Meta introduces new large AI model in search of universal translator
August 23, 2023
0
Meta wants to build an AI model that works like a universal translator in Star Trek. A new model, freely available to researchers, already shows promising capabilities. Meta
Meta wants to build an AI model that works like a universal translator in Star Trek. A new model, freely available to researchers, already shows promising capabilities.
Meta launches a new AI model that speaks and understands multiple languages: SeamlessM4T. M4T stands for Massively multilingual and multimodal machine translation. The model is capable of translating text to speech, speech to text, text to text and speech to speech.
If the output is to be text, then the model already understands almost a hundred languages. When the intent is to instantly convert text or speech to spoken output, SeamlessM4T understands the same hundred languages as the input, but can respond in “only” one of 35 languages.
Everything to everything
Meta’s goal is to develop a model where spoken language is instantly converted into another language so that people who don’t speak each other’s language can talk to each other. When designing, consider the following universal translator from Star Trek. Meta itself prefers the comparison with the Babylonian fish out of The Hitchhiker’s Guide to the Galaxy.
Meta’s approach is innovative as the company is working on a model that can do the translation in one pass, both for text and speech and for input and output in different languages. The model automatically understands what language is being spoken and even knows how to deal with speakers who switch languages. So theoretically it could even work in Brussels.
sensitivities
When training the model, Meta takes any sensitivities into account. For example, the company ensures that translations do not accidentally become rude or offensive, and that incorrect gender information does not sneak into words (e.g. when a word is female in the source language but there is no equivalent in the target language).
SeamlessM4T is licensed under a Creative Commons license, so it can also be used by other researchers.
As an experienced journalist and author, Mary has been reporting on the latest news and trends for over 5 years. With a passion for uncovering the stories behind the headlines, Mary has earned a reputation as a trusted voice in the world of journalism. Her writing style is insightful, engaging and thought-provoking, as she takes a deep dive into the most pressing issues of our time.