Analisadores sintáticos de linguagens naturais: uma comparação do estado da arte

Abstract

Parsers, whose area comprises the Natural Language Processing field, has become more studied and explored lately. Besides its applications being in a evolution process, techniques such as deep learning, for instance, has been incorporated in parsers in order to improve learning. This means that every moment parsers come up with new solutions and it makes hard to conduct a survey about what is trending in parsing field at the moment or which good solutions were proposed so far. Moreover, surveys can become lagged or small due to the fast evolution of this area. This work shows our experiments with dependency and constituency parsers in order to run an survey about it. Related to dependency parsers, we tested them into 12 languages and regarding to the constituency ones we ran with 3. It was possible to run tests and analyse dependencies projectivities and non projectivities from parsers and corpus. Unfortunately we did not achieve the same analysis and results regarding constituency parsers due to the lack of corpus, however we finish the work pointing out some discussions about this problem.