Extração de relações hiponímicas em corpora de língua portuguesa

Abstract

Natural Language Processing (NLP) is a Computer Science area featured by its relevance to the development of applications that process large amounts of text or speech. In this paper we focus on texts in Portuguese, extracting from them hyponymic relations between entities, using a rules-based approach adapted from Hearst to English, and Freitas and Quental and Taba and Caseli to Portuguese. The prototype was executed over a corpus of Portuguese texts and the output was analyzed according to the reference author and rule sets. The evaluation process followed the one proposed by Freitas and Quental with human judgment, and the results are compared to those reported in the main references. The dissertation also studies in detail the most common errors identified.