NEBEL: Never-ending bilingual equivalent learner

Abstract

In this paper, we present NEBEL: an automatic system able to learn bilingual equivalents (translations) using the never-ending machine learning (NEML) strategy. Motivated by the way humans learn, the NEML is a continuous learning strategy which uses the knowledge already acquired to learn new information and, therefore, to improve its performance. The NEML was chosen to be applied in our context because it has two desirable features to deal with our intended problem: (i) it uses the Internet as knowledge source and (ii) it combines different extractions methods to improve the final result. In the experiments presented in this paper, NEBEL reached 65% accuracy in the English-Portuguese pair of languages.