Facebook AI can translate directly between any of 100 languages

 Facebook has developed a man-made intelligence capable of accurately translating between any pair of 100 languages without wishing on first translating to English, as many existing systems do.

The AI outperforms such systems by 10 points on a 100-point scale utilized by academics to automatically evaluate the standard of machine translations. Translations produced by the model were also assessed by humans, who scored it as around 90 per cent accurate.

Facebook’s system was trained on a knowledge set of seven.5 billion sentence pairs gathered from the online across 100 languages, though not all the languages had an equal number of sentence pairs. “What I actually was curious about was operation English as a middle man. Globally there are many regions where they speak two languages that aren’t English,” says Angela Fan of Facebook AI, who led the work.

The model was trained by specializing in languages that are commonly translated to and from one another, grouping languages into 14 separate collections supported geography and cultural similarities. This was done to make sure top quality translation of more commonly used connections and to coach the model more accurately.

For some language pairs, the new system shows significant improvements over existing translation quality. for instance, translating from Spanish to Portuguese is especially strong because Spanish is that the second-most spoken tongue worldwide, meaning the researchers had access to an oversized amount of coaching data. Translation between English and Belarusian also improved over existing efforts because the AI learns from translating Russian, which shares similarities with Belarusian.

While the system isn’t yet in use on the social network site, Facebook plans to place it to figure soon to handle the 20 billion translations made each day when people click “Translate” on posts written in additional than 160 languages. Future work is done on other languages, says Fan, “especially for languages where we don’t have lots of information, like South-East Asian and African languages”.

The work “breaks faraway from the English-centric models and tries to create more diverse multilingual ones”, says Sheila Castilho of the ADAPT Centre at Dublin University, Ireland. “That’s refreshing.” But, says Castilho, the human assessments only checked out a tiny low fraction of examples, making it hard to understand if this can be an accurate judgement of how the AI performs.

She also worries that the evaluation was done by bilingual volunteers, instead of professional translators. “Non-professionals lack knowledge of translation so may not notice subtle differences that make one translation better than another,” she says.

Her colleague at the ADAPT Centre, Andy Way, suggests Facebook isn’t making a good comparison with state-of-the-art translation systems. “Their claim to possess such an oversized improvement over ‘English-centric’ models could be a bit empty, as most of the time, people don’t do that anymore,” he says. Facebook disagrees, saying translation through English continues to be commonplace.