Building a sequence-to-sequence model for text translation