Microsoft announces milestone in machine translation

Microsoft researchers announced that they have created what they believe to be ‘the first machine translation system that can translate sentences of news articles from Chinese to English with the same quality and accuracy as a person’. To develop this translation system, the researchers used several methods of training artificial intelligence (AI) systems: dual learning, deliberation networks, joint training, and agreement regularisation. Microsoft’s translation system, explained in details in a research paper, achieved the so-called human parity on a commonly used dataset of news stories – newstest2017. The translation results were compared with two independently produced human reference translations, and the comparison showed that the machine translation results were accurate and on par with the translations carried out by humans. While the researchers are enthusiastic about achieving the human parity milestone on the dataset, they cautioned, that the achievement does not mean that machine translation is a solved problem. Challenges remain, and the translation system needs to be tested on real-time news stories.