Conclusion

In this paper we have evaluated the different tasks carried out in our MT approach (for Spanish and English languages) that allowed the correct pronominal anaphora translation into the target language. We have shown the importance of the resolution of anaphoric expressions in any MT system for correct translations into the target language, and how the main MT systems do not conveniently resolve this phenomenon.

Our approach, called AGIR, works on unrestricted texts to which partial-parsing techniques have been applied. After parsing and solving NLP problems, an interlingua representation of the entire text is obtained. This fact is one of the main advantages of our system since several problems (hardly solved by the majority of MT systems) can be treated and solved. These problems are the translation of intersentential anaphora, the detection of co-reference chains, and the translation of Spanish zero pronouns into English.

In the evaluation, we obtained a precision of 80.4% and 84.8% in the translation of Spanish and English pronominal anaphora, respectively. Previously, Spanish zero pronouns had been resolved (with a precision of 81.4%) and anaphoric personal pronouns had been resolved in English (with precisions of 86.6% and 76.8% in the SemCor corpus with semantic information and in the MTI corpus without it, respectively) and in Spanish (with a precision of 82.2%).

In addition, we carried out an exhaustive comparison with some well-known anaphora-resolution algorithms. Finally, we also compared pronoun translation with one of the most representative MT systems at the moment: Systran. In all of these comparisons, AGIR was shown to perform better.

A very important conclusion was extracted during the evaluation phase: the adding of semantic information improves the precision of the anaphora-resolution process considerably, and therefore the corresponding precision of the anaphora-translation process. Currently, the addition of this kind of information in the different stages of the AGIR system is being studied in order to improve the overall performance of the system.

The resolution and translation of new types of references, such as definite descriptions or anaphora originated by demonstrative pronouns, will be studied in the future. Moreover, the addition of new languages to the interlingua approach will be taken into account.