XML Rules for Enclitic Segmentation

摘要

Sentence word segmentation is a very complex and important task in almost all natural language processing applications. Several works conceal or obviate the difficulties evolved in this process. In some cases, they adopt an easy partial solution acceptable for certain languages and applications, and, in others, they rely on a later or previous phase for solving it. However, there are hardly any papers with explanations describing how this later or previous phases have to be done.In this paper we have described these problems, focusing on part-of-speech tagging tasks, and propose a solution for one of them: the segmentation of verbal forms which contain enclitic pronouns. We have presented a generic verb processing system, which segments and pretags verbs which have enclitic pronouns joined to them.As we have seen, the system does not limit its function to segmentation, since it pretags the different linguistic components of a verbal form with enclitics, and removes invalid tags for its context. This innovative issue will be useful forpart-of-speech taggers, which can use this information to avoid making certain errors, thus improving its results.Although we have applied it to the Galician language, it can be easily adapted to other romance languages. The generic rule system we have designed allows rules to be written on the basis of XML files. This, combined with the use of lexicons, makes this adaptation simple and independent of the system internals.