Improvements on Automatic Word Codification for Connectionist Machine Translation

Encouragingly accurate translations have recently been obtained using a connectionist translator called RECONTRA (Recurrent Connectionist Translator). In order to deal with tasks of medium or large vocabularies, distributed representations of the lexicons are required in this translator. A simple connectionist model has been recently designed to automatically obtain word distributed representations. In this paper several learning algorithms were used to train this connectionist encoder aiming to improve the translation rates achieved with the corresponding obtained codifications of the vocabularies involved.