A THEMATIC CONNECTIONIST APPROACH TO PORTUGUESE LANGUAGE PROCESSING

In the symbolic approach to Natural Language Processing (NLP), a system can only parse grammatically well constructed sentences. Within such a context, several linguistic phenomena, e.g. the thematic pattern relationships between the sentence constituents, can be accounted for (these pattern relationships are explained by a rule-based linguistic theory called thematic theory [1]). An alternative approach to NLP is a parallel distributed processing model, which has the benefits of learning and generalization. McClelland and Kawamoto [2] and Waltz and Pollack [3] proposed connectionist NLP models in which semantic microfeature representations of the words are used, in order to account for the relationship patterns between words in a sentence. In this paper, an adaptation of these connectionist systems is presented. Here, the microfeatures are based on thematic relationships between the words in the Portuguese sentences. However, will the output of such connectionist system account for the relationship patterns between the words in a sentence in a way which is comparable to that provided by the rulebased approach? The aim of this system is to investigate whether and how connectionist modeling can handle thematic relationships in sentences. An additional aim is to bring the Portuguese language to the scene, for comparative purposes.