Meta translator will use artificial intelligence (AI) models with the aim of translating 200 different languages in real-time and will be accessible worldwide under Meta platform services such as Facebook, Instagram, or even WhatsApp.
Do you want to know more about international commerce and e-commerce, and also get all the tips and news that might give your company an edge? Subscribe to Cross-Border Magazine and get your digital copy now for free!
The Internet giant has succeeded in presenting this new functionality called NLLB-200, a pioneering system that is capable of translating 200 languages in real-time, twice as many as previous systems. This, with the goal of facilitating connections between people from different cultures.
NLLB-200 is based on the M2M-100 that was presented in 2020, but incorporates a substantial improvement: in this new alternative, translations are made directly from the source language to the target language, without the need to go through English, as was necessary in the previous case.
M2M-100 is the most common system on the Internet today and is the one on which language processing systems are based. This is what makes translators go from the different languages into English and then translate them into the language that is being requested. According to Meta's researchers, this leads to large losses in meaning and also in the nuances of expressions. At the same time, they have found that certain languages are underrepresented on the Internet today.
In order to provide an answer to this situation, the researchers have perfected a model that is able to extract good performance from each parsed sentence in a language. In addition, they have increased the size of the databases on which the algorithm is based.
The tool developed by Meta is capable of translating both written and spoken language. Today, this translator has 200 operational languages and, according to the company, is aimed at the 4 billion people in the world who speak languages that are underrepresented on the Internet. To this end, 55 African languages have been included in the 200 languages, many of which have not been available in any translator until now.
Another highlight is that Meta has decided to publish the NLLB-200 model in open source and its training code. This is in order to help other researchers to improve their translation tools using the Meta Translator as a starting point.
Looking to the future, Meta aims to eventually make the company's augmented reality glasses capable of real-time translation, as well as serving up subtitles that are visible to wearers. It's a goal that Google has also been working towards, after unveiling a similar prototype of the glasses in May.
e-commerce, and also get all the tips and news that might give your company an edge? Subscribe to Cross-Border Magazine and get your digital copy now for free!
The Internet giant has succeeded in presenting this new functionality called NLLB-200, a pioneering system that is capable of translating 200 languages in real-time, twice as many as previous systems. This, with the goal of facilitating connections between people from different cultures.
The NLLB-200 Meta Translator is based on the previous M2M-100 that was presented in 2020, but incorporates a substantial improvement: in this new alternative, translations are made directly from the source language to the target language, without the need to go through English, as was necessary in the previous case.
M2M-100 is the most common system on the Internet today and is the one on which language processing systems are based. This is what makes translators go from the different languages into English and then translate them into the language that is being requested. According to Meta's researchers, this leads to large losses in meaning and also in the nuances of expressions. At the same time, they have found that certain languages are underrepresented on the Internet today.
In order to provide an answer to this situation, the researchers have perfected a model that is able to extract good performance from each parsed sentence in a language. In addition, they have increased the size of the databases on which the algorithm is based.
The tool developed by Meta is capable of translating both written and spoken language. Today, this translator has 200 operational languages and, according to the company, is aimed at the 4 billion people in the world who speak languages that are underrepresented on the Internet. To this end, 55 African languages have been included in the 200 languages, many of which have not been available in any translator until now.
Another highlight is that Meta has decided to publish the NLLB-200 model in open source and its training code. This is in order to help other researchers to improve their translation tools.
Looking to the future, Meta aims to eventually make the company's augmented reality glasses capable of real-time translation, as well as serving up subtitles that are visible to wearers. It's a goal that Google has also been working towards, after unveiling a similar prototype of the glasses in May.
