Zero-Resource Translation with Multi-Lingual Neural Machine Translation
This enables translation for language pairs without direct parallel data, addressing a key bottleneck in multilingual NLP.
The paper tackles zero-resource machine translation by proposing a novel finetuning algorithm for multilingual neural machine translation, achieving performance comparable to models trained with up to 1M parallel sentences and better than pivot-based strategies.
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.