In this paper, we propose a novel finetuning algorithm for the recently introduced multi-way, mulitlingual neural machine translate that enables zero-resource machine translation. When used together with novel many-to-one translation strategies, we empirically show that this finetuning algorithm allows the multi-way, multilingual model to translate a zero-resource language pair (1) as well as a single-pair neural translation model trained with up to 1M direct parallel sentences of the same language pair and (2) better than pivot-based translation strategy, while keeping only one additional copy of attention-related parameters.
@article{arxiv.1606.04164,
title = {Zero-Resource Translation with Multi-Lingual Neural Machine Translation},
author = {Orhan Firat and Baskaran Sankaran and Yaser Al-Onaizan and Fatos T. Yarman Vural and Kyunghyun Cho},
journal= {arXiv preprint arXiv:1606.04164},
year = {2016}
}