Evaluating polarity for verbal phraseological units

Abstract

Fixation in linguistic expressions is an inherent property of natural language that plays a central role in their description. Verbal phraseological units are phrases made up of two or more words characterized for presenting certain degree of fixation or idiomaticity (at least one of these words is a verb that plays the role of the predicate). Phraseological units do not appear so frequently in manually constructed lexical resources as they do in real-word text, and this problem of coverage may impact the performance of many natural language processing tasks. Therefore, the construction of automatic understanding systems for these types of linguistic structures is very important, since they are a standard way of expressing a concept or idea. In this paper we present a set of experiments towards the automatic identification of the polarity of verbal phraseological units. We obtained a maximum performance of 80% for this particular task when the contextual information of a phraseological unit is considered, in comparison with a 62% when the VPU alone is only used. These results highlight the importance of analyzing automatically this type of linguistic structures. It should be stressed at the outset that these experiments are intended as a preliminary study rather than as a comprehensive analysis or solution of the aforementioned problem.