A New Approach to Codifications for the Recontra Neural Translator

G.A. Casañ and M.A. Castaño (Spain)


Machine Translation, Neural Networks, Automatic Word Codification


Encouragingly accurate translations have recently been obtained using a connectionist translator called RECONTRA (Recurrent Connectionist Translator). This paper approaches a text-to-text machine translation task more complex than those previously tackled with this translator. The distributed codifications of the lexicons involved in the task were automatically extracted of the hidden layer of a multilayer perceptron with output delays. The use of a simple pruning method determined the size of the extracted representations, and adopting distributed codifications for the input and output of the multilayer perceptrons, the size and training time of the networks were reduced.

Important Links:

Go Back