Meta is starting a new program in partnership with UNESCO to collect speech recordings and transcriptions the company said will help the development of future openly available AI.
The Language Technology Partner Program is looking for partners who can provide over 10 hours of recorded speech with transcripts, a lot of written text, and collections of translated words in various languages. Meta says that partners will collaborate with their AI teams to add these languages to AI speech recognition and translation models, which will be made available to everyone once completed.
So far, partners include the government of Nunavut, which is a mostly empty area in Northern Canada. Some people in Nunavut speak a group of languages called Inuktut.
Meta stated in a blog post shared with Parhlo World that they are mainly focusing on helping languages that are not well represented, in line with UNESCO’s mission. Our aim is to build smart systems that can understand and react to people’s needs, no matter their language or culture.
Complementary to the new program, Meta said that it’s releasing an open source machine translation benchmark to rate the performance of language translation models. The benchmark, made up of sentences written by language experts, includes seven languages. You can view it and add to it on the AI development platform Hugging Face.
Meta is presenting both projects as charitable efforts. The company will gain from improved speech recognition and translation systems.
Meta is increasing the number of languages its AI assistant, Meta AI, can understand and is testing new features like automatic translation for authors. In September, Meta said it would start trying a tool for Instagram Reels that translates voices, letting creators dub their speech and automatically sync their lips.
Meta has faced a lot of criticism for how it handles material in languages other than English on its platforms. A study says that Facebook did not mark about 70% of COVID misinformation in Italian and Spanish, while it flagged 29% of similar misinformation in English. Leaked company papers show that Arabic-language posts are often wrongly marked as hate speech.
Meta has announced that it is working to make its translation and control tools better.
What do you say about this story? Visit Parhlo World For more.
