Generating from a discourse model
Rodolfo Delmonte | Dario Bianchi | Emanuele Pianta
Proceedings of the International Conference on Machine Translation and Multilingual Applications in the new Millennium: MT 2000
This paper describes an implemented algorithm for handling pronominal reference and anaphoric control within an LFG framework. At first there is a brief description of the grammar implemented in Prolog using XGs (extraposition grammars) introduced by Pereira (1981;1983). Then the algorithm mapping binding equations is discussed at length. In particular the algorithm makes use of f-command together with the obviation principle, rather than c-command which is shown to be insufficient to explain the facts of binding of both English and Italian. Previous work (Ingria,1989;Hobbs,1978) was based on English and the classes of pronominals to account for were two: personal and possessive pronouns and anaphors - reflexives and reciprocals. In Italian, and in other languages of the world, the classes are many more. We dealt with four: a.pronouns - personal and independent pronouns, epithets, possessive pronouns; b.clitic pronouns and Morphologically Unexpressed PRO/pros; c.long distance anaphors; short distance anaphors. Binding of anaphors and coreference of pronouns is extensively shown to depend on structural properties of f-structures, on thematic roles and grammatical functions associated with the antecedents or controller, on definiteness of NPs and mood of clausal f-structures. The algorithm uses feature matrixes to tell pronominal classes apart and scores to determine the ranking of candidates for antecedenthood, as well as for restricting the behaviour of proforms and anaphors.