Pythia NLG – um modelo para integração de recursos voltados à geração de linguagem natural

Abstract

The natural language generation is a field of study designed to analyze the structure of language to finally construct text in a language understandable to the human being. The natural language generation has different application niches, either to generate answers to questions, creative content such as narratives and poetry or to refine sentences to achieve a friendlier interaction with its users, among others. It is observed the growth of the area, with research focused on these different niches of use. There is, however, a lack of tools to use and exploit such applications together, which could provide benefits of reuse and performance comparison. The analysis of recent works in the area allows observing techniques used in only one application or purpose, but, for the best of our knowledge, there is not a tool that can attend different applications or promote the comparison of techniques. This study aims to present proposal of a model for an environment that allows the integration of tools already known in the field of natural language generation and facilitates the exploration and comparison of new techniques and algorithms. The methodology used is exploratory, and we sought to identify solutions to a problem initially presented. The development of the research involved the accomplishment of a bibliographic study in the area of natural language generation, in order to identify information related to the area of interest. Finally, an analysis was made of the possibilities for the construction of a model that contemplates the needs evidenced by the problem identified, related to the integration and reuse of resources for the natural language generation (NLG). As a contribution, we propose a model nominated Pythia NLG, which promotes an open approach to allow the continuous integration of resources in this area. As a way of moving towards its evaluation, use scenarios are described that allow verifying its potential. The implemented Pythia NLG prototype presented a favorable evaluation, through use scenarios, for an environment that allows the integration and reuse of algorithms and techniques in complex NLG tasks. It is concluded that the development of the Pythia NLG prototype proposes a tool with potential for exploration of new experiments in the area of natural language generation, facilitating the reuse and sharing of techniques, as well as the possibility of integration of algorithms for evaluation and comparison of results. The implemented Pythia NLG prototype allows NLG scholars and researchers to benefit from new research with an environment for NLG task trials.