Identificação e extração semiautomática de contextos definitórios em corpus com vistas à redação da definição terminológica: proposta de sistematização linguística para a língua portuguesa

Abstract

Computational tools in Natural Language Processing (NLP) are essential in handling electronic texts. Some of the resources commonly used are: frequency counters, word lists, keywords, and concordancers. It is noteworthy that this last tool, which terminologists use to view and extract defining contexts for certain terms, is useful in the stage of writing terminological definitions. Depending on the term and corpus size, the list of concordances may exceed a few hundred lines, making the task of defining the term extremely time-consuming. Yet, while concordancers facilitate the task of writing definitions, studies in terminology (ALARCÓN, 2009) and NLP (KLAVANS; MURESAN, 2001) have shown that it is possible to develop linguistic formalism that can be a substrate for the generation or enrichment of a system capable of detecting such contexts. While research in this direction has already been undertaken for English, Spanish, German, and French, among other languagesǡ research on Portuguese needs a more accurate linguistic explanation of defining context, in order to serve as a base for the development of similar systems for the Portuguese language. Therefore, the goals of this research are as follows: 1) to investigate the patterns of defining contexts found in technical corpora in Portuguese, 2) to provide linguistic knowledge which can be formalised computationally to create a system of semi-automatic extraction of candidate defining contexts, and 3) to evaluate the results generated. As the study’s corpus, we used scientific articles from the Bank of Portuguese (LAEL-PUC), and for the analysis, the following verbs were chosen: nomear ‘to name’, conceber ‘to conceive’, chamar ‘to call’, entender ‘to understand’, conhecer ‘to know’ and denominar ‘to denominate’. It was possible to do: 1) a quantitative and qualitative description of each verbal definitory pattern, 2) a local grammar for the chosen verbs with the purpose of aiding in the semiautomatic retrieval of definitory contexts, 3) an exclusion grammar to serve as a stoplist for local grammars and 4) a set of heuristics for a semiautomatic definitory context classifier. The evaluation of the lexical-syntactic rules of these six verbs showed 64% accuracy and 92% coverage in the global average, which represents an optimistic result in comparison to the results of previous studies. As a result, it was possible to 1) validate the methodology used, making it possible to extend it to other lexical-syntactic patterns and 2) obtain linguistic knowledge in order to integrate a semiautomatic computational system for definitory context candidates extraction for the Portuguese language.