Posted · Add Comment

Christopher Manning has described neural machine translation (NMT) as “the approach of modeling the entire [machine translation] process via one big artificial neural network” (Neural Machine Translation and Models with Attention 3 April 2017, emphasis in original). Google has replaced its earlier statistical methods with a NMT system of its own, although not in all its language pairs (Lewis-Kraus 2016). Microsoft Translator has also migrated to an NMT system. On its website, the company promises “major advances in translation quality” compared to statistical methods: “Neural networks better capture the context of full sentences [-] providing [-] more human-sounding output” (Microsoft Translator launching Neural Network based translations for all its speech languages 15 November 2016).