Hi, everyone, 

Junyoung Chung will talk today about recent work on Character-Level Neural Machine Translation:

A Character-Level Decoder without Explicit Segmentation for Neural Machine Translation

Who: Junyoung Chung
When: Fri. May 6th, 14:30 
Where: AA3195


Abstract: The existing machine translation systems, whether phrase-based or neural, have relied almost exclusively on word-level modelling with explicit segmentation. In this paper, we ask a fundamental question: can neural machine translation generate a character sequence without any explicit segmentation? To answer this question, we evaluate an attention-based encoder– decoder with a subword-level encoder and a character-level decoder on four language pairs–En-Cs, En-De, En-Ru and En-Fi– using the parallel corpora from WMT’15. Our experiments show that the models with a character-level decoder outperform the ones with a subword-level decoder on all of the four language pairs. Furthermore, the ensembles of neural models with a character-level decoder outperform the state-of-the-art non-neural machine translation systems on En-Cs, En-De and En-Fi and perform comparably on En-Ru.
---