Neural Machine Translation Quality Comparison
Image Source: Google Research Blog
The year 2017 looks promising for the translation and localization industries. The recent advances in Neural Machine Translation (NMT) have been a popular topic over the past year.
Neural Machine Translation
Neural Machine Translation is a different approach to machine learning. Krzysztof Wołk and Krzysztof Marasek define neural machine translation as “an approach to machine translation that uses a large neural network. It departs from phrase-based statistical translation approaches that use separately engineered subcomponents.”
Neural networks are quite successful at grasping the context of sentences before their translation, thus improving the natural, human, sound of the outcome and the overall quality of the final translation.
The Big Giants
The big giants Google, Microsoft, and Facebook use neural machine technologies. Neural machine translation is used to produce higher quality translations for various language pairs, which sound more natural than those resulting from statistical machine translation.
Google has developed its own Google Neural Machine Translation system (GNMT) using an artificial neural network to further develop Google Translate. According to Research at Google, “Neural Machine Translation (NMT) is an end-to-end learning approach for automated translation, with the potential to overcome many of the weaknesses of conventional phrase-based translation systems.”
Last year Microsoft launched their Neural Network-based translations for all its speech languages. Skype Translator and the Microsoft Translator speech app for mobile devices use neural network technology to provide customers with better quality translations. Microsoft cognitive services use neural learning technology in multiple artificial intelligence scenarios such as speech and image processing.
Facebook is also using neural networks to improve the automatic translation of news feed posts and to master informal forms of language used in casual conversation. Alan Packer, director of engineering for Facebook’s language technology team said, “neural networks may also be better at learning how to translate idioms and metaphors into their equivalents in other languages.”
What do Google, Microsoft, and Facebook have in common?
They all have access to huge amounts of stored data with which they train language-processing software. These companies are looking to perfect machine translation within their systems in order to provide better translations to their customers.
Language service companies are also offering machine translation technology services such KantanMT. KantanMT recently announced the KantanNeural™ engines, part of their KantanFleet™ pre-built machine translation engines, which can be accessed directly by customers on the KantanMT platform. They will be able to create their own Custom Neural Machine Translation engines to translate entire documents in any of the language pairs supported by KantanMT.
Neural machine translation promises to be a great success in the years to come. We will keep you posted about the new developments in the latest trends in translation and localization. In the meantime, follow us on Facebook, Twitter, and LinkedIn for more!
- Neural-based Machine Translation for Medical Text Domain. Based on European Medicines Agency Leaflet Texts
- Google’s Neural Machine Translation System: Bridging the Gap between Human and Machine Translation
- Microsoft Translator launching Neural Network based translations for all its speech languages
- Facebook Plans to Boost Its Translations Using Neural Networks This Year
- Machine Translation Trends in 2027