next up previous
Next: Acknowledgments Up: Computational Approach to Anaphora Previous: Final evaluation (test corpus)


Conclusion

In this paper we have presented an algorithm for identifying the noun phrase antecedents of pronouns and adjectival anaphors in Spanish dialogues. This algorithm exploits different kinds of information: linguistic knowledge, discourse/dialogue structure information, and discourse topic knowledge. It is based on a set of constraints and preferences which depend on all available knowledge in order to resolve anaphora.

In addition, a definition of the anaphoric accessibility space based on discourse/dialogue structure information was presented. We have shown the importance of this accessibility space in anaphora resolution, in contrast to algorithms that do not rely on any such space. Results show that 95.9% of the antecedents were located in the proposed space.

Finally, we described a set of experiments concerning this algorithm and accessibility space using a corpus of 204 dialogues. The algorithm was implemented using Prolog. In our final experiment, a precision of 81.3% was achieved for pronominal anaphora resolution and a precision of 81.5% was achieved for adjectival anaphora resolution.

As a tool for resolving pronominal and adjectival anaphora for Spanish dialogues, this system can be used in support of various NLP tasks, including machine translation, information extraction, retrieval information, or question-answering.

Currently, the authors are working on incorporating semantic information into the algorithm.


next up previous
Next: Acknowledgments Up: Computational Approach to Anaphora Previous: Final evaluation (test corpus)
patricio 2001-10-17