|
Briefing
|
- Machine Translation System – Microsoft researchers in Beijing and Redmond, Washington developed first machine translation system that can translate Chinese news articles to English with human parity
- Dataset Used – Used publicly available newstest2017 dataset, released by group of industry and academic organizations, comprised of news stories to test machine translation systems
- Human Evaluators – Hired external bilingual human evaluators to judge and compare machine’s results to actual human translations
- AI Techniques Used – Include dual learning for fact-checking, deliberation networks for running translation repeatedly to refine outcomes, joint training which enhances Chinese-to-English translation and back to English-to-Chinese, and agreement regularization for translating words from left-to-right and then right-to-left
- Microsoft Translator – Makes machine translation system, called Machine Translator, available to public, but with limited functionality compared to system used in research
|
