Kyoto University Participation to WAT 2016

This paper describes the experiments we did for the WAT 2016 (Nakazawa et al., 2016a) shared translation task. We used two different machine translation (MT) approaches. On one hand, we used an incremental improvement to an example-based MT (EBMT) system: the KyotoEBMT system we used for WAT 2015. On the other hand, we implemented a neural MT (NMT) system that makes use of the recent results obtained by researchers in the field. We found that the NMT approach works very well on the ASPEC (Nakazawa et al., 2016b) data, especially for the Chinese-to-Japanese direction. Overall, we could obtain the best results reported for several language directions. This paper is decomposed as such: in Section 2, we describe the incremental improvements to our EBMT system compared with the WAT 2015 workshop. We then describe our NMT implementation and the settings we used for our experiments in Section 3. Finally, we discuss the results obtained in the shared task.