Meta have launched their new artificial intelligence model that seeks to translate more than 200 languages. The initiative, called No Language Left Behind (NLLB), has been able to decipher 202 languages so far, including 55 African languages. This latter point is important to Meta, who noted that fewer than 25 African languages were covered by widely accepted translation tools previously.
While the increased number of languages available is impressive, the most important aspect to the success of this is the accuracy of the translations. Bilingual Evaluation Understudy (BLEU) is an algorithm-based evaluation platform designed to measure the quality of machine translated languages from one language to another. In a recent BLEU review, NLLB saw an average of a 44 percent improvement in translation accuracy over other translation solutions. This increased to 70 percent for some African and Indian languages.
Meta can look to use NLLB to help improve translation functionality across their owned properties of Facebook, Instagram, and Wikipedia. If the functionality can continue to evolve, allowing for real time translations, this could be hugely important for Meta’s metaverse ambitions. The benefits are not exclusive to Meta, as they are making NLLB open source, allowing everyone to access and use this platform. In fact, Meta is providing up to $200,000 USD grants for non profit organisations to use NLLB for real world applications.
Not everyone is keen to ingest languages into NLLB, however, with New Zealand’s Māori community fighting against the move. In 2018, Te Hiku Media, a non profit radio station in New Zealand’s Far North, began compiling an audio library of te reo Māori. The ambition was to capture as much te reo as possible to help retain the language as they saw its usage slipping away. Within 10 days they had over 300 hours of annotated audio of te reo Māori, submitted by Māori speakers across New Zealand. With this data, Te Hiku were able to begin building their own speech recognition and speech-to-text capabilities.
The threat of losing the language altogether is a very real concern, with the UN estimating that one indigenous language dies off every two weeks. This was the driving force behind Te Hiku’s motivation, but also saw them capture the attention of big tech companies around the world, such as Meta, who were looking to add te reo Māori into their own language processing AI. Peter-Lucus Jones, Te Hiku CEO, and Keoni Mahelona, Te Hiku Chief Technology Officer, vehemently opposed giving their data away to any corporation, maintaining that the only people who could profit from the Māori language, were the Māori people.
This opens up a highly contentious argument around ownership of language, and the role in which technology can be used to retain or shape the usage of language into the future. For many, cultural identity relies on indiginous languages. By having this language readily available globally, does it reduce the power, or mana, that it has within the community? Is it better to maintain culture, despite the threat of losing the language completely, than to potentially weaken the connection with the origins and history of the language and release it to the world?