Artificial intelligence systems can now translate an incredible number of different languages, bringing the world closer to its long-lasting dream of eliminating the language barrier among the planet’s inhabitants.
“Meta”, the parent company of Facebook, has announced that its artificial intelligence systems are now able to fully translate two hundred languages, whatever the combination, doubling the number of interchangeably translatable languages from the previous 100.
In a blog post, Meta founder and Facebook CEO Mark Zuckerberg said that “many” of the languages involved in this new advancement “were previously unavailable for machine translation.”
This announcement represents an example of the race for machine translation between the major Internet groups seeking to offer their services and products to the entire population of the world.
Last May, researchers at Google published a scientific article entitled “Building machine translation systems for the next thousand languages”, which reflects the great ambitions in this field.
The artificial intelligence systems developed by Google, Microsoft, and Meta can now translate languages with scarce parallel data, meaning they are rarely translated into other languages.
This technology allows for the translation of texts in languages that have limited reach on the planet, such as Quechua (largely limited to Peru) and the language of the Fulani people (in West Africa), even though no human has ever been involved in such a task.
Francois Yvonne, a researcher at the National Center for Scientific Research in France specializing in linguistics, says that machine translation is “particularly important for Facebook, which needs to monitor hate messages” that are spreading all over the world in every language.
Yvonne explained that Machine translation could allow English-speaking moderators to evaluate content published in other languages, however, the reliability of these tools remains to be determined.
In this context, Meta noted that its new system can offer a 44% better performance compared to the previous model that could only translate 100 languages.”
According to META, this improvement exceeds 70% compared to the recently used translation systems in some African and Indian languages.
However, François Yvonne explained that the quality of machine translations provided by Google or Facebook engines will undoubtedly remain unequal depending on the individual language, saying: “Languages that are highly translated, such as European languages, will probably retain the preference.”
The same opinion is expressed by Vincent Godard, CEO of the French company “SYSTRAN”, which specializes in machine translation, and operates in 56 languages.
Godard pointed out that the technology used by this group was initially the same as Meta and Google, but has been enhanced by the work of real linguists to avoid errors, explaining “When you’re translating a fighter jet assembly manual, there is no room for one mistake,” however mistakes could be tolerated when it comes to translating a restaurant review.
Answering the age-old question of whether we are any closer to having machine speech translation that enables us to speak directly with anyone on the planet, for example in the future world of Metaverse, Antoine Bord, director general of the Meta Research Laboratory of Artificial Intelligence (VIR) said, “We are not there yet, but we are working on it”.
He added, “We have another project regarding machine translation of speech, currently working with a much smaller number of languages, the trick is merging the two projects so that one day we can speak two hundred languages while maintaining the intended tone, emotion, and dialect”.