Amazon gives FBK the AWS Machine Learning Research Award

$ 100,000 worth of computing resources to be used on the AWS cloud platform to refine the system and $ 50,000 to further project research activities. This is the prize that the MT – Machine Translation Unit has received by winning the AWS Machine Learning Research Award, a competition held by Amazon that provides funding for the most promising technologies in the industry.

The solution studied by researchers at Fondazione Bruno Kessler represents a remarkable evolution compared to the current technology and allows to acquire an audio signal in a language and provide its automatic written translation into another language. This means, for instance, that users could follow the English audio live streaming of an event while at the same time read the subtitles being automatically generated in Italian.

“The vast majority of video content produced every day is in English,” Marco Turchi, head of the MT Unit and author of the project with FBK researcher Matteo Negri, explained “and enabling its use in different languages is a priority in order to spread knowledge across different cultures. From a technical standpoint, current systems use two components that involve two steps: audio-to-text transcription in the source language and, after that, text-to-text translation from the source language into the target language. This combination, in addition to being complex, is prone to error propagation issues that can significantly compromise the quality of the text being translated. The alternative we suggest is instead based on artificial intelligence techniques, thanks to which the system is able to extract information from the audio content and use it to directly generate the text in another language, thus avoiding the accumulation of possible errors”