July 29, 2019
Improved Zero-shot Neural Machine Translation via Ignoring Spurious Correlations
Association for Computational Linguistics (ACL)
Zero-shot translation, translating between language pairs on which a Neural Machine Translation (NMT) system has never been trained, is an emergent property when training the system in multilingual settings. However, naïve training for zero-shot NMT easily fails, and is sensitive to hyper-parameter setting. The performance typically lags far behind the more conventional pivot-based approach which translates twice using a third language as a pivot.
By: Jiatao Gu, Yong Wang, Kyunghyun Cho, Victor O.K. Li
Facebook AI Research
Natural Language Processing & Speech