Tuning word codifications for the RECONTRA translator
Gustavo Adolfo Casañ Núñez, Asunción Castaño Álvarez · Dialnet (Universidad de la Rioja) · 2007
In order to approach a Machine Translation task with large vocabularies using the RECONTRA connectionist model, compact (distributed) codifications of such vocabularies are required. These codifications can be extracted from Multilayer Perceptrons with output delays trained as encoders of the vocabularies. In this paper we explore different mechanisms that can be adopted for tuning up such codifications for the translation task.