Meta advances towards a universal translator for the metaverse: this will be its AI capable of translating 200 languages ​​- Marketing 4 Ecommerce – Your online marketing magazine for e-commerce

Meta has made a new ad looking for break the language barriers present all over the worldthrough Artificial Intelligence (AI), for which it has presented an ambitious project: the a universal translator, capable of translating 200 different languages, including minority languages ​​such as Kambra, Lao or Igbo, spoken in different parts of Africa. NLLB-200 is part of the tech giant’s initiative called “No Language Left Behind”which can be translated as «no language is left behind».

The idea is not only to be able to translate the most spoken languages ​​in the world, such as English, Spanish or Portuguese, to minority languagesbut to achieve quality reverse translations (there are expressions specific to each region or words that do not make sense in another language) and to break down one of the most difficult barriers for translation tools.

“A handful of languages, including English, Mandarin, Spanish, and Arabic, dominate the web. Native speakers of these widely spoken languages ​​may take for granted how meaningful it is to read something in their own mother tongue. NLLB will help more people read things in their native languageinstead of always requiring an intermediate language that is often wrong in sentiment or content” explains a of the company on its official blog.

NLLB-200 Background

Getting to this AI model has not been easy for Meta, which has faced several challenges. First, the translation services are trained with data, a training that is based on paired sentences between combinations with other languages. But there are many combinations that do not have parallel sentences that can serve as a translationso the translations may have grammatical errors or be inconsistent.

See also  What is GMV: a metric increasingly used by large marketplaces - Marketing 4 Ecommerce - Your online marketing magazine for e-commerce

Previously, Meta researchers developed M2M-100, a 100 language translation model, with which the creation of new methods to collect data and improve the data was promoted. But this model was not capable of covering even half of the current languages, which made it impossible for thousands of users to understand the content that was published on Facebook and Instagram.

The new engine has taken a different approach: the AI ​​has a machine learning system powered by a large set of sentences. Since there are different possible combinations, the AI ​​compares the contents and decides the most appropriate in the context.

In order to measure the quality of translations, Meta IA presented its technology FLOWERS-200, an evaluation data set that makes it possible for researchers to evaluate the performance of the NLLB-200 in more than 40,000 addresses between different languages.

Meta states that, “When comparing translation quality to previous AI research, NLLB-200 averaged 44% higher. For some African and Indian languages, NLLB-200 translations were 70% more accurate”

A new tool to democratize the use of the metaverse

But Meta doesn’t want to keep his work a secret, so he has published the model NLLB-200 and FLORES-200 in open source, so that any researcher can improve their own translation tools. Likewise, he announced the granting of grants up to $200,000 (196,000 euros), awarded to independent researchers or non-profit organizations that want to apply technology in social initiatives.

With this, he hopes that the advances in research can support over 25 billion translations published every day in the Facebook feed, Instagram and other company platforms.

See also  The 3 best tools to professionalize your sales via Whatsapp - Marketing 4 Ecommerce - Your online marketing magazine for e-commerce

Although there are still challenges to face, such as the lack of grammatical rules of some languages ​​created on the basis of speech, but this translation model It promises to reduce the access gap to the virtual world and that it has more uses than simple interactions or games.

As the company itself explains “As the world begins to take shape, the ability to create technologies that work well in a broader range of languages ​​will help democratize access to immersive experiences in virtual worlds.”

A real-time translation tool like this could become key to facilitating fluid communication in virtual environments such as the metaversewhere users from different countries and cultures seek to interact in the most “real” way possible and where voice commands are part of the strategy to popularize its use.

Image

Stay informed of the most relevant news on our news channel

Loading Facebook Comments ...
Loading Disqus Comments ...