Meta founder Mark Zuckerberg said in a blog post that “many” of the languages involved in this new extension “are not available for automatic translation.”
This announcement is an example of a career for automatic translation between major Internet groups that aim to bring their services and products to the entire world’s population.
In May, Google researchers published a research paper entitled “Building Automatic Translation Systems for the Next Thousand Languages,” highlighting great ambitions in this area.
AI systems developed by Google, Microsoft and Meta now have the ability to translate languages where there is little parallel data, meaning they rarely translate into other languages.
This allows texts to be translated into languages with a limited distribution on the planet, such as Quechua (which is mostly limited to Eblero) and the language of the Fulani people (in West Africa), although no one has yet participated. in this work. .
François Yvonne, a researcher at the National Center for Scientific Research in France, who specializes in linguistics, said that automatic translation is “especially important for Facebook, which needs to monitor hate messages” distributed around the world and in all languages.
He explained that automatic translation could, in particular, allow English moderators to interfere with content published in other languages. But the reliability of these tools remains to be determined. In this context, Meta points out that its new system can deliver “44% better performance compared to the previous model, which can translate 100 languages.
In some African and Indian languages, this improvement is more than 70% compared to recently used translation systems, according to the world’s leading social network.
But François Yvonne believes that the quality of automatic translations provided by Google or Facebook engines will undoubtedly remain uneven depending on the language. “Languages with multiple translations, such as European languages, are likely to keep the edge,” he said.
Vincent Godard, CEO of leading French company Systran, which operates in 56 languages, expresses the same opinion.
Godard said the technology used by this group was originally the same as that of Meta and Google, but it was refined through the work of real linguists to avoid mistakes.
“When you translate a fighter jet manufacturing manual, you can’t make a mistake,” he explains, while mistakes can be made when it comes to translating an opinion about a restaurant.
So, are we any closer to automatic speech translation to communicate directly with anyone on this planet, for example, in the future Metaverse universe?
“We haven’t arrived yet, but we’re doing it,” said Antoine Bord, CEO of the Meta-Research Laboratory for Artificial Intelligence. “We have another project for machine translation of speech, which now works in a smaller number of languages,” he said.
But the benefit is “by connecting the two projects so that one day we can speak two hundred languages while maintaining the tone of speech, emotions and dialects…”, according to Burd.
Source: Al Akhbar
