Fb’s Engaged on a Common Translation Engine that Can Perceive 128 Completely different Languages
Are you familiar with those futuristic translation devices they have in science fiction films that allow characters from completely different universes to converse in real time without the need for any prior knowledge of the local language?
That could soon become a reality as Facebook’s AI research team today unveils its latest work on an improved language translation system that can currently translate up to 128 different languages into English in a single application.
Called XLS-R is the process capable of performing speech recognition, language translation and speech recognition at a higher rate than any comparable system available.
As explained by Facebook AI:
“XLS-R is based on more than 436,000 hours of publicly available voice recordings and is based on wav2vec 2.0, our approach for the self-monitored learning of language representations. By using language data from various sources, from parliamentary sessions to audio books, we have expanded to 128 different languages, covering almost two and a half times more languages than its predecessor. “
In fact, Facebook says it tested XLS-R against four major multilingual speech recognition benchmarks where it outperformed on most of the languages tested.
“In particular, we tried it in five languages from BABEL, ten languages from CommonVoice, eight languages from MLS and the 14 languages from VoxPopuli.”
As you can see here, the error rates of these systems are still relatively high for some languages, but the XLS-R shows a significant improvement in accuracy, which Facebook is still improving in the development of the process.
This could eventually open up a range of new uses, including cross-border connections and commerce, expanding the possibilities for businesses around the world.
Facebook also points out that such systems could be of significant value for its future Metaverse applications as they enable a more universal connection in these digital, open world communities.
As mentioned before, Facebook is taking the system even further, and Facebook is also finding that there are more than 7,000 languages spoken in the world, so it’s far from a truly universal translator. But it’s another step toward that next level and making Facebook a more important utility in connection.
You can read more about Facebook’s XLS-R research here.