Exploiting the Discourse Structure for Anaphora Generation

In this paper we propose a model for anaphora generation that fruitfully takes advantage of the information on the discourse structure available during the generation process. We focus our discussion on pronominal referring expressions. We consider in particular the problem of generating a null pronoun vs. an overt pronoun in Italian. First, we propose a general rule based simply on centers comparisons and we show that this rule agrees with the results of a corpus analysis presented in the literature. Then, we propose that exceptions to the rule can be dealt with using knowledge on the rhetorical structure of the discourse