorithm successf ully identifies the antecedent of t he p ronoun f or 86% of these p ronoun occurrences. The relative contributions of t he algorithm s compon ents to its overall success rate in this blind test are examined. Exp eriments were conducted with an enhancement of t he algorithm which contributes statistically modelled inf ormation concerning semantic and real world relations to the algorithm s decision procedure. Interestingly, this enhancement only marginally imp roves the algorithm s p erf ormance (by 2%) . The algorithm is compa red with other approaches to anaphora resolution which have been proposed in the literature. In particular, the search procedure of Hobbs algorithm was imp lemented in the Slot Grammarf ramework and applied to the sentences in the blind test set. The authors algorithm achieves a higher rate of success (4%) than Hobbs algorithm. The relation of t he algorithm to the centering approach is discussed, as well as to models of anaphora resolution which