SDL just announced a new machine translation (MT) platform, dubbed XMT. Machine translation has been around for quite some time but its uses have been somewhat limited as the qualities of the translations aren’t as good as those done by human beings. This is especially true for difficult language pairs such as English to Japanese and Chinese. XMT is a complete re-write of MT by SDL Research Labs, and one of its primary goals is to improve the quality of hard to translate language pairs.
Current machine translation technologies apply a one-size-fits-all translation algorithm universally to all language pairs, but XMT allows different algorithms to be used for different languages. This means, for example if you have to change the word order prior to translation between Japanese and English you can do it just for that language pair without affecting any others. This also means that you can create algorithms for social media formats such as Twitter and informal languages that are now popping up in different social apps.
This new platform is also a base for new MT capabilities. In addition to enabling higher quality translations it also brings artificial intelligence to machine translation by allowing the engines to learn and adapt. SDL Language Learning is the ability for the MT engine to learn and remember individual user’s language translation preferences. For example if the MT user is a translator for a drug company and uses the word “dressing", and the MT engine has not been trained in Life Science vocabulary, then it may come back with the translation of “salad dressing". When the user corrects the translation to “medical dressing" the engine will remember their preference and future translations of the word will be medical. SDL announced and demonstrated SDL Language Learning capability but it will be late this year or early next before the capability is productized and available for use. SDL still offers industry or company specific pre-trained engines, but SDL Language Learning is passive and improves translation quality the more an individual uses the technology.
Please don’t misunderstand me; MT is still not as good as a human translator and won’t be for a long time. But we are a big step closer to having a tool to quickly and cheaply translate all types of content that was either too expensive or time consuming to be done by humans. While this announcement may not be Star Trek’s Universal Translator, it does take us a large step closer.
Find out more in our press release.