Related work on anaphora resolution in dialogues

For anaphora resolution in dialogues, a proliferation of methods based on dialogue structure (discourse-oriented approaches) have been developed. Among these, we should like to especially acknowledge the work of Grosz 1977, 1981, in which the influence of dialogue structure in anaphora resolution is justified. Grosz's work focuses specifically on task-oriented dialogues. Other studies, such as those published by Grosz et al. 1983, 1995, present a centering framework as a model to explain the coherence of local discourse segments in which the speaker's focus of attention is related to referring expressions. This model has achieved successful results in anaphora resolution in monologues, but would require certain modifications to be successfully applied to dialogues. Along those lines, Byron and Stent 1998 have developed extensions of the centering method for application to dialogues. They conclude that centering is as consistent in dialogues as it is in monologues.

Nevertheless, according to Strube and Hahn 1999, the crucial point of the centering model is the candidate list. Grosz et al. 1995 state that this list may be ordered using different factors, but they only use information about grammatical roles. However, it is difficult to define grammatical roles in free-word-order languages like German or Spanish without using semantic information.³

On the other hand, work carried out by Eckert and Strube 1999 on English details a method for resolving pronominal anaphora in dialogues with a precision of 66.2% and a recall rate of 68.2%. This method is based on the distribution of dialogue acts as an alternative to the centering method.

Furthermore, Martínez-Barco et al. 1999 emphasize the importance of discourse-topic knowledge as a complementary method for anaphora resolution in dialogues in which such knowledge is necessary for long-distance anaphora resolution.